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