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 bool streams;
  61 module_param(streams, bool, 0644);
  62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  63
  64 /*
  65  * nvme_wq - hosts nvme related works that are not reset or delete
  66  * nvme_reset_wq - hosts nvme reset works
  67  * nvme_delete_wq - hosts nvme delete works
  68  *
  69  * nvme_wq will host works such as scan, aen handling, fw activation,
  70  * keep-alive, periodic reconnects etc. nvme_reset_wq
  71  * runs reset works which also flush works hosted on nvme_wq for
  72  * serialization purposes. nvme_delete_wq host controller deletion
  73  * works which flush reset works for serialization.
  74  */
  75 struct workqueue_struct *nvme_wq;
  76 EXPORT_SYMBOL_GPL(nvme_wq);
  77
  78 struct workqueue_struct *nvme_reset_wq;
  79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  80
  81 struct workqueue_struct *nvme_delete_wq;
  82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  83
  84 static LIST_HEAD(nvme_subsystems);
  85 static DEFINE_MUTEX(nvme_subsystems_lock);
  86
  87 static DEFINE_IDA(nvme_instance_ida);
  88 static dev_t nvme_ctrl_base_chr_devt;
  89 static struct class *nvme_class;
  90 static struct class *nvme_subsys_class;
  91
  92 static DEFINE_IDA(nvme_ns_chr_minor_ida);
  93 static dev_t nvme_ns_chr_devt;
  94 static struct class *nvme_ns_chr_class;
  95
  96 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
  97 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
  98                                            unsigned nsid);
  99
 100 /*
 101  * Prepare a queue for teardown.
 102  *
 103  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
 104  * the capacity to 0 after that to avoid blocking dispatchers that may be
 105  * holding bd_butex.  This will end buffered writers dirtying pages that can't
 106  * be synced.
 107  */
 108 static void nvme_set_queue_dying(struct nvme_ns *ns)
 109 {
 110         if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 111                 return;
 112
 113         blk_set_queue_dying(ns->queue);
 114         blk_mq_unquiesce_queue(ns->queue);
 115
 116         set_capacity_and_notify(ns->disk, 0);
 117 }
 118
 119 void nvme_queue_scan(struct nvme_ctrl *ctrl)
 120 {
 121         /*
 122          * Only new queue scan work when admin and IO queues are both alive
 123          */
 124         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 125                 queue_work(nvme_wq, &ctrl->scan_work);
 126 }
 127
 128 /*
 129  * Use this function to proceed with scheduling reset_work for a controller
 130  * that had previously been set to the resetting state. This is intended for
 131  * code paths that can't be interrupted by other reset attempts. A hot removal
 132  * may prevent this from succeeding.
 133  */
 134 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 135 {
 136         if (ctrl->state != NVME_CTRL_RESETTING)
 137                 return -EBUSY;
 138         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 139                 return -EBUSY;
 140         return 0;
 141 }
 142 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 143
 144 static void nvme_failfast_work(struct work_struct *work)
 145 {
 146         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 147                         struct nvme_ctrl, failfast_work);
 148
 149         if (ctrl->state != NVME_CTRL_CONNECTING)
 150                 return;
 151
 152         set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 153         dev_info(ctrl->device, "failfast expired\n");
 154         nvme_kick_requeue_lists(ctrl);
 155 }
 156
 157 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
 158 {
 159         if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
 160                 return;
 161
 162         schedule_delayed_work(&ctrl->failfast_work,
 163                               ctrl->opts->fast_io_fail_tmo * HZ);
 164 }
 165
 166 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
 167 {
 168         if (!ctrl->opts)
 169                 return;
 170
 171         cancel_delayed_work_sync(&ctrl->failfast_work);
 172         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 173 }
 174
 175
 176 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 177 {
 178         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 179                 return -EBUSY;
 180         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 181                 return -EBUSY;
 182         return 0;
 183 }
 184 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 185
 186 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 187 {
 188         int ret;
 189
 190         ret = nvme_reset_ctrl(ctrl);
 191         if (!ret) {
 192                 flush_work(&ctrl->reset_work);
 193                 if (ctrl->state != NVME_CTRL_LIVE)
 194                         ret = -ENETRESET;
 195         }
 196
 197         return ret;
 198 }
 199
 200 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 201 {
 202         dev_info(ctrl->device,
 203                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 204
 205         flush_work(&ctrl->reset_work);
 206         nvme_stop_ctrl(ctrl);
 207         nvme_remove_namespaces(ctrl);
 208         ctrl->ops->delete_ctrl(ctrl);
 209         nvme_uninit_ctrl(ctrl);
 210 }
 211
 212 static void nvme_delete_ctrl_work(struct work_struct *work)
 213 {
 214         struct nvme_ctrl *ctrl =
 215                 container_of(work, struct nvme_ctrl, delete_work);
 216
 217         nvme_do_delete_ctrl(ctrl);
 218 }
 219
 220 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 221 {
 222         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 223                 return -EBUSY;
 224         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 225                 return -EBUSY;
 226         return 0;
 227 }
 228 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 229
 230 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 231 {
 232         /*
 233          * Keep a reference until nvme_do_delete_ctrl() complete,
 234          * since ->delete_ctrl can free the controller.
 235          */
 236         nvme_get_ctrl(ctrl);
 237         if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 238                 nvme_do_delete_ctrl(ctrl);
 239         nvme_put_ctrl(ctrl);
 240 }
 241
 242 static blk_status_t nvme_error_status(u16 status)
 243 {
 244         switch (status & 0x7ff) {
 245         case NVME_SC_SUCCESS:
 246                 return BLK_STS_OK;
 247         case NVME_SC_CAP_EXCEEDED:
 248                 return BLK_STS_NOSPC;
 249         case NVME_SC_LBA_RANGE:
 250         case NVME_SC_CMD_INTERRUPTED:
 251         case NVME_SC_NS_NOT_READY:
 252                 return BLK_STS_TARGET;
 253         case NVME_SC_BAD_ATTRIBUTES:
 254         case NVME_SC_ONCS_NOT_SUPPORTED:
 255         case NVME_SC_INVALID_OPCODE:
 256         case NVME_SC_INVALID_FIELD:
 257         case NVME_SC_INVALID_NS:
 258                 return BLK_STS_NOTSUPP;
 259         case NVME_SC_WRITE_FAULT:
 260         case NVME_SC_READ_ERROR:
 261         case NVME_SC_UNWRITTEN_BLOCK:
 262         case NVME_SC_ACCESS_DENIED:
 263         case NVME_SC_READ_ONLY:
 264         case NVME_SC_COMPARE_FAILED:
 265                 return BLK_STS_MEDIUM;
 266         case NVME_SC_GUARD_CHECK:
 267         case NVME_SC_APPTAG_CHECK:
 268         case NVME_SC_REFTAG_CHECK:
 269         case NVME_SC_INVALID_PI:
 270                 return BLK_STS_PROTECTION;
 271         case NVME_SC_RESERVATION_CONFLICT:
 272                 return BLK_STS_NEXUS;
 273         case NVME_SC_HOST_PATH_ERROR:
 274                 return BLK_STS_TRANSPORT;
 275         case NVME_SC_ZONE_TOO_MANY_ACTIVE:
 276                 return BLK_STS_ZONE_ACTIVE_RESOURCE;
 277         case NVME_SC_ZONE_TOO_MANY_OPEN:
 278                 return BLK_STS_ZONE_OPEN_RESOURCE;
 279         default:
 280                 return BLK_STS_IOERR;
 281         }
 282 }
 283
 284 static void nvme_retry_req(struct request *req)
 285 {
 286         unsigned long delay = 0;
 287         u16 crd;
 288
 289         /* The mask and shift result must be <= 3 */
 290         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 291         if (crd)
 292                 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
 293
 294         nvme_req(req)->retries++;
 295         blk_mq_requeue_request(req, false);
 296         blk_mq_delay_kick_requeue_list(req->q, delay);
 297 }
 298
 299 enum nvme_disposition {
 300         COMPLETE,
 301         RETRY,
 302         FAILOVER,
 303 };
 304
 305 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 306 {
 307         if (likely(nvme_req(req)->status == 0))
 308                 return COMPLETE;
 309
 310         if (blk_noretry_request(req) ||
 311             (nvme_req(req)->status & NVME_SC_DNR) ||
 312             nvme_req(req)->retries >= nvme_max_retries)
 313                 return COMPLETE;
 314
 315         if (req->cmd_flags & REQ_NVME_MPATH) {
 316                 if (nvme_is_path_error(nvme_req(req)->status) ||
 317                     blk_queue_dying(req->q))
 318                         return FAILOVER;
 319         } else {
 320                 if (blk_queue_dying(req->q))
 321                         return COMPLETE;
 322         }
 323
 324         return RETRY;
 325 }
 326
 327 static inline void nvme_end_req(struct request *req)
 328 {
 329         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 330
 331         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
 332             req_op(req) == REQ_OP_ZONE_APPEND)
 333                 req->__sector = nvme_lba_to_sect(req->q->queuedata,
 334                         le64_to_cpu(nvme_req(req)->result.u64));
 335
 336         nvme_trace_bio_complete(req);
 337         blk_mq_end_request(req, status);
 338 }
 339
 340 void nvme_complete_rq(struct request *req)
 341 {
 342         trace_nvme_complete_rq(req);
 343         nvme_cleanup_cmd(req);
 344
 345         if (nvme_req(req)->ctrl->kas)
 346                 nvme_req(req)->ctrl->comp_seen = true;
 347
 348         switch (nvme_decide_disposition(req)) {
 349         case COMPLETE:
 350                 nvme_end_req(req);
 351                 return;
 352         case RETRY:
 353                 nvme_retry_req(req);
 354                 return;
 355         case FAILOVER:
 356                 nvme_failover_req(req);
 357                 return;
 358         }
 359 }
 360 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 361
 362 /*
 363  * Called to unwind from ->queue_rq on a failed command submission so that the
 364  * multipathing code gets called to potentially failover to another path.
 365  * The caller needs to unwind all transport specific resource allocations and
 366  * must return propagate the return value.
 367  */
 368 blk_status_t nvme_host_path_error(struct request *req)
 369 {
 370         nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
 371         blk_mq_set_request_complete(req);
 372         nvme_complete_rq(req);
 373         return BLK_STS_OK;
 374 }
 375 EXPORT_SYMBOL_GPL(nvme_host_path_error);
 376
 377 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 378 {
 379         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 380                                 "Cancelling I/O %d", req->tag);
 381
 382         /* don't abort one completed request */
 383         if (blk_mq_request_completed(req))
 384                 return true;
 385
 386         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 387         nvme_req(req)->flags |= NVME_REQ_CANCELLED;
 388         blk_mq_complete_request(req);
 389         return true;
 390 }
 391 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 392
 393 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
 394 {
 395         if (ctrl->tagset) {
 396                 blk_mq_tagset_busy_iter(ctrl->tagset,
 397                                 nvme_cancel_request, ctrl);
 398                 blk_mq_tagset_wait_completed_request(ctrl->tagset);
 399         }
 400 }
 401 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
 402
 403 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
 404 {
 405         if (ctrl->admin_tagset) {
 406                 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
 407                                 nvme_cancel_request, ctrl);
 408                 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
 409         }
 410 }
 411 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
 412
 413 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 414                 enum nvme_ctrl_state new_state)
 415 {
 416         enum nvme_ctrl_state old_state;
 417         unsigned long flags;
 418         bool changed = false;
 419
 420         spin_lock_irqsave(&ctrl->lock, flags);
 421
 422         old_state = ctrl->state;
 423         switch (new_state) {
 424         case NVME_CTRL_LIVE:
 425                 switch (old_state) {
 426                 case NVME_CTRL_NEW:
 427                 case NVME_CTRL_RESETTING:
 428                 case NVME_CTRL_CONNECTING:
 429                         changed = true;
 430                         fallthrough;
 431                 default:
 432                         break;
 433                 }
 434                 break;
 435         case NVME_CTRL_RESETTING:
 436                 switch (old_state) {
 437                 case NVME_CTRL_NEW:
 438                 case NVME_CTRL_LIVE:
 439                         changed = true;
 440                         fallthrough;
 441                 default:
 442                         break;
 443                 }
 444                 break;
 445         case NVME_CTRL_CONNECTING:
 446                 switch (old_state) {
 447                 case NVME_CTRL_NEW:
 448                 case NVME_CTRL_RESETTING:
 449                         changed = true;
 450                         fallthrough;
 451                 default:
 452                         break;
 453                 }
 454                 break;
 455         case NVME_CTRL_DELETING:
 456                 switch (old_state) {
 457                 case NVME_CTRL_LIVE:
 458                 case NVME_CTRL_RESETTING:
 459                 case NVME_CTRL_CONNECTING:
 460                         changed = true;
 461                         fallthrough;
 462                 default:
 463                         break;
 464                 }
 465                 break;
 466         case NVME_CTRL_DELETING_NOIO:
 467                 switch (old_state) {
 468                 case NVME_CTRL_DELETING:
 469                 case NVME_CTRL_DEAD:
 470                         changed = true;
 471                         fallthrough;
 472                 default:
 473                         break;
 474                 }
 475                 break;
 476         case NVME_CTRL_DEAD:
 477                 switch (old_state) {
 478                 case NVME_CTRL_DELETING:
 479                         changed = true;
 480                         fallthrough;
 481                 default:
 482                         break;
 483                 }
 484                 break;
 485         default:
 486                 break;
 487         }
 488
 489         if (changed) {
 490                 ctrl->state = new_state;
 491                 wake_up_all(&ctrl->state_wq);
 492         }
 493
 494         spin_unlock_irqrestore(&ctrl->lock, flags);
 495         if (!changed)
 496                 return false;
 497
 498         if (ctrl->state == NVME_CTRL_LIVE) {
 499                 if (old_state == NVME_CTRL_CONNECTING)
 500                         nvme_stop_failfast_work(ctrl);
 501                 nvme_kick_requeue_lists(ctrl);
 502         } else if (ctrl->state == NVME_CTRL_CONNECTING &&
 503                 old_state == NVME_CTRL_RESETTING) {
 504                 nvme_start_failfast_work(ctrl);
 505         }
 506         return changed;
 507 }
 508 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 509
 510 /*
 511  * Returns true for sink states that can't ever transition back to live.
 512  */
 513 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 514 {
 515         switch (ctrl->state) {
 516         case NVME_CTRL_NEW:
 517         case NVME_CTRL_LIVE:
 518         case NVME_CTRL_RESETTING:
 519         case NVME_CTRL_CONNECTING:
 520                 return false;
 521         case NVME_CTRL_DELETING:
 522         case NVME_CTRL_DELETING_NOIO:
 523         case NVME_CTRL_DEAD:
 524                 return true;
 525         default:
 526                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 527                 return true;
 528         }
 529 }
 530
 531 /*
 532  * Waits for the controller state to be resetting, or returns false if it is
 533  * not possible to ever transition to that state.
 534  */
 535 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 536 {
 537         wait_event(ctrl->state_wq,
 538                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 539                    nvme_state_terminal(ctrl));
 540         return ctrl->state == NVME_CTRL_RESETTING;
 541 }
 542 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 543
 544 static void nvme_free_ns_head(struct kref *ref)
 545 {
 546         struct nvme_ns_head *head =
 547                 container_of(ref, struct nvme_ns_head, ref);
 548
 549         nvme_mpath_remove_disk(head);
 550         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 551         cleanup_srcu_struct(&head->srcu);
 552         nvme_put_subsystem(head->subsys);
 553         kfree(head);
 554 }
 555
 556 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
 557 {
 558         return kref_get_unless_zero(&head->ref);
 559 }
 560
 561 void nvme_put_ns_head(struct nvme_ns_head *head)
 562 {
 563         kref_put(&head->ref, nvme_free_ns_head);
 564 }
 565
 566 static void nvme_free_ns(struct kref *kref)
 567 {
 568         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 569
 570         if (ns->ndev)
 571                 nvme_nvm_unregister(ns);
 572
 573         put_disk(ns->disk);
 574         nvme_put_ns_head(ns->head);
 575         nvme_put_ctrl(ns->ctrl);
 576         kfree(ns);
 577 }
 578
 579 static inline bool nvme_get_ns(struct nvme_ns *ns)
 580 {
 581         return kref_get_unless_zero(&ns->kref);
 582 }
 583
 584 void nvme_put_ns(struct nvme_ns *ns)
 585 {
 586         kref_put(&ns->kref, nvme_free_ns);
 587 }
 588 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
 589
 590 static inline void nvme_clear_nvme_request(struct request *req)
 591 {
 592         nvme_req(req)->retries = 0;
 593         nvme_req(req)->flags = 0;
 594         req->rq_flags |= RQF_DONTPREP;
 595 }
 596
 597 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
 598 {
 599         return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 600 }
 601
 602 static inline void nvme_init_request(struct request *req,
 603                 struct nvme_command *cmd)
 604 {
 605         if (req->q->queuedata)
 606                 req->timeout = NVME_IO_TIMEOUT;
 607         else /* no queuedata implies admin queue */
 608                 req->timeout = NVME_ADMIN_TIMEOUT;
 609
 610         /* passthru commands should let the driver set the SGL flags */
 611         cmd->common.flags &= ~NVME_CMD_SGL_ALL;
 612
 613         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 614         nvme_clear_nvme_request(req);
 615         memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
 616 }
 617
 618 struct request *nvme_alloc_request(struct request_queue *q,
 619                 struct nvme_command *cmd, blk_mq_req_flags_t flags)
 620 {
 621         struct request *req;
 622
 623         req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
 624         if (!IS_ERR(req))
 625                 nvme_init_request(req, cmd);
 626         return req;
 627 }
 628 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 629
 630 static struct request *nvme_alloc_request_qid(struct request_queue *q,
 631                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 632 {
 633         struct request *req;
 634
 635         req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
 636                         qid ? qid - 1 : 0);
 637         if (!IS_ERR(req))
 638                 nvme_init_request(req, cmd);
 639         return req;
 640 }
 641
 642 /*
 643  * For something we're not in a state to send to the device the default action
 644  * is to busy it and retry it after the controller state is recovered.  However,
 645  * if the controller is deleting or if anything is marked for failfast or
 646  * nvme multipath it is immediately failed.
 647  *
 648  * Note: commands used to initialize the controller will be marked for failfast.
 649  * Note: nvme cli/ioctl commands are marked for failfast.
 650  */
 651 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
 652                 struct request *rq)
 653 {
 654         if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
 655             ctrl->state != NVME_CTRL_DEAD &&
 656             !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
 657             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
 658                 return BLK_STS_RESOURCE;
 659         return nvme_host_path_error(rq);
 660 }
 661 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
 662
 663 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
 664                 bool queue_live)
 665 {
 666         struct nvme_request *req = nvme_req(rq);
 667
 668         /*
 669          * currently we have a problem sending passthru commands
 670          * on the admin_q if the controller is not LIVE because we can't
 671          * make sure that they are going out after the admin connect,
 672          * controller enable and/or other commands in the initialization
 673          * sequence. until the controller will be LIVE, fail with
 674          * BLK_STS_RESOURCE so that they will be rescheduled.
 675          */
 676         if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
 677                 return false;
 678
 679         if (ctrl->ops->flags & NVME_F_FABRICS) {
 680                 /*
 681                  * Only allow commands on a live queue, except for the connect
 682                  * command, which is require to set the queue live in the
 683                  * appropinquate states.
 684                  */
 685                 switch (ctrl->state) {
 686                 case NVME_CTRL_CONNECTING:
 687                         if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
 688                             req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
 689                                 return true;
 690                         break;
 691                 default:
 692                         break;
 693                 case NVME_CTRL_DEAD:
 694                         return false;
 695                 }
 696         }
 697
 698         return queue_live;
 699 }
 700 EXPORT_SYMBOL_GPL(__nvme_check_ready);
 701
 702 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 703 {
 704         struct nvme_command c;
 705
 706         memset(&c, 0, sizeof(c));
 707
 708         c.directive.opcode = nvme_admin_directive_send;
 709         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 710         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 711         c.directive.dtype = NVME_DIR_IDENTIFY;
 712         c.directive.tdtype = NVME_DIR_STREAMS;
 713         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 714
 715         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 716 }
 717
 718 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 719 {
 720         return nvme_toggle_streams(ctrl, false);
 721 }
 722
 723 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 724 {
 725         return nvme_toggle_streams(ctrl, true);
 726 }
 727
 728 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 729                                   struct streams_directive_params *s, u32 nsid)
 730 {
 731         struct nvme_command c;
 732
 733         memset(&c, 0, sizeof(c));
 734         memset(s, 0, sizeof(*s));
 735
 736         c.directive.opcode = nvme_admin_directive_recv;
 737         c.directive.nsid = cpu_to_le32(nsid);
 738         c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
 739         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 740         c.directive.dtype = NVME_DIR_STREAMS;
 741
 742         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 743 }
 744
 745 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 746 {
 747         struct streams_directive_params s;
 748         int ret;
 749
 750         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 751                 return 0;
 752         if (!streams)
 753                 return 0;
 754
 755         ret = nvme_enable_streams(ctrl);
 756         if (ret)
 757                 return ret;
 758
 759         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 760         if (ret)
 761                 goto out_disable_stream;
 762
 763         ctrl->nssa = le16_to_cpu(s.nssa);
 764         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 765                 dev_info(ctrl->device, "too few streams (%u) available\n",
 766                                         ctrl->nssa);
 767                 goto out_disable_stream;
 768         }
 769
 770         ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 771         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 772         return 0;
 773
 774 out_disable_stream:
 775         nvme_disable_streams(ctrl);
 776         return ret;
 777 }
 778
 779 /*
 780  * Check if 'req' has a write hint associated with it. If it does, assign
 781  * a valid namespace stream to the write.
 782  */
 783 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 784                                      struct request *req, u16 *control,
 785                                      u32 *dsmgmt)
 786 {
 787         enum rw_hint streamid = req->write_hint;
 788
 789         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 790                 streamid = 0;
 791         else {
 792                 streamid--;
 793                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 794                         return;
 795
 796                 *control |= NVME_RW_DTYPE_STREAMS;
 797                 *dsmgmt |= streamid << 16;
 798         }
 799
 800         if (streamid < ARRAY_SIZE(req->q->write_hints))
 801                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 802 }
 803
 804 static inline void nvme_setup_flush(struct nvme_ns *ns,
 805                 struct nvme_command *cmnd)
 806 {
 807         cmnd->common.opcode = nvme_cmd_flush;
 808         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 809 }
 810
 811 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 812                 struct nvme_command *cmnd)
 813 {
 814         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 815         struct nvme_dsm_range *range;
 816         struct bio *bio;
 817
 818         /*
 819          * Some devices do not consider the DSM 'Number of Ranges' field when
 820          * determining how much data to DMA. Always allocate memory for maximum
 821          * number of segments to prevent device reading beyond end of buffer.
 822          */
 823         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 824
 825         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 826         if (!range) {
 827                 /*
 828                  * If we fail allocation our range, fallback to the controller
 829                  * discard page. If that's also busy, it's safe to return
 830                  * busy, as we know we can make progress once that's freed.
 831                  */
 832                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 833                         return BLK_STS_RESOURCE;
 834
 835                 range = page_address(ns->ctrl->discard_page);
 836         }
 837
 838         __rq_for_each_bio(bio, req) {
 839                 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 840                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 841
 842                 if (n < segments) {
 843                         range[n].cattr = cpu_to_le32(0);
 844                         range[n].nlb = cpu_to_le32(nlb);
 845                         range[n].slba = cpu_to_le64(slba);
 846                 }
 847                 n++;
 848         }
 849
 850         if (WARN_ON_ONCE(n != segments)) {
 851                 if (virt_to_page(range) == ns->ctrl->discard_page)
 852                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 853                 else
 854                         kfree(range);
 855                 return BLK_STS_IOERR;
 856         }
 857
 858         cmnd->dsm.opcode = nvme_cmd_dsm;
 859         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 860         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 861         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 862
 863         req->special_vec.bv_page = virt_to_page(range);
 864         req->special_vec.bv_offset = offset_in_page(range);
 865         req->special_vec.bv_len = alloc_size;
 866         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 867
 868         return BLK_STS_OK;
 869 }
 870
 871 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 872                 struct request *req, struct nvme_command *cmnd)
 873 {
 874         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 875                 return nvme_setup_discard(ns, req, cmnd);
 876
 877         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 878         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 879         cmnd->write_zeroes.slba =
 880                 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 881         cmnd->write_zeroes.length =
 882                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 883         cmnd->write_zeroes.control = 0;
 884         return BLK_STS_OK;
 885 }
 886
 887 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 888                 struct request *req, struct nvme_command *cmnd,
 889                 enum nvme_opcode op)
 890 {
 891         struct nvme_ctrl *ctrl = ns->ctrl;
 892         u16 control = 0;
 893         u32 dsmgmt = 0;
 894
 895         if (req->cmd_flags & REQ_FUA)
 896                 control |= NVME_RW_FUA;
 897         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 898                 control |= NVME_RW_LR;
 899
 900         if (req->cmd_flags & REQ_RAHEAD)
 901                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 902
 903         cmnd->rw.opcode = op;
 904         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 905         cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 906         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 907
 908         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 909                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 910
 911         if (ns->ms) {
 912                 /*
 913                  * If formated with metadata, the block layer always provides a
 914                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 915                  * we enable the PRACT bit for protection information or set the
 916                  * namespace capacity to zero to prevent any I/O.
 917                  */
 918                 if (!blk_integrity_rq(req)) {
 919                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 920                                 return BLK_STS_NOTSUPP;
 921                         control |= NVME_RW_PRINFO_PRACT;
 922                 }
 923
 924                 switch (ns->pi_type) {
 925                 case NVME_NS_DPS_PI_TYPE3:
 926                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 927                         break;
 928                 case NVME_NS_DPS_PI_TYPE1:
 929                 case NVME_NS_DPS_PI_TYPE2:
 930                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 931                                         NVME_RW_PRINFO_PRCHK_REF;
 932                         if (op == nvme_cmd_zone_append)
 933                                 control |= NVME_RW_APPEND_PIREMAP;
 934                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 935                         break;
 936                 }
 937         }
 938
 939         cmnd->rw.control = cpu_to_le16(control);
 940         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 941         return 0;
 942 }
 943
 944 void nvme_cleanup_cmd(struct request *req)
 945 {
 946         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 947                 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
 948                 struct page *page = req->special_vec.bv_page;
 949
 950                 if (page == ctrl->discard_page)
 951                         clear_bit_unlock(0, &ctrl->discard_page_busy);
 952                 else
 953                         kfree(page_address(page) + req->special_vec.bv_offset);
 954         }
 955 }
 956 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 957
 958 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
 959 {
 960         struct nvme_command *cmd = nvme_req(req)->cmd;
 961         blk_status_t ret = BLK_STS_OK;
 962
 963         if (!(req->rq_flags & RQF_DONTPREP)) {
 964                 nvme_clear_nvme_request(req);
 965                 memset(cmd, 0, sizeof(*cmd));
 966         }
 967
 968         switch (req_op(req)) {
 969         case REQ_OP_DRV_IN:
 970         case REQ_OP_DRV_OUT:
 971                 /* these are setup prior to execution in nvme_init_request() */
 972                 break;
 973         case REQ_OP_FLUSH:
 974                 nvme_setup_flush(ns, cmd);
 975                 break;
 976         case REQ_OP_ZONE_RESET_ALL:
 977         case REQ_OP_ZONE_RESET:
 978                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
 979                 break;
 980         case REQ_OP_ZONE_OPEN:
 981                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
 982                 break;
 983         case REQ_OP_ZONE_CLOSE:
 984                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
 985                 break;
 986         case REQ_OP_ZONE_FINISH:
 987                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
 988                 break;
 989         case REQ_OP_WRITE_ZEROES:
 990                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 991                 break;
 992         case REQ_OP_DISCARD:
 993                 ret = nvme_setup_discard(ns, req, cmd);
 994                 break;
 995         case REQ_OP_READ:
 996                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
 997                 break;
 998         case REQ_OP_WRITE:
 999                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1000                 break;
1001         case REQ_OP_ZONE_APPEND:
1002                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1003                 break;
1004         default:
1005                 WARN_ON_ONCE(1);
1006                 return BLK_STS_IOERR;
1007         }
1008
1009         cmd->common.command_id = req->tag;
1010         trace_nvme_setup_cmd(req, cmd);
1011         return ret;
1012 }
1013 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1014
1015 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
1016 {
1017         struct completion *waiting = rq->end_io_data;
1018
1019         rq->end_io_data = NULL;
1020         complete(waiting);
1021 }
1022
1023 static void nvme_execute_rq_polled(struct request_queue *q,
1024                 struct gendisk *bd_disk, struct request *rq, int at_head)
1025 {
1026         DECLARE_COMPLETION_ONSTACK(wait);
1027
1028         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
1029
1030         rq->cmd_flags |= REQ_HIPRI;
1031         rq->end_io_data = &wait;
1032         blk_execute_rq_nowait(bd_disk, rq, at_head, nvme_end_sync_rq);
1033
1034         while (!completion_done(&wait)) {
1035                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
1036                 cond_resched();
1037         }
1038 }
1039
1040 /*
1041  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1042  * if the result is positive, it's an NVM Express status code
1043  */
1044 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1045                 union nvme_result *result, void *buffer, unsigned bufflen,
1046                 unsigned timeout, int qid, int at_head,
1047                 blk_mq_req_flags_t flags, bool poll)
1048 {
1049         struct request *req;
1050         int ret;
1051
1052         if (qid == NVME_QID_ANY)
1053                 req = nvme_alloc_request(q, cmd, flags);
1054         else
1055                 req = nvme_alloc_request_qid(q, cmd, flags, qid);
1056         if (IS_ERR(req))
1057                 return PTR_ERR(req);
1058
1059         if (timeout)
1060                 req->timeout = timeout;
1061
1062         if (buffer && bufflen) {
1063                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1064                 if (ret)
1065                         goto out;
1066         }
1067
1068         if (poll)
1069                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
1070         else
1071                 blk_execute_rq(NULL, req, at_head);
1072         if (result)
1073                 *result = nvme_req(req)->result;
1074         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1075                 ret = -EINTR;
1076         else
1077                 ret = nvme_req(req)->status;
1078  out:
1079         blk_mq_free_request(req);
1080         return ret;
1081 }
1082 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1083
1084 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1085                 void *buffer, unsigned bufflen)
1086 {
1087         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1088                         NVME_QID_ANY, 0, 0, false);
1089 }
1090 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1091
1092 static u32 nvme_known_admin_effects(u8 opcode)
1093 {
1094         switch (opcode) {
1095         case nvme_admin_format_nvm:
1096                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1097                         NVME_CMD_EFFECTS_CSE_MASK;
1098         case nvme_admin_sanitize_nvm:
1099                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1100         default:
1101                 break;
1102         }
1103         return 0;
1104 }
1105
1106 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1107 {
1108         u32 effects = 0;
1109
1110         if (ns) {
1111                 if (ns->head->effects)
1112                         effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1113                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1114                         dev_warn_once(ctrl->device,
1115                                 "IO command:%02x has unhandled effects:%08x\n",
1116                                 opcode, effects);
1117                 return 0;
1118         }
1119
1120         if (ctrl->effects)
1121                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1122         effects |= nvme_known_admin_effects(opcode);
1123
1124         return effects;
1125 }
1126 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1127
1128 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1129                                u8 opcode)
1130 {
1131         u32 effects = nvme_command_effects(ctrl, ns, opcode);
1132
1133         /*
1134          * For simplicity, IO to all namespaces is quiesced even if the command
1135          * effects say only one namespace is affected.
1136          */
1137         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1138                 mutex_lock(&ctrl->scan_lock);
1139                 mutex_lock(&ctrl->subsys->lock);
1140                 nvme_mpath_start_freeze(ctrl->subsys);
1141                 nvme_mpath_wait_freeze(ctrl->subsys);
1142                 nvme_start_freeze(ctrl);
1143                 nvme_wait_freeze(ctrl);
1144         }
1145         return effects;
1146 }
1147
1148 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1149 {
1150         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1151                 nvme_unfreeze(ctrl);
1152                 nvme_mpath_unfreeze(ctrl->subsys);
1153                 mutex_unlock(&ctrl->subsys->lock);
1154                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1155                 mutex_unlock(&ctrl->scan_lock);
1156         }
1157         if (effects & NVME_CMD_EFFECTS_CCC)
1158                 nvme_init_ctrl_finish(ctrl);
1159         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1160                 nvme_queue_scan(ctrl);
1161                 flush_work(&ctrl->scan_work);
1162         }
1163 }
1164
1165 void nvme_execute_passthru_rq(struct request *rq)
1166 {
1167         struct nvme_command *cmd = nvme_req(rq)->cmd;
1168         struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1169         struct nvme_ns *ns = rq->q->queuedata;
1170         struct gendisk *disk = ns ? ns->disk : NULL;
1171         u32 effects;
1172
1173         effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1174         blk_execute_rq(disk, rq, 0);
1175         if (effects) /* nothing to be done for zero cmd effects */
1176                 nvme_passthru_end(ctrl, effects);
1177 }
1178 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1179
1180 /*
1181  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1182  *
1183  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1184  *   accounting for transport roundtrip times [..].
1185  */
1186 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1187 {
1188         queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1189 }
1190
1191 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1192 {
1193         struct nvme_ctrl *ctrl = rq->end_io_data;
1194         unsigned long flags;
1195         bool startka = false;
1196
1197         blk_mq_free_request(rq);
1198
1199         if (status) {
1200                 dev_err(ctrl->device,
1201                         "failed nvme_keep_alive_end_io error=%d\n",
1202                                 status);
1203                 return;
1204         }
1205
1206         ctrl->comp_seen = false;
1207         spin_lock_irqsave(&ctrl->lock, flags);
1208         if (ctrl->state == NVME_CTRL_LIVE ||
1209             ctrl->state == NVME_CTRL_CONNECTING)
1210                 startka = true;
1211         spin_unlock_irqrestore(&ctrl->lock, flags);
1212         if (startka)
1213                 nvme_queue_keep_alive_work(ctrl);
1214 }
1215
1216 static void nvme_keep_alive_work(struct work_struct *work)
1217 {
1218         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1219                         struct nvme_ctrl, ka_work);
1220         bool comp_seen = ctrl->comp_seen;
1221         struct request *rq;
1222
1223         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1224                 dev_dbg(ctrl->device,
1225                         "reschedule traffic based keep-alive timer\n");
1226                 ctrl->comp_seen = false;
1227                 nvme_queue_keep_alive_work(ctrl);
1228                 return;
1229         }
1230
1231         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1232                                 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1233         if (IS_ERR(rq)) {
1234                 /* allocation failure, reset the controller */
1235                 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1236                 nvme_reset_ctrl(ctrl);
1237                 return;
1238         }
1239
1240         rq->timeout = ctrl->kato * HZ;
1241         rq->end_io_data = ctrl;
1242         blk_execute_rq_nowait(NULL, rq, 0, nvme_keep_alive_end_io);
1243 }
1244
1245 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1246 {
1247         if (unlikely(ctrl->kato == 0))
1248                 return;
1249
1250         nvme_queue_keep_alive_work(ctrl);
1251 }
1252
1253 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1254 {
1255         if (unlikely(ctrl->kato == 0))
1256                 return;
1257
1258         cancel_delayed_work_sync(&ctrl->ka_work);
1259 }
1260 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1261
1262 /*
1263  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1264  * flag, thus sending any new CNS opcodes has a big chance of not working.
1265  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1266  * (but not for any later version).
1267  */
1268 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1269 {
1270         if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1271                 return ctrl->vs < NVME_VS(1, 2, 0);
1272         return ctrl->vs < NVME_VS(1, 1, 0);
1273 }
1274
1275 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1276 {
1277         struct nvme_command c = { };
1278         int error;
1279
1280         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1281         c.identify.opcode = nvme_admin_identify;
1282         c.identify.cns = NVME_ID_CNS_CTRL;
1283
1284         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1285         if (!*id)
1286                 return -ENOMEM;
1287
1288         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1289                         sizeof(struct nvme_id_ctrl));
1290         if (error)
1291                 kfree(*id);
1292         return error;
1293 }
1294
1295 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1296 {
1297         return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1298 }
1299
1300 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1301                 struct nvme_ns_id_desc *cur, bool *csi_seen)
1302 {
1303         const char *warn_str = "ctrl returned bogus length:";
1304         void *data = cur;
1305
1306         switch (cur->nidt) {
1307         case NVME_NIDT_EUI64:
1308                 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1309                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1310                                  warn_str, cur->nidl);
1311                         return -1;
1312                 }
1313                 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1314                 return NVME_NIDT_EUI64_LEN;
1315         case NVME_NIDT_NGUID:
1316                 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1317                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1318                                  warn_str, cur->nidl);
1319                         return -1;
1320                 }
1321                 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1322                 return NVME_NIDT_NGUID_LEN;
1323         case NVME_NIDT_UUID:
1324                 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1325                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1326                                  warn_str, cur->nidl);
1327                         return -1;
1328                 }
1329                 uuid_copy(&ids->uuid, data + sizeof(*cur));
1330                 return NVME_NIDT_UUID_LEN;
1331         case NVME_NIDT_CSI:
1332                 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1333                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1334                                  warn_str, cur->nidl);
1335                         return -1;
1336                 }
1337                 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1338                 *csi_seen = true;
1339                 return NVME_NIDT_CSI_LEN;
1340         default:
1341                 /* Skip unknown types */
1342                 return cur->nidl;
1343         }
1344 }
1345
1346 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1347                 struct nvme_ns_ids *ids)
1348 {
1349         struct nvme_command c = { };
1350         bool csi_seen = false;
1351         int status, pos, len;
1352         void *data;
1353
1354         if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1355                 return 0;
1356         if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1357                 return 0;
1358
1359         c.identify.opcode = nvme_admin_identify;
1360         c.identify.nsid = cpu_to_le32(nsid);
1361         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1362
1363         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1364         if (!data)
1365                 return -ENOMEM;
1366
1367         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1368                                       NVME_IDENTIFY_DATA_SIZE);
1369         if (status) {
1370                 dev_warn(ctrl->device,
1371                         "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1372                         nsid, status);
1373                 goto free_data;
1374         }
1375
1376         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1377                 struct nvme_ns_id_desc *cur = data + pos;
1378
1379                 if (cur->nidl == 0)
1380                         break;
1381
1382                 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1383                 if (len < 0)
1384                         break;
1385
1386                 len += sizeof(*cur);
1387         }
1388
1389         if (nvme_multi_css(ctrl) && !csi_seen) {
1390                 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1391                          nsid);
1392                 status = -EINVAL;
1393         }
1394
1395 free_data:
1396         kfree(data);
1397         return status;
1398 }
1399
1400 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1401                         struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1402 {
1403         struct nvme_command c = { };
1404         int error;
1405
1406         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1407         c.identify.opcode = nvme_admin_identify;
1408         c.identify.nsid = cpu_to_le32(nsid);
1409         c.identify.cns = NVME_ID_CNS_NS;
1410
1411         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1412         if (!*id)
1413                 return -ENOMEM;
1414
1415         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1416         if (error) {
1417                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1418                 goto out_free_id;
1419         }
1420
1421         error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1422         if ((*id)->ncap == 0) /* namespace not allocated or attached */
1423                 goto out_free_id;
1424
1425         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1426             !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1427                 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1428         if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1429             !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1430                 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1431
1432         return 0;
1433
1434 out_free_id:
1435         kfree(*id);
1436         return error;
1437 }
1438
1439 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1440                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1441 {
1442         union nvme_result res = { 0 };
1443         struct nvme_command c;
1444         int ret;
1445
1446         memset(&c, 0, sizeof(c));
1447         c.features.opcode = op;
1448         c.features.fid = cpu_to_le32(fid);
1449         c.features.dword11 = cpu_to_le32(dword11);
1450
1451         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1452                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1453         if (ret >= 0 && result)
1454                 *result = le32_to_cpu(res.u32);
1455         return ret;
1456 }
1457
1458 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1459                       unsigned int dword11, void *buffer, size_t buflen,
1460                       u32 *result)
1461 {
1462         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1463                              buflen, result);
1464 }
1465 EXPORT_SYMBOL_GPL(nvme_set_features);
1466
1467 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1468                       unsigned int dword11, void *buffer, size_t buflen,
1469                       u32 *result)
1470 {
1471         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1472                              buflen, result);
1473 }
1474 EXPORT_SYMBOL_GPL(nvme_get_features);
1475
1476 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1477 {
1478         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1479         u32 result;
1480         int status, nr_io_queues;
1481
1482         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1483                         &result);
1484         if (status < 0)
1485                 return status;
1486
1487         /*
1488          * Degraded controllers might return an error when setting the queue
1489          * count.  We still want to be able to bring them online and offer
1490          * access to the admin queue, as that might be only way to fix them up.
1491          */
1492         if (status > 0) {
1493                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1494                 *count = 0;
1495         } else {
1496                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1497                 *count = min(*count, nr_io_queues);
1498         }
1499
1500         return 0;
1501 }
1502 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1503
1504 #define NVME_AEN_SUPPORTED \
1505         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1506          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1507
1508 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1509 {
1510         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1511         int status;
1512
1513         if (!supported_aens)
1514                 return;
1515
1516         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1517                         NULL, 0, &result);
1518         if (status)
1519                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1520                          supported_aens);
1521
1522         queue_work(nvme_wq, &ctrl->async_event_work);
1523 }
1524
1525 /*
1526  * Issue ioctl requests on the first available path.  Note that unlike normal
1527  * block layer requests we will not retry failed request on another controller.
1528  */
1529 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1530                 struct nvme_ns_head **head, int *srcu_idx)
1531 {
1532 #ifdef CONFIG_NVME_MULTIPATH
1533         if (disk->fops == &nvme_ns_head_ops) {
1534                 struct nvme_ns *ns;
1535
1536                 *head = disk->private_data;
1537                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1538                 ns = nvme_find_path(*head);
1539                 if (!ns)
1540                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1541                 return ns;
1542         }
1543 #endif
1544         *head = NULL;
1545         *srcu_idx = -1;
1546         return disk->private_data;
1547 }
1548
1549 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1550 {
1551         if (head)
1552                 srcu_read_unlock(&head->srcu, idx);
1553 }
1554
1555 static int nvme_ns_open(struct nvme_ns *ns)
1556 {
1557
1558         /* should never be called due to GENHD_FL_HIDDEN */
1559         if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1560                 goto fail;
1561         if (!nvme_get_ns(ns))
1562                 goto fail;
1563         if (!try_module_get(ns->ctrl->ops->module))
1564                 goto fail_put_ns;
1565
1566         return 0;
1567
1568 fail_put_ns:
1569         nvme_put_ns(ns);
1570 fail:
1571         return -ENXIO;
1572 }
1573
1574 static void nvme_ns_release(struct nvme_ns *ns)
1575 {
1576
1577         module_put(ns->ctrl->ops->module);
1578         nvme_put_ns(ns);
1579 }
1580
1581 static int nvme_open(struct block_device *bdev, fmode_t mode)
1582 {
1583         return nvme_ns_open(bdev->bd_disk->private_data);
1584 }
1585
1586 static void nvme_release(struct gendisk *disk, fmode_t mode)
1587 {
1588         nvme_ns_release(disk->private_data);
1589 }
1590
1591 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1592 {
1593         /* some standard values */
1594         geo->heads = 1 << 6;
1595         geo->sectors = 1 << 5;
1596         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1597         return 0;
1598 }
1599
1600 #ifdef CONFIG_BLK_DEV_INTEGRITY
1601 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1602                                 u32 max_integrity_segments)
1603 {
1604         struct blk_integrity integrity;
1605
1606         memset(&integrity, 0, sizeof(integrity));
1607         switch (pi_type) {
1608         case NVME_NS_DPS_PI_TYPE3:
1609                 integrity.profile = &t10_pi_type3_crc;
1610                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1611                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1612                 break;
1613         case NVME_NS_DPS_PI_TYPE1:
1614         case NVME_NS_DPS_PI_TYPE2:
1615                 integrity.profile = &t10_pi_type1_crc;
1616                 integrity.tag_size = sizeof(u16);
1617                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1618                 break;
1619         default:
1620                 integrity.profile = NULL;
1621                 break;
1622         }
1623         integrity.tuple_size = ms;
1624         blk_integrity_register(disk, &integrity);
1625         blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1626 }
1627 #else
1628 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1629                                 u32 max_integrity_segments)
1630 {
1631 }
1632 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1633
1634 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1635 {
1636         struct nvme_ctrl *ctrl = ns->ctrl;
1637         struct request_queue *queue = disk->queue;
1638         u32 size = queue_logical_block_size(queue);
1639
1640         if (ctrl->max_discard_sectors == 0) {
1641                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1642                 return;
1643         }
1644
1645         if (ctrl->nr_streams && ns->sws && ns->sgs)
1646                 size *= ns->sws * ns->sgs;
1647
1648         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1649                         NVME_DSM_MAX_RANGES);
1650
1651         queue->limits.discard_alignment = 0;
1652         queue->limits.discard_granularity = size;
1653
1654         /* If discard is already enabled, don't reset queue limits */
1655         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1656                 return;
1657
1658         blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1659         blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1660
1661         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1662                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1663 }
1664
1665 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1666 {
1667         return !uuid_is_null(&ids->uuid) ||
1668                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1669                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1670 }
1671
1672 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1673 {
1674         return uuid_equal(&a->uuid, &b->uuid) &&
1675                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1676                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1677                 a->csi == b->csi;
1678 }
1679
1680 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1681                                  u32 *phys_bs, u32 *io_opt)
1682 {
1683         struct streams_directive_params s;
1684         int ret;
1685
1686         if (!ctrl->nr_streams)
1687                 return 0;
1688
1689         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1690         if (ret)
1691                 return ret;
1692
1693         ns->sws = le32_to_cpu(s.sws);
1694         ns->sgs = le16_to_cpu(s.sgs);
1695
1696         if (ns->sws) {
1697                 *phys_bs = ns->sws * (1 << ns->lba_shift);
1698                 if (ns->sgs)
1699                         *io_opt = *phys_bs * ns->sgs;
1700         }
1701
1702         return 0;
1703 }
1704
1705 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1706 {
1707         struct nvme_ctrl *ctrl = ns->ctrl;
1708
1709         /*
1710          * The PI implementation requires the metadata size to be equal to the
1711          * t10 pi tuple size.
1712          */
1713         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1714         if (ns->ms == sizeof(struct t10_pi_tuple))
1715                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1716         else
1717                 ns->pi_type = 0;
1718
1719         ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1720         if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1721                 return 0;
1722         if (ctrl->ops->flags & NVME_F_FABRICS) {
1723                 /*
1724                  * The NVMe over Fabrics specification only supports metadata as
1725                  * part of the extended data LBA.  We rely on HCA/HBA support to
1726                  * remap the separate metadata buffer from the block layer.
1727                  */
1728                 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1729                         return -EINVAL;
1730                 if (ctrl->max_integrity_segments)
1731                         ns->features |=
1732                                 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1733         } else {
1734                 /*
1735                  * For PCIe controllers, we can't easily remap the separate
1736                  * metadata buffer from the block layer and thus require a
1737                  * separate metadata buffer for block layer metadata/PI support.
1738                  * We allow extended LBAs for the passthrough interface, though.
1739                  */
1740                 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1741                         ns->features |= NVME_NS_EXT_LBAS;
1742                 else
1743                         ns->features |= NVME_NS_METADATA_SUPPORTED;
1744         }
1745
1746         return 0;
1747 }
1748
1749 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1750                 struct request_queue *q)
1751 {
1752         bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1753
1754         if (ctrl->max_hw_sectors) {
1755                 u32 max_segments =
1756                         (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1757
1758                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1759                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1760                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1761         }
1762         blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1763         blk_queue_dma_alignment(q, 7);
1764         blk_queue_write_cache(q, vwc, vwc);
1765 }
1766
1767 static void nvme_update_disk_info(struct gendisk *disk,
1768                 struct nvme_ns *ns, struct nvme_id_ns *id)
1769 {
1770         sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1771         unsigned short bs = 1 << ns->lba_shift;
1772         u32 atomic_bs, phys_bs, io_opt = 0;
1773
1774         /*
1775          * The block layer can't support LBA sizes larger than the page size
1776          * yet, so catch this early and don't allow block I/O.
1777          */
1778         if (ns->lba_shift > PAGE_SHIFT) {
1779                 capacity = 0;
1780                 bs = (1 << 9);
1781         }
1782
1783         blk_integrity_unregister(disk);
1784
1785         atomic_bs = phys_bs = bs;
1786         nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1787         if (id->nabo == 0) {
1788                 /*
1789                  * Bit 1 indicates whether NAWUPF is defined for this namespace
1790                  * and whether it should be used instead of AWUPF. If NAWUPF ==
1791                  * 0 then AWUPF must be used instead.
1792                  */
1793                 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1794                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1795                 else
1796                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1797         }
1798
1799         if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1800                 /* NPWG = Namespace Preferred Write Granularity */
1801                 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1802                 /* NOWS = Namespace Optimal Write Size */
1803                 io_opt = bs * (1 + le16_to_cpu(id->nows));
1804         }
1805
1806         blk_queue_logical_block_size(disk->queue, bs);
1807         /*
1808          * Linux filesystems assume writing a single physical block is
1809          * an atomic operation. Hence limit the physical block size to the
1810          * value of the Atomic Write Unit Power Fail parameter.
1811          */
1812         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1813         blk_queue_io_min(disk->queue, phys_bs);
1814         blk_queue_io_opt(disk->queue, io_opt);
1815
1816         /*
1817          * Register a metadata profile for PI, or the plain non-integrity NVMe
1818          * metadata masquerading as Type 0 if supported, otherwise reject block
1819          * I/O to namespaces with metadata except when the namespace supports
1820          * PI, as it can strip/insert in that case.
1821          */
1822         if (ns->ms) {
1823                 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1824                     (ns->features & NVME_NS_METADATA_SUPPORTED))
1825                         nvme_init_integrity(disk, ns->ms, ns->pi_type,
1826                                             ns->ctrl->max_integrity_segments);
1827                 else if (!nvme_ns_has_pi(ns))
1828                         capacity = 0;
1829         }
1830
1831         set_capacity_and_notify(disk, capacity);
1832
1833         nvme_config_discard(disk, ns);
1834         blk_queue_max_write_zeroes_sectors(disk->queue,
1835                                            ns->ctrl->max_zeroes_sectors);
1836
1837         set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1838                 test_bit(NVME_NS_FORCE_RO, &ns->flags));
1839 }
1840
1841 static inline bool nvme_first_scan(struct gendisk *disk)
1842 {
1843         /* nvme_alloc_ns() scans the disk prior to adding it */
1844         return !(disk->flags & GENHD_FL_UP);
1845 }
1846
1847 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1848 {
1849         struct nvme_ctrl *ctrl = ns->ctrl;
1850         u32 iob;
1851
1852         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1853             is_power_of_2(ctrl->max_hw_sectors))
1854                 iob = ctrl->max_hw_sectors;
1855         else
1856                 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1857
1858         if (!iob)
1859                 return;
1860
1861         if (!is_power_of_2(iob)) {
1862                 if (nvme_first_scan(ns->disk))
1863                         pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1864                                 ns->disk->disk_name, iob);
1865                 return;
1866         }
1867
1868         if (blk_queue_is_zoned(ns->disk->queue)) {
1869                 if (nvme_first_scan(ns->disk))
1870                         pr_warn("%s: ignoring zoned namespace IO boundary\n",
1871                                 ns->disk->disk_name);
1872                 return;
1873         }
1874
1875         blk_queue_chunk_sectors(ns->queue, iob);
1876 }
1877
1878 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1879 {
1880         unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1881         int ret;
1882
1883         blk_mq_freeze_queue(ns->disk->queue);
1884         ns->lba_shift = id->lbaf[lbaf].ds;
1885         nvme_set_queue_limits(ns->ctrl, ns->queue);
1886
1887         ret = nvme_configure_metadata(ns, id);
1888         if (ret)
1889                 goto out_unfreeze;
1890         nvme_set_chunk_sectors(ns, id);
1891         nvme_update_disk_info(ns->disk, ns, id);
1892
1893         if (ns->head->ids.csi == NVME_CSI_ZNS) {
1894                 ret = nvme_update_zone_info(ns, lbaf);
1895                 if (ret)
1896                         goto out_unfreeze;
1897         }
1898
1899         blk_mq_unfreeze_queue(ns->disk->queue);
1900
1901         if (blk_queue_is_zoned(ns->queue)) {
1902                 ret = nvme_revalidate_zones(ns);
1903                 if (ret && !nvme_first_scan(ns->disk))
1904                         goto out;
1905         }
1906
1907         if (nvme_ns_head_multipath(ns->head)) {
1908                 blk_mq_freeze_queue(ns->head->disk->queue);
1909                 nvme_update_disk_info(ns->head->disk, ns, id);
1910                 blk_stack_limits(&ns->head->disk->queue->limits,
1911                                  &ns->queue->limits, 0);
1912                 blk_queue_update_readahead(ns->head->disk->queue);
1913                 blk_mq_unfreeze_queue(ns->head->disk->queue);
1914         }
1915         return 0;
1916
1917 out_unfreeze:
1918         blk_mq_unfreeze_queue(ns->disk->queue);
1919 out:
1920         /*
1921          * If probing fails due an unsupported feature, hide the block device,
1922          * but still allow other access.
1923          */
1924         if (ret == -ENODEV) {
1925                 ns->disk->flags |= GENHD_FL_HIDDEN;
1926                 ret = 0;
1927         }
1928         return ret;
1929 }
1930
1931 static char nvme_pr_type(enum pr_type type)
1932 {
1933         switch (type) {
1934         case PR_WRITE_EXCLUSIVE:
1935                 return 1;
1936         case PR_EXCLUSIVE_ACCESS:
1937                 return 2;
1938         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1939                 return 3;
1940         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1941                 return 4;
1942         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1943                 return 5;
1944         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1945                 return 6;
1946         default:
1947                 return 0;
1948         }
1949 };
1950
1951 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1952                                 u64 key, u64 sa_key, u8 op)
1953 {
1954         struct nvme_ns_head *head = NULL;
1955         struct nvme_ns *ns;
1956         struct nvme_command c;
1957         int srcu_idx, ret;
1958         u8 data[16] = { 0, };
1959
1960         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1961         if (unlikely(!ns))
1962                 return -EWOULDBLOCK;
1963
1964         put_unaligned_le64(key, &data[0]);
1965         put_unaligned_le64(sa_key, &data[8]);
1966
1967         memset(&c, 0, sizeof(c));
1968         c.common.opcode = op;
1969         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1970         c.common.cdw10 = cpu_to_le32(cdw10);
1971
1972         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1973         nvme_put_ns_from_disk(head, srcu_idx);
1974         return ret;
1975 }
1976
1977 static int nvme_pr_register(struct block_device *bdev, u64 old,
1978                 u64 new, unsigned flags)
1979 {
1980         u32 cdw10;
1981
1982         if (flags & ~PR_FL_IGNORE_KEY)
1983                 return -EOPNOTSUPP;
1984
1985         cdw10 = old ? 2 : 0;
1986         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1987         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1988         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1989 }
1990
1991 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1992                 enum pr_type type, unsigned flags)
1993 {
1994         u32 cdw10;
1995
1996         if (flags & ~PR_FL_IGNORE_KEY)
1997                 return -EOPNOTSUPP;
1998
1999         cdw10 = nvme_pr_type(type) << 8;
2000         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2001         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2002 }
2003
2004 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2005                 enum pr_type type, bool abort)
2006 {
2007         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2008
2009         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2010 }
2011
2012 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2013 {
2014         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2015
2016         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2017 }
2018
2019 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2020 {
2021         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2022
2023         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2024 }
2025
2026 const struct pr_ops nvme_pr_ops = {
2027         .pr_register    = nvme_pr_register,
2028         .pr_reserve     = nvme_pr_reserve,
2029         .pr_release     = nvme_pr_release,
2030         .pr_preempt     = nvme_pr_preempt,
2031         .pr_clear       = nvme_pr_clear,
2032 };
2033
2034 #ifdef CONFIG_BLK_SED_OPAL
2035 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2036                 bool send)
2037 {
2038         struct nvme_ctrl *ctrl = data;
2039         struct nvme_command cmd;
2040
2041         memset(&cmd, 0, sizeof(cmd));
2042         if (send)
2043                 cmd.common.opcode = nvme_admin_security_send;
2044         else
2045                 cmd.common.opcode = nvme_admin_security_recv;
2046         cmd.common.nsid = 0;
2047         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2048         cmd.common.cdw11 = cpu_to_le32(len);
2049
2050         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2051                         NVME_QID_ANY, 1, 0, false);
2052 }
2053 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2054 #endif /* CONFIG_BLK_SED_OPAL */
2055
2056 static const struct block_device_operations nvme_bdev_ops = {
2057         .owner          = THIS_MODULE,
2058         .ioctl          = nvme_ioctl,
2059         .open           = nvme_open,
2060         .release        = nvme_release,
2061         .getgeo         = nvme_getgeo,
2062         .report_zones   = nvme_report_zones,
2063         .pr_ops         = &nvme_pr_ops,
2064 };
2065
2066 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2067 {
2068         unsigned long timeout =
2069                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2070         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2071         int ret;
2072
2073         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2074                 if (csts == ~0)
2075                         return -ENODEV;
2076                 if ((csts & NVME_CSTS_RDY) == bit)
2077                         break;
2078
2079                 usleep_range(1000, 2000);
2080                 if (fatal_signal_pending(current))
2081                         return -EINTR;
2082                 if (time_after(jiffies, timeout)) {
2083                         dev_err(ctrl->device,
2084                                 "Device not ready; aborting %s, CSTS=0x%x\n",
2085                                 enabled ? "initialisation" : "reset", csts);
2086                         return -ENODEV;
2087                 }
2088         }
2089
2090         return ret;
2091 }
2092
2093 /*
2094  * If the device has been passed off to us in an enabled state, just clear
2095  * the enabled bit.  The spec says we should set the 'shutdown notification
2096  * bits', but doing so may cause the device to complete commands to the
2097  * admin queue ... and we don't know what memory that might be pointing at!
2098  */
2099 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2100 {
2101         int ret;
2102
2103         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2104         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2105
2106         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2107         if (ret)
2108                 return ret;
2109
2110         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2111                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2112
2113         return nvme_wait_ready(ctrl, ctrl->cap, false);
2114 }
2115 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2116
2117 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2118 {
2119         unsigned dev_page_min;
2120         int ret;
2121
2122         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2123         if (ret) {
2124                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2125                 return ret;
2126         }
2127         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2128
2129         if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2130                 dev_err(ctrl->device,
2131                         "Minimum device page size %u too large for host (%u)\n",
2132                         1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2133                 return -ENODEV;
2134         }
2135
2136         if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2137                 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2138         else
2139                 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2140         ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2141         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2142         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2143         ctrl->ctrl_config |= NVME_CC_ENABLE;
2144
2145         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2146         if (ret)
2147                 return ret;
2148         return nvme_wait_ready(ctrl, ctrl->cap, true);
2149 }
2150 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2151
2152 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2153 {
2154         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2155         u32 csts;
2156         int ret;
2157
2158         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2159         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2160
2161         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2162         if (ret)
2163                 return ret;
2164
2165         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2166                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2167                         break;
2168
2169                 msleep(100);
2170                 if (fatal_signal_pending(current))
2171                         return -EINTR;
2172                 if (time_after(jiffies, timeout)) {
2173                         dev_err(ctrl->device,
2174                                 "Device shutdown incomplete; abort shutdown\n");
2175                         return -ENODEV;
2176                 }
2177         }
2178
2179         return ret;
2180 }
2181 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2182
2183 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2184 {
2185         __le64 ts;
2186         int ret;
2187
2188         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2189                 return 0;
2190
2191         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2192         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2193                         NULL);
2194         if (ret)
2195                 dev_warn_once(ctrl->device,
2196                         "could not set timestamp (%d)\n", ret);
2197         return ret;
2198 }
2199
2200 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2201 {
2202         struct nvme_feat_host_behavior *host;
2203         int ret;
2204
2205         /* Don't bother enabling the feature if retry delay is not reported */
2206         if (!ctrl->crdt[0])
2207                 return 0;
2208
2209         host = kzalloc(sizeof(*host), GFP_KERNEL);
2210         if (!host)
2211                 return 0;
2212
2213         host->acre = NVME_ENABLE_ACRE;
2214         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2215                                 host, sizeof(*host), NULL);
2216         kfree(host);
2217         return ret;
2218 }
2219
2220 /*
2221  * APST (Autonomous Power State Transition) lets us program a table of power
2222  * state transitions that the controller will perform automatically.
2223  * We configure it with a simple heuristic: we are willing to spend at most 2%
2224  * of the time transitioning between power states.  Therefore, when running in
2225  * any given state, we will enter the next lower-power non-operational state
2226  * after waiting 50 * (enlat + exlat) microseconds, as long as that state's exit
2227  * latency is under the requested maximum latency.
2228  *
2229  * We will not autonomously enter any non-operational state for which the total
2230  * latency exceeds ps_max_latency_us.
2231  *
2232  * Users can set ps_max_latency_us to zero to turn off APST.
2233  */
2234 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2235 {
2236         struct nvme_feat_auto_pst *table;
2237         unsigned apste = 0;
2238         u64 max_lat_us = 0;
2239         __le64 target = 0;
2240         int max_ps = -1;
2241         int state;
2242         int ret;
2243
2244         /*
2245          * If APST isn't supported or if we haven't been initialized yet,
2246          * then don't do anything.
2247          */
2248         if (!ctrl->apsta)
2249                 return 0;
2250
2251         if (ctrl->npss > 31) {
2252                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2253                 return 0;
2254         }
2255
2256         table = kzalloc(sizeof(*table), GFP_KERNEL);
2257         if (!table)
2258                 return 0;
2259
2260         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2261                 /* Turn off APST. */
2262                 dev_dbg(ctrl->device, "APST disabled\n");
2263                 goto done;
2264         }
2265
2266         /*
2267          * Walk through all states from lowest- to highest-power.
2268          * According to the spec, lower-numbered states use more power.  NPSS,
2269          * despite the name, is the index of the lowest-power state, not the
2270          * number of states.
2271          */
2272         for (state = (int)ctrl->npss; state >= 0; state--) {
2273                 u64 total_latency_us, exit_latency_us, transition_ms;
2274
2275                 if (target)
2276                         table->entries[state] = target;
2277
2278                 /*
2279                  * Don't allow transitions to the deepest state if it's quirked
2280                  * off.
2281                  */
2282                 if (state == ctrl->npss &&
2283                     (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2284                         continue;
2285
2286                 /*
2287                  * Is this state a useful non-operational state for higher-power
2288                  * states to autonomously transition to?
2289                  */
2290                 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2291                         continue;
2292
2293                 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2294                 if (exit_latency_us > ctrl->ps_max_latency_us)
2295                         continue;
2296
2297                 total_latency_us = exit_latency_us +
2298                         le32_to_cpu(ctrl->psd[state].entry_lat);
2299
2300                 /*
2301                  * This state is good.  Use it as the APST idle target for
2302                  * higher power states.
2303                  */
2304                 transition_ms = total_latency_us + 19;
2305                 do_div(transition_ms, 20);
2306                 if (transition_ms > (1 << 24) - 1)
2307                         transition_ms = (1 << 24) - 1;
2308
2309                 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2310                 if (max_ps == -1)
2311                         max_ps = state;
2312                 if (total_latency_us > max_lat_us)
2313                         max_lat_us = total_latency_us;
2314         }
2315
2316         if (max_ps == -1)
2317                 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2318         else
2319                 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2320                         max_ps, max_lat_us, (int)sizeof(*table), table);
2321         apste = 1;
2322
2323 done:
2324         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2325                                 table, sizeof(*table), NULL);
2326         if (ret)
2327                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2328         kfree(table);
2329         return ret;
2330 }
2331
2332 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2333 {
2334         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2335         u64 latency;
2336
2337         switch (val) {
2338         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2339         case PM_QOS_LATENCY_ANY:
2340                 latency = U64_MAX;
2341                 break;
2342
2343         default:
2344                 latency = val;
2345         }
2346
2347         if (ctrl->ps_max_latency_us != latency) {
2348                 ctrl->ps_max_latency_us = latency;
2349                 if (ctrl->state == NVME_CTRL_LIVE)
2350                         nvme_configure_apst(ctrl);
2351         }
2352 }
2353
2354 struct nvme_core_quirk_entry {
2355         /*
2356          * NVMe model and firmware strings are padded with spaces.  For
2357          * simplicity, strings in the quirk table are padded with NULLs
2358          * instead.
2359          */
2360         u16 vid;
2361         const char *mn;
2362         const char *fr;
2363         unsigned long quirks;
2364 };
2365
2366 static const struct nvme_core_quirk_entry core_quirks[] = {
2367         {
2368                 /*
2369                  * This Toshiba device seems to die using any APST states.  See:
2370                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2371                  */
2372                 .vid = 0x1179,
2373                 .mn = "THNSF5256GPUK TOSHIBA",
2374                 .quirks = NVME_QUIRK_NO_APST,
2375         },
2376         {
2377                 /*
2378                  * This LiteON CL1-3D*-Q11 firmware version has a race
2379                  * condition associated with actions related to suspend to idle
2380                  * LiteON has resolved the problem in future firmware
2381                  */
2382                 .vid = 0x14a4,
2383                 .fr = "22301111",
2384                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2385         }
2386 };
2387
2388 /* match is null-terminated but idstr is space-padded. */
2389 static bool string_matches(const char *idstr, const char *match, size_t len)
2390 {
2391         size_t matchlen;
2392
2393         if (!match)
2394                 return true;
2395
2396         matchlen = strlen(match);
2397         WARN_ON_ONCE(matchlen > len);
2398
2399         if (memcmp(idstr, match, matchlen))
2400                 return false;
2401
2402         for (; matchlen < len; matchlen++)
2403                 if (idstr[matchlen] != ' ')
2404                         return false;
2405
2406         return true;
2407 }
2408
2409 static bool quirk_matches(const struct nvme_id_ctrl *id,
2410                           const struct nvme_core_quirk_entry *q)
2411 {
2412         return q->vid == le16_to_cpu(id->vid) &&
2413                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2414                 string_matches(id->fr, q->fr, sizeof(id->fr));
2415 }
2416
2417 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2418                 struct nvme_id_ctrl *id)
2419 {
2420         size_t nqnlen;
2421         int off;
2422
2423         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2424                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2425                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2426                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2427                         return;
2428                 }
2429
2430                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2431                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2432         }
2433
2434         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2435         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2436                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2437                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2438         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2439         off += sizeof(id->sn);
2440         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2441         off += sizeof(id->mn);
2442         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2443 }
2444
2445 static void nvme_release_subsystem(struct device *dev)
2446 {
2447         struct nvme_subsystem *subsys =
2448                 container_of(dev, struct nvme_subsystem, dev);
2449
2450         if (subsys->instance >= 0)
2451                 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2452         kfree(subsys);
2453 }
2454
2455 static void nvme_destroy_subsystem(struct kref *ref)
2456 {
2457         struct nvme_subsystem *subsys =
2458                         container_of(ref, struct nvme_subsystem, ref);
2459
2460         mutex_lock(&nvme_subsystems_lock);
2461         list_del(&subsys->entry);
2462         mutex_unlock(&nvme_subsystems_lock);
2463
2464         ida_destroy(&subsys->ns_ida);
2465         device_del(&subsys->dev);
2466         put_device(&subsys->dev);
2467 }
2468
2469 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2470 {
2471         kref_put(&subsys->ref, nvme_destroy_subsystem);
2472 }
2473
2474 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2475 {
2476         struct nvme_subsystem *subsys;
2477
2478         lockdep_assert_held(&nvme_subsystems_lock);
2479
2480         /*
2481          * Fail matches for discovery subsystems. This results
2482          * in each discovery controller bound to a unique subsystem.
2483          * This avoids issues with validating controller values
2484          * that can only be true when there is a single unique subsystem.
2485          * There may be multiple and completely independent entities
2486          * that provide discovery controllers.
2487          */
2488         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2489                 return NULL;
2490
2491         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2492                 if (strcmp(subsys->subnqn, subsysnqn))
2493                         continue;
2494                 if (!kref_get_unless_zero(&subsys->ref))
2495                         continue;
2496                 return subsys;
2497         }
2498
2499         return NULL;
2500 }
2501
2502 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2503         struct device_attribute subsys_attr_##_name = \
2504                 __ATTR(_name, _mode, _show, NULL)
2505
2506 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2507                                     struct device_attribute *attr,
2508                                     char *buf)
2509 {
2510         struct nvme_subsystem *subsys =
2511                 container_of(dev, struct nvme_subsystem, dev);
2512
2513         return sysfs_emit(buf, "%s\n", subsys->subnqn);
2514 }
2515 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2516
2517 #define nvme_subsys_show_str_function(field)                            \
2518 static ssize_t subsys_##field##_show(struct device *dev,                \
2519                             struct device_attribute *attr, char *buf)   \
2520 {                                                                       \
2521         struct nvme_subsystem *subsys =                                 \
2522                 container_of(dev, struct nvme_subsystem, dev);          \
2523         return sysfs_emit(buf, "%.*s\n",                                \
2524                            (int)sizeof(subsys->field), subsys->field);  \
2525 }                                                                       \
2526 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2527
2528 nvme_subsys_show_str_function(model);
2529 nvme_subsys_show_str_function(serial);
2530 nvme_subsys_show_str_function(firmware_rev);
2531
2532 static struct attribute *nvme_subsys_attrs[] = {
2533         &subsys_attr_model.attr,
2534         &subsys_attr_serial.attr,
2535         &subsys_attr_firmware_rev.attr,
2536         &subsys_attr_subsysnqn.attr,
2537 #ifdef CONFIG_NVME_MULTIPATH
2538         &subsys_attr_iopolicy.attr,
2539 #endif
2540         NULL,
2541 };
2542
2543 static const struct attribute_group nvme_subsys_attrs_group = {
2544         .attrs = nvme_subsys_attrs,
2545 };
2546
2547 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2548         &nvme_subsys_attrs_group,
2549         NULL,
2550 };
2551
2552 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2553 {
2554         return ctrl->opts && ctrl->opts->discovery_nqn;
2555 }
2556
2557 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2558                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2559 {
2560         struct nvme_ctrl *tmp;
2561
2562         lockdep_assert_held(&nvme_subsystems_lock);
2563
2564         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2565                 if (nvme_state_terminal(tmp))
2566                         continue;
2567
2568                 if (tmp->cntlid == ctrl->cntlid) {
2569                         dev_err(ctrl->device,
2570                                 "Duplicate cntlid %u with %s, rejecting\n",
2571                                 ctrl->cntlid, dev_name(tmp->device));
2572                         return false;
2573                 }
2574
2575                 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2576                     nvme_discovery_ctrl(ctrl))
2577                         continue;
2578
2579                 dev_err(ctrl->device,
2580                         "Subsystem does not support multiple controllers\n");
2581                 return false;
2582         }
2583
2584         return true;
2585 }
2586
2587 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2588 {
2589         struct nvme_subsystem *subsys, *found;
2590         int ret;
2591
2592         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2593         if (!subsys)
2594                 return -ENOMEM;
2595
2596         subsys->instance = -1;
2597         mutex_init(&subsys->lock);
2598         kref_init(&subsys->ref);
2599         INIT_LIST_HEAD(&subsys->ctrls);
2600         INIT_LIST_HEAD(&subsys->nsheads);
2601         nvme_init_subnqn(subsys, ctrl, id);
2602         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2603         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2604         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2605         subsys->vendor_id = le16_to_cpu(id->vid);
2606         subsys->cmic = id->cmic;
2607         subsys->awupf = le16_to_cpu(id->awupf);
2608 #ifdef CONFIG_NVME_MULTIPATH
2609         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2610 #endif
2611
2612         subsys->dev.class = nvme_subsys_class;
2613         subsys->dev.release = nvme_release_subsystem;
2614         subsys->dev.groups = nvme_subsys_attrs_groups;
2615         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2616         device_initialize(&subsys->dev);
2617
2618         mutex_lock(&nvme_subsystems_lock);
2619         found = __nvme_find_get_subsystem(subsys->subnqn);
2620         if (found) {
2621                 put_device(&subsys->dev);
2622                 subsys = found;
2623
2624                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2625                         ret = -EINVAL;
2626                         goto out_put_subsystem;
2627                 }
2628         } else {
2629                 ret = device_add(&subsys->dev);
2630                 if (ret) {
2631                         dev_err(ctrl->device,
2632                                 "failed to register subsystem device.\n");
2633                         put_device(&subsys->dev);
2634                         goto out_unlock;
2635                 }
2636                 ida_init(&subsys->ns_ida);
2637                 list_add_tail(&subsys->entry, &nvme_subsystems);
2638         }
2639
2640         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2641                                 dev_name(ctrl->device));
2642         if (ret) {
2643                 dev_err(ctrl->device,
2644                         "failed to create sysfs link from subsystem.\n");
2645                 goto out_put_subsystem;
2646         }
2647
2648         if (!found)
2649                 subsys->instance = ctrl->instance;
2650         ctrl->subsys = subsys;
2651         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2652         mutex_unlock(&nvme_subsystems_lock);
2653         return 0;
2654
2655 out_put_subsystem:
2656         nvme_put_subsystem(subsys);
2657 out_unlock:
2658         mutex_unlock(&nvme_subsystems_lock);
2659         return ret;
2660 }
2661
2662 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2663                 void *log, size_t size, u64 offset)
2664 {
2665         struct nvme_command c = { };
2666         u32 dwlen = nvme_bytes_to_numd(size);
2667
2668         c.get_log_page.opcode = nvme_admin_get_log_page;
2669         c.get_log_page.nsid = cpu_to_le32(nsid);
2670         c.get_log_page.lid = log_page;
2671         c.get_log_page.lsp = lsp;
2672         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2673         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2674         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2675         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2676         c.get_log_page.csi = csi;
2677
2678         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2679 }
2680
2681 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2682                                 struct nvme_effects_log **log)
2683 {
2684         struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2685         int ret;
2686
2687         if (cel)
2688                 goto out;
2689
2690         cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2691         if (!cel)
2692                 return -ENOMEM;
2693
2694         ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2695                         cel, sizeof(*cel), 0);
2696         if (ret) {
2697                 kfree(cel);
2698                 return ret;
2699         }
2700
2701         xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2702 out:
2703         *log = cel;
2704         return 0;
2705 }
2706
2707 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2708 {
2709         u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2710
2711         if (check_shl_overflow(1U, units + page_shift - 9, &val))
2712                 return UINT_MAX;
2713         return val;
2714 }
2715
2716 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2717 {
2718         struct nvme_command c = { };
2719         struct nvme_id_ctrl_nvm *id;
2720         int ret;
2721
2722         if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2723                 ctrl->max_discard_sectors = UINT_MAX;
2724                 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2725         } else {
2726                 ctrl->max_discard_sectors = 0;
2727                 ctrl->max_discard_segments = 0;
2728         }
2729
2730         /*
2731          * Even though NVMe spec explicitly states that MDTS is not applicable
2732          * to the write-zeroes, we are cautious and limit the size to the
2733          * controllers max_hw_sectors value, which is based on the MDTS field
2734          * and possibly other limiting factors.
2735          */
2736         if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2737             !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2738                 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2739         else
2740                 ctrl->max_zeroes_sectors = 0;
2741
2742         if (nvme_ctrl_limited_cns(ctrl))
2743                 return 0;
2744
2745         id = kzalloc(sizeof(*id), GFP_KERNEL);
2746         if (!id)
2747                 return 0;
2748
2749         c.identify.opcode = nvme_admin_identify;
2750         c.identify.cns = NVME_ID_CNS_CS_CTRL;
2751         c.identify.csi = NVME_CSI_NVM;
2752
2753         ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2754         if (ret)
2755                 goto free_data;
2756
2757         if (id->dmrl)
2758                 ctrl->max_discard_segments = id->dmrl;
2759         if (id->dmrsl)
2760                 ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
2761         if (id->wzsl)
2762                 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2763
2764 free_data:
2765         kfree(id);
2766         return ret;
2767 }
2768
2769 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2770 {
2771         struct nvme_id_ctrl *id;
2772         u32 max_hw_sectors;
2773         bool prev_apst_enabled;
2774         int ret;
2775
2776         ret = nvme_identify_ctrl(ctrl, &id);
2777         if (ret) {
2778                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2779                 return -EIO;
2780         }
2781
2782         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2783                 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2784                 if (ret < 0)
2785                         goto out_free;
2786         }
2787
2788         if (!(ctrl->ops->flags & NVME_F_FABRICS))
2789                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2790
2791         if (!ctrl->identified) {
2792                 unsigned int i;
2793
2794                 ret = nvme_init_subsystem(ctrl, id);
2795                 if (ret)
2796                         goto out_free;
2797
2798                 /*
2799                  * Check for quirks.  Quirk can depend on firmware version,
2800                  * so, in principle, the set of quirks present can change
2801                  * across a reset.  As a possible future enhancement, we
2802                  * could re-scan for quirks every time we reinitialize
2803                  * the device, but we'd have to make sure that the driver
2804                  * behaves intelligently if the quirks change.
2805                  */
2806                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2807                         if (quirk_matches(id, &core_quirks[i]))
2808                                 ctrl->quirks |= core_quirks[i].quirks;
2809                 }
2810         }
2811
2812         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2813                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2814                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2815         }
2816
2817         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2818         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2819         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2820
2821         ctrl->oacs = le16_to_cpu(id->oacs);
2822         ctrl->oncs = le16_to_cpu(id->oncs);
2823         ctrl->mtfa = le16_to_cpu(id->mtfa);
2824         ctrl->oaes = le32_to_cpu(id->oaes);
2825         ctrl->wctemp = le16_to_cpu(id->wctemp);
2826         ctrl->cctemp = le16_to_cpu(id->cctemp);
2827
2828         atomic_set(&ctrl->abort_limit, id->acl + 1);
2829         ctrl->vwc = id->vwc;
2830         if (id->mdts)
2831                 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
2832         else
2833                 max_hw_sectors = UINT_MAX;
2834         ctrl->max_hw_sectors =
2835                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2836
2837         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2838         ctrl->sgls = le32_to_cpu(id->sgls);
2839         ctrl->kas = le16_to_cpu(id->kas);
2840         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2841         ctrl->ctratt = le32_to_cpu(id->ctratt);
2842
2843         if (id->rtd3e) {
2844                 /* us -> s */
2845                 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
2846
2847                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2848                                                  shutdown_timeout, 60);
2849
2850                 if (ctrl->shutdown_timeout != shutdown_timeout)
2851                         dev_info(ctrl->device,
2852                                  "Shutdown timeout set to %u seconds\n",
2853                                  ctrl->shutdown_timeout);
2854         } else
2855                 ctrl->shutdown_timeout = shutdown_timeout;
2856
2857         ctrl->npss = id->npss;
2858         ctrl->apsta = id->apsta;
2859         prev_apst_enabled = ctrl->apst_enabled;
2860         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2861                 if (force_apst && id->apsta) {
2862                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2863                         ctrl->apst_enabled = true;
2864                 } else {
2865                         ctrl->apst_enabled = false;
2866                 }
2867         } else {
2868                 ctrl->apst_enabled = id->apsta;
2869         }
2870         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2871
2872         if (ctrl->ops->flags & NVME_F_FABRICS) {
2873                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2874                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2875                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2876                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2877
2878                 /*
2879                  * In fabrics we need to verify the cntlid matches the
2880                  * admin connect
2881                  */
2882                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2883                         dev_err(ctrl->device,
2884                                 "Mismatching cntlid: Connect %u vs Identify "
2885                                 "%u, rejecting\n",
2886                                 ctrl->cntlid, le16_to_cpu(id->cntlid));
2887                         ret = -EINVAL;
2888                         goto out_free;
2889                 }
2890
2891                 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
2892                         dev_err(ctrl->device,
2893                                 "keep-alive support is mandatory for fabrics\n");
2894                         ret = -EINVAL;
2895                         goto out_free;
2896                 }
2897         } else {
2898                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2899                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2900                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2901                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2902         }
2903
2904         ret = nvme_mpath_init_identify(ctrl, id);
2905         if (ret < 0)
2906                 goto out_free;
2907
2908         if (ctrl->apst_enabled && !prev_apst_enabled)
2909                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2910         else if (!ctrl->apst_enabled && prev_apst_enabled)
2911                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2912
2913 out_free:
2914         kfree(id);
2915         return ret;
2916 }
2917
2918 /*
2919  * Initialize the cached copies of the Identify data and various controller
2920  * register in our nvme_ctrl structure.  This should be called as soon as
2921  * the admin queue is fully up and running.
2922  */
2923 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
2924 {
2925         int ret;
2926
2927         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2928         if (ret) {
2929                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2930                 return ret;
2931         }
2932
2933         ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2934
2935         if (ctrl->vs >= NVME_VS(1, 1, 0))
2936                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2937
2938         ret = nvme_init_identify(ctrl);
2939         if (ret)
2940                 return ret;
2941
2942         ret = nvme_init_non_mdts_limits(ctrl);
2943         if (ret < 0)
2944                 return ret;
2945
2946         ret = nvme_configure_apst(ctrl);
2947         if (ret < 0)
2948                 return ret;
2949
2950         ret = nvme_configure_timestamp(ctrl);
2951         if (ret < 0)
2952                 return ret;
2953
2954         ret = nvme_configure_directives(ctrl);
2955         if (ret < 0)
2956                 return ret;
2957
2958         ret = nvme_configure_acre(ctrl);
2959         if (ret < 0)
2960                 return ret;
2961
2962         if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
2963                 ret = nvme_hwmon_init(ctrl);
2964                 if (ret < 0)
2965                         return ret;
2966         }
2967
2968         ctrl->identified = true;
2969
2970         return 0;
2971 }
2972 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
2973
2974 static int nvme_dev_open(struct inode *inode, struct file *file)
2975 {
2976         struct nvme_ctrl *ctrl =
2977                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2978
2979         switch (ctrl->state) {
2980         case NVME_CTRL_LIVE:
2981                 break;
2982         default:
2983                 return -EWOULDBLOCK;
2984         }
2985
2986         nvme_get_ctrl(ctrl);
2987         if (!try_module_get(ctrl->ops->module)) {
2988                 nvme_put_ctrl(ctrl);
2989                 return -EINVAL;
2990         }
2991
2992         file->private_data = ctrl;
2993         return 0;
2994 }
2995
2996 static int nvme_dev_release(struct inode *inode, struct file *file)
2997 {
2998         struct nvme_ctrl *ctrl =
2999                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3000
3001         module_put(ctrl->ops->module);
3002         nvme_put_ctrl(ctrl);
3003         return 0;
3004 }
3005
3006 static const struct file_operations nvme_dev_fops = {
3007         .owner          = THIS_MODULE,
3008         .open           = nvme_dev_open,
3009         .release        = nvme_dev_release,
3010         .unlocked_ioctl = nvme_dev_ioctl,
3011         .compat_ioctl   = compat_ptr_ioctl,
3012 };
3013
3014 static ssize_t nvme_sysfs_reset(struct device *dev,
3015                                 struct device_attribute *attr, const char *buf,
3016                                 size_t count)
3017 {
3018         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3019         int ret;
3020
3021         ret = nvme_reset_ctrl_sync(ctrl);
3022         if (ret < 0)
3023                 return ret;
3024         return count;
3025 }
3026 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3027
3028 static ssize_t nvme_sysfs_rescan(struct device *dev,
3029                                 struct device_attribute *attr, const char *buf,
3030                                 size_t count)
3031 {
3032         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3033
3034         nvme_queue_scan(ctrl);
3035         return count;
3036 }
3037 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3038
3039 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3040 {
3041         struct gendisk *disk = dev_to_disk(dev);
3042
3043         if (disk->fops == &nvme_bdev_ops)
3044                 return nvme_get_ns_from_dev(dev)->head;
3045         else
3046                 return disk->private_data;
3047 }
3048
3049 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3050                 char *buf)
3051 {
3052         struct nvme_ns_head *head = dev_to_ns_head(dev);
3053         struct nvme_ns_ids *ids = &head->ids;
3054         struct nvme_subsystem *subsys = head->subsys;
3055         int serial_len = sizeof(subsys->serial);
3056         int model_len = sizeof(subsys->model);
3057
3058         if (!uuid_is_null(&ids->uuid))
3059                 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3060
3061         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3062                 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3063
3064         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3065                 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3066
3067         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3068                                   subsys->serial[serial_len - 1] == '\0'))
3069                 serial_len--;
3070         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3071                                  subsys->model[model_len - 1] == '\0'))
3072                 model_len--;
3073
3074         return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3075                 serial_len, subsys->serial, model_len, subsys->model,
3076                 head->ns_id);
3077 }
3078 static DEVICE_ATTR_RO(wwid);
3079
3080 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3081                 char *buf)
3082 {
3083         return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3084 }
3085 static DEVICE_ATTR_RO(nguid);
3086
3087 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3088                 char *buf)
3089 {
3090         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3091
3092         /* For backward compatibility expose the NGUID to userspace if
3093          * we have no UUID set
3094          */
3095         if (uuid_is_null(&ids->uuid)) {
3096                 printk_ratelimited(KERN_WARNING
3097                                    "No UUID available providing old NGUID\n");
3098                 return sysfs_emit(buf, "%pU\n", ids->nguid);
3099         }
3100         return sysfs_emit(buf, "%pU\n", &ids->uuid);
3101 }
3102 static DEVICE_ATTR_RO(uuid);
3103
3104 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3105                 char *buf)
3106 {
3107         return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3108 }
3109 static DEVICE_ATTR_RO(eui);
3110
3111 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3112                 char *buf)
3113 {
3114         return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3115 }
3116 static DEVICE_ATTR_RO(nsid);
3117
3118 static struct attribute *nvme_ns_id_attrs[] = {
3119         &dev_attr_wwid.attr,
3120         &dev_attr_uuid.attr,
3121         &dev_attr_nguid.attr,
3122         &dev_attr_eui.attr,
3123         &dev_attr_nsid.attr,
3124 #ifdef CONFIG_NVME_MULTIPATH
3125         &dev_attr_ana_grpid.attr,
3126         &dev_attr_ana_state.attr,
3127 #endif
3128         NULL,
3129 };
3130
3131 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3132                 struct attribute *a, int n)
3133 {
3134         struct device *dev = container_of(kobj, struct device, kobj);
3135         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3136
3137         if (a == &dev_attr_uuid.attr) {
3138                 if (uuid_is_null(&ids->uuid) &&
3139                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3140                         return 0;
3141         }
3142         if (a == &dev_attr_nguid.attr) {
3143                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3144                         return 0;
3145         }
3146         if (a == &dev_attr_eui.attr) {
3147                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3148                         return 0;
3149         }
3150 #ifdef CONFIG_NVME_MULTIPATH
3151         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3152                 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3153                         return 0;
3154                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3155                         return 0;
3156         }
3157 #endif
3158         return a->mode;
3159 }
3160
3161 static const struct attribute_group nvme_ns_id_attr_group = {
3162         .attrs          = nvme_ns_id_attrs,
3163         .is_visible     = nvme_ns_id_attrs_are_visible,
3164 };
3165
3166 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3167         &nvme_ns_id_attr_group,
3168 #ifdef CONFIG_NVM
3169         &nvme_nvm_attr_group,
3170 #endif
3171         NULL,
3172 };
3173
3174 #define nvme_show_str_function(field)                                           \
3175 static ssize_t  field##_show(struct device *dev,                                \
3176                             struct device_attribute *attr, char *buf)           \
3177 {                                                                               \
3178         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3179         return sysfs_emit(buf, "%.*s\n",                                        \
3180                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3181 }                                                                               \
3182 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3183
3184 nvme_show_str_function(model);
3185 nvme_show_str_function(serial);
3186 nvme_show_str_function(firmware_rev);
3187
3188 #define nvme_show_int_function(field)                                           \
3189 static ssize_t  field##_show(struct device *dev,                                \
3190                             struct device_attribute *attr, char *buf)           \
3191 {                                                                               \
3192         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3193         return sysfs_emit(buf, "%d\n", ctrl->field);                            \
3194 }                                                                               \
3195 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3196
3197 nvme_show_int_function(cntlid);
3198 nvme_show_int_function(numa_node);
3199 nvme_show_int_function(queue_count);
3200 nvme_show_int_function(sqsize);
3201 nvme_show_int_function(kato);
3202
3203 static ssize_t nvme_sysfs_delete(struct device *dev,
3204                                 struct device_attribute *attr, const char *buf,
3205                                 size_t count)
3206 {
3207         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3208
3209         if (device_remove_file_self(dev, attr))
3210                 nvme_delete_ctrl_sync(ctrl);
3211         return count;
3212 }
3213 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3214
3215 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3216                                          struct device_attribute *attr,
3217                                          char *buf)
3218 {
3219         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3220
3221         return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3222 }
3223 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3224
3225 static ssize_t nvme_sysfs_show_state(struct device *dev,
3226                                      struct device_attribute *attr,
3227                                      char *buf)
3228 {
3229         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3230         static const char *const state_name[] = {
3231                 [NVME_CTRL_NEW]         = "new",
3232                 [NVME_CTRL_LIVE]        = "live",
3233                 [NVME_CTRL_RESETTING]   = "resetting",
3234                 [NVME_CTRL_CONNECTING]  = "connecting",
3235                 [NVME_CTRL_DELETING]    = "deleting",
3236                 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3237                 [NVME_CTRL_DEAD]        = "dead",
3238         };
3239
3240         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3241             state_name[ctrl->state])
3242                 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3243
3244         return sysfs_emit(buf, "unknown state\n");
3245 }
3246
3247 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3248
3249 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3250                                          struct device_attribute *attr,
3251                                          char *buf)
3252 {
3253         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3254
3255         return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3256 }
3257 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3258
3259 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3260                                         struct device_attribute *attr,
3261                                         char *buf)
3262 {
3263         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3264
3265         return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3266 }
3267 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3268
3269 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3270                                         struct device_attribute *attr,
3271                                         char *buf)
3272 {
3273         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3274
3275         return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3276 }
3277 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3278
3279 static ssize_t nvme_sysfs_show_address(struct device *dev,
3280                                          struct device_attribute *attr,
3281                                          char *buf)
3282 {
3283         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3284
3285         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3286 }
3287 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3288
3289 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3290                 struct device_attribute *attr, char *buf)
3291 {
3292         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3293         struct nvmf_ctrl_options *opts = ctrl->opts;
3294
3295         if (ctrl->opts->max_reconnects == -1)
3296                 return sysfs_emit(buf, "off\n");
3297         return sysfs_emit(buf, "%d\n",
3298                           opts->max_reconnects * opts->reconnect_delay);
3299 }
3300
3301 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3302                 struct device_attribute *attr, const char *buf, size_t count)
3303 {
3304         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3305         struct nvmf_ctrl_options *opts = ctrl->opts;
3306         int ctrl_loss_tmo, err;
3307
3308         err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3309         if (err)
3310                 return -EINVAL;
3311
3312         if (ctrl_loss_tmo < 0)
3313                 opts->max_reconnects = -1;
3314         else
3315                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3316                                                 opts->reconnect_delay);
3317         return count;
3318 }
3319 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3320         nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3321
3322 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3323                 struct device_attribute *attr, char *buf)
3324 {
3325         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3326
3327         if (ctrl->opts->reconnect_delay == -1)
3328                 return sysfs_emit(buf, "off\n");
3329         return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3330 }
3331
3332 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3333                 struct device_attribute *attr, const char *buf, size_t count)
3334 {
3335         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3336         unsigned int v;
3337         int err;
3338
3339         err = kstrtou32(buf, 10, &v);
3340         if (err)
3341                 return err;
3342
3343         ctrl->opts->reconnect_delay = v;
3344         return count;
3345 }
3346 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3347         nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3348
3349 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3350                 struct device_attribute *attr, char *buf)
3351 {
3352         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3353
3354         if (ctrl->opts->fast_io_fail_tmo == -1)
3355                 return sysfs_emit(buf, "off\n");
3356         return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3357 }
3358
3359 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3360                 struct device_attribute *attr, const char *buf, size_t count)
3361 {
3362         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3363         struct nvmf_ctrl_options *opts = ctrl->opts;
3364         int fast_io_fail_tmo, err;
3365
3366         err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3367         if (err)
3368                 return -EINVAL;
3369
3370         if (fast_io_fail_tmo < 0)
3371                 opts->fast_io_fail_tmo = -1;
3372         else
3373                 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3374         return count;
3375 }
3376 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3377         nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3378
3379 static struct attribute *nvme_dev_attrs[] = {
3380         &dev_attr_reset_controller.attr,
3381         &dev_attr_rescan_controller.attr,
3382         &dev_attr_model.attr,
3383         &dev_attr_serial.attr,
3384         &dev_attr_firmware_rev.attr,
3385         &dev_attr_cntlid.attr,
3386         &dev_attr_delete_controller.attr,
3387         &dev_attr_transport.attr,
3388         &dev_attr_subsysnqn.attr,
3389         &dev_attr_address.attr,
3390         &dev_attr_state.attr,
3391         &dev_attr_numa_node.attr,
3392         &dev_attr_queue_count.attr,
3393         &dev_attr_sqsize.attr,
3394         &dev_attr_hostnqn.attr,
3395         &dev_attr_hostid.attr,
3396         &dev_attr_ctrl_loss_tmo.attr,
3397         &dev_attr_reconnect_delay.attr,
3398         &dev_attr_fast_io_fail_tmo.attr,
3399         &dev_attr_kato.attr,
3400         NULL
3401 };
3402
3403 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3404                 struct attribute *a, int n)
3405 {
3406         struct device *dev = container_of(kobj, struct device, kobj);
3407         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3408
3409         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3410                 return 0;
3411         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3412                 return 0;
3413         if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3414                 return 0;
3415         if (a == &dev_attr_hostid.attr && !ctrl->opts)
3416                 return 0;
3417         if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3418                 return 0;
3419         if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3420                 return 0;
3421         if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3422                 return 0;
3423
3424         return a->mode;
3425 }
3426
3427 static const struct attribute_group nvme_dev_attrs_group = {
3428         .attrs          = nvme_dev_attrs,
3429         .is_visible     = nvme_dev_attrs_are_visible,
3430 };
3431
3432 static const struct attribute_group *nvme_dev_attr_groups[] = {
3433         &nvme_dev_attrs_group,
3434         NULL,
3435 };
3436
3437 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3438                 unsigned nsid)
3439 {
3440         struct nvme_ns_head *h;
3441
3442         lockdep_assert_held(&subsys->lock);
3443
3444         list_for_each_entry(h, &subsys->nsheads, entry) {
3445                 if (h->ns_id == nsid && nvme_tryget_ns_head(h))
3446                         return h;
3447         }
3448
3449         return NULL;
3450 }
3451
3452 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3453                 struct nvme_ns_head *new)
3454 {
3455         struct nvme_ns_head *h;
3456
3457         lockdep_assert_held(&subsys->lock);
3458
3459         list_for_each_entry(h, &subsys->nsheads, entry) {
3460                 if (nvme_ns_ids_valid(&new->ids) &&
3461                     nvme_ns_ids_equal(&new->ids, &h->ids))
3462                         return -EINVAL;
3463         }
3464
3465         return 0;
3466 }
3467
3468 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3469 {
3470         cdev_device_del(cdev, cdev_device);
3471         ida_simple_remove(&nvme_ns_chr_minor_ida, MINOR(cdev_device->devt));
3472 }
3473
3474 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3475                 const struct file_operations *fops, struct module *owner)
3476 {
3477         int minor, ret;
3478
3479         minor = ida_simple_get(&nvme_ns_chr_minor_ida, 0, 0, GFP_KERNEL);
3480         if (minor < 0)
3481                 return minor;
3482         cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3483         cdev_device->class = nvme_ns_chr_class;
3484         device_initialize(cdev_device);
3485         cdev_init(cdev, fops);
3486         cdev->owner = owner;
3487         ret = cdev_device_add(cdev, cdev_device);
3488         if (ret)
3489                 ida_simple_remove(&nvme_ns_chr_minor_ida, minor);
3490         return ret;
3491 }
3492
3493 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3494 {
3495         return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3496 }
3497
3498 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3499 {
3500         nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3501         return 0;
3502 }
3503
3504 static const struct file_operations nvme_ns_chr_fops = {
3505         .owner          = THIS_MODULE,
3506         .open           = nvme_ns_chr_open,
3507         .release        = nvme_ns_chr_release,
3508         .unlocked_ioctl = nvme_ns_chr_ioctl,
3509         .compat_ioctl   = compat_ptr_ioctl,
3510 };
3511
3512 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3513 {
3514         int ret;
3515
3516         ns->cdev_device.parent = ns->ctrl->device;
3517         ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3518                            ns->ctrl->instance, ns->head->instance);
3519         if (ret)
3520                 return ret;
3521         ret = nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3522                             ns->ctrl->ops->module);
3523         if (ret)
3524                 kfree_const(ns->cdev_device.kobj.name);
3525         return ret;
3526 }
3527
3528 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3529                 unsigned nsid, struct nvme_ns_ids *ids)
3530 {
3531         struct nvme_ns_head *head;
3532         size_t size = sizeof(*head);
3533         int ret = -ENOMEM;
3534
3535 #ifdef CONFIG_NVME_MULTIPATH
3536         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3537 #endif
3538
3539         head = kzalloc(size, GFP_KERNEL);
3540         if (!head)
3541                 goto out;
3542         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3543         if (ret < 0)
3544                 goto out_free_head;
3545         head->instance = ret;
3546         INIT_LIST_HEAD(&head->list);
3547         ret = init_srcu_struct(&head->srcu);
3548         if (ret)
3549                 goto out_ida_remove;
3550         head->subsys = ctrl->subsys;
3551         head->ns_id = nsid;
3552         head->ids = *ids;
3553         kref_init(&head->ref);
3554
3555         ret = __nvme_check_ids(ctrl->subsys, head);
3556         if (ret) {
3557                 dev_err(ctrl->device,
3558                         "duplicate IDs for nsid %d\n", nsid);
3559                 goto out_cleanup_srcu;
3560         }
3561
3562         if (head->ids.csi) {
3563                 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3564                 if (ret)
3565                         goto out_cleanup_srcu;
3566         } else
3567                 head->effects = ctrl->effects;
3568
3569         ret = nvme_mpath_alloc_disk(ctrl, head);
3570         if (ret)
3571                 goto out_cleanup_srcu;
3572
3573         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3574
3575         kref_get(&ctrl->subsys->ref);
3576
3577         return head;
3578 out_cleanup_srcu:
3579         cleanup_srcu_struct(&head->srcu);
3580 out_ida_remove:
3581         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3582 out_free_head:
3583         kfree(head);
3584 out:
3585         if (ret > 0)
3586                 ret = blk_status_to_errno(nvme_error_status(ret));
3587         return ERR_PTR(ret);
3588 }
3589
3590 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3591                 struct nvme_ns_ids *ids, bool is_shared)
3592 {
3593         struct nvme_ctrl *ctrl = ns->ctrl;
3594         struct nvme_ns_head *head = NULL;
3595         int ret = 0;
3596
3597         mutex_lock(&ctrl->subsys->lock);
3598         head = nvme_find_ns_head(ctrl->subsys, nsid);
3599         if (!head) {
3600                 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3601                 if (IS_ERR(head)) {
3602                         ret = PTR_ERR(head);
3603                         goto out_unlock;
3604                 }
3605                 head->shared = is_shared;
3606         } else {
3607                 ret = -EINVAL;
3608                 if (!is_shared || !head->shared) {
3609                         dev_err(ctrl->device,
3610                                 "Duplicate unshared namespace %d\n", nsid);
3611                         goto out_put_ns_head;
3612                 }
3613                 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3614                         dev_err(ctrl->device,
3615                                 "IDs don't match for shared namespace %d\n",
3616                                         nsid);
3617                         goto out_put_ns_head;
3618                 }
3619         }
3620
3621         list_add_tail_rcu(&ns->siblings, &head->list);
3622         ns->head = head;
3623         mutex_unlock(&ctrl->subsys->lock);
3624         return 0;
3625
3626 out_put_ns_head:
3627         nvme_put_ns_head(head);
3628 out_unlock:
3629         mutex_unlock(&ctrl->subsys->lock);
3630         return ret;
3631 }
3632
3633 static int ns_cmp(void *priv, const struct list_head *a,
3634                 const struct list_head *b)
3635 {
3636         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3637         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3638
3639         return nsa->head->ns_id - nsb->head->ns_id;
3640 }
3641
3642 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3643 {
3644         struct nvme_ns *ns, *ret = NULL;
3645
3646         down_read(&ctrl->namespaces_rwsem);
3647         list_for_each_entry(ns, &ctrl->namespaces, list) {
3648                 if (ns->head->ns_id == nsid) {
3649                         if (!nvme_get_ns(ns))
3650                                 continue;
3651                         ret = ns;
3652                         break;
3653                 }
3654                 if (ns->head->ns_id > nsid)
3655                         break;
3656         }
3657         up_read(&ctrl->namespaces_rwsem);
3658         return ret;
3659 }
3660 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3661
3662 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3663                 struct nvme_ns_ids *ids)
3664 {
3665         struct nvme_ns *ns;
3666         struct gendisk *disk;
3667         struct nvme_id_ns *id;
3668         int node = ctrl->numa_node;
3669
3670         if (nvme_identify_ns(ctrl, nsid, ids, &id))
3671                 return;
3672
3673         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3674         if (!ns)
3675                 goto out_free_id;
3676
3677         ns->queue = blk_mq_init_queue(ctrl->tagset);
3678         if (IS_ERR(ns->queue))
3679                 goto out_free_ns;
3680
3681         if (ctrl->opts && ctrl->opts->data_digest)
3682                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3683
3684         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3685         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3686                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3687
3688         ns->queue->queuedata = ns;
3689         ns->ctrl = ctrl;
3690         kref_init(&ns->kref);
3691
3692         if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3693                 goto out_free_queue;
3694
3695         disk = alloc_disk_node(0, node);
3696         if (!disk)
3697                 goto out_unlink_ns;
3698
3699         disk->fops = &nvme_bdev_ops;
3700         disk->private_data = ns;
3701         disk->queue = ns->queue;
3702         disk->flags = GENHD_FL_EXT_DEVT;
3703         /*
3704          * Without the multipath code enabled, multiple controller per
3705          * subsystems are visible as devices and thus we cannot use the
3706          * subsystem instance.
3707          */
3708         if (!nvme_mpath_set_disk_name(ns, disk->disk_name, &disk->flags))
3709                 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3710                         ns->head->instance);
3711         ns->disk = disk;
3712
3713         if (nvme_update_ns_info(ns, id))
3714                 goto out_put_disk;
3715
3716         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3717                 if (nvme_nvm_register(ns, disk->disk_name, node)) {
3718                         dev_warn(ctrl->device, "LightNVM init failure\n");
3719                         goto out_put_disk;
3720                 }
3721         }
3722
3723         down_write(&ctrl->namespaces_rwsem);
3724         list_add_tail(&ns->list, &ctrl->namespaces);
3725         up_write(&ctrl->namespaces_rwsem);
3726
3727         nvme_get_ctrl(ctrl);
3728
3729         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3730         if (!nvme_ns_head_multipath(ns->head))
3731                 nvme_add_ns_cdev(ns);
3732
3733         nvme_mpath_add_disk(ns, id);
3734         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3735         kfree(id);
3736
3737         return;
3738  out_put_disk:
3739         /* prevent double queue cleanup */
3740         ns->disk->queue = NULL;
3741         put_disk(ns->disk);
3742  out_unlink_ns:
3743         mutex_lock(&ctrl->subsys->lock);
3744         list_del_rcu(&ns->siblings);
3745         if (list_empty(&ns->head->list))
3746                 list_del_init(&ns->head->entry);
3747         mutex_unlock(&ctrl->subsys->lock);
3748         nvme_put_ns_head(ns->head);
3749  out_free_queue:
3750         blk_cleanup_queue(ns->queue);
3751  out_free_ns:
3752         kfree(ns);
3753  out_free_id:
3754         kfree(id);
3755 }
3756
3757 static void nvme_ns_remove(struct nvme_ns *ns)
3758 {
3759         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3760                 return;
3761
3762         set_capacity(ns->disk, 0);
3763         nvme_fault_inject_fini(&ns->fault_inject);
3764
3765         mutex_lock(&ns->ctrl->subsys->lock);
3766         list_del_rcu(&ns->siblings);
3767         if (list_empty(&ns->head->list))
3768                 list_del_init(&ns->head->entry);
3769         mutex_unlock(&ns->ctrl->subsys->lock);
3770
3771         synchronize_rcu(); /* guarantee not available in head->list */
3772         nvme_mpath_clear_current_path(ns);
3773         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3774
3775         if (ns->disk->flags & GENHD_FL_UP) {
3776                 if (!nvme_ns_head_multipath(ns->head))
3777                         nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3778                 del_gendisk(ns->disk);
3779                 blk_cleanup_queue(ns->queue);
3780                 if (blk_get_integrity(ns->disk))
3781                         blk_integrity_unregister(ns->disk);
3782         }
3783
3784         down_write(&ns->ctrl->namespaces_rwsem);
3785         list_del_init(&ns->list);
3786         up_write(&ns->ctrl->namespaces_rwsem);
3787
3788         nvme_mpath_check_last_path(ns);
3789         nvme_put_ns(ns);
3790 }
3791
3792 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3793 {
3794         struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3795
3796         if (ns) {
3797                 nvme_ns_remove(ns);
3798                 nvme_put_ns(ns);
3799         }
3800 }
3801
3802 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
3803 {
3804         struct nvme_id_ns *id;
3805         int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3806
3807         if (test_bit(NVME_NS_DEAD, &ns->flags))
3808                 goto out;
3809
3810         ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
3811         if (ret)
3812                 goto out;
3813
3814         ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3815         if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
3816                 dev_err(ns->ctrl->device,
3817                         "identifiers changed for nsid %d\n", ns->head->ns_id);
3818                 goto out_free_id;
3819         }
3820
3821         ret = nvme_update_ns_info(ns, id);
3822
3823 out_free_id:
3824         kfree(id);
3825 out:
3826         /*
3827          * Only remove the namespace if we got a fatal error back from the
3828          * device, otherwise ignore the error and just move on.
3829          *
3830          * TODO: we should probably schedule a delayed retry here.
3831          */
3832         if (ret > 0 && (ret & NVME_SC_DNR))
3833                 nvme_ns_remove(ns);
3834 }
3835
3836 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3837 {
3838         struct nvme_ns_ids ids = { };
3839         struct nvme_ns *ns;
3840
3841         if (nvme_identify_ns_descs(ctrl, nsid, &ids))
3842                 return;
3843
3844         ns = nvme_find_get_ns(ctrl, nsid);
3845         if (ns) {
3846                 nvme_validate_ns(ns, &ids);
3847                 nvme_put_ns(ns);
3848                 return;
3849         }
3850
3851         switch (ids.csi) {
3852         case NVME_CSI_NVM:
3853                 nvme_alloc_ns(ctrl, nsid, &ids);
3854                 break;
3855         case NVME_CSI_ZNS:
3856                 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
3857                         dev_warn(ctrl->device,
3858                                 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
3859                                 nsid);
3860                         break;
3861                 }
3862                 if (!nvme_multi_css(ctrl)) {
3863                         dev_warn(ctrl->device,
3864                                 "command set not reported for nsid: %d\n",
3865                                 nsid);
3866                         break;
3867                 }
3868                 nvme_alloc_ns(ctrl, nsid, &ids);
3869                 break;
3870         default:
3871                 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
3872                         ids.csi, nsid);
3873                 break;
3874         }
3875 }
3876
3877 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3878                                         unsigned nsid)
3879 {
3880         struct nvme_ns *ns, *next;
3881         LIST_HEAD(rm_list);
3882
3883         down_write(&ctrl->namespaces_rwsem);
3884         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3885                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3886                         list_move_tail(&ns->list, &rm_list);
3887         }
3888         up_write(&ctrl->namespaces_rwsem);
3889
3890         list_for_each_entry_safe(ns, next, &rm_list, list)
3891                 nvme_ns_remove(ns);
3892
3893 }
3894
3895 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3896 {
3897         const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3898         __le32 *ns_list;
3899         u32 prev = 0;
3900         int ret = 0, i;
3901
3902         if (nvme_ctrl_limited_cns(ctrl))
3903                 return -EOPNOTSUPP;
3904
3905         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3906         if (!ns_list)
3907                 return -ENOMEM;
3908
3909         for (;;) {
3910                 struct nvme_command cmd = {
3911                         .identify.opcode        = nvme_admin_identify,
3912                         .identify.cns           = NVME_ID_CNS_NS_ACTIVE_LIST,
3913                         .identify.nsid          = cpu_to_le32(prev),
3914                 };
3915
3916                 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
3917                                             NVME_IDENTIFY_DATA_SIZE);
3918                 if (ret) {
3919                         dev_warn(ctrl->device,
3920                                 "Identify NS List failed (status=0x%x)\n", ret);
3921                         goto free;
3922                 }
3923
3924                 for (i = 0; i < nr_entries; i++) {
3925                         u32 nsid = le32_to_cpu(ns_list[i]);
3926
3927                         if (!nsid)      /* end of the list? */
3928                                 goto out;
3929                         nvme_validate_or_alloc_ns(ctrl, nsid);
3930                         while (++prev < nsid)
3931                                 nvme_ns_remove_by_nsid(ctrl, prev);
3932                 }
3933         }
3934  out:
3935         nvme_remove_invalid_namespaces(ctrl, prev);
3936  free:
3937         kfree(ns_list);
3938         return ret;
3939 }
3940
3941 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3942 {
3943         struct nvme_id_ctrl *id;
3944         u32 nn, i;
3945
3946         if (nvme_identify_ctrl(ctrl, &id))
3947                 return;
3948         nn = le32_to_cpu(id->nn);
3949         kfree(id);
3950
3951         for (i = 1; i <= nn; i++)
3952                 nvme_validate_or_alloc_ns(ctrl, i);
3953
3954         nvme_remove_invalid_namespaces(ctrl, nn);
3955 }
3956
3957 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3958 {
3959         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3960         __le32 *log;
3961         int error;
3962
3963         log = kzalloc(log_size, GFP_KERNEL);
3964         if (!log)
3965                 return;
3966
3967         /*
3968          * We need to read the log to clear the AEN, but we don't want to rely
3969          * on it for the changed namespace information as userspace could have
3970          * raced with us in reading the log page, which could cause us to miss
3971          * updates.
3972          */
3973         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
3974                         NVME_CSI_NVM, log, log_size, 0);
3975         if (error)
3976                 dev_warn(ctrl->device,
3977                         "reading changed ns log failed: %d\n", error);
3978
3979         kfree(log);
3980 }
3981
3982 static void nvme_scan_work(struct work_struct *work)
3983 {
3984         struct nvme_ctrl *ctrl =
3985                 container_of(work, struct nvme_ctrl, scan_work);
3986
3987         /* No tagset on a live ctrl means IO queues could not created */
3988         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3989                 return;
3990
3991         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3992                 dev_info(ctrl->device, "rescanning namespaces.\n");
3993                 nvme_clear_changed_ns_log(ctrl);
3994         }
3995
3996         mutex_lock(&ctrl->scan_lock);
3997         if (nvme_scan_ns_list(ctrl) != 0)
3998                 nvme_scan_ns_sequential(ctrl);
3999         mutex_unlock(&ctrl->scan_lock);
4000
4001         down_write(&ctrl->namespaces_rwsem);
4002         list_sort(NULL, &ctrl->namespaces, ns_cmp);
4003         up_write(&ctrl->namespaces_rwsem);
4004 }
4005
4006 /*
4007  * This function iterates the namespace list unlocked to allow recovery from
4008  * controller failure. It is up to the caller to ensure the namespace list is
4009  * not modified by scan work while this function is executing.
4010  */
4011 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4012 {
4013         struct nvme_ns *ns, *next;
4014         LIST_HEAD(ns_list);
4015
4016         /*
4017          * make sure to requeue I/O to all namespaces as these
4018          * might result from the scan itself and must complete
4019          * for the scan_work to make progress
4020          */
4021         nvme_mpath_clear_ctrl_paths(ctrl);
4022
4023         /* prevent racing with ns scanning */
4024         flush_work(&ctrl->scan_work);
4025
4026         /*
4027          * The dead states indicates the controller was not gracefully
4028          * disconnected. In that case, we won't be able to flush any data while
4029          * removing the namespaces' disks; fail all the queues now to avoid
4030          * potentially having to clean up the failed sync later.
4031          */
4032         if (ctrl->state == NVME_CTRL_DEAD)
4033                 nvme_kill_queues(ctrl);
4034
4035         /* this is a no-op when called from the controller reset handler */
4036         nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4037
4038         down_write(&ctrl->namespaces_rwsem);
4039         list_splice_init(&ctrl->namespaces, &ns_list);
4040         up_write(&ctrl->namespaces_rwsem);
4041
4042         list_for_each_entry_safe(ns, next, &ns_list, list)
4043                 nvme_ns_remove(ns);
4044 }
4045 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4046
4047 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4048 {
4049         struct nvme_ctrl *ctrl =
4050                 container_of(dev, struct nvme_ctrl, ctrl_device);
4051         struct nvmf_ctrl_options *opts = ctrl->opts;
4052         int ret;
4053
4054         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4055         if (ret)
4056                 return ret;
4057
4058         if (opts) {
4059                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4060                 if (ret)
4061                         return ret;
4062
4063                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4064                                 opts->trsvcid ?: "none");
4065                 if (ret)
4066                         return ret;
4067
4068                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4069                                 opts->host_traddr ?: "none");
4070         }
4071         return ret;
4072 }
4073
4074 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4075 {
4076         char *envp[2] = { NULL, NULL };
4077         u32 aen_result = ctrl->aen_result;
4078
4079         ctrl->aen_result = 0;
4080         if (!aen_result)
4081                 return;
4082
4083         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4084         if (!envp[0])
4085                 return;
4086         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4087         kfree(envp[0]);
4088 }
4089
4090 static void nvme_async_event_work(struct work_struct *work)
4091 {
4092         struct nvme_ctrl *ctrl =
4093                 container_of(work, struct nvme_ctrl, async_event_work);
4094
4095         nvme_aen_uevent(ctrl);
4096         ctrl->ops->submit_async_event(ctrl);
4097 }
4098
4099 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4100 {
4101
4102         u32 csts;
4103
4104         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4105                 return false;
4106
4107         if (csts == ~0)
4108                 return false;
4109
4110         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4111 }
4112
4113 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4114 {
4115         struct nvme_fw_slot_info_log *log;
4116
4117         log = kmalloc(sizeof(*log), GFP_KERNEL);
4118         if (!log)
4119                 return;
4120
4121         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4122                         log, sizeof(*log), 0))
4123                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4124         kfree(log);
4125 }
4126
4127 static void nvme_fw_act_work(struct work_struct *work)
4128 {
4129         struct nvme_ctrl *ctrl = container_of(work,
4130                                 struct nvme_ctrl, fw_act_work);
4131         unsigned long fw_act_timeout;
4132
4133         if (ctrl->mtfa)
4134                 fw_act_timeout = jiffies +
4135                                 msecs_to_jiffies(ctrl->mtfa * 100);
4136         else
4137                 fw_act_timeout = jiffies +
4138                                 msecs_to_jiffies(admin_timeout * 1000);
4139
4140         nvme_stop_queues(ctrl);
4141         while (nvme_ctrl_pp_status(ctrl)) {
4142                 if (time_after(jiffies, fw_act_timeout)) {
4143                         dev_warn(ctrl->device,
4144                                 "Fw activation timeout, reset controller\n");
4145                         nvme_try_sched_reset(ctrl);
4146                         return;
4147                 }
4148                 msleep(100);
4149         }
4150
4151         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4152                 return;
4153
4154         nvme_start_queues(ctrl);
4155         /* read FW slot information to clear the AER */
4156         nvme_get_fw_slot_info(ctrl);
4157 }
4158
4159 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4160 {
4161         u32 aer_notice_type = (result & 0xff00) >> 8;
4162
4163         trace_nvme_async_event(ctrl, aer_notice_type);
4164
4165         switch (aer_notice_type) {
4166         case NVME_AER_NOTICE_NS_CHANGED:
4167                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4168                 nvme_queue_scan(ctrl);
4169                 break;
4170         case NVME_AER_NOTICE_FW_ACT_STARTING:
4171                 /*
4172                  * We are (ab)using the RESETTING state to prevent subsequent
4173                  * recovery actions from interfering with the controller's
4174                  * firmware activation.
4175                  */
4176                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4177                         queue_work(nvme_wq, &ctrl->fw_act_work);
4178                 break;
4179 #ifdef CONFIG_NVME_MULTIPATH
4180         case NVME_AER_NOTICE_ANA:
4181                 if (!ctrl->ana_log_buf)
4182                         break;
4183                 queue_work(nvme_wq, &ctrl->ana_work);
4184                 break;
4185 #endif
4186         case NVME_AER_NOTICE_DISC_CHANGED:
4187                 ctrl->aen_result = result;
4188                 break;
4189         default:
4190                 dev_warn(ctrl->device, "async event result %08x\n", result);
4191         }
4192 }
4193
4194 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4195                 volatile union nvme_result *res)
4196 {
4197         u32 result = le32_to_cpu(res->u32);
4198         u32 aer_type = result & 0x07;
4199
4200         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4201                 return;
4202
4203         switch (aer_type) {
4204         case NVME_AER_NOTICE:
4205                 nvme_handle_aen_notice(ctrl, result);
4206                 break;
4207         case NVME_AER_ERROR:
4208         case NVME_AER_SMART:
4209         case NVME_AER_CSS:
4210         case NVME_AER_VS:
4211                 trace_nvme_async_event(ctrl, aer_type);
4212                 ctrl->aen_result = result;
4213                 break;
4214         default:
4215                 break;
4216         }
4217         queue_work(nvme_wq, &ctrl->async_event_work);
4218 }
4219 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4220
4221 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4222 {
4223         nvme_mpath_stop(ctrl);
4224         nvme_stop_keep_alive(ctrl);
4225         nvme_stop_failfast_work(ctrl);
4226         flush_work(&ctrl->async_event_work);
4227         cancel_work_sync(&ctrl->fw_act_work);
4228 }
4229 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4230
4231 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4232 {
4233         nvme_start_keep_alive(ctrl);
4234
4235         nvme_enable_aen(ctrl);
4236
4237         if (ctrl->queue_count > 1) {
4238                 nvme_queue_scan(ctrl);
4239                 nvme_start_queues(ctrl);
4240         }
4241 }
4242 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4243
4244 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4245 {
4246         nvme_hwmon_exit(ctrl);
4247         nvme_fault_inject_fini(&ctrl->fault_inject);
4248         dev_pm_qos_hide_latency_tolerance(ctrl->device);
4249         cdev_device_del(&ctrl->cdev, ctrl->device);
4250         nvme_put_ctrl(ctrl);
4251 }
4252 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4253
4254 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4255 {
4256         struct nvme_effects_log *cel;
4257         unsigned long i;
4258
4259         xa_for_each(&ctrl->cels, i, cel) {
4260                 xa_erase(&ctrl->cels, i);
4261                 kfree(cel);
4262         }
4263
4264         xa_destroy(&ctrl->cels);
4265 }
4266
4267 static void nvme_free_ctrl(struct device *dev)
4268 {
4269         struct nvme_ctrl *ctrl =
4270                 container_of(dev, struct nvme_ctrl, ctrl_device);
4271         struct nvme_subsystem *subsys = ctrl->subsys;
4272
4273         if (!subsys || ctrl->instance != subsys->instance)
4274                 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4275
4276         nvme_free_cels(ctrl);
4277         nvme_mpath_uninit(ctrl);
4278         __free_page(ctrl->discard_page);
4279
4280         if (subsys) {
4281                 mutex_lock(&nvme_subsystems_lock);
4282                 list_del(&ctrl->subsys_entry);
4283                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4284                 mutex_unlock(&nvme_subsystems_lock);
4285         }
4286
4287         ctrl->ops->free_ctrl(ctrl);
4288
4289         if (subsys)
4290                 nvme_put_subsystem(subsys);
4291 }
4292
4293 /*
4294  * Initialize a NVMe controller structures.  This needs to be called during
4295  * earliest initialization so that we have the initialized structured around
4296  * during probing.
4297  */
4298 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4299                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4300 {
4301         int ret;
4302
4303         ctrl->state = NVME_CTRL_NEW;
4304         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4305         spin_lock_init(&ctrl->lock);
4306         mutex_init(&ctrl->scan_lock);
4307         INIT_LIST_HEAD(&ctrl->namespaces);
4308         xa_init(&ctrl->cels);
4309         init_rwsem(&ctrl->namespaces_rwsem);
4310         ctrl->dev = dev;
4311         ctrl->ops = ops;
4312         ctrl->quirks = quirks;
4313         ctrl->numa_node = NUMA_NO_NODE;
4314         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4315         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4316         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4317         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4318         init_waitqueue_head(&ctrl->state_wq);
4319
4320         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4321         INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4322         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4323         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4324
4325         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4326                         PAGE_SIZE);
4327         ctrl->discard_page = alloc_page(GFP_KERNEL);
4328         if (!ctrl->discard_page) {
4329                 ret = -ENOMEM;
4330                 goto out;
4331         }
4332
4333         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4334         if (ret < 0)
4335                 goto out;
4336         ctrl->instance = ret;
4337
4338         device_initialize(&ctrl->ctrl_device);
4339         ctrl->device = &ctrl->ctrl_device;
4340         ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4341                         ctrl->instance);
4342         ctrl->device->class = nvme_class;
4343         ctrl->device->parent = ctrl->dev;
4344         ctrl->device->groups = nvme_dev_attr_groups;
4345         ctrl->device->release = nvme_free_ctrl;
4346         dev_set_drvdata(ctrl->device, ctrl);
4347         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4348         if (ret)
4349                 goto out_release_instance;
4350
4351         nvme_get_ctrl(ctrl);
4352         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4353         ctrl->cdev.owner = ops->module;
4354         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4355         if (ret)
4356                 goto out_free_name;
4357
4358         /*
4359          * Initialize latency tolerance controls.  The sysfs files won't
4360          * be visible to userspace unless the device actually supports APST.
4361          */
4362         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4363         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4364                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4365
4366         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4367         nvme_mpath_init_ctrl(ctrl);
4368
4369         return 0;
4370 out_free_name:
4371         nvme_put_ctrl(ctrl);
4372         kfree_const(ctrl->device->kobj.name);
4373 out_release_instance:
4374         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4375 out:
4376         if (ctrl->discard_page)
4377                 __free_page(ctrl->discard_page);
4378         return ret;
4379 }
4380 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4381
4382 /**
4383  * nvme_kill_queues(): Ends all namespace queues
4384  * @ctrl: the dead controller that needs to end
4385  *
4386  * Call this function when the driver determines it is unable to get the
4387  * controller in a state capable of servicing IO.
4388  */
4389 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4390 {
4391         struct nvme_ns *ns;
4392
4393         down_read(&ctrl->namespaces_rwsem);
4394
4395         /* Forcibly unquiesce queues to avoid blocking dispatch */
4396         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4397                 blk_mq_unquiesce_queue(ctrl->admin_q);
4398
4399         list_for_each_entry(ns, &ctrl->namespaces, list)
4400                 nvme_set_queue_dying(ns);
4401
4402         up_read(&ctrl->namespaces_rwsem);
4403 }
4404 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4405
4406 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4407 {
4408         struct nvme_ns *ns;
4409
4410         down_read(&ctrl->namespaces_rwsem);
4411         list_for_each_entry(ns, &ctrl->namespaces, list)
4412                 blk_mq_unfreeze_queue(ns->queue);
4413         up_read(&ctrl->namespaces_rwsem);
4414 }
4415 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4416
4417 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4418 {
4419         struct nvme_ns *ns;
4420
4421         down_read(&ctrl->namespaces_rwsem);
4422         list_for_each_entry(ns, &ctrl->namespaces, list) {
4423                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4424                 if (timeout <= 0)
4425                         break;
4426         }
4427         up_read(&ctrl->namespaces_rwsem);
4428         return timeout;
4429 }
4430 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4431
4432 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4433 {
4434         struct nvme_ns *ns;
4435
4436         down_read(&ctrl->namespaces_rwsem);
4437         list_for_each_entry(ns, &ctrl->namespaces, list)
4438                 blk_mq_freeze_queue_wait(ns->queue);
4439         up_read(&ctrl->namespaces_rwsem);
4440 }
4441 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4442
4443 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4444 {
4445         struct nvme_ns *ns;
4446
4447         down_read(&ctrl->namespaces_rwsem);
4448         list_for_each_entry(ns, &ctrl->namespaces, list)
4449                 blk_freeze_queue_start(ns->queue);
4450         up_read(&ctrl->namespaces_rwsem);
4451 }
4452 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4453
4454 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4455 {
4456         struct nvme_ns *ns;
4457
4458         down_read(&ctrl->namespaces_rwsem);
4459         list_for_each_entry(ns, &ctrl->namespaces, list)
4460                 blk_mq_quiesce_queue(ns->queue);
4461         up_read(&ctrl->namespaces_rwsem);
4462 }
4463 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4464
4465 void nvme_start_queues(struct nvme_ctrl *ctrl)
4466 {
4467         struct nvme_ns *ns;
4468
4469         down_read(&ctrl->namespaces_rwsem);
4470         list_for_each_entry(ns, &ctrl->namespaces, list)
4471                 blk_mq_unquiesce_queue(ns->queue);
4472         up_read(&ctrl->namespaces_rwsem);
4473 }
4474 EXPORT_SYMBOL_GPL(nvme_start_queues);
4475
4476 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4477 {
4478         struct nvme_ns *ns;
4479
4480         down_read(&ctrl->namespaces_rwsem);
4481         list_for_each_entry(ns, &ctrl->namespaces, list)
4482                 blk_sync_queue(ns->queue);
4483         up_read(&ctrl->namespaces_rwsem);
4484 }
4485 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4486
4487 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4488 {
4489         nvme_sync_io_queues(ctrl);
4490         if (ctrl->admin_q)
4491                 blk_sync_queue(ctrl->admin_q);
4492 }
4493 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4494
4495 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4496 {
4497         if (file->f_op != &nvme_dev_fops)
4498                 return NULL;
4499         return file->private_data;
4500 }
4501 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4502
4503 /*
4504  * Check we didn't inadvertently grow the command structure sizes:
4505  */
4506 static inline void _nvme_check_size(void)
4507 {
4508         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4509         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4510         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4511         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4512         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4513         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4514         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4515         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4516         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4517         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4518         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4519         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4520         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4521         BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4522         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4523         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4524         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4525         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4526         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4527         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4528 }
4529
4530
4531 static int __init nvme_core_init(void)
4532 {
4533         int result = -ENOMEM;
4534
4535         _nvme_check_size();
4536
4537         nvme_wq = alloc_workqueue("nvme-wq",
4538                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4539         if (!nvme_wq)
4540                 goto out;
4541
4542         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4543                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4544         if (!nvme_reset_wq)
4545                 goto destroy_wq;
4546
4547         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4548                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4549         if (!nvme_delete_wq)
4550                 goto destroy_reset_wq;
4551
4552         result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4553                         NVME_MINORS, "nvme");
4554         if (result < 0)
4555                 goto destroy_delete_wq;
4556
4557         nvme_class = class_create(THIS_MODULE, "nvme");
4558         if (IS_ERR(nvme_class)) {
4559                 result = PTR_ERR(nvme_class);
4560                 goto unregister_chrdev;
4561         }
4562         nvme_class->dev_uevent = nvme_class_uevent;
4563
4564         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4565         if (IS_ERR(nvme_subsys_class)) {
4566                 result = PTR_ERR(nvme_subsys_class);
4567                 goto destroy_class;
4568         }
4569
4570         result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4571                                      "nvme-generic");
4572         if (result < 0)
4573                 goto destroy_subsys_class;
4574
4575         nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
4576         if (IS_ERR(nvme_ns_chr_class)) {
4577                 result = PTR_ERR(nvme_ns_chr_class);
4578                 goto unregister_generic_ns;
4579         }
4580
4581         return 0;
4582
4583 unregister_generic_ns:
4584         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4585 destroy_subsys_class:
4586         class_destroy(nvme_subsys_class);
4587 destroy_class:
4588         class_destroy(nvme_class);
4589 unregister_chrdev:
4590         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4591 destroy_delete_wq:
4592         destroy_workqueue(nvme_delete_wq);
4593 destroy_reset_wq:
4594         destroy_workqueue(nvme_reset_wq);
4595 destroy_wq:
4596         destroy_workqueue(nvme_wq);
4597 out:
4598         return result;
4599 }
4600
4601 static void __exit nvme_core_exit(void)
4602 {
4603         class_destroy(nvme_ns_chr_class);
4604         class_destroy(nvme_subsys_class);
4605         class_destroy(nvme_class);
4606         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4607         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4608         destroy_workqueue(nvme_delete_wq);
4609         destroy_workqueue(nvme_reset_wq);
4610         destroy_workqueue(nvme_wq);
4611         ida_destroy(&nvme_ns_chr_minor_ida);
4612         ida_destroy(&nvme_instance_ida);
4613 }
4614
4615 MODULE_LICENSE("GPL");
4616 MODULE_VERSION("1.0");
4617 module_init(nvme_core_init);
4618 module_exit(nvme_core_exit);