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/compat.h>
  10 #include <linux/delay.h>
  11 #include <linux/errno.h>
  12 #include <linux/hdreg.h>
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/list_sort.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
 120 /*
 121  * Prepare a queue for teardown.
 122  *
 123  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
 124  * the capacity to 0 after that to avoid blocking dispatchers that may be
 125  * holding bd_butex.  This will end buffered writers dirtying pages that can't
 126  * be synced.
 127  */
 128 static void nvme_set_queue_dying(struct nvme_ns *ns)
 129 {
 130         if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 131                 return;
 132
 133         blk_set_queue_dying(ns->queue);
 134         blk_mq_unquiesce_queue(ns->queue);
 135
 136         set_capacity_and_notify(ns->disk, 0);
 137 }
 138
 139 void nvme_queue_scan(struct nvme_ctrl *ctrl)
 140 {
 141         /*
 142          * Only new queue scan work when admin and IO queues are both alive
 143          */
 144         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 145                 queue_work(nvme_wq, &ctrl->scan_work);
 146 }
 147
 148 /*
 149  * Use this function to proceed with scheduling reset_work for a controller
 150  * that had previously been set to the resetting state. This is intended for
 151  * code paths that can't be interrupted by other reset attempts. A hot removal
 152  * may prevent this from succeeding.
 153  */
 154 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 155 {
 156         if (ctrl->state != NVME_CTRL_RESETTING)
 157                 return -EBUSY;
 158         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 159                 return -EBUSY;
 160         return 0;
 161 }
 162 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 163
 164 static void nvme_failfast_work(struct work_struct *work)
 165 {
 166         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 167                         struct nvme_ctrl, failfast_work);
 168
 169         if (ctrl->state != NVME_CTRL_CONNECTING)
 170                 return;
 171
 172         set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 173         dev_info(ctrl->device, "failfast expired\n");
 174         nvme_kick_requeue_lists(ctrl);
 175 }
 176
 177 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
 178 {
 179         if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
 180                 return;
 181
 182         schedule_delayed_work(&ctrl->failfast_work,
 183                               ctrl->opts->fast_io_fail_tmo * HZ);
 184 }
 185
 186 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
 187 {
 188         if (!ctrl->opts)
 189                 return;
 190
 191         cancel_delayed_work_sync(&ctrl->failfast_work);
 192         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 193 }
 194
 195
 196 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 197 {
 198         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 199                 return -EBUSY;
 200         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 201                 return -EBUSY;
 202         return 0;
 203 }
 204 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 205
 206 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 207 {
 208         int ret;
 209
 210         ret = nvme_reset_ctrl(ctrl);
 211         if (!ret) {
 212                 flush_work(&ctrl->reset_work);
 213                 if (ctrl->state != NVME_CTRL_LIVE)
 214                         ret = -ENETRESET;
 215         }
 216
 217         return ret;
 218 }
 219
 220 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 221 {
 222         dev_info(ctrl->device,
 223                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 224
 225         flush_work(&ctrl->reset_work);
 226         nvme_stop_ctrl(ctrl);
 227         nvme_remove_namespaces(ctrl);
 228         ctrl->ops->delete_ctrl(ctrl);
 229         nvme_uninit_ctrl(ctrl);
 230 }
 231
 232 static void nvme_delete_ctrl_work(struct work_struct *work)
 233 {
 234         struct nvme_ctrl *ctrl =
 235                 container_of(work, struct nvme_ctrl, delete_work);
 236
 237         nvme_do_delete_ctrl(ctrl);
 238 }
 239
 240 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 241 {
 242         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 243                 return -EBUSY;
 244         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 245                 return -EBUSY;
 246         return 0;
 247 }
 248 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 249
 250 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 251 {
 252         /*
 253          * Keep a reference until nvme_do_delete_ctrl() complete,
 254          * since ->delete_ctrl can free the controller.
 255          */
 256         nvme_get_ctrl(ctrl);
 257         if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 258                 nvme_do_delete_ctrl(ctrl);
 259         nvme_put_ctrl(ctrl);
 260 }
 261
 262 static blk_status_t nvme_error_status(u16 status)
 263 {
 264         switch (status & 0x7ff) {
 265         case NVME_SC_SUCCESS:
 266                 return BLK_STS_OK;
 267         case NVME_SC_CAP_EXCEEDED:
 268                 return BLK_STS_NOSPC;
 269         case NVME_SC_LBA_RANGE:
 270         case NVME_SC_CMD_INTERRUPTED:
 271         case NVME_SC_NS_NOT_READY:
 272                 return BLK_STS_TARGET;
 273         case NVME_SC_BAD_ATTRIBUTES:
 274         case NVME_SC_ONCS_NOT_SUPPORTED:
 275         case NVME_SC_INVALID_OPCODE:
 276         case NVME_SC_INVALID_FIELD:
 277         case NVME_SC_INVALID_NS:
 278                 return BLK_STS_NOTSUPP;
 279         case NVME_SC_WRITE_FAULT:
 280         case NVME_SC_READ_ERROR:
 281         case NVME_SC_UNWRITTEN_BLOCK:
 282         case NVME_SC_ACCESS_DENIED:
 283         case NVME_SC_READ_ONLY:
 284         case NVME_SC_COMPARE_FAILED:
 285                 return BLK_STS_MEDIUM;
 286         case NVME_SC_GUARD_CHECK:
 287         case NVME_SC_APPTAG_CHECK:
 288         case NVME_SC_REFTAG_CHECK:
 289         case NVME_SC_INVALID_PI:
 290                 return BLK_STS_PROTECTION;
 291         case NVME_SC_RESERVATION_CONFLICT:
 292                 return BLK_STS_NEXUS;
 293         case NVME_SC_HOST_PATH_ERROR:
 294                 return BLK_STS_TRANSPORT;
 295         case NVME_SC_ZONE_TOO_MANY_ACTIVE:
 296                 return BLK_STS_ZONE_ACTIVE_RESOURCE;
 297         case NVME_SC_ZONE_TOO_MANY_OPEN:
 298                 return BLK_STS_ZONE_OPEN_RESOURCE;
 299         default:
 300                 return BLK_STS_IOERR;
 301         }
 302 }
 303
 304 static void nvme_retry_req(struct request *req)
 305 {
 306         unsigned long delay = 0;
 307         u16 crd;
 308
 309         /* The mask and shift result must be <= 3 */
 310         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 311         if (crd)
 312                 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
 313
 314         nvme_req(req)->retries++;
 315         blk_mq_requeue_request(req, false);
 316         blk_mq_delay_kick_requeue_list(req->q, delay);
 317 }
 318
 319 enum nvme_disposition {
 320         COMPLETE,
 321         RETRY,
 322         FAILOVER,
 323 };
 324
 325 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 326 {
 327         if (likely(nvme_req(req)->status == 0))
 328                 return COMPLETE;
 329
 330         if (blk_noretry_request(req) ||
 331             (nvme_req(req)->status & NVME_SC_DNR) ||
 332             nvme_req(req)->retries >= nvme_max_retries)
 333                 return COMPLETE;
 334
 335         if (req->cmd_flags & REQ_NVME_MPATH) {
 336                 if (nvme_is_path_error(nvme_req(req)->status) ||
 337                     blk_queue_dying(req->q))
 338                         return FAILOVER;
 339         } else {
 340                 if (blk_queue_dying(req->q))
 341                         return COMPLETE;
 342         }
 343
 344         return RETRY;
 345 }
 346
 347 static inline void nvme_end_req(struct request *req)
 348 {
 349         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 350
 351         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
 352             req_op(req) == REQ_OP_ZONE_APPEND)
 353                 req->__sector = nvme_lba_to_sect(req->q->queuedata,
 354                         le64_to_cpu(nvme_req(req)->result.u64));
 355
 356         nvme_trace_bio_complete(req);
 357         blk_mq_end_request(req, status);
 358 }
 359
 360 void nvme_complete_rq(struct request *req)
 361 {
 362         trace_nvme_complete_rq(req);
 363         nvme_cleanup_cmd(req);
 364
 365         if (nvme_req(req)->ctrl->kas)
 366                 nvme_req(req)->ctrl->comp_seen = true;
 367
 368         switch (nvme_decide_disposition(req)) {
 369         case COMPLETE:
 370                 nvme_end_req(req);
 371                 return;
 372         case RETRY:
 373                 nvme_retry_req(req);
 374                 return;
 375         case FAILOVER:
 376                 nvme_failover_req(req);
 377                 return;
 378         }
 379 }
 380 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 381
 382 /*
 383  * Called to unwind from ->queue_rq on a failed command submission so that the
 384  * multipathing code gets called to potentially failover to another path.
 385  * The caller needs to unwind all transport specific resource allocations and
 386  * must return propagate the return value.
 387  */
 388 blk_status_t nvme_host_path_error(struct request *req)
 389 {
 390         nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
 391         blk_mq_set_request_complete(req);
 392         nvme_complete_rq(req);
 393         return BLK_STS_OK;
 394 }
 395 EXPORT_SYMBOL_GPL(nvme_host_path_error);
 396
 397 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 398 {
 399         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 400                                 "Cancelling I/O %d", req->tag);
 401
 402         /* don't abort one completed request */
 403         if (blk_mq_request_completed(req))
 404                 return true;
 405
 406         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 407         nvme_req(req)->flags |= NVME_REQ_CANCELLED;
 408         blk_mq_complete_request(req);
 409         return true;
 410 }
 411 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 412
 413 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
 414 {
 415         if (ctrl->tagset) {
 416                 blk_mq_tagset_busy_iter(ctrl->tagset,
 417                                 nvme_cancel_request, ctrl);
 418                 blk_mq_tagset_wait_completed_request(ctrl->tagset);
 419         }
 420 }
 421 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
 422
 423 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
 424 {
 425         if (ctrl->admin_tagset) {
 426                 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
 427                                 nvme_cancel_request, ctrl);
 428                 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
 429         }
 430 }
 431 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
 432
 433 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 434                 enum nvme_ctrl_state new_state)
 435 {
 436         enum nvme_ctrl_state old_state;
 437         unsigned long flags;
 438         bool changed = false;
 439
 440         spin_lock_irqsave(&ctrl->lock, flags);
 441
 442         old_state = ctrl->state;
 443         switch (new_state) {
 444         case NVME_CTRL_LIVE:
 445                 switch (old_state) {
 446                 case NVME_CTRL_NEW:
 447                 case NVME_CTRL_RESETTING:
 448                 case NVME_CTRL_CONNECTING:
 449                         changed = true;
 450                         fallthrough;
 451                 default:
 452                         break;
 453                 }
 454                 break;
 455         case NVME_CTRL_RESETTING:
 456                 switch (old_state) {
 457                 case NVME_CTRL_NEW:
 458                 case NVME_CTRL_LIVE:
 459                         changed = true;
 460                         fallthrough;
 461                 default:
 462                         break;
 463                 }
 464                 break;
 465         case NVME_CTRL_CONNECTING:
 466                 switch (old_state) {
 467                 case NVME_CTRL_NEW:
 468                 case NVME_CTRL_RESETTING:
 469                         changed = true;
 470                         fallthrough;
 471                 default:
 472                         break;
 473                 }
 474                 break;
 475         case NVME_CTRL_DELETING:
 476                 switch (old_state) {
 477                 case NVME_CTRL_LIVE:
 478                 case NVME_CTRL_RESETTING:
 479                 case NVME_CTRL_CONNECTING:
 480                         changed = true;
 481                         fallthrough;
 482                 default:
 483                         break;
 484                 }
 485                 break;
 486         case NVME_CTRL_DELETING_NOIO:
 487                 switch (old_state) {
 488                 case NVME_CTRL_DELETING:
 489                 case NVME_CTRL_DEAD:
 490                         changed = true;
 491                         fallthrough;
 492                 default:
 493                         break;
 494                 }
 495                 break;
 496         case NVME_CTRL_DEAD:
 497                 switch (old_state) {
 498                 case NVME_CTRL_DELETING:
 499                         changed = true;
 500                         fallthrough;
 501                 default:
 502                         break;
 503                 }
 504                 break;
 505         default:
 506                 break;
 507         }
 508
 509         if (changed) {
 510                 ctrl->state = new_state;
 511                 wake_up_all(&ctrl->state_wq);
 512         }
 513
 514         spin_unlock_irqrestore(&ctrl->lock, flags);
 515         if (!changed)
 516                 return false;
 517
 518         if (ctrl->state == NVME_CTRL_LIVE) {
 519                 if (old_state == NVME_CTRL_CONNECTING)
 520                         nvme_stop_failfast_work(ctrl);
 521                 nvme_kick_requeue_lists(ctrl);
 522         } else if (ctrl->state == NVME_CTRL_CONNECTING &&
 523                 old_state == NVME_CTRL_RESETTING) {
 524                 nvme_start_failfast_work(ctrl);
 525         }
 526         return changed;
 527 }
 528 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 529
 530 /*
 531  * Returns true for sink states that can't ever transition back to live.
 532  */
 533 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 534 {
 535         switch (ctrl->state) {
 536         case NVME_CTRL_NEW:
 537         case NVME_CTRL_LIVE:
 538         case NVME_CTRL_RESETTING:
 539         case NVME_CTRL_CONNECTING:
 540                 return false;
 541         case NVME_CTRL_DELETING:
 542         case NVME_CTRL_DELETING_NOIO:
 543         case NVME_CTRL_DEAD:
 544                 return true;
 545         default:
 546                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 547                 return true;
 548         }
 549 }
 550
 551 /*
 552  * Waits for the controller state to be resetting, or returns false if it is
 553  * not possible to ever transition to that state.
 554  */
 555 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 556 {
 557         wait_event(ctrl->state_wq,
 558                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 559                    nvme_state_terminal(ctrl));
 560         return ctrl->state == NVME_CTRL_RESETTING;
 561 }
 562 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 563
 564 static void nvme_free_ns_head(struct kref *ref)
 565 {
 566         struct nvme_ns_head *head =
 567                 container_of(ref, struct nvme_ns_head, ref);
 568
 569         nvme_mpath_remove_disk(head);
 570         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 571         cleanup_srcu_struct(&head->srcu);
 572         nvme_put_subsystem(head->subsys);
 573         kfree(head);
 574 }
 575
 576 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
 577 {
 578         return kref_get_unless_zero(&head->ref);
 579 }
 580
 581 void nvme_put_ns_head(struct nvme_ns_head *head)
 582 {
 583         kref_put(&head->ref, nvme_free_ns_head);
 584 }
 585
 586 static void nvme_free_ns(struct kref *kref)
 587 {
 588         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 589
 590         if (ns->ndev)
 591                 nvme_nvm_unregister(ns);
 592
 593         put_disk(ns->disk);
 594         nvme_put_ns_head(ns->head);
 595         nvme_put_ctrl(ns->ctrl);
 596         kfree(ns);
 597 }
 598
 599 static inline bool nvme_get_ns(struct nvme_ns *ns)
 600 {
 601         return kref_get_unless_zero(&ns->kref);
 602 }
 603
 604 void nvme_put_ns(struct nvme_ns *ns)
 605 {
 606         kref_put(&ns->kref, nvme_free_ns);
 607 }
 608 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
 609
 610 static inline void nvme_clear_nvme_request(struct request *req)
 611 {
 612         nvme_req(req)->status = 0;
 613         nvme_req(req)->retries = 0;
 614         nvme_req(req)->flags = 0;
 615         req->rq_flags |= RQF_DONTPREP;
 616 }
 617
 618 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
 619 {
 620         return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 621 }
 622
 623 static inline void nvme_init_request(struct request *req,
 624                 struct nvme_command *cmd)
 625 {
 626         if (req->q->queuedata)
 627                 req->timeout = NVME_IO_TIMEOUT;
 628         else /* no queuedata implies admin queue */
 629                 req->timeout = NVME_ADMIN_TIMEOUT;
 630
 631         /* passthru commands should let the driver set the SGL flags */
 632         cmd->common.flags &= ~NVME_CMD_SGL_ALL;
 633
 634         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 635         if (req->mq_hctx->type == HCTX_TYPE_POLL)
 636                 req->cmd_flags |= REQ_HIPRI;
 637         nvme_clear_nvme_request(req);
 638         memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
 639 }
 640
 641 struct request *nvme_alloc_request(struct request_queue *q,
 642                 struct nvme_command *cmd, blk_mq_req_flags_t flags)
 643 {
 644         struct request *req;
 645
 646         req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
 647         if (!IS_ERR(req))
 648                 nvme_init_request(req, cmd);
 649         return req;
 650 }
 651 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 652
 653 static struct request *nvme_alloc_request_qid(struct request_queue *q,
 654                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 655 {
 656         struct request *req;
 657
 658         req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
 659                         qid ? qid - 1 : 0);
 660         if (!IS_ERR(req))
 661                 nvme_init_request(req, cmd);
 662         return req;
 663 }
 664
 665 /*
 666  * For something we're not in a state to send to the device the default action
 667  * is to busy it and retry it after the controller state is recovered.  However,
 668  * if the controller is deleting or if anything is marked for failfast or
 669  * nvme multipath it is immediately failed.
 670  *
 671  * Note: commands used to initialize the controller will be marked for failfast.
 672  * Note: nvme cli/ioctl commands are marked for failfast.
 673  */
 674 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
 675                 struct request *rq)
 676 {
 677         if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
 678             ctrl->state != NVME_CTRL_DEAD &&
 679             !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
 680             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
 681                 return BLK_STS_RESOURCE;
 682         return nvme_host_path_error(rq);
 683 }
 684 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
 685
 686 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
 687                 bool queue_live)
 688 {
 689         struct nvme_request *req = nvme_req(rq);
 690
 691         /*
 692          * currently we have a problem sending passthru commands
 693          * on the admin_q if the controller is not LIVE because we can't
 694          * make sure that they are going out after the admin connect,
 695          * controller enable and/or other commands in the initialization
 696          * sequence. until the controller will be LIVE, fail with
 697          * BLK_STS_RESOURCE so that they will be rescheduled.
 698          */
 699         if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
 700                 return false;
 701
 702         if (ctrl->ops->flags & NVME_F_FABRICS) {
 703                 /*
 704                  * Only allow commands on a live queue, except for the connect
 705                  * command, which is require to set the queue live in the
 706                  * appropinquate states.
 707                  */
 708                 switch (ctrl->state) {
 709                 case NVME_CTRL_CONNECTING:
 710                         if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
 711                             req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
 712                                 return true;
 713                         break;
 714                 default:
 715                         break;
 716                 case NVME_CTRL_DEAD:
 717                         return false;
 718                 }
 719         }
 720
 721         return queue_live;
 722 }
 723 EXPORT_SYMBOL_GPL(__nvme_check_ready);
 724
 725 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 726 {
 727         struct nvme_command c = { };
 728
 729         c.directive.opcode = nvme_admin_directive_send;
 730         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 731         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 732         c.directive.dtype = NVME_DIR_IDENTIFY;
 733         c.directive.tdtype = NVME_DIR_STREAMS;
 734         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 735
 736         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 737 }
 738
 739 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 740 {
 741         return nvme_toggle_streams(ctrl, false);
 742 }
 743
 744 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 745 {
 746         return nvme_toggle_streams(ctrl, true);
 747 }
 748
 749 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 750                                   struct streams_directive_params *s, u32 nsid)
 751 {
 752         struct nvme_command c = { };
 753
 754         memset(s, 0, sizeof(*s));
 755
 756         c.directive.opcode = nvme_admin_directive_recv;
 757         c.directive.nsid = cpu_to_le32(nsid);
 758         c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
 759         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 760         c.directive.dtype = NVME_DIR_STREAMS;
 761
 762         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 763 }
 764
 765 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 766 {
 767         struct streams_directive_params s;
 768         int ret;
 769
 770         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 771                 return 0;
 772         if (!streams)
 773                 return 0;
 774
 775         ret = nvme_enable_streams(ctrl);
 776         if (ret)
 777                 return ret;
 778
 779         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 780         if (ret)
 781                 goto out_disable_stream;
 782
 783         ctrl->nssa = le16_to_cpu(s.nssa);
 784         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 785                 dev_info(ctrl->device, "too few streams (%u) available\n",
 786                                         ctrl->nssa);
 787                 goto out_disable_stream;
 788         }
 789
 790         ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 791         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 792         return 0;
 793
 794 out_disable_stream:
 795         nvme_disable_streams(ctrl);
 796         return ret;
 797 }
 798
 799 /*
 800  * Check if 'req' has a write hint associated with it. If it does, assign
 801  * a valid namespace stream to the write.
 802  */
 803 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 804                                      struct request *req, u16 *control,
 805                                      u32 *dsmgmt)
 806 {
 807         enum rw_hint streamid = req->write_hint;
 808
 809         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 810                 streamid = 0;
 811         else {
 812                 streamid--;
 813                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 814                         return;
 815
 816                 *control |= NVME_RW_DTYPE_STREAMS;
 817                 *dsmgmt |= streamid << 16;
 818         }
 819
 820         if (streamid < ARRAY_SIZE(req->q->write_hints))
 821                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 822 }
 823
 824 static inline void nvme_setup_flush(struct nvme_ns *ns,
 825                 struct nvme_command *cmnd)
 826 {
 827         cmnd->common.opcode = nvme_cmd_flush;
 828         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 829 }
 830
 831 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 832                 struct nvme_command *cmnd)
 833 {
 834         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 835         struct nvme_dsm_range *range;
 836         struct bio *bio;
 837
 838         /*
 839          * Some devices do not consider the DSM 'Number of Ranges' field when
 840          * determining how much data to DMA. Always allocate memory for maximum
 841          * number of segments to prevent device reading beyond end of buffer.
 842          */
 843         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 844
 845         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 846         if (!range) {
 847                 /*
 848                  * If we fail allocation our range, fallback to the controller
 849                  * discard page. If that's also busy, it's safe to return
 850                  * busy, as we know we can make progress once that's freed.
 851                  */
 852                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 853                         return BLK_STS_RESOURCE;
 854
 855                 range = page_address(ns->ctrl->discard_page);
 856         }
 857
 858         __rq_for_each_bio(bio, req) {
 859                 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 860                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 861
 862                 if (n < segments) {
 863                         range[n].cattr = cpu_to_le32(0);
 864                         range[n].nlb = cpu_to_le32(nlb);
 865                         range[n].slba = cpu_to_le64(slba);
 866                 }
 867                 n++;
 868         }
 869
 870         if (WARN_ON_ONCE(n != segments)) {
 871                 if (virt_to_page(range) == ns->ctrl->discard_page)
 872                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 873                 else
 874                         kfree(range);
 875                 return BLK_STS_IOERR;
 876         }
 877
 878         cmnd->dsm.opcode = nvme_cmd_dsm;
 879         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 880         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 881         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 882
 883         req->special_vec.bv_page = virt_to_page(range);
 884         req->special_vec.bv_offset = offset_in_page(range);
 885         req->special_vec.bv_len = alloc_size;
 886         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 887
 888         return BLK_STS_OK;
 889 }
 890
 891 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 892                 struct request *req, struct nvme_command *cmnd)
 893 {
 894         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 895                 return nvme_setup_discard(ns, req, cmnd);
 896
 897         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 898         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 899         cmnd->write_zeroes.slba =
 900                 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 901         cmnd->write_zeroes.length =
 902                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 903         cmnd->write_zeroes.control = 0;
 904         return BLK_STS_OK;
 905 }
 906
 907 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 908                 struct request *req, struct nvme_command *cmnd,
 909                 enum nvme_opcode op)
 910 {
 911         struct nvme_ctrl *ctrl = ns->ctrl;
 912         u16 control = 0;
 913         u32 dsmgmt = 0;
 914
 915         if (req->cmd_flags & REQ_FUA)
 916                 control |= NVME_RW_FUA;
 917         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 918                 control |= NVME_RW_LR;
 919
 920         if (req->cmd_flags & REQ_RAHEAD)
 921                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 922
 923         cmnd->rw.opcode = op;
 924         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 925         cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 926         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 927
 928         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 929                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 930
 931         if (ns->ms) {
 932                 /*
 933                  * If formated with metadata, the block layer always provides a
 934                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 935                  * we enable the PRACT bit for protection information or set the
 936                  * namespace capacity to zero to prevent any I/O.
 937                  */
 938                 if (!blk_integrity_rq(req)) {
 939                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 940                                 return BLK_STS_NOTSUPP;
 941                         control |= NVME_RW_PRINFO_PRACT;
 942                 }
 943
 944                 switch (ns->pi_type) {
 945                 case NVME_NS_DPS_PI_TYPE3:
 946                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 947                         break;
 948                 case NVME_NS_DPS_PI_TYPE1:
 949                 case NVME_NS_DPS_PI_TYPE2:
 950                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 951                                         NVME_RW_PRINFO_PRCHK_REF;
 952                         if (op == nvme_cmd_zone_append)
 953                                 control |= NVME_RW_APPEND_PIREMAP;
 954                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 955                         break;
 956                 }
 957         }
 958
 959         cmnd->rw.control = cpu_to_le16(control);
 960         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 961         return 0;
 962 }
 963
 964 void nvme_cleanup_cmd(struct request *req)
 965 {
 966         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 967                 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
 968                 struct page *page = req->special_vec.bv_page;
 969
 970                 if (page == ctrl->discard_page)
 971                         clear_bit_unlock(0, &ctrl->discard_page_busy);
 972                 else
 973                         kfree(page_address(page) + req->special_vec.bv_offset);
 974         }
 975 }
 976 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 977
 978 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
 979 {
 980         struct nvme_command *cmd = nvme_req(req)->cmd;
 981         blk_status_t ret = BLK_STS_OK;
 982
 983         if (!(req->rq_flags & RQF_DONTPREP)) {
 984                 nvme_clear_nvme_request(req);
 985                 memset(cmd, 0, sizeof(*cmd));
 986         }
 987
 988         switch (req_op(req)) {
 989         case REQ_OP_DRV_IN:
 990         case REQ_OP_DRV_OUT:
 991                 /* these are setup prior to execution in nvme_init_request() */
 992                 break;
 993         case REQ_OP_FLUSH:
 994                 nvme_setup_flush(ns, cmd);
 995                 break;
 996         case REQ_OP_ZONE_RESET_ALL:
 997         case REQ_OP_ZONE_RESET:
 998                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
 999                 break;
1000         case REQ_OP_ZONE_OPEN:
1001                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
1002                 break;
1003         case REQ_OP_ZONE_CLOSE:
1004                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
1005                 break;
1006         case REQ_OP_ZONE_FINISH:
1007                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
1008                 break;
1009         case REQ_OP_WRITE_ZEROES:
1010                 ret = nvme_setup_write_zeroes(ns, req, cmd);
1011                 break;
1012         case REQ_OP_DISCARD:
1013                 ret = nvme_setup_discard(ns, req, cmd);
1014                 break;
1015         case REQ_OP_READ:
1016                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1017                 break;
1018         case REQ_OP_WRITE:
1019                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1020                 break;
1021         case REQ_OP_ZONE_APPEND:
1022                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1023                 break;
1024         default:
1025                 WARN_ON_ONCE(1);
1026                 return BLK_STS_IOERR;
1027         }
1028
1029         cmd->common.command_id = req->tag;
1030         trace_nvme_setup_cmd(req, cmd);
1031         return ret;
1032 }
1033 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1034
1035 /*
1036  * Return values:
1037  * 0:  success
1038  * >0: nvme controller's cqe status response
1039  * <0: kernel error in lieu of controller response
1040  */
1041 static int nvme_execute_rq(struct gendisk *disk, struct request *rq,
1042                 bool at_head)
1043 {
1044         blk_status_t status;
1045
1046         status = blk_execute_rq(disk, rq, at_head);
1047         if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1048                 return -EINTR;
1049         if (nvme_req(rq)->status)
1050                 return nvme_req(rq)->status;
1051         return blk_status_to_errno(status);
1052 }
1053
1054 /*
1055  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1056  * if the result is positive, it's an NVM Express status code
1057  */
1058 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1059                 union nvme_result *result, void *buffer, unsigned bufflen,
1060                 unsigned timeout, int qid, int at_head,
1061                 blk_mq_req_flags_t flags)
1062 {
1063         struct request *req;
1064         int ret;
1065
1066         if (qid == NVME_QID_ANY)
1067                 req = nvme_alloc_request(q, cmd, flags);
1068         else
1069                 req = nvme_alloc_request_qid(q, cmd, flags, qid);
1070         if (IS_ERR(req))
1071                 return PTR_ERR(req);
1072
1073         if (timeout)
1074                 req->timeout = timeout;
1075
1076         if (buffer && bufflen) {
1077                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1078                 if (ret)
1079                         goto out;
1080         }
1081
1082         ret = nvme_execute_rq(NULL, req, at_head);
1083         if (result && ret >= 0)
1084                 *result = nvme_req(req)->result;
1085  out:
1086         blk_mq_free_request(req);
1087         return ret;
1088 }
1089 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1090
1091 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1092                 void *buffer, unsigned bufflen)
1093 {
1094         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1095                         NVME_QID_ANY, 0, 0);
1096 }
1097 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1098
1099 static u32 nvme_known_admin_effects(u8 opcode)
1100 {
1101         switch (opcode) {
1102         case nvme_admin_format_nvm:
1103                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1104                         NVME_CMD_EFFECTS_CSE_MASK;
1105         case nvme_admin_sanitize_nvm:
1106                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1107         default:
1108                 break;
1109         }
1110         return 0;
1111 }
1112
1113 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1114 {
1115         u32 effects = 0;
1116
1117         if (ns) {
1118                 if (ns->head->effects)
1119                         effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1120                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1121                         dev_warn_once(ctrl->device,
1122                                 "IO command:%02x has unhandled effects:%08x\n",
1123                                 opcode, effects);
1124                 return 0;
1125         }
1126
1127         if (ctrl->effects)
1128                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1129         effects |= nvme_known_admin_effects(opcode);
1130
1131         return effects;
1132 }
1133 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1134
1135 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1136                                u8 opcode)
1137 {
1138         u32 effects = nvme_command_effects(ctrl, ns, opcode);
1139
1140         /*
1141          * For simplicity, IO to all namespaces is quiesced even if the command
1142          * effects say only one namespace is affected.
1143          */
1144         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1145                 mutex_lock(&ctrl->scan_lock);
1146                 mutex_lock(&ctrl->subsys->lock);
1147                 nvme_mpath_start_freeze(ctrl->subsys);
1148                 nvme_mpath_wait_freeze(ctrl->subsys);
1149                 nvme_start_freeze(ctrl);
1150                 nvme_wait_freeze(ctrl);
1151         }
1152         return effects;
1153 }
1154
1155 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1156 {
1157         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1158                 nvme_unfreeze(ctrl);
1159                 nvme_mpath_unfreeze(ctrl->subsys);
1160                 mutex_unlock(&ctrl->subsys->lock);
1161                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1162                 mutex_unlock(&ctrl->scan_lock);
1163         }
1164         if (effects & NVME_CMD_EFFECTS_CCC)
1165                 nvme_init_ctrl_finish(ctrl);
1166         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1167                 nvme_queue_scan(ctrl);
1168                 flush_work(&ctrl->scan_work);
1169         }
1170 }
1171
1172 int nvme_execute_passthru_rq(struct request *rq)
1173 {
1174         struct nvme_command *cmd = nvme_req(rq)->cmd;
1175         struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1176         struct nvme_ns *ns = rq->q->queuedata;
1177         struct gendisk *disk = ns ? ns->disk : NULL;
1178         u32 effects;
1179         int  ret;
1180
1181         effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1182         ret = nvme_execute_rq(disk, rq, false);
1183         if (effects) /* nothing to be done for zero cmd effects */
1184                 nvme_passthru_end(ctrl, effects);
1185
1186         return ret;
1187 }
1188 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1189
1190 /*
1191  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1192  *
1193  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1194  *   accounting for transport roundtrip times [..].
1195  */
1196 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1197 {
1198         queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1199 }
1200
1201 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1202 {
1203         struct nvme_ctrl *ctrl = rq->end_io_data;
1204         unsigned long flags;
1205         bool startka = false;
1206
1207         blk_mq_free_request(rq);
1208
1209         if (status) {
1210                 dev_err(ctrl->device,
1211                         "failed nvme_keep_alive_end_io error=%d\n",
1212                                 status);
1213                 return;
1214         }
1215
1216         ctrl->comp_seen = false;
1217         spin_lock_irqsave(&ctrl->lock, flags);
1218         if (ctrl->state == NVME_CTRL_LIVE ||
1219             ctrl->state == NVME_CTRL_CONNECTING)
1220                 startka = true;
1221         spin_unlock_irqrestore(&ctrl->lock, flags);
1222         if (startka)
1223                 nvme_queue_keep_alive_work(ctrl);
1224 }
1225
1226 static void nvme_keep_alive_work(struct work_struct *work)
1227 {
1228         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1229                         struct nvme_ctrl, ka_work);
1230         bool comp_seen = ctrl->comp_seen;
1231         struct request *rq;
1232
1233         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1234                 dev_dbg(ctrl->device,
1235                         "reschedule traffic based keep-alive timer\n");
1236                 ctrl->comp_seen = false;
1237                 nvme_queue_keep_alive_work(ctrl);
1238                 return;
1239         }
1240
1241         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1242                                 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1243         if (IS_ERR(rq)) {
1244                 /* allocation failure, reset the controller */
1245                 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1246                 nvme_reset_ctrl(ctrl);
1247                 return;
1248         }
1249
1250         rq->timeout = ctrl->kato * HZ;
1251         rq->end_io_data = ctrl;
1252         blk_execute_rq_nowait(NULL, rq, 0, nvme_keep_alive_end_io);
1253 }
1254
1255 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1256 {
1257         if (unlikely(ctrl->kato == 0))
1258                 return;
1259
1260         nvme_queue_keep_alive_work(ctrl);
1261 }
1262
1263 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1264 {
1265         if (unlikely(ctrl->kato == 0))
1266                 return;
1267
1268         cancel_delayed_work_sync(&ctrl->ka_work);
1269 }
1270 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1271
1272 /*
1273  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1274  * flag, thus sending any new CNS opcodes has a big chance of not working.
1275  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1276  * (but not for any later version).
1277  */
1278 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1279 {
1280         if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1281                 return ctrl->vs < NVME_VS(1, 2, 0);
1282         return ctrl->vs < NVME_VS(1, 1, 0);
1283 }
1284
1285 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1286 {
1287         struct nvme_command c = { };
1288         int error;
1289
1290         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1291         c.identify.opcode = nvme_admin_identify;
1292         c.identify.cns = NVME_ID_CNS_CTRL;
1293
1294         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1295         if (!*id)
1296                 return -ENOMEM;
1297
1298         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1299                         sizeof(struct nvme_id_ctrl));
1300         if (error)
1301                 kfree(*id);
1302         return error;
1303 }
1304
1305 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1306 {
1307         return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1308 }
1309
1310 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1311                 struct nvme_ns_id_desc *cur, bool *csi_seen)
1312 {
1313         const char *warn_str = "ctrl returned bogus length:";
1314         void *data = cur;
1315
1316         switch (cur->nidt) {
1317         case NVME_NIDT_EUI64:
1318                 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1319                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1320                                  warn_str, cur->nidl);
1321                         return -1;
1322                 }
1323                 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1324                 return NVME_NIDT_EUI64_LEN;
1325         case NVME_NIDT_NGUID:
1326                 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1327                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1328                                  warn_str, cur->nidl);
1329                         return -1;
1330                 }
1331                 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1332                 return NVME_NIDT_NGUID_LEN;
1333         case NVME_NIDT_UUID:
1334                 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1335                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1336                                  warn_str, cur->nidl);
1337                         return -1;
1338                 }
1339                 uuid_copy(&ids->uuid, data + sizeof(*cur));
1340                 return NVME_NIDT_UUID_LEN;
1341         case NVME_NIDT_CSI:
1342                 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1343                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1344                                  warn_str, cur->nidl);
1345                         return -1;
1346                 }
1347                 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1348                 *csi_seen = true;
1349                 return NVME_NIDT_CSI_LEN;
1350         default:
1351                 /* Skip unknown types */
1352                 return cur->nidl;
1353         }
1354 }
1355
1356 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1357                 struct nvme_ns_ids *ids)
1358 {
1359         struct nvme_command c = { };
1360         bool csi_seen = false;
1361         int status, pos, len;
1362         void *data;
1363
1364         if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1365                 return 0;
1366         if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1367                 return 0;
1368
1369         c.identify.opcode = nvme_admin_identify;
1370         c.identify.nsid = cpu_to_le32(nsid);
1371         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1372
1373         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1374         if (!data)
1375                 return -ENOMEM;
1376
1377         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1378                                       NVME_IDENTIFY_DATA_SIZE);
1379         if (status) {
1380                 dev_warn(ctrl->device,
1381                         "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1382                         nsid, status);
1383                 goto free_data;
1384         }
1385
1386         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1387                 struct nvme_ns_id_desc *cur = data + pos;
1388
1389                 if (cur->nidl == 0)
1390                         break;
1391
1392                 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1393                 if (len < 0)
1394                         break;
1395
1396                 len += sizeof(*cur);
1397         }
1398
1399         if (nvme_multi_css(ctrl) && !csi_seen) {
1400                 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1401                          nsid);
1402                 status = -EINVAL;
1403         }
1404
1405 free_data:
1406         kfree(data);
1407         return status;
1408 }
1409
1410 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1411                         struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1412 {
1413         struct nvme_command c = { };
1414         int error;
1415
1416         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1417         c.identify.opcode = nvme_admin_identify;
1418         c.identify.nsid = cpu_to_le32(nsid);
1419         c.identify.cns = NVME_ID_CNS_NS;
1420
1421         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1422         if (!*id)
1423                 return -ENOMEM;
1424
1425         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1426         if (error) {
1427                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1428                 goto out_free_id;
1429         }
1430
1431         error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1432         if ((*id)->ncap == 0) /* namespace not allocated or attached */
1433                 goto out_free_id;
1434
1435         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1436             !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1437                 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1438         if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1439             !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1440                 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1441
1442         return 0;
1443
1444 out_free_id:
1445         kfree(*id);
1446         return error;
1447 }
1448
1449 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1450                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1451 {
1452         union nvme_result res = { 0 };
1453         struct nvme_command c = { };
1454         int ret;
1455
1456         c.features.opcode = op;
1457         c.features.fid = cpu_to_le32(fid);
1458         c.features.dword11 = cpu_to_le32(dword11);
1459
1460         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1461                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1462         if (ret >= 0 && result)
1463                 *result = le32_to_cpu(res.u32);
1464         return ret;
1465 }
1466
1467 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1468                       unsigned int dword11, void *buffer, size_t buflen,
1469                       u32 *result)
1470 {
1471         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1472                              buflen, result);
1473 }
1474 EXPORT_SYMBOL_GPL(nvme_set_features);
1475
1476 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1477                       unsigned int dword11, void *buffer, size_t buflen,
1478                       u32 *result)
1479 {
1480         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1481                              buflen, result);
1482 }
1483 EXPORT_SYMBOL_GPL(nvme_get_features);
1484
1485 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1486 {
1487         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1488         u32 result;
1489         int status, nr_io_queues;
1490
1491         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1492                         &result);
1493         if (status < 0)
1494                 return status;
1495
1496         /*
1497          * Degraded controllers might return an error when setting the queue
1498          * count.  We still want to be able to bring them online and offer
1499          * access to the admin queue, as that might be only way to fix them up.
1500          */
1501         if (status > 0) {
1502                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1503                 *count = 0;
1504         } else {
1505                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1506                 *count = min(*count, nr_io_queues);
1507         }
1508
1509         return 0;
1510 }
1511 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1512
1513 #define NVME_AEN_SUPPORTED \
1514         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1515          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1516
1517 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1518 {
1519         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1520         int status;
1521
1522         if (!supported_aens)
1523                 return;
1524
1525         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1526                         NULL, 0, &result);
1527         if (status)
1528                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1529                          supported_aens);
1530
1531         queue_work(nvme_wq, &ctrl->async_event_work);
1532 }
1533
1534 static int nvme_ns_open(struct nvme_ns *ns)
1535 {
1536
1537         /* should never be called due to GENHD_FL_HIDDEN */
1538         if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1539                 goto fail;
1540         if (!nvme_get_ns(ns))
1541                 goto fail;
1542         if (!try_module_get(ns->ctrl->ops->module))
1543                 goto fail_put_ns;
1544
1545         return 0;
1546
1547 fail_put_ns:
1548         nvme_put_ns(ns);
1549 fail:
1550         return -ENXIO;
1551 }
1552
1553 static void nvme_ns_release(struct nvme_ns *ns)
1554 {
1555
1556         module_put(ns->ctrl->ops->module);
1557         nvme_put_ns(ns);
1558 }
1559
1560 static int nvme_open(struct block_device *bdev, fmode_t mode)
1561 {
1562         return nvme_ns_open(bdev->bd_disk->private_data);
1563 }
1564
1565 static void nvme_release(struct gendisk *disk, fmode_t mode)
1566 {
1567         nvme_ns_release(disk->private_data);
1568 }
1569
1570 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1571 {
1572         /* some standard values */
1573         geo->heads = 1 << 6;
1574         geo->sectors = 1 << 5;
1575         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1576         return 0;
1577 }
1578
1579 #ifdef CONFIG_BLK_DEV_INTEGRITY
1580 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1581                                 u32 max_integrity_segments)
1582 {
1583         struct blk_integrity integrity = { };
1584
1585         switch (pi_type) {
1586         case NVME_NS_DPS_PI_TYPE3:
1587                 integrity.profile = &t10_pi_type3_crc;
1588                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1589                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1590                 break;
1591         case NVME_NS_DPS_PI_TYPE1:
1592         case NVME_NS_DPS_PI_TYPE2:
1593                 integrity.profile = &t10_pi_type1_crc;
1594                 integrity.tag_size = sizeof(u16);
1595                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1596                 break;
1597         default:
1598                 integrity.profile = NULL;
1599                 break;
1600         }
1601         integrity.tuple_size = ms;
1602         blk_integrity_register(disk, &integrity);
1603         blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1604 }
1605 #else
1606 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1607                                 u32 max_integrity_segments)
1608 {
1609 }
1610 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1611
1612 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1613 {
1614         struct nvme_ctrl *ctrl = ns->ctrl;
1615         struct request_queue *queue = disk->queue;
1616         u32 size = queue_logical_block_size(queue);
1617
1618         if (ctrl->max_discard_sectors == 0) {
1619                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1620                 return;
1621         }
1622
1623         if (ctrl->nr_streams && ns->sws && ns->sgs)
1624                 size *= ns->sws * ns->sgs;
1625
1626         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1627                         NVME_DSM_MAX_RANGES);
1628
1629         queue->limits.discard_alignment = 0;
1630         queue->limits.discard_granularity = size;
1631
1632         /* If discard is already enabled, don't reset queue limits */
1633         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1634                 return;
1635
1636         blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1637         blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1638
1639         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1640                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1641 }
1642
1643 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1644 {
1645         return !uuid_is_null(&ids->uuid) ||
1646                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1647                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1648 }
1649
1650 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1651 {
1652         return uuid_equal(&a->uuid, &b->uuid) &&
1653                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1654                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1655                 a->csi == b->csi;
1656 }
1657
1658 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1659                                  u32 *phys_bs, u32 *io_opt)
1660 {
1661         struct streams_directive_params s;
1662         int ret;
1663
1664         if (!ctrl->nr_streams)
1665                 return 0;
1666
1667         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1668         if (ret)
1669                 return ret;
1670
1671         ns->sws = le32_to_cpu(s.sws);
1672         ns->sgs = le16_to_cpu(s.sgs);
1673
1674         if (ns->sws) {
1675                 *phys_bs = ns->sws * (1 << ns->lba_shift);
1676                 if (ns->sgs)
1677                         *io_opt = *phys_bs * ns->sgs;
1678         }
1679
1680         return 0;
1681 }
1682
1683 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1684 {
1685         struct nvme_ctrl *ctrl = ns->ctrl;
1686
1687         /*
1688          * The PI implementation requires the metadata size to be equal to the
1689          * t10 pi tuple size.
1690          */
1691         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1692         if (ns->ms == sizeof(struct t10_pi_tuple))
1693                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1694         else
1695                 ns->pi_type = 0;
1696
1697         ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1698         if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1699                 return 0;
1700         if (ctrl->ops->flags & NVME_F_FABRICS) {
1701                 /*
1702                  * The NVMe over Fabrics specification only supports metadata as
1703                  * part of the extended data LBA.  We rely on HCA/HBA support to
1704                  * remap the separate metadata buffer from the block layer.
1705                  */
1706                 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1707                         return -EINVAL;
1708                 if (ctrl->max_integrity_segments)
1709                         ns->features |=
1710                                 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1711         } else {
1712                 /*
1713                  * For PCIe controllers, we can't easily remap the separate
1714                  * metadata buffer from the block layer and thus require a
1715                  * separate metadata buffer for block layer metadata/PI support.
1716                  * We allow extended LBAs for the passthrough interface, though.
1717                  */
1718                 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1719                         ns->features |= NVME_NS_EXT_LBAS;
1720                 else
1721                         ns->features |= NVME_NS_METADATA_SUPPORTED;
1722         }
1723
1724         return 0;
1725 }
1726
1727 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1728                 struct request_queue *q)
1729 {
1730         bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1731
1732         if (ctrl->max_hw_sectors) {
1733                 u32 max_segments =
1734                         (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1735
1736                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1737                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1738                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1739         }
1740         blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1741         blk_queue_dma_alignment(q, 7);
1742         blk_queue_write_cache(q, vwc, vwc);
1743 }
1744
1745 static void nvme_update_disk_info(struct gendisk *disk,
1746                 struct nvme_ns *ns, struct nvme_id_ns *id)
1747 {
1748         sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1749         unsigned short bs = 1 << ns->lba_shift;
1750         u32 atomic_bs, phys_bs, io_opt = 0;
1751
1752         /*
1753          * The block layer can't support LBA sizes larger than the page size
1754          * yet, so catch this early and don't allow block I/O.
1755          */
1756         if (ns->lba_shift > PAGE_SHIFT) {
1757                 capacity = 0;
1758                 bs = (1 << 9);
1759         }
1760
1761         blk_integrity_unregister(disk);
1762
1763         atomic_bs = phys_bs = bs;
1764         nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1765         if (id->nabo == 0) {
1766                 /*
1767                  * Bit 1 indicates whether NAWUPF is defined for this namespace
1768                  * and whether it should be used instead of AWUPF. If NAWUPF ==
1769                  * 0 then AWUPF must be used instead.
1770                  */
1771                 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1772                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1773                 else
1774                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1775         }
1776
1777         if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1778                 /* NPWG = Namespace Preferred Write Granularity */
1779                 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1780                 /* NOWS = Namespace Optimal Write Size */
1781                 io_opt = bs * (1 + le16_to_cpu(id->nows));
1782         }
1783
1784         blk_queue_logical_block_size(disk->queue, bs);
1785         /*
1786          * Linux filesystems assume writing a single physical block is
1787          * an atomic operation. Hence limit the physical block size to the
1788          * value of the Atomic Write Unit Power Fail parameter.
1789          */
1790         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1791         blk_queue_io_min(disk->queue, phys_bs);
1792         blk_queue_io_opt(disk->queue, io_opt);
1793
1794         /*
1795          * Register a metadata profile for PI, or the plain non-integrity NVMe
1796          * metadata masquerading as Type 0 if supported, otherwise reject block
1797          * I/O to namespaces with metadata except when the namespace supports
1798          * PI, as it can strip/insert in that case.
1799          */
1800         if (ns->ms) {
1801                 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1802                     (ns->features & NVME_NS_METADATA_SUPPORTED))
1803                         nvme_init_integrity(disk, ns->ms, ns->pi_type,
1804                                             ns->ctrl->max_integrity_segments);
1805                 else if (!nvme_ns_has_pi(ns))
1806                         capacity = 0;
1807         }
1808
1809         set_capacity_and_notify(disk, capacity);
1810
1811         nvme_config_discard(disk, ns);
1812         blk_queue_max_write_zeroes_sectors(disk->queue,
1813                                            ns->ctrl->max_zeroes_sectors);
1814
1815         set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1816                 test_bit(NVME_NS_FORCE_RO, &ns->flags));
1817 }
1818
1819 static inline bool nvme_first_scan(struct gendisk *disk)
1820 {
1821         /* nvme_alloc_ns() scans the disk prior to adding it */
1822         return !(disk->flags & GENHD_FL_UP);
1823 }
1824
1825 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1826 {
1827         struct nvme_ctrl *ctrl = ns->ctrl;
1828         u32 iob;
1829
1830         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1831             is_power_of_2(ctrl->max_hw_sectors))
1832                 iob = ctrl->max_hw_sectors;
1833         else
1834                 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1835
1836         if (!iob)
1837                 return;
1838
1839         if (!is_power_of_2(iob)) {
1840                 if (nvme_first_scan(ns->disk))
1841                         pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1842                                 ns->disk->disk_name, iob);
1843                 return;
1844         }
1845
1846         if (blk_queue_is_zoned(ns->disk->queue)) {
1847                 if (nvme_first_scan(ns->disk))
1848                         pr_warn("%s: ignoring zoned namespace IO boundary\n",
1849                                 ns->disk->disk_name);
1850                 return;
1851         }
1852
1853         blk_queue_chunk_sectors(ns->queue, iob);
1854 }
1855
1856 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1857 {
1858         unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1859         int ret;
1860
1861         blk_mq_freeze_queue(ns->disk->queue);
1862         ns->lba_shift = id->lbaf[lbaf].ds;
1863         nvme_set_queue_limits(ns->ctrl, ns->queue);
1864
1865         ret = nvme_configure_metadata(ns, id);
1866         if (ret)
1867                 goto out_unfreeze;
1868         nvme_set_chunk_sectors(ns, id);
1869         nvme_update_disk_info(ns->disk, ns, id);
1870
1871         if (ns->head->ids.csi == NVME_CSI_ZNS) {
1872                 ret = nvme_update_zone_info(ns, lbaf);
1873                 if (ret)
1874                         goto out_unfreeze;
1875         }
1876
1877         blk_mq_unfreeze_queue(ns->disk->queue);
1878
1879         if (blk_queue_is_zoned(ns->queue)) {
1880                 ret = nvme_revalidate_zones(ns);
1881                 if (ret && !nvme_first_scan(ns->disk))
1882                         goto out;
1883         }
1884
1885         if (nvme_ns_head_multipath(ns->head)) {
1886                 blk_mq_freeze_queue(ns->head->disk->queue);
1887                 nvme_update_disk_info(ns->head->disk, ns, id);
1888                 blk_stack_limits(&ns->head->disk->queue->limits,
1889                                  &ns->queue->limits, 0);
1890                 blk_queue_update_readahead(ns->head->disk->queue);
1891                 blk_mq_unfreeze_queue(ns->head->disk->queue);
1892         }
1893         return 0;
1894
1895 out_unfreeze:
1896         blk_mq_unfreeze_queue(ns->disk->queue);
1897 out:
1898         /*
1899          * If probing fails due an unsupported feature, hide the block device,
1900          * but still allow other access.
1901          */
1902         if (ret == -ENODEV) {
1903                 ns->disk->flags |= GENHD_FL_HIDDEN;
1904                 ret = 0;
1905         }
1906         return ret;
1907 }
1908
1909 static char nvme_pr_type(enum pr_type type)
1910 {
1911         switch (type) {
1912         case PR_WRITE_EXCLUSIVE:
1913                 return 1;
1914         case PR_EXCLUSIVE_ACCESS:
1915                 return 2;
1916         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1917                 return 3;
1918         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1919                 return 4;
1920         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1921                 return 5;
1922         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1923                 return 6;
1924         default:
1925                 return 0;
1926         }
1927 };
1928
1929 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
1930                 struct nvme_command *c, u8 data[16])
1931 {
1932         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1933         int srcu_idx = srcu_read_lock(&head->srcu);
1934         struct nvme_ns *ns = nvme_find_path(head);
1935         int ret = -EWOULDBLOCK;
1936
1937         if (ns) {
1938                 c->common.nsid = cpu_to_le32(ns->head->ns_id);
1939                 ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
1940         }
1941         srcu_read_unlock(&head->srcu, srcu_idx);
1942         return ret;
1943 }
1944
1945 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
1946                 u8 data[16])
1947 {
1948         c->common.nsid = cpu_to_le32(ns->head->ns_id);
1949         return nvme_submit_sync_cmd(ns->queue, c, data, 16);
1950 }
1951
1952 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1953                                 u64 key, u64 sa_key, u8 op)
1954 {
1955         struct nvme_command c = { };
1956         u8 data[16] = { 0, };
1957
1958         put_unaligned_le64(key, &data[0]);
1959         put_unaligned_le64(sa_key, &data[8]);
1960
1961         c.common.opcode = op;
1962         c.common.cdw10 = cpu_to_le32(cdw10);
1963
1964         if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
1965             bdev->bd_disk->fops == &nvme_ns_head_ops)
1966                 return nvme_send_ns_head_pr_command(bdev, &c, data);
1967         return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
1968 }
1969
1970 static int nvme_pr_register(struct block_device *bdev, u64 old,
1971                 u64 new, unsigned flags)
1972 {
1973         u32 cdw10;
1974
1975         if (flags & ~PR_FL_IGNORE_KEY)
1976                 return -EOPNOTSUPP;
1977
1978         cdw10 = old ? 2 : 0;
1979         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1980         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1981         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1982 }
1983
1984 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1985                 enum pr_type type, unsigned flags)
1986 {
1987         u32 cdw10;
1988
1989         if (flags & ~PR_FL_IGNORE_KEY)
1990                 return -EOPNOTSUPP;
1991
1992         cdw10 = nvme_pr_type(type) << 8;
1993         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1994         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1995 }
1996
1997 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1998                 enum pr_type type, bool abort)
1999 {
2000         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2001
2002         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2003 }
2004
2005 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2006 {
2007         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2008
2009         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2010 }
2011
2012 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2013 {
2014         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2015
2016         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2017 }
2018
2019 const struct pr_ops nvme_pr_ops = {
2020         .pr_register    = nvme_pr_register,
2021         .pr_reserve     = nvme_pr_reserve,
2022         .pr_release     = nvme_pr_release,
2023         .pr_preempt     = nvme_pr_preempt,
2024         .pr_clear       = nvme_pr_clear,
2025 };
2026
2027 #ifdef CONFIG_BLK_SED_OPAL
2028 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2029                 bool send)
2030 {
2031         struct nvme_ctrl *ctrl = data;
2032         struct nvme_command cmd = { };
2033
2034         if (send)
2035                 cmd.common.opcode = nvme_admin_security_send;
2036         else
2037                 cmd.common.opcode = nvme_admin_security_recv;
2038         cmd.common.nsid = 0;
2039         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2040         cmd.common.cdw11 = cpu_to_le32(len);
2041
2042         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2043                         NVME_QID_ANY, 1, 0);
2044 }
2045 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2046 #endif /* CONFIG_BLK_SED_OPAL */
2047
2048 #ifdef CONFIG_BLK_DEV_ZONED
2049 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2050                 unsigned int nr_zones, report_zones_cb cb, void *data)
2051 {
2052         return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2053                         data);
2054 }
2055 #else
2056 #define nvme_report_zones       NULL
2057 #endif /* CONFIG_BLK_DEV_ZONED */
2058
2059 static const struct block_device_operations nvme_bdev_ops = {
2060         .owner          = THIS_MODULE,
2061         .ioctl          = nvme_ioctl,
2062         .open           = nvme_open,
2063         .release        = nvme_release,
2064         .getgeo         = nvme_getgeo,
2065         .report_zones   = nvme_report_zones,
2066         .pr_ops         = &nvme_pr_ops,
2067 };
2068
2069 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2070 {
2071         unsigned long timeout =
2072                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2073         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2074         int ret;
2075
2076         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2077                 if (csts == ~0)
2078                         return -ENODEV;
2079                 if ((csts & NVME_CSTS_RDY) == bit)
2080                         break;
2081
2082                 usleep_range(1000, 2000);
2083                 if (fatal_signal_pending(current))
2084                         return -EINTR;
2085                 if (time_after(jiffies, timeout)) {
2086                         dev_err(ctrl->device,
2087                                 "Device not ready; aborting %s, CSTS=0x%x\n",
2088                                 enabled ? "initialisation" : "reset", csts);
2089                         return -ENODEV;
2090                 }
2091         }
2092
2093         return ret;
2094 }
2095
2096 /*
2097  * If the device has been passed off to us in an enabled state, just clear
2098  * the enabled bit.  The spec says we should set the 'shutdown notification
2099  * bits', but doing so may cause the device to complete commands to the
2100  * admin queue ... and we don't know what memory that might be pointing at!
2101  */
2102 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2103 {
2104         int ret;
2105
2106         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2107         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2108
2109         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2110         if (ret)
2111                 return ret;
2112
2113         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2114                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2115
2116         return nvme_wait_ready(ctrl, ctrl->cap, false);
2117 }
2118 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2119
2120 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2121 {
2122         unsigned dev_page_min;
2123         int ret;
2124
2125         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2126         if (ret) {
2127                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2128                 return ret;
2129         }
2130         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2131
2132         if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2133                 dev_err(ctrl->device,
2134                         "Minimum device page size %u too large for host (%u)\n",
2135                         1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2136                 return -ENODEV;
2137         }
2138
2139         if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2140                 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2141         else
2142                 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2143         ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2144         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2145         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2146         ctrl->ctrl_config |= NVME_CC_ENABLE;
2147
2148         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2149         if (ret)
2150                 return ret;
2151         return nvme_wait_ready(ctrl, ctrl->cap, true);
2152 }
2153 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2154
2155 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2156 {
2157         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2158         u32 csts;
2159         int ret;
2160
2161         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2162         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2163
2164         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2165         if (ret)
2166                 return ret;
2167
2168         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2169                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2170                         break;
2171
2172                 msleep(100);
2173                 if (fatal_signal_pending(current))
2174                         return -EINTR;
2175                 if (time_after(jiffies, timeout)) {
2176                         dev_err(ctrl->device,
2177                                 "Device shutdown incomplete; abort shutdown\n");
2178                         return -ENODEV;
2179                 }
2180         }
2181
2182         return ret;
2183 }
2184 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2185
2186 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2187 {
2188         __le64 ts;
2189         int ret;
2190
2191         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2192                 return 0;
2193
2194         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2195         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2196                         NULL);
2197         if (ret)
2198                 dev_warn_once(ctrl->device,
2199                         "could not set timestamp (%d)\n", ret);
2200         return ret;
2201 }
2202
2203 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2204 {
2205         struct nvme_feat_host_behavior *host;
2206         int ret;
2207
2208         /* Don't bother enabling the feature if retry delay is not reported */
2209         if (!ctrl->crdt[0])
2210                 return 0;
2211
2212         host = kzalloc(sizeof(*host), GFP_KERNEL);
2213         if (!host)
2214                 return 0;
2215
2216         host->acre = NVME_ENABLE_ACRE;
2217         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2218                                 host, sizeof(*host), NULL);
2219         kfree(host);
2220         return ret;
2221 }
2222
2223 /*
2224  * The function checks whether the given total (exlat + enlat) latency of
2225  * a power state allows the latter to be used as an APST transition target.
2226  * It does so by comparing the latency to the primary and secondary latency
2227  * tolerances defined by module params. If there's a match, the corresponding
2228  * timeout value is returned and the matching tolerance index (1 or 2) is
2229  * reported.
2230  */
2231 static bool nvme_apst_get_transition_time(u64 total_latency,
2232                 u64 *transition_time, unsigned *last_index)
2233 {
2234         if (total_latency <= apst_primary_latency_tol_us) {
2235                 if (*last_index == 1)
2236                         return false;
2237                 *last_index = 1;
2238                 *transition_time = apst_primary_timeout_ms;
2239                 return true;
2240         }
2241         if (apst_secondary_timeout_ms &&
2242                 total_latency <= apst_secondary_latency_tol_us) {
2243                 if (*last_index <= 2)
2244                         return false;
2245                 *last_index = 2;
2246                 *transition_time = apst_secondary_timeout_ms;
2247                 return true;
2248         }
2249         return false;
2250 }
2251
2252 /*
2253  * APST (Autonomous Power State Transition) lets us program a table of power
2254  * state transitions that the controller will perform automatically.
2255  *
2256  * Depending on module params, one of the two supported techniques will be used:
2257  *
2258  * - If the parameters provide explicit timeouts and tolerances, they will be
2259  *   used to build a table with up to 2 non-operational states to transition to.
2260  *   The default parameter values were selected based on the values used by
2261  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2262  *   regeneration of the APST table in the event of switching between external
2263  *   and battery power, the timeouts and tolerances reflect a compromise
2264  *   between values used by Microsoft for AC and battery scenarios.
2265  * - If not, we'll configure the table with a simple heuristic: we are willing
2266  *   to spend at most 2% of the time transitioning between power states.
2267  *   Therefore, when running in any given state, we will enter the next
2268  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2269  *   microseconds, as long as that state's exit latency is under the requested
2270  *   maximum latency.
2271  *
2272  * We will not autonomously enter any non-operational state for which the total
2273  * latency exceeds ps_max_latency_us.
2274  *
2275  * Users can set ps_max_latency_us to zero to turn off APST.
2276  */
2277 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2278 {
2279         struct nvme_feat_auto_pst *table;
2280         unsigned apste = 0;
2281         u64 max_lat_us = 0;
2282         __le64 target = 0;
2283         int max_ps = -1;
2284         int state;
2285         int ret;
2286         unsigned last_lt_index = UINT_MAX;
2287
2288         /*
2289          * If APST isn't supported or if we haven't been initialized yet,
2290          * then don't do anything.
2291          */
2292         if (!ctrl->apsta)
2293                 return 0;
2294
2295         if (ctrl->npss > 31) {
2296                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2297                 return 0;
2298         }
2299
2300         table = kzalloc(sizeof(*table), GFP_KERNEL);
2301         if (!table)
2302                 return 0;
2303
2304         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2305                 /* Turn off APST. */
2306                 dev_dbg(ctrl->device, "APST disabled\n");
2307                 goto done;
2308         }
2309
2310         /*
2311          * Walk through all states from lowest- to highest-power.
2312          * According to the spec, lower-numbered states use more power.  NPSS,
2313          * despite the name, is the index of the lowest-power state, not the
2314          * number of states.
2315          */
2316         for (state = (int)ctrl->npss; state >= 0; state--) {
2317                 u64 total_latency_us, exit_latency_us, transition_ms;
2318
2319                 if (target)
2320                         table->entries[state] = target;
2321
2322                 /*
2323                  * Don't allow transitions to the deepest state if it's quirked
2324                  * off.
2325                  */
2326                 if (state == ctrl->npss &&
2327                     (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2328                         continue;
2329
2330                 /*
2331                  * Is this state a useful non-operational state for higher-power
2332                  * states to autonomously transition to?
2333                  */
2334                 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2335                         continue;
2336
2337                 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2338                 if (exit_latency_us > ctrl->ps_max_latency_us)
2339                         continue;
2340
2341                 total_latency_us = exit_latency_us +
2342                         le32_to_cpu(ctrl->psd[state].entry_lat);
2343
2344                 /*
2345                  * This state is good. It can be used as the APST idle target
2346                  * for higher power states.
2347                  */
2348                 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2349                         if (!nvme_apst_get_transition_time(total_latency_us,
2350                                         &transition_ms, &last_lt_index))
2351                                 continue;
2352                 } else {
2353                         transition_ms = total_latency_us + 19;
2354                         do_div(transition_ms, 20);
2355                         if (transition_ms > (1 << 24) - 1)
2356                                 transition_ms = (1 << 24) - 1;
2357                 }
2358
2359                 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2360                 if (max_ps == -1)
2361                         max_ps = state;
2362                 if (total_latency_us > max_lat_us)
2363                         max_lat_us = total_latency_us;
2364         }
2365
2366         if (max_ps == -1)
2367                 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2368         else
2369                 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2370                         max_ps, max_lat_us, (int)sizeof(*table), table);
2371         apste = 1;
2372
2373 done:
2374         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2375                                 table, sizeof(*table), NULL);
2376         if (ret)
2377                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2378         kfree(table);
2379         return ret;
2380 }
2381
2382 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2383 {
2384         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2385         u64 latency;
2386
2387         switch (val) {
2388         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2389         case PM_QOS_LATENCY_ANY:
2390                 latency = U64_MAX;
2391                 break;
2392
2393         default:
2394                 latency = val;
2395         }
2396
2397         if (ctrl->ps_max_latency_us != latency) {
2398                 ctrl->ps_max_latency_us = latency;
2399                 if (ctrl->state == NVME_CTRL_LIVE)
2400                         nvme_configure_apst(ctrl);
2401         }
2402 }
2403
2404 struct nvme_core_quirk_entry {
2405         /*
2406          * NVMe model and firmware strings are padded with spaces.  For
2407          * simplicity, strings in the quirk table are padded with NULLs
2408          * instead.
2409          */
2410         u16 vid;
2411         const char *mn;
2412         const char *fr;
2413         unsigned long quirks;
2414 };
2415
2416 static const struct nvme_core_quirk_entry core_quirks[] = {
2417         {
2418                 /*
2419                  * This Toshiba device seems to die using any APST states.  See:
2420                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2421                  */
2422                 .vid = 0x1179,
2423                 .mn = "THNSF5256GPUK TOSHIBA",
2424                 .quirks = NVME_QUIRK_NO_APST,
2425         },
2426         {
2427                 /*
2428                  * This LiteON CL1-3D*-Q11 firmware version has a race
2429                  * condition associated with actions related to suspend to idle
2430                  * LiteON has resolved the problem in future firmware
2431                  */
2432                 .vid = 0x14a4,
2433                 .fr = "22301111",
2434                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2435         }
2436 };
2437
2438 /* match is null-terminated but idstr is space-padded. */
2439 static bool string_matches(const char *idstr, const char *match, size_t len)
2440 {
2441         size_t matchlen;
2442
2443         if (!match)
2444                 return true;
2445
2446         matchlen = strlen(match);
2447         WARN_ON_ONCE(matchlen > len);
2448
2449         if (memcmp(idstr, match, matchlen))
2450                 return false;
2451
2452         for (; matchlen < len; matchlen++)
2453                 if (idstr[matchlen] != ' ')
2454                         return false;
2455
2456         return true;
2457 }
2458
2459 static bool quirk_matches(const struct nvme_id_ctrl *id,
2460                           const struct nvme_core_quirk_entry *q)
2461 {
2462         return q->vid == le16_to_cpu(id->vid) &&
2463                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2464                 string_matches(id->fr, q->fr, sizeof(id->fr));
2465 }
2466
2467 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2468                 struct nvme_id_ctrl *id)
2469 {
2470         size_t nqnlen;
2471         int off;
2472
2473         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2474                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2475                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2476                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2477                         return;
2478                 }
2479
2480                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2481                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2482         }
2483
2484         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2485         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2486                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2487                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2488         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2489         off += sizeof(id->sn);
2490         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2491         off += sizeof(id->mn);
2492         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2493 }
2494
2495 static void nvme_release_subsystem(struct device *dev)
2496 {
2497         struct nvme_subsystem *subsys =
2498                 container_of(dev, struct nvme_subsystem, dev);
2499
2500         if (subsys->instance >= 0)
2501                 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2502         kfree(subsys);
2503 }
2504
2505 static void nvme_destroy_subsystem(struct kref *ref)
2506 {
2507         struct nvme_subsystem *subsys =
2508                         container_of(ref, struct nvme_subsystem, ref);
2509
2510         mutex_lock(&nvme_subsystems_lock);
2511         list_del(&subsys->entry);
2512         mutex_unlock(&nvme_subsystems_lock);
2513
2514         ida_destroy(&subsys->ns_ida);
2515         device_del(&subsys->dev);
2516         put_device(&subsys->dev);
2517 }
2518
2519 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2520 {
2521         kref_put(&subsys->ref, nvme_destroy_subsystem);
2522 }
2523
2524 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2525 {
2526         struct nvme_subsystem *subsys;
2527
2528         lockdep_assert_held(&nvme_subsystems_lock);
2529
2530         /*
2531          * Fail matches for discovery subsystems. This results
2532          * in each discovery controller bound to a unique subsystem.
2533          * This avoids issues with validating controller values
2534          * that can only be true when there is a single unique subsystem.
2535          * There may be multiple and completely independent entities
2536          * that provide discovery controllers.
2537          */
2538         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2539                 return NULL;
2540
2541         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2542                 if (strcmp(subsys->subnqn, subsysnqn))
2543                         continue;
2544                 if (!kref_get_unless_zero(&subsys->ref))
2545                         continue;
2546                 return subsys;
2547         }
2548
2549         return NULL;
2550 }
2551
2552 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2553         struct device_attribute subsys_attr_##_name = \
2554                 __ATTR(_name, _mode, _show, NULL)
2555
2556 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2557                                     struct device_attribute *attr,
2558                                     char *buf)
2559 {
2560         struct nvme_subsystem *subsys =
2561                 container_of(dev, struct nvme_subsystem, dev);
2562
2563         return sysfs_emit(buf, "%s\n", subsys->subnqn);
2564 }
2565 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2566
2567 #define nvme_subsys_show_str_function(field)                            \
2568 static ssize_t subsys_##field##_show(struct device *dev,                \
2569                             struct device_attribute *attr, char *buf)   \
2570 {                                                                       \
2571         struct nvme_subsystem *subsys =                                 \
2572                 container_of(dev, struct nvme_subsystem, dev);          \
2573         return sysfs_emit(buf, "%.*s\n",                                \
2574                            (int)sizeof(subsys->field), subsys->field);  \
2575 }                                                                       \
2576 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2577
2578 nvme_subsys_show_str_function(model);
2579 nvme_subsys_show_str_function(serial);
2580 nvme_subsys_show_str_function(firmware_rev);
2581
2582 static struct attribute *nvme_subsys_attrs[] = {
2583         &subsys_attr_model.attr,
2584         &subsys_attr_serial.attr,
2585         &subsys_attr_firmware_rev.attr,
2586         &subsys_attr_subsysnqn.attr,
2587 #ifdef CONFIG_NVME_MULTIPATH
2588         &subsys_attr_iopolicy.attr,
2589 #endif
2590         NULL,
2591 };
2592
2593 static const struct attribute_group nvme_subsys_attrs_group = {
2594         .attrs = nvme_subsys_attrs,
2595 };
2596
2597 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2598         &nvme_subsys_attrs_group,
2599         NULL,
2600 };
2601
2602 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2603 {
2604         return ctrl->opts && ctrl->opts->discovery_nqn;
2605 }
2606
2607 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2608                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2609 {
2610         struct nvme_ctrl *tmp;
2611
2612         lockdep_assert_held(&nvme_subsystems_lock);
2613
2614         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2615                 if (nvme_state_terminal(tmp))
2616                         continue;
2617
2618                 if (tmp->cntlid == ctrl->cntlid) {
2619                         dev_err(ctrl->device,
2620                                 "Duplicate cntlid %u with %s, rejecting\n",
2621                                 ctrl->cntlid, dev_name(tmp->device));
2622                         return false;
2623                 }
2624
2625                 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2626                     nvme_discovery_ctrl(ctrl))
2627                         continue;
2628
2629                 dev_err(ctrl->device,
2630                         "Subsystem does not support multiple controllers\n");
2631                 return false;
2632         }
2633
2634         return true;
2635 }
2636
2637 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2638 {
2639         struct nvme_subsystem *subsys, *found;
2640         int ret;
2641
2642         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2643         if (!subsys)
2644                 return -ENOMEM;
2645
2646         subsys->instance = -1;
2647         mutex_init(&subsys->lock);
2648         kref_init(&subsys->ref);
2649         INIT_LIST_HEAD(&subsys->ctrls);
2650         INIT_LIST_HEAD(&subsys->nsheads);
2651         nvme_init_subnqn(subsys, ctrl, id);
2652         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2653         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2654         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2655         subsys->vendor_id = le16_to_cpu(id->vid);
2656         subsys->cmic = id->cmic;
2657         subsys->awupf = le16_to_cpu(id->awupf);
2658 #ifdef CONFIG_NVME_MULTIPATH
2659         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2660 #endif
2661
2662         subsys->dev.class = nvme_subsys_class;
2663         subsys->dev.release = nvme_release_subsystem;
2664         subsys->dev.groups = nvme_subsys_attrs_groups;
2665         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2666         device_initialize(&subsys->dev);
2667
2668         mutex_lock(&nvme_subsystems_lock);
2669         found = __nvme_find_get_subsystem(subsys->subnqn);
2670         if (found) {
2671                 put_device(&subsys->dev);
2672                 subsys = found;
2673
2674                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2675                         ret = -EINVAL;
2676                         goto out_put_subsystem;
2677                 }
2678         } else {
2679                 ret = device_add(&subsys->dev);
2680                 if (ret) {
2681                         dev_err(ctrl->device,
2682                                 "failed to register subsystem device.\n");
2683                         put_device(&subsys->dev);
2684                         goto out_unlock;
2685                 }
2686                 ida_init(&subsys->ns_ida);
2687                 list_add_tail(&subsys->entry, &nvme_subsystems);
2688         }
2689
2690         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2691                                 dev_name(ctrl->device));
2692         if (ret) {
2693                 dev_err(ctrl->device,
2694                         "failed to create sysfs link from subsystem.\n");
2695                 goto out_put_subsystem;
2696         }
2697
2698         if (!found)
2699                 subsys->instance = ctrl->instance;
2700         ctrl->subsys = subsys;
2701         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2702         mutex_unlock(&nvme_subsystems_lock);
2703         return 0;
2704
2705 out_put_subsystem:
2706         nvme_put_subsystem(subsys);
2707 out_unlock:
2708         mutex_unlock(&nvme_subsystems_lock);
2709         return ret;
2710 }
2711
2712 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2713                 void *log, size_t size, u64 offset)
2714 {
2715         struct nvme_command c = { };
2716         u32 dwlen = nvme_bytes_to_numd(size);
2717
2718         c.get_log_page.opcode = nvme_admin_get_log_page;
2719         c.get_log_page.nsid = cpu_to_le32(nsid);
2720         c.get_log_page.lid = log_page;
2721         c.get_log_page.lsp = lsp;
2722         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2723         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2724         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2725         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2726         c.get_log_page.csi = csi;
2727
2728         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2729 }
2730
2731 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2732                                 struct nvme_effects_log **log)
2733 {
2734         struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2735         int ret;
2736
2737         if (cel)
2738                 goto out;
2739
2740         cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2741         if (!cel)
2742                 return -ENOMEM;
2743
2744         ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2745                         cel, sizeof(*cel), 0);
2746         if (ret) {
2747                 kfree(cel);
2748                 return ret;
2749         }
2750
2751         xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2752 out:
2753         *log = cel;
2754         return 0;
2755 }
2756
2757 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2758 {
2759         u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2760
2761         if (check_shl_overflow(1U, units + page_shift - 9, &val))
2762                 return UINT_MAX;
2763         return val;
2764 }
2765
2766 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2767 {
2768         struct nvme_command c = { };
2769         struct nvme_id_ctrl_nvm *id;
2770         int ret;
2771
2772         if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2773                 ctrl->max_discard_sectors = UINT_MAX;
2774                 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2775         } else {
2776                 ctrl->max_discard_sectors = 0;
2777                 ctrl->max_discard_segments = 0;
2778         }
2779
2780         /*
2781          * Even though NVMe spec explicitly states that MDTS is not applicable
2782          * to the write-zeroes, we are cautious and limit the size to the
2783          * controllers max_hw_sectors value, which is based on the MDTS field
2784          * and possibly other limiting factors.
2785          */
2786         if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2787             !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2788                 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2789         else
2790                 ctrl->max_zeroes_sectors = 0;
2791
2792         if (nvme_ctrl_limited_cns(ctrl))
2793                 return 0;
2794
2795         id = kzalloc(sizeof(*id), GFP_KERNEL);
2796         if (!id)
2797                 return 0;
2798
2799         c.identify.opcode = nvme_admin_identify;
2800         c.identify.cns = NVME_ID_CNS_CS_CTRL;
2801         c.identify.csi = NVME_CSI_NVM;
2802
2803         ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2804         if (ret)
2805                 goto free_data;
2806
2807         if (id->dmrl)
2808                 ctrl->max_discard_segments = id->dmrl;
2809         if (id->dmrsl)
2810                 ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
2811         if (id->wzsl)
2812                 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2813
2814 free_data:
2815         kfree(id);
2816         return ret;
2817 }
2818
2819 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2820 {
2821         struct nvme_id_ctrl *id;
2822         u32 max_hw_sectors;
2823         bool prev_apst_enabled;
2824         int ret;
2825
2826         ret = nvme_identify_ctrl(ctrl, &id);
2827         if (ret) {
2828                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2829                 return -EIO;
2830         }
2831
2832         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2833                 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2834                 if (ret < 0)
2835                         goto out_free;
2836         }
2837
2838         if (!(ctrl->ops->flags & NVME_F_FABRICS))
2839                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2840
2841         if (!ctrl->identified) {
2842                 unsigned int i;
2843
2844                 ret = nvme_init_subsystem(ctrl, id);
2845                 if (ret)
2846                         goto out_free;
2847
2848                 /*
2849                  * Check for quirks.  Quirk can depend on firmware version,
2850                  * so, in principle, the set of quirks present can change
2851                  * across a reset.  As a possible future enhancement, we
2852                  * could re-scan for quirks every time we reinitialize
2853                  * the device, but we'd have to make sure that the driver
2854                  * behaves intelligently if the quirks change.
2855                  */
2856                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2857                         if (quirk_matches(id, &core_quirks[i]))
2858                                 ctrl->quirks |= core_quirks[i].quirks;
2859                 }
2860         }
2861
2862         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2863                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2864                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2865         }
2866
2867         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2868         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2869         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2870
2871         ctrl->oacs = le16_to_cpu(id->oacs);
2872         ctrl->oncs = le16_to_cpu(id->oncs);
2873         ctrl->mtfa = le16_to_cpu(id->mtfa);
2874         ctrl->oaes = le32_to_cpu(id->oaes);
2875         ctrl->wctemp = le16_to_cpu(id->wctemp);
2876         ctrl->cctemp = le16_to_cpu(id->cctemp);
2877
2878         atomic_set(&ctrl->abort_limit, id->acl + 1);
2879         ctrl->vwc = id->vwc;
2880         if (id->mdts)
2881                 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
2882         else
2883                 max_hw_sectors = UINT_MAX;
2884         ctrl->max_hw_sectors =
2885                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2886
2887         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2888         ctrl->sgls = le32_to_cpu(id->sgls);
2889         ctrl->kas = le16_to_cpu(id->kas);
2890         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2891         ctrl->ctratt = le32_to_cpu(id->ctratt);
2892
2893         if (id->rtd3e) {
2894                 /* us -> s */
2895                 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
2896
2897                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2898                                                  shutdown_timeout, 60);
2899
2900                 if (ctrl->shutdown_timeout != shutdown_timeout)
2901                         dev_info(ctrl->device,
2902                                  "Shutdown timeout set to %u seconds\n",
2903                                  ctrl->shutdown_timeout);
2904         } else
2905                 ctrl->shutdown_timeout = shutdown_timeout;
2906
2907         ctrl->npss = id->npss;
2908         ctrl->apsta = id->apsta;
2909         prev_apst_enabled = ctrl->apst_enabled;
2910         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2911                 if (force_apst && id->apsta) {
2912                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2913                         ctrl->apst_enabled = true;
2914                 } else {
2915                         ctrl->apst_enabled = false;
2916                 }
2917         } else {
2918                 ctrl->apst_enabled = id->apsta;
2919         }
2920         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2921
2922         if (ctrl->ops->flags & NVME_F_FABRICS) {
2923                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2924                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2925                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2926                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2927
2928                 /*
2929                  * In fabrics we need to verify the cntlid matches the
2930                  * admin connect
2931                  */
2932                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2933                         dev_err(ctrl->device,
2934                                 "Mismatching cntlid: Connect %u vs Identify "
2935                                 "%u, rejecting\n",
2936                                 ctrl->cntlid, le16_to_cpu(id->cntlid));
2937                         ret = -EINVAL;
2938                         goto out_free;
2939                 }
2940
2941                 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
2942                         dev_err(ctrl->device,
2943                                 "keep-alive support is mandatory for fabrics\n");
2944                         ret = -EINVAL;
2945                         goto out_free;
2946                 }
2947         } else {
2948                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2949                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2950                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2951                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2952         }
2953
2954         ret = nvme_mpath_init_identify(ctrl, id);
2955         if (ret < 0)
2956                 goto out_free;
2957
2958         if (ctrl->apst_enabled && !prev_apst_enabled)
2959                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2960         else if (!ctrl->apst_enabled && prev_apst_enabled)
2961                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2962
2963 out_free:
2964         kfree(id);
2965         return ret;
2966 }
2967
2968 /*
2969  * Initialize the cached copies of the Identify data and various controller
2970  * register in our nvme_ctrl structure.  This should be called as soon as
2971  * the admin queue is fully up and running.
2972  */
2973 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
2974 {
2975         int ret;
2976
2977         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2978         if (ret) {
2979                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2980                 return ret;
2981         }
2982
2983         ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2984
2985         if (ctrl->vs >= NVME_VS(1, 1, 0))
2986                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2987
2988         ret = nvme_init_identify(ctrl);
2989         if (ret)
2990                 return ret;
2991
2992         ret = nvme_init_non_mdts_limits(ctrl);
2993         if (ret < 0)
2994                 return ret;
2995
2996         ret = nvme_configure_apst(ctrl);
2997         if (ret < 0)
2998                 return ret;
2999
3000         ret = nvme_configure_timestamp(ctrl);
3001         if (ret < 0)
3002                 return ret;
3003
3004         ret = nvme_configure_directives(ctrl);
3005         if (ret < 0)
3006                 return ret;
3007
3008         ret = nvme_configure_acre(ctrl);
3009         if (ret < 0)
3010                 return ret;
3011
3012         if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3013                 ret = nvme_hwmon_init(ctrl);
3014                 if (ret < 0)
3015                         return ret;
3016         }
3017
3018         ctrl->identified = true;
3019
3020         return 0;
3021 }
3022 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3023
3024 static int nvme_dev_open(struct inode *inode, struct file *file)
3025 {
3026         struct nvme_ctrl *ctrl =
3027                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3028
3029         switch (ctrl->state) {
3030         case NVME_CTRL_LIVE:
3031                 break;
3032         default:
3033                 return -EWOULDBLOCK;
3034         }
3035
3036         nvme_get_ctrl(ctrl);
3037         if (!try_module_get(ctrl->ops->module)) {
3038                 nvme_put_ctrl(ctrl);
3039                 return -EINVAL;
3040         }
3041
3042         file->private_data = ctrl;
3043         return 0;
3044 }
3045
3046 static int nvme_dev_release(struct inode *inode, struct file *file)
3047 {
3048         struct nvme_ctrl *ctrl =
3049                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3050
3051         module_put(ctrl->ops->module);
3052         nvme_put_ctrl(ctrl);
3053         return 0;
3054 }
3055
3056 static const struct file_operations nvme_dev_fops = {
3057         .owner          = THIS_MODULE,
3058         .open           = nvme_dev_open,
3059         .release        = nvme_dev_release,
3060         .unlocked_ioctl = nvme_dev_ioctl,
3061         .compat_ioctl   = compat_ptr_ioctl,
3062 };
3063
3064 static ssize_t nvme_sysfs_reset(struct device *dev,
3065                                 struct device_attribute *attr, const char *buf,
3066                                 size_t count)
3067 {
3068         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3069         int ret;
3070
3071         ret = nvme_reset_ctrl_sync(ctrl);
3072         if (ret < 0)
3073                 return ret;
3074         return count;
3075 }
3076 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3077
3078 static ssize_t nvme_sysfs_rescan(struct device *dev,
3079                                 struct device_attribute *attr, const char *buf,
3080                                 size_t count)
3081 {
3082         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3083
3084         nvme_queue_scan(ctrl);
3085         return count;
3086 }
3087 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3088
3089 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3090 {
3091         struct gendisk *disk = dev_to_disk(dev);
3092
3093         if (disk->fops == &nvme_bdev_ops)
3094                 return nvme_get_ns_from_dev(dev)->head;
3095         else
3096                 return disk->private_data;
3097 }
3098
3099 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3100                 char *buf)
3101 {
3102         struct nvme_ns_head *head = dev_to_ns_head(dev);
3103         struct nvme_ns_ids *ids = &head->ids;
3104         struct nvme_subsystem *subsys = head->subsys;
3105         int serial_len = sizeof(subsys->serial);
3106         int model_len = sizeof(subsys->model);
3107
3108         if (!uuid_is_null(&ids->uuid))
3109                 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3110
3111         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3112                 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3113
3114         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3115                 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3116
3117         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3118                                   subsys->serial[serial_len - 1] == '\0'))
3119                 serial_len--;
3120         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3121                                  subsys->model[model_len - 1] == '\0'))
3122                 model_len--;
3123
3124         return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3125                 serial_len, subsys->serial, model_len, subsys->model,
3126                 head->ns_id);
3127 }
3128 static DEVICE_ATTR_RO(wwid);
3129
3130 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3131                 char *buf)
3132 {
3133         return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3134 }
3135 static DEVICE_ATTR_RO(nguid);
3136
3137 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3138                 char *buf)
3139 {
3140         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3141
3142         /* For backward compatibility expose the NGUID to userspace if
3143          * we have no UUID set
3144          */
3145         if (uuid_is_null(&ids->uuid)) {
3146                 printk_ratelimited(KERN_WARNING
3147                                    "No UUID available providing old NGUID\n");
3148                 return sysfs_emit(buf, "%pU\n", ids->nguid);
3149         }
3150         return sysfs_emit(buf, "%pU\n", &ids->uuid);
3151 }
3152 static DEVICE_ATTR_RO(uuid);
3153
3154 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3155                 char *buf)
3156 {
3157         return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3158 }
3159 static DEVICE_ATTR_RO(eui);
3160
3161 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3162                 char *buf)
3163 {
3164         return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3165 }
3166 static DEVICE_ATTR_RO(nsid);
3167
3168 static struct attribute *nvme_ns_id_attrs[] = {
3169         &dev_attr_wwid.attr,
3170         &dev_attr_uuid.attr,
3171         &dev_attr_nguid.attr,
3172         &dev_attr_eui.attr,
3173         &dev_attr_nsid.attr,
3174 #ifdef CONFIG_NVME_MULTIPATH
3175         &dev_attr_ana_grpid.attr,
3176         &dev_attr_ana_state.attr,
3177 #endif
3178         NULL,
3179 };
3180
3181 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3182                 struct attribute *a, int n)
3183 {
3184         struct device *dev = container_of(kobj, struct device, kobj);
3185         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3186
3187         if (a == &dev_attr_uuid.attr) {
3188                 if (uuid_is_null(&ids->uuid) &&
3189                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3190                         return 0;
3191         }
3192         if (a == &dev_attr_nguid.attr) {
3193                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3194                         return 0;
3195         }
3196         if (a == &dev_attr_eui.attr) {
3197                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3198                         return 0;
3199         }
3200 #ifdef CONFIG_NVME_MULTIPATH
3201         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3202                 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3203                         return 0;
3204                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3205                         return 0;
3206         }
3207 #endif
3208         return a->mode;
3209 }
3210
3211 static const struct attribute_group nvme_ns_id_attr_group = {
3212         .attrs          = nvme_ns_id_attrs,
3213         .is_visible     = nvme_ns_id_attrs_are_visible,
3214 };
3215
3216 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3217         &nvme_ns_id_attr_group,
3218 #ifdef CONFIG_NVM
3219         &nvme_nvm_attr_group,
3220 #endif
3221         NULL,
3222 };
3223
3224 #define nvme_show_str_function(field)                                           \
3225 static ssize_t  field##_show(struct device *dev,                                \
3226                             struct device_attribute *attr, char *buf)           \
3227 {                                                                               \
3228         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3229         return sysfs_emit(buf, "%.*s\n",                                        \
3230                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3231 }                                                                               \
3232 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3233
3234 nvme_show_str_function(model);
3235 nvme_show_str_function(serial);
3236 nvme_show_str_function(firmware_rev);
3237
3238 #define nvme_show_int_function(field)                                           \
3239 static ssize_t  field##_show(struct device *dev,                                \
3240                             struct device_attribute *attr, char *buf)           \
3241 {                                                                               \
3242         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3243         return sysfs_emit(buf, "%d\n", ctrl->field);                            \
3244 }                                                                               \
3245 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3246
3247 nvme_show_int_function(cntlid);
3248 nvme_show_int_function(numa_node);
3249 nvme_show_int_function(queue_count);
3250 nvme_show_int_function(sqsize);
3251 nvme_show_int_function(kato);
3252
3253 static ssize_t nvme_sysfs_delete(struct device *dev,
3254                                 struct device_attribute *attr, const char *buf,
3255                                 size_t count)
3256 {
3257         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3258
3259         if (device_remove_file_self(dev, attr))
3260                 nvme_delete_ctrl_sync(ctrl);
3261         return count;
3262 }
3263 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3264
3265 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3266                                          struct device_attribute *attr,
3267                                          char *buf)
3268 {
3269         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3270
3271         return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3272 }
3273 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3274
3275 static ssize_t nvme_sysfs_show_state(struct device *dev,
3276                                      struct device_attribute *attr,
3277                                      char *buf)
3278 {
3279         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3280         static const char *const state_name[] = {
3281                 [NVME_CTRL_NEW]         = "new",
3282                 [NVME_CTRL_LIVE]        = "live",
3283                 [NVME_CTRL_RESETTING]   = "resetting",
3284                 [NVME_CTRL_CONNECTING]  = "connecting",
3285                 [NVME_CTRL_DELETING]    = "deleting",
3286                 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3287                 [NVME_CTRL_DEAD]        = "dead",
3288         };
3289
3290         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3291             state_name[ctrl->state])
3292                 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3293
3294         return sysfs_emit(buf, "unknown state\n");
3295 }
3296
3297 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3298
3299 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3300                                          struct device_attribute *attr,
3301                                          char *buf)
3302 {
3303         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3304
3305         return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3306 }
3307 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3308
3309 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3310                                         struct device_attribute *attr,
3311                                         char *buf)
3312 {
3313         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3314
3315         return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3316 }
3317 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3318
3319 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3320                                         struct device_attribute *attr,
3321                                         char *buf)
3322 {
3323         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3324
3325         return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3326 }
3327 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3328
3329 static ssize_t nvme_sysfs_show_address(struct device *dev,
3330                                          struct device_attribute *attr,
3331                                          char *buf)
3332 {
3333         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3334
3335         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3336 }
3337 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3338
3339 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3340                 struct device_attribute *attr, char *buf)
3341 {
3342         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3343         struct nvmf_ctrl_options *opts = ctrl->opts;
3344
3345         if (ctrl->opts->max_reconnects == -1)
3346                 return sysfs_emit(buf, "off\n");
3347         return sysfs_emit(buf, "%d\n",
3348                           opts->max_reconnects * opts->reconnect_delay);
3349 }
3350
3351 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3352                 struct device_attribute *attr, const char *buf, size_t count)
3353 {
3354         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3355         struct nvmf_ctrl_options *opts = ctrl->opts;
3356         int ctrl_loss_tmo, err;
3357
3358         err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3359         if (err)
3360                 return -EINVAL;
3361
3362         if (ctrl_loss_tmo < 0)
3363                 opts->max_reconnects = -1;
3364         else
3365                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3366                                                 opts->reconnect_delay);
3367         return count;
3368 }
3369 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3370         nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3371
3372 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3373                 struct device_attribute *attr, char *buf)
3374 {
3375         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3376
3377         if (ctrl->opts->reconnect_delay == -1)
3378                 return sysfs_emit(buf, "off\n");
3379         return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3380 }
3381
3382 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3383                 struct device_attribute *attr, const char *buf, size_t count)
3384 {
3385         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3386         unsigned int v;
3387         int err;
3388
3389         err = kstrtou32(buf, 10, &v);
3390         if (err)
3391                 return err;
3392
3393         ctrl->opts->reconnect_delay = v;
3394         return count;
3395 }
3396 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3397         nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3398
3399 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3400                 struct device_attribute *attr, char *buf)
3401 {
3402         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3403
3404         if (ctrl->opts->fast_io_fail_tmo == -1)
3405                 return sysfs_emit(buf, "off\n");
3406         return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3407 }
3408
3409 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3410                 struct device_attribute *attr, const char *buf, size_t count)
3411 {
3412         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3413         struct nvmf_ctrl_options *opts = ctrl->opts;
3414         int fast_io_fail_tmo, err;
3415
3416         err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3417         if (err)
3418                 return -EINVAL;
3419
3420         if (fast_io_fail_tmo < 0)
3421                 opts->fast_io_fail_tmo = -1;
3422         else
3423                 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3424         return count;
3425 }
3426 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3427         nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3428
3429 static struct attribute *nvme_dev_attrs[] = {
3430         &dev_attr_reset_controller.attr,
3431         &dev_attr_rescan_controller.attr,
3432         &dev_attr_model.attr,
3433         &dev_attr_serial.attr,
3434         &dev_attr_firmware_rev.attr,
3435         &dev_attr_cntlid.attr,
3436         &dev_attr_delete_controller.attr,
3437         &dev_attr_transport.attr,
3438         &dev_attr_subsysnqn.attr,
3439         &dev_attr_address.attr,
3440         &dev_attr_state.attr,
3441         &dev_attr_numa_node.attr,
3442         &dev_attr_queue_count.attr,
3443         &dev_attr_sqsize.attr,
3444         &dev_attr_hostnqn.attr,
3445         &dev_attr_hostid.attr,
3446         &dev_attr_ctrl_loss_tmo.attr,
3447         &dev_attr_reconnect_delay.attr,
3448         &dev_attr_fast_io_fail_tmo.attr,
3449         &dev_attr_kato.attr,
3450         NULL
3451 };
3452
3453 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3454                 struct attribute *a, int n)
3455 {
3456         struct device *dev = container_of(kobj, struct device, kobj);
3457         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3458
3459         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3460                 return 0;
3461         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3462                 return 0;
3463         if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3464                 return 0;
3465         if (a == &dev_attr_hostid.attr && !ctrl->opts)
3466                 return 0;
3467         if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3468                 return 0;
3469         if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3470                 return 0;
3471         if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3472                 return 0;
3473
3474         return a->mode;
3475 }
3476
3477 static const struct attribute_group nvme_dev_attrs_group = {
3478         .attrs          = nvme_dev_attrs,
3479         .is_visible     = nvme_dev_attrs_are_visible,
3480 };
3481
3482 static const struct attribute_group *nvme_dev_attr_groups[] = {
3483         &nvme_dev_attrs_group,
3484         NULL,
3485 };
3486
3487 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3488                 unsigned nsid)
3489 {
3490         struct nvme_ns_head *h;
3491
3492         lockdep_assert_held(&subsys->lock);
3493
3494         list_for_each_entry(h, &subsys->nsheads, entry) {
3495                 if (h->ns_id == nsid && nvme_tryget_ns_head(h))
3496                         return h;
3497         }
3498
3499         return NULL;
3500 }
3501
3502 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3503                 struct nvme_ns_head *new)
3504 {
3505         struct nvme_ns_head *h;
3506
3507         lockdep_assert_held(&subsys->lock);
3508
3509         list_for_each_entry(h, &subsys->nsheads, entry) {
3510                 if (nvme_ns_ids_valid(&new->ids) &&
3511                     nvme_ns_ids_equal(&new->ids, &h->ids))
3512                         return -EINVAL;
3513         }
3514
3515         return 0;
3516 }
3517
3518 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3519 {
3520         cdev_device_del(cdev, cdev_device);
3521         ida_simple_remove(&nvme_ns_chr_minor_ida, MINOR(cdev_device->devt));
3522 }
3523
3524 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3525                 const struct file_operations *fops, struct module *owner)
3526 {
3527         int minor, ret;
3528
3529         minor = ida_simple_get(&nvme_ns_chr_minor_ida, 0, 0, GFP_KERNEL);
3530         if (minor < 0)
3531                 return minor;
3532         cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3533         cdev_device->class = nvme_ns_chr_class;
3534         device_initialize(cdev_device);
3535         cdev_init(cdev, fops);
3536         cdev->owner = owner;
3537         ret = cdev_device_add(cdev, cdev_device);
3538         if (ret) {
3539                 put_device(cdev_device);
3540                 ida_simple_remove(&nvme_ns_chr_minor_ida, minor);
3541         }
3542         return ret;
3543 }
3544
3545 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3546 {
3547         return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3548 }
3549
3550 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3551 {
3552         nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3553         return 0;
3554 }
3555
3556 static const struct file_operations nvme_ns_chr_fops = {
3557         .owner          = THIS_MODULE,
3558         .open           = nvme_ns_chr_open,
3559         .release        = nvme_ns_chr_release,
3560         .unlocked_ioctl = nvme_ns_chr_ioctl,
3561         .compat_ioctl   = compat_ptr_ioctl,
3562 };
3563
3564 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3565 {
3566         int ret;
3567
3568         ns->cdev_device.parent = ns->ctrl->device;
3569         ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3570                            ns->ctrl->instance, ns->head->instance);
3571         if (ret)
3572                 return ret;
3573         ret = nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3574                             ns->ctrl->ops->module);
3575         if (ret)
3576                 kfree_const(ns->cdev_device.kobj.name);
3577         return ret;
3578 }
3579
3580 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3581                 unsigned nsid, struct nvme_ns_ids *ids)
3582 {
3583         struct nvme_ns_head *head;
3584         size_t size = sizeof(*head);
3585         int ret = -ENOMEM;
3586
3587 #ifdef CONFIG_NVME_MULTIPATH
3588         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3589 #endif
3590
3591         head = kzalloc(size, GFP_KERNEL);
3592         if (!head)
3593                 goto out;
3594         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3595         if (ret < 0)
3596                 goto out_free_head;
3597         head->instance = ret;
3598         INIT_LIST_HEAD(&head->list);
3599         ret = init_srcu_struct(&head->srcu);
3600         if (ret)
3601                 goto out_ida_remove;
3602         head->subsys = ctrl->subsys;
3603         head->ns_id = nsid;
3604         head->ids = *ids;
3605         kref_init(&head->ref);
3606
3607         ret = __nvme_check_ids(ctrl->subsys, head);
3608         if (ret) {
3609                 dev_err(ctrl->device,
3610                         "duplicate IDs for nsid %d\n", nsid);
3611                 goto out_cleanup_srcu;
3612         }
3613
3614         if (head->ids.csi) {
3615                 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3616                 if (ret)
3617                         goto out_cleanup_srcu;
3618         } else
3619                 head->effects = ctrl->effects;
3620
3621         ret = nvme_mpath_alloc_disk(ctrl, head);
3622         if (ret)
3623                 goto out_cleanup_srcu;
3624
3625         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3626
3627         kref_get(&ctrl->subsys->ref);
3628
3629         return head;
3630 out_cleanup_srcu:
3631         cleanup_srcu_struct(&head->srcu);
3632 out_ida_remove:
3633         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3634 out_free_head:
3635         kfree(head);
3636 out:
3637         if (ret > 0)
3638                 ret = blk_status_to_errno(nvme_error_status(ret));
3639         return ERR_PTR(ret);
3640 }
3641
3642 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3643                 struct nvme_ns_ids *ids, bool is_shared)
3644 {
3645         struct nvme_ctrl *ctrl = ns->ctrl;
3646         struct nvme_ns_head *head = NULL;
3647         int ret = 0;
3648
3649         mutex_lock(&ctrl->subsys->lock);
3650         head = nvme_find_ns_head(ctrl->subsys, nsid);
3651         if (!head) {
3652                 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3653                 if (IS_ERR(head)) {
3654                         ret = PTR_ERR(head);
3655                         goto out_unlock;
3656                 }
3657                 head->shared = is_shared;
3658         } else {
3659                 ret = -EINVAL;
3660                 if (!is_shared || !head->shared) {
3661                         dev_err(ctrl->device,
3662                                 "Duplicate unshared namespace %d\n", nsid);
3663                         goto out_put_ns_head;
3664                 }
3665                 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3666                         dev_err(ctrl->device,
3667                                 "IDs don't match for shared namespace %d\n",
3668                                         nsid);
3669                         goto out_put_ns_head;
3670                 }
3671         }
3672
3673         list_add_tail_rcu(&ns->siblings, &head->list);
3674         ns->head = head;
3675         mutex_unlock(&ctrl->subsys->lock);
3676         return 0;
3677
3678 out_put_ns_head:
3679         nvme_put_ns_head(head);
3680 out_unlock:
3681         mutex_unlock(&ctrl->subsys->lock);
3682         return ret;
3683 }
3684
3685 static int ns_cmp(void *priv, const struct list_head *a,
3686                 const struct list_head *b)
3687 {
3688         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3689         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3690
3691         return nsa->head->ns_id - nsb->head->ns_id;
3692 }
3693
3694 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3695 {
3696         struct nvme_ns *ns, *ret = NULL;
3697
3698         down_read(&ctrl->namespaces_rwsem);
3699         list_for_each_entry(ns, &ctrl->namespaces, list) {
3700                 if (ns->head->ns_id == nsid) {
3701                         if (!nvme_get_ns(ns))
3702                                 continue;
3703                         ret = ns;
3704                         break;
3705                 }
3706                 if (ns->head->ns_id > nsid)
3707                         break;
3708         }
3709         up_read(&ctrl->namespaces_rwsem);
3710         return ret;
3711 }
3712 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3713
3714 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3715                 struct nvme_ns_ids *ids)
3716 {
3717         struct nvme_ns *ns;
3718         struct gendisk *disk;
3719         struct nvme_id_ns *id;
3720         int node = ctrl->numa_node;
3721
3722         if (nvme_identify_ns(ctrl, nsid, ids, &id))
3723                 return;
3724
3725         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3726         if (!ns)
3727                 goto out_free_id;
3728
3729         ns->queue = blk_mq_init_queue(ctrl->tagset);
3730         if (IS_ERR(ns->queue))
3731                 goto out_free_ns;
3732
3733         if (ctrl->opts && ctrl->opts->data_digest)
3734                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3735
3736         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3737         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3738                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3739
3740         ns->queue->queuedata = ns;
3741         ns->ctrl = ctrl;
3742         kref_init(&ns->kref);
3743
3744         if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3745                 goto out_free_queue;
3746
3747         disk = alloc_disk_node(0, node);
3748         if (!disk)
3749                 goto out_unlink_ns;
3750
3751         disk->fops = &nvme_bdev_ops;
3752         disk->private_data = ns;
3753         disk->queue = ns->queue;
3754         /*
3755          * Without the multipath code enabled, multiple controller per
3756          * subsystems are visible as devices and thus we cannot use the
3757          * subsystem instance.
3758          */
3759         if (!nvme_mpath_set_disk_name(ns, disk->disk_name, &disk->flags))
3760                 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3761                         ns->head->instance);
3762         ns->disk = disk;
3763
3764         if (nvme_update_ns_info(ns, id))
3765                 goto out_put_disk;
3766
3767         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3768                 if (nvme_nvm_register(ns, disk->disk_name, node)) {
3769                         dev_warn(ctrl->device, "LightNVM init failure\n");
3770                         goto out_put_disk;
3771                 }
3772         }
3773
3774         down_write(&ctrl->namespaces_rwsem);
3775         list_add_tail(&ns->list, &ctrl->namespaces);
3776         up_write(&ctrl->namespaces_rwsem);
3777
3778         nvme_get_ctrl(ctrl);
3779
3780         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3781         if (!nvme_ns_head_multipath(ns->head))
3782                 nvme_add_ns_cdev(ns);
3783
3784         nvme_mpath_add_disk(ns, id);
3785         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3786         kfree(id);
3787
3788         return;
3789  out_put_disk:
3790         /* prevent double queue cleanup */
3791         ns->disk->queue = NULL;
3792         put_disk(ns->disk);
3793  out_unlink_ns:
3794         mutex_lock(&ctrl->subsys->lock);
3795         list_del_rcu(&ns->siblings);
3796         if (list_empty(&ns->head->list))
3797                 list_del_init(&ns->head->entry);
3798         mutex_unlock(&ctrl->subsys->lock);
3799         nvme_put_ns_head(ns->head);
3800  out_free_queue:
3801         blk_cleanup_queue(ns->queue);
3802  out_free_ns:
3803         kfree(ns);
3804  out_free_id:
3805         kfree(id);
3806 }
3807
3808 static void nvme_ns_remove(struct nvme_ns *ns)
3809 {
3810         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3811                 return;
3812
3813         set_capacity(ns->disk, 0);
3814         nvme_fault_inject_fini(&ns->fault_inject);
3815
3816         mutex_lock(&ns->ctrl->subsys->lock);
3817         list_del_rcu(&ns->siblings);
3818         if (list_empty(&ns->head->list))
3819                 list_del_init(&ns->head->entry);
3820         mutex_unlock(&ns->ctrl->subsys->lock);
3821
3822         synchronize_rcu(); /* guarantee not available in head->list */
3823         nvme_mpath_clear_current_path(ns);
3824         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3825
3826         if (ns->disk->flags & GENHD_FL_UP) {
3827                 if (!nvme_ns_head_multipath(ns->head))
3828                         nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3829                 del_gendisk(ns->disk);
3830                 blk_cleanup_queue(ns->queue);
3831                 if (blk_get_integrity(ns->disk))
3832                         blk_integrity_unregister(ns->disk);
3833         }
3834
3835         down_write(&ns->ctrl->namespaces_rwsem);
3836         list_del_init(&ns->list);
3837         up_write(&ns->ctrl->namespaces_rwsem);
3838
3839         nvme_mpath_check_last_path(ns);
3840         nvme_put_ns(ns);
3841 }
3842
3843 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3844 {
3845         struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3846
3847         if (ns) {
3848                 nvme_ns_remove(ns);
3849                 nvme_put_ns(ns);
3850         }
3851 }
3852
3853 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
3854 {
3855         struct nvme_id_ns *id;
3856         int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3857
3858         if (test_bit(NVME_NS_DEAD, &ns->flags))
3859                 goto out;
3860
3861         ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
3862         if (ret)
3863                 goto out;
3864
3865         ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3866         if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
3867                 dev_err(ns->ctrl->device,
3868                         "identifiers changed for nsid %d\n", ns->head->ns_id);
3869                 goto out_free_id;
3870         }
3871
3872         ret = nvme_update_ns_info(ns, id);
3873
3874 out_free_id:
3875         kfree(id);
3876 out:
3877         /*
3878          * Only remove the namespace if we got a fatal error back from the
3879          * device, otherwise ignore the error and just move on.
3880          *
3881          * TODO: we should probably schedule a delayed retry here.
3882          */
3883         if (ret > 0 && (ret & NVME_SC_DNR))
3884                 nvme_ns_remove(ns);
3885 }
3886
3887 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3888 {
3889         struct nvme_ns_ids ids = { };
3890         struct nvme_ns *ns;
3891
3892         if (nvme_identify_ns_descs(ctrl, nsid, &ids))
3893                 return;
3894
3895         ns = nvme_find_get_ns(ctrl, nsid);
3896         if (ns) {
3897                 nvme_validate_ns(ns, &ids);
3898                 nvme_put_ns(ns);
3899                 return;
3900         }
3901
3902         switch (ids.csi) {
3903         case NVME_CSI_NVM:
3904                 nvme_alloc_ns(ctrl, nsid, &ids);
3905                 break;
3906         case NVME_CSI_ZNS:
3907                 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
3908                         dev_warn(ctrl->device,
3909                                 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
3910                                 nsid);
3911                         break;
3912                 }
3913                 if (!nvme_multi_css(ctrl)) {
3914                         dev_warn(ctrl->device,
3915                                 "command set not reported for nsid: %d\n",
3916                                 nsid);
3917                         break;
3918                 }
3919                 nvme_alloc_ns(ctrl, nsid, &ids);
3920                 break;
3921         default:
3922                 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
3923                         ids.csi, nsid);
3924                 break;
3925         }
3926 }
3927
3928 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3929                                         unsigned nsid)
3930 {
3931         struct nvme_ns *ns, *next;
3932         LIST_HEAD(rm_list);
3933
3934         down_write(&ctrl->namespaces_rwsem);
3935         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3936                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3937                         list_move_tail(&ns->list, &rm_list);
3938         }
3939         up_write(&ctrl->namespaces_rwsem);
3940
3941         list_for_each_entry_safe(ns, next, &rm_list, list)
3942                 nvme_ns_remove(ns);
3943
3944 }
3945
3946 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3947 {
3948         const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3949         __le32 *ns_list;
3950         u32 prev = 0;
3951         int ret = 0, i;
3952
3953         if (nvme_ctrl_limited_cns(ctrl))
3954                 return -EOPNOTSUPP;
3955
3956         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3957         if (!ns_list)
3958                 return -ENOMEM;
3959
3960         for (;;) {
3961                 struct nvme_command cmd = {
3962                         .identify.opcode        = nvme_admin_identify,
3963                         .identify.cns           = NVME_ID_CNS_NS_ACTIVE_LIST,
3964                         .identify.nsid          = cpu_to_le32(prev),
3965                 };
3966
3967                 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
3968                                             NVME_IDENTIFY_DATA_SIZE);
3969                 if (ret) {
3970                         dev_warn(ctrl->device,
3971                                 "Identify NS List failed (status=0x%x)\n", ret);
3972                         goto free;
3973                 }
3974
3975                 for (i = 0; i < nr_entries; i++) {
3976                         u32 nsid = le32_to_cpu(ns_list[i]);
3977
3978                         if (!nsid)      /* end of the list? */
3979                                 goto out;
3980                         nvme_validate_or_alloc_ns(ctrl, nsid);
3981                         while (++prev < nsid)
3982                                 nvme_ns_remove_by_nsid(ctrl, prev);
3983                 }
3984         }
3985  out:
3986         nvme_remove_invalid_namespaces(ctrl, prev);
3987  free:
3988         kfree(ns_list);
3989         return ret;
3990 }
3991
3992 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3993 {
3994         struct nvme_id_ctrl *id;
3995         u32 nn, i;
3996
3997         if (nvme_identify_ctrl(ctrl, &id))
3998                 return;
3999         nn = le32_to_cpu(id->nn);
4000         kfree(id);
4001
4002         for (i = 1; i <= nn; i++)
4003                 nvme_validate_or_alloc_ns(ctrl, i);
4004
4005         nvme_remove_invalid_namespaces(ctrl, nn);
4006 }
4007
4008 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4009 {
4010         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4011         __le32 *log;
4012         int error;
4013
4014         log = kzalloc(log_size, GFP_KERNEL);
4015         if (!log)
4016                 return;
4017
4018         /*
4019          * We need to read the log to clear the AEN, but we don't want to rely
4020          * on it for the changed namespace information as userspace could have
4021          * raced with us in reading the log page, which could cause us to miss
4022          * updates.
4023          */
4024         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4025                         NVME_CSI_NVM, log, log_size, 0);
4026         if (error)
4027                 dev_warn(ctrl->device,
4028                         "reading changed ns log failed: %d\n", error);
4029
4030         kfree(log);
4031 }
4032
4033 static void nvme_scan_work(struct work_struct *work)
4034 {
4035         struct nvme_ctrl *ctrl =
4036                 container_of(work, struct nvme_ctrl, scan_work);
4037
4038         /* No tagset on a live ctrl means IO queues could not created */
4039         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4040                 return;
4041
4042         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4043                 dev_info(ctrl->device, "rescanning namespaces.\n");
4044                 nvme_clear_changed_ns_log(ctrl);
4045         }
4046
4047         mutex_lock(&ctrl->scan_lock);
4048         if (nvme_scan_ns_list(ctrl) != 0)
4049                 nvme_scan_ns_sequential(ctrl);
4050         mutex_unlock(&ctrl->scan_lock);
4051
4052         down_write(&ctrl->namespaces_rwsem);
4053         list_sort(NULL, &ctrl->namespaces, ns_cmp);
4054         up_write(&ctrl->namespaces_rwsem);
4055 }
4056
4057 /*
4058  * This function iterates the namespace list unlocked to allow recovery from
4059  * controller failure. It is up to the caller to ensure the namespace list is
4060  * not modified by scan work while this function is executing.
4061  */
4062 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4063 {
4064         struct nvme_ns *ns, *next;
4065         LIST_HEAD(ns_list);
4066
4067         /*
4068          * make sure to requeue I/O to all namespaces as these
4069          * might result from the scan itself and must complete
4070          * for the scan_work to make progress
4071          */
4072         nvme_mpath_clear_ctrl_paths(ctrl);
4073
4074         /* prevent racing with ns scanning */
4075         flush_work(&ctrl->scan_work);
4076
4077         /*
4078          * The dead states indicates the controller was not gracefully
4079          * disconnected. In that case, we won't be able to flush any data while
4080          * removing the namespaces' disks; fail all the queues now to avoid
4081          * potentially having to clean up the failed sync later.
4082          */
4083         if (ctrl->state == NVME_CTRL_DEAD)
4084                 nvme_kill_queues(ctrl);
4085
4086         /* this is a no-op when called from the controller reset handler */
4087         nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4088
4089         down_write(&ctrl->namespaces_rwsem);
4090         list_splice_init(&ctrl->namespaces, &ns_list);
4091         up_write(&ctrl->namespaces_rwsem);
4092
4093         list_for_each_entry_safe(ns, next, &ns_list, list)
4094                 nvme_ns_remove(ns);
4095 }
4096 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4097
4098 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4099 {
4100         struct nvme_ctrl *ctrl =
4101                 container_of(dev, struct nvme_ctrl, ctrl_device);
4102         struct nvmf_ctrl_options *opts = ctrl->opts;
4103         int ret;
4104
4105         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4106         if (ret)
4107                 return ret;
4108
4109         if (opts) {
4110                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4111                 if (ret)
4112                         return ret;
4113
4114                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4115                                 opts->trsvcid ?: "none");
4116                 if (ret)
4117                         return ret;
4118
4119                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4120                                 opts->host_traddr ?: "none");
4121                 if (ret)
4122                         return ret;
4123
4124                 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4125                                 opts->host_iface ?: "none");
4126         }
4127         return ret;
4128 }
4129
4130 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4131 {
4132         char *envp[2] = { NULL, NULL };
4133         u32 aen_result = ctrl->aen_result;
4134
4135         ctrl->aen_result = 0;
4136         if (!aen_result)
4137                 return;
4138
4139         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4140         if (!envp[0])
4141                 return;
4142         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4143         kfree(envp[0]);
4144 }
4145
4146 static void nvme_async_event_work(struct work_struct *work)
4147 {
4148         struct nvme_ctrl *ctrl =
4149                 container_of(work, struct nvme_ctrl, async_event_work);
4150
4151         nvme_aen_uevent(ctrl);
4152         ctrl->ops->submit_async_event(ctrl);
4153 }
4154
4155 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4156 {
4157
4158         u32 csts;
4159
4160         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4161                 return false;
4162
4163         if (csts == ~0)
4164                 return false;
4165
4166         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4167 }
4168
4169 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4170 {
4171         struct nvme_fw_slot_info_log *log;
4172
4173         log = kmalloc(sizeof(*log), GFP_KERNEL);
4174         if (!log)
4175                 return;
4176
4177         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4178                         log, sizeof(*log), 0))
4179                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4180         kfree(log);
4181 }
4182
4183 static void nvme_fw_act_work(struct work_struct *work)
4184 {
4185         struct nvme_ctrl *ctrl = container_of(work,
4186                                 struct nvme_ctrl, fw_act_work);
4187         unsigned long fw_act_timeout;
4188
4189         if (ctrl->mtfa)
4190                 fw_act_timeout = jiffies +
4191                                 msecs_to_jiffies(ctrl->mtfa * 100);
4192         else
4193                 fw_act_timeout = jiffies +
4194                                 msecs_to_jiffies(admin_timeout * 1000);
4195
4196         nvme_stop_queues(ctrl);
4197         while (nvme_ctrl_pp_status(ctrl)) {
4198                 if (time_after(jiffies, fw_act_timeout)) {
4199                         dev_warn(ctrl->device,
4200                                 "Fw activation timeout, reset controller\n");
4201                         nvme_try_sched_reset(ctrl);
4202                         return;
4203                 }
4204                 msleep(100);
4205         }
4206
4207         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4208                 return;
4209
4210         nvme_start_queues(ctrl);
4211         /* read FW slot information to clear the AER */
4212         nvme_get_fw_slot_info(ctrl);
4213 }
4214
4215 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4216 {
4217         u32 aer_notice_type = (result & 0xff00) >> 8;
4218
4219         trace_nvme_async_event(ctrl, aer_notice_type);
4220
4221         switch (aer_notice_type) {
4222         case NVME_AER_NOTICE_NS_CHANGED:
4223                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4224                 nvme_queue_scan(ctrl);
4225                 break;
4226         case NVME_AER_NOTICE_FW_ACT_STARTING:
4227                 /*
4228                  * We are (ab)using the RESETTING state to prevent subsequent
4229                  * recovery actions from interfering with the controller's
4230                  * firmware activation.
4231                  */
4232                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4233                         queue_work(nvme_wq, &ctrl->fw_act_work);
4234                 break;
4235 #ifdef CONFIG_NVME_MULTIPATH
4236         case NVME_AER_NOTICE_ANA:
4237                 if (!ctrl->ana_log_buf)
4238                         break;
4239                 queue_work(nvme_wq, &ctrl->ana_work);
4240                 break;
4241 #endif
4242         case NVME_AER_NOTICE_DISC_CHANGED:
4243                 ctrl->aen_result = result;
4244                 break;
4245         default:
4246                 dev_warn(ctrl->device, "async event result %08x\n", result);
4247         }
4248 }
4249
4250 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4251                 volatile union nvme_result *res)
4252 {
4253         u32 result = le32_to_cpu(res->u32);
4254         u32 aer_type = result & 0x07;
4255
4256         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4257                 return;
4258
4259         switch (aer_type) {
4260         case NVME_AER_NOTICE:
4261                 nvme_handle_aen_notice(ctrl, result);
4262                 break;
4263         case NVME_AER_ERROR:
4264         case NVME_AER_SMART:
4265         case NVME_AER_CSS:
4266         case NVME_AER_VS:
4267                 trace_nvme_async_event(ctrl, aer_type);
4268                 ctrl->aen_result = result;
4269                 break;
4270         default:
4271                 break;
4272         }
4273         queue_work(nvme_wq, &ctrl->async_event_work);
4274 }
4275 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4276
4277 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4278 {
4279         nvme_mpath_stop(ctrl);
4280         nvme_stop_keep_alive(ctrl);
4281         nvme_stop_failfast_work(ctrl);
4282         flush_work(&ctrl->async_event_work);
4283         cancel_work_sync(&ctrl->fw_act_work);
4284 }
4285 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4286
4287 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4288 {
4289         nvme_start_keep_alive(ctrl);
4290
4291         nvme_enable_aen(ctrl);
4292
4293         if (ctrl->queue_count > 1) {
4294                 nvme_queue_scan(ctrl);
4295                 nvme_start_queues(ctrl);
4296         }
4297 }
4298 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4299
4300 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4301 {
4302         nvme_hwmon_exit(ctrl);
4303         nvme_fault_inject_fini(&ctrl->fault_inject);
4304         dev_pm_qos_hide_latency_tolerance(ctrl->device);
4305         cdev_device_del(&ctrl->cdev, ctrl->device);
4306         nvme_put_ctrl(ctrl);
4307 }
4308 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4309
4310 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4311 {
4312         struct nvme_effects_log *cel;
4313         unsigned long i;
4314
4315         xa_for_each(&ctrl->cels, i, cel) {
4316                 xa_erase(&ctrl->cels, i);
4317                 kfree(cel);
4318         }
4319
4320         xa_destroy(&ctrl->cels);
4321 }
4322
4323 static void nvme_free_ctrl(struct device *dev)
4324 {
4325         struct nvme_ctrl *ctrl =
4326                 container_of(dev, struct nvme_ctrl, ctrl_device);
4327         struct nvme_subsystem *subsys = ctrl->subsys;
4328
4329         if (!subsys || ctrl->instance != subsys->instance)
4330                 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4331
4332         nvme_free_cels(ctrl);
4333         nvme_mpath_uninit(ctrl);
4334         __free_page(ctrl->discard_page);
4335
4336         if (subsys) {
4337                 mutex_lock(&nvme_subsystems_lock);
4338                 list_del(&ctrl->subsys_entry);
4339                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4340                 mutex_unlock(&nvme_subsystems_lock);
4341         }
4342
4343         ctrl->ops->free_ctrl(ctrl);
4344
4345         if (subsys)
4346                 nvme_put_subsystem(subsys);
4347 }
4348
4349 /*
4350  * Initialize a NVMe controller structures.  This needs to be called during
4351  * earliest initialization so that we have the initialized structured around
4352  * during probing.
4353  */
4354 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4355                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4356 {
4357         int ret;
4358
4359         ctrl->state = NVME_CTRL_NEW;
4360         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4361         spin_lock_init(&ctrl->lock);
4362         mutex_init(&ctrl->scan_lock);
4363         INIT_LIST_HEAD(&ctrl->namespaces);
4364         xa_init(&ctrl->cels);
4365         init_rwsem(&ctrl->namespaces_rwsem);
4366         ctrl->dev = dev;
4367         ctrl->ops = ops;
4368         ctrl->quirks = quirks;
4369         ctrl->numa_node = NUMA_NO_NODE;
4370         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4371         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4372         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4373         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4374         init_waitqueue_head(&ctrl->state_wq);
4375
4376         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4377         INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4378         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4379         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4380
4381         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4382                         PAGE_SIZE);
4383         ctrl->discard_page = alloc_page(GFP_KERNEL);
4384         if (!ctrl->discard_page) {
4385                 ret = -ENOMEM;
4386                 goto out;
4387         }
4388
4389         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4390         if (ret < 0)
4391                 goto out;
4392         ctrl->instance = ret;
4393
4394         device_initialize(&ctrl->ctrl_device);
4395         ctrl->device = &ctrl->ctrl_device;
4396         ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4397                         ctrl->instance);
4398         ctrl->device->class = nvme_class;
4399         ctrl->device->parent = ctrl->dev;
4400         ctrl->device->groups = nvme_dev_attr_groups;
4401         ctrl->device->release = nvme_free_ctrl;
4402         dev_set_drvdata(ctrl->device, ctrl);
4403         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4404         if (ret)
4405                 goto out_release_instance;
4406
4407         nvme_get_ctrl(ctrl);
4408         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4409         ctrl->cdev.owner = ops->module;
4410         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4411         if (ret)
4412                 goto out_free_name;
4413
4414         /*
4415          * Initialize latency tolerance controls.  The sysfs files won't
4416          * be visible to userspace unless the device actually supports APST.
4417          */
4418         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4419         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4420                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4421
4422         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4423         nvme_mpath_init_ctrl(ctrl);
4424
4425         return 0;
4426 out_free_name:
4427         nvme_put_ctrl(ctrl);
4428         kfree_const(ctrl->device->kobj.name);
4429 out_release_instance:
4430         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4431 out:
4432         if (ctrl->discard_page)
4433                 __free_page(ctrl->discard_page);
4434         return ret;
4435 }
4436 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4437
4438 /**
4439  * nvme_kill_queues(): Ends all namespace queues
4440  * @ctrl: the dead controller that needs to end
4441  *
4442  * Call this function when the driver determines it is unable to get the
4443  * controller in a state capable of servicing IO.
4444  */
4445 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4446 {
4447         struct nvme_ns *ns;
4448
4449         down_read(&ctrl->namespaces_rwsem);
4450
4451         /* Forcibly unquiesce queues to avoid blocking dispatch */
4452         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4453                 blk_mq_unquiesce_queue(ctrl->admin_q);
4454
4455         list_for_each_entry(ns, &ctrl->namespaces, list)
4456                 nvme_set_queue_dying(ns);
4457
4458         up_read(&ctrl->namespaces_rwsem);
4459 }
4460 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4461
4462 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4463 {
4464         struct nvme_ns *ns;
4465
4466         down_read(&ctrl->namespaces_rwsem);
4467         list_for_each_entry(ns, &ctrl->namespaces, list)
4468                 blk_mq_unfreeze_queue(ns->queue);
4469         up_read(&ctrl->namespaces_rwsem);
4470 }
4471 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4472
4473 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4474 {
4475         struct nvme_ns *ns;
4476
4477         down_read(&ctrl->namespaces_rwsem);
4478         list_for_each_entry(ns, &ctrl->namespaces, list) {
4479                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4480                 if (timeout <= 0)
4481                         break;
4482         }
4483         up_read(&ctrl->namespaces_rwsem);
4484         return timeout;
4485 }
4486 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4487
4488 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4489 {
4490         struct nvme_ns *ns;
4491
4492         down_read(&ctrl->namespaces_rwsem);
4493         list_for_each_entry(ns, &ctrl->namespaces, list)
4494                 blk_mq_freeze_queue_wait(ns->queue);
4495         up_read(&ctrl->namespaces_rwsem);
4496 }
4497 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4498
4499 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4500 {
4501         struct nvme_ns *ns;
4502
4503         down_read(&ctrl->namespaces_rwsem);
4504         list_for_each_entry(ns, &ctrl->namespaces, list)
4505                 blk_freeze_queue_start(ns->queue);
4506         up_read(&ctrl->namespaces_rwsem);
4507 }
4508 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4509
4510 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4511 {
4512         struct nvme_ns *ns;
4513
4514         down_read(&ctrl->namespaces_rwsem);
4515         list_for_each_entry(ns, &ctrl->namespaces, list)
4516                 blk_mq_quiesce_queue(ns->queue);
4517         up_read(&ctrl->namespaces_rwsem);
4518 }
4519 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4520
4521 void nvme_start_queues(struct nvme_ctrl *ctrl)
4522 {
4523         struct nvme_ns *ns;
4524
4525         down_read(&ctrl->namespaces_rwsem);
4526         list_for_each_entry(ns, &ctrl->namespaces, list)
4527                 blk_mq_unquiesce_queue(ns->queue);
4528         up_read(&ctrl->namespaces_rwsem);
4529 }
4530 EXPORT_SYMBOL_GPL(nvme_start_queues);
4531
4532 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4533 {
4534         struct nvme_ns *ns;
4535
4536         down_read(&ctrl->namespaces_rwsem);
4537         list_for_each_entry(ns, &ctrl->namespaces, list)
4538                 blk_sync_queue(ns->queue);
4539         up_read(&ctrl->namespaces_rwsem);
4540 }
4541 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4542
4543 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4544 {
4545         nvme_sync_io_queues(ctrl);
4546         if (ctrl->admin_q)
4547                 blk_sync_queue(ctrl->admin_q);
4548 }
4549 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4550
4551 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4552 {
4553         if (file->f_op != &nvme_dev_fops)
4554                 return NULL;
4555         return file->private_data;
4556 }
4557 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4558
4559 /*
4560  * Check we didn't inadvertently grow the command structure sizes:
4561  */
4562 static inline void _nvme_check_size(void)
4563 {
4564         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4565         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4566         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4567         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4568         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4569         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4570         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4571         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4572         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4573         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4574         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4575         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4576         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4577         BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4578         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4579         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4580         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4581         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4582         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4583         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4584 }
4585
4586
4587 static int __init nvme_core_init(void)
4588 {
4589         int result = -ENOMEM;
4590
4591         _nvme_check_size();
4592
4593         nvme_wq = alloc_workqueue("nvme-wq",
4594                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4595         if (!nvme_wq)
4596                 goto out;
4597
4598         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4599                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4600         if (!nvme_reset_wq)
4601                 goto destroy_wq;
4602
4603         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4604                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4605         if (!nvme_delete_wq)
4606                 goto destroy_reset_wq;
4607
4608         result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4609                         NVME_MINORS, "nvme");
4610         if (result < 0)
4611                 goto destroy_delete_wq;
4612
4613         nvme_class = class_create(THIS_MODULE, "nvme");
4614         if (IS_ERR(nvme_class)) {
4615                 result = PTR_ERR(nvme_class);
4616                 goto unregister_chrdev;
4617         }
4618         nvme_class->dev_uevent = nvme_class_uevent;
4619
4620         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4621         if (IS_ERR(nvme_subsys_class)) {
4622                 result = PTR_ERR(nvme_subsys_class);
4623                 goto destroy_class;
4624         }
4625
4626         result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4627                                      "nvme-generic");
4628         if (result < 0)
4629                 goto destroy_subsys_class;
4630
4631         nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
4632         if (IS_ERR(nvme_ns_chr_class)) {
4633                 result = PTR_ERR(nvme_ns_chr_class);
4634                 goto unregister_generic_ns;
4635         }
4636
4637         return 0;
4638
4639 unregister_generic_ns:
4640         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4641 destroy_subsys_class:
4642         class_destroy(nvme_subsys_class);
4643 destroy_class:
4644         class_destroy(nvme_class);
4645 unregister_chrdev:
4646         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4647 destroy_delete_wq:
4648         destroy_workqueue(nvme_delete_wq);
4649 destroy_reset_wq:
4650         destroy_workqueue(nvme_reset_wq);
4651 destroy_wq:
4652         destroy_workqueue(nvme_wq);
4653 out:
4654         return result;
4655 }
4656
4657 static void __exit nvme_core_exit(void)
4658 {
4659         class_destroy(nvme_ns_chr_class);
4660         class_destroy(nvme_subsys_class);
4661         class_destroy(nvme_class);
4662         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4663         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4664         destroy_workqueue(nvme_delete_wq);
4665         destroy_workqueue(nvme_reset_wq);
4666         destroy_workqueue(nvme_wq);
4667         ida_destroy(&nvme_ns_chr_minor_ida);
4668         ida_destroy(&nvme_instance_ida);
4669 }
4670
4671 MODULE_LICENSE("GPL");
4672 MODULE_VERSION("1.0");
4673 module_init(nvme_core_init);
4674 module_exit(nvme_core_exit);