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