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/t10-pi.h>
  23 #include <linux/pm_qos.h>
  24 #include <asm/unaligned.h>
  25
  26 #include "nvme.h"
  27 #include "fabrics.h"
  28
  29 #define CREATE_TRACE_POINTS
  30 #include "trace.h"
  31
  32 #define NVME_MINORS             (1U << MINORBITS)
  33
  34 unsigned int admin_timeout = 60;
  35 module_param(admin_timeout, uint, 0644);
  36 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  37 EXPORT_SYMBOL_GPL(admin_timeout);
  38
  39 unsigned int nvme_io_timeout = 30;
  40 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  41 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  42 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  43
  44 static unsigned char shutdown_timeout = 5;
  45 module_param(shutdown_timeout, byte, 0644);
  46 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  47
  48 static u8 nvme_max_retries = 5;
  49 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  50 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  51
  52 static unsigned long default_ps_max_latency_us = 100000;
  53 module_param(default_ps_max_latency_us, ulong, 0644);
  54 MODULE_PARM_DESC(default_ps_max_latency_us,
  55                  "max power saving latency for new devices; use PM QOS to change per device");
  56
  57 static bool force_apst;
  58 module_param(force_apst, bool, 0644);
  59 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  60
  61 static bool streams;
  62 module_param(streams, bool, 0644);
  63 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  64
  65 /*
  66  * nvme_wq - hosts nvme related works that are not reset or delete
  67  * nvme_reset_wq - hosts nvme reset works
  68  * nvme_delete_wq - hosts nvme delete works
  69  *
  70  * nvme_wq will host works such as scan, aen handling, fw activation,
  71  * keep-alive, periodic reconnects etc. nvme_reset_wq
  72  * runs reset works which also flush works hosted on nvme_wq for
  73  * serialization purposes. nvme_delete_wq host controller deletion
  74  * works which flush reset works for serialization.
  75  */
  76 struct workqueue_struct *nvme_wq;
  77 EXPORT_SYMBOL_GPL(nvme_wq);
  78
  79 struct workqueue_struct *nvme_reset_wq;
  80 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  81
  82 struct workqueue_struct *nvme_delete_wq;
  83 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  84
  85 static LIST_HEAD(nvme_subsystems);
  86 static DEFINE_MUTEX(nvme_subsystems_lock);
  87
  88 static DEFINE_IDA(nvme_instance_ida);
  89 static dev_t nvme_chr_devt;
  90 static struct class *nvme_class;
  91 static struct class *nvme_subsys_class;
  92
  93 static int nvme_revalidate_disk(struct gendisk *disk);
  94 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
  95 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
  96                                            unsigned nsid);
  97
  98 static void nvme_set_queue_dying(struct nvme_ns *ns)
  99 {
 100         /*
 101          * Revalidating a dead namespace sets capacity to 0. This will end
 102          * buffered writers dirtying pages that can't be synced.
 103          */
 104         if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 105                 return;
 106         blk_set_queue_dying(ns->queue);
 107         /* Forcibly unquiesce queues to avoid blocking dispatch */
 108         blk_mq_unquiesce_queue(ns->queue);
 109         /*
 110          * Revalidate after unblocking dispatchers that may be holding bd_butex
 111          */
 112         revalidate_disk(ns->disk);
 113 }
 114
 115 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 116 {
 117         /*
 118          * Only new queue scan work when admin and IO queues are both alive
 119          */
 120         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 121                 queue_work(nvme_wq, &ctrl->scan_work);
 122 }
 123
 124 /*
 125  * Use this function to proceed with scheduling reset_work for a controller
 126  * that had previously been set to the resetting state. This is intended for
 127  * code paths that can't be interrupted by other reset attempts. A hot removal
 128  * may prevent this from succeeding.
 129  */
 130 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 131 {
 132         if (ctrl->state != NVME_CTRL_RESETTING)
 133                 return -EBUSY;
 134         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 135                 return -EBUSY;
 136         return 0;
 137 }
 138 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 139
 140 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 141 {
 142         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 143                 return -EBUSY;
 144         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 145                 return -EBUSY;
 146         return 0;
 147 }
 148 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 149
 150 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 151 {
 152         int ret;
 153
 154         ret = nvme_reset_ctrl(ctrl);
 155         if (!ret) {
 156                 flush_work(&ctrl->reset_work);
 157                 if (ctrl->state != NVME_CTRL_LIVE)
 158                         ret = -ENETRESET;
 159         }
 160
 161         return ret;
 162 }
 163 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 164
 165 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 166 {
 167         dev_info(ctrl->device,
 168                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 169
 170         flush_work(&ctrl->reset_work);
 171         nvme_stop_ctrl(ctrl);
 172         nvme_remove_namespaces(ctrl);
 173         ctrl->ops->delete_ctrl(ctrl);
 174         nvme_uninit_ctrl(ctrl);
 175 }
 176
 177 static void nvme_delete_ctrl_work(struct work_struct *work)
 178 {
 179         struct nvme_ctrl *ctrl =
 180                 container_of(work, struct nvme_ctrl, delete_work);
 181
 182         nvme_do_delete_ctrl(ctrl);
 183 }
 184
 185 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 186 {
 187         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 188                 return -EBUSY;
 189         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 190                 return -EBUSY;
 191         return 0;
 192 }
 193 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 194
 195 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 196 {
 197         /*
 198          * Keep a reference until nvme_do_delete_ctrl() complete,
 199          * since ->delete_ctrl can free the controller.
 200          */
 201         nvme_get_ctrl(ctrl);
 202         if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 203                 nvme_do_delete_ctrl(ctrl);
 204         nvme_put_ctrl(ctrl);
 205 }
 206
 207 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 208 {
 209         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 210 }
 211
 212 static blk_status_t nvme_error_status(u16 status)
 213 {
 214         switch (status & 0x7ff) {
 215         case NVME_SC_SUCCESS:
 216                 return BLK_STS_OK;
 217         case NVME_SC_CAP_EXCEEDED:
 218                 return BLK_STS_NOSPC;
 219         case NVME_SC_LBA_RANGE:
 220         case NVME_SC_CMD_INTERRUPTED:
 221         case NVME_SC_NS_NOT_READY:
 222                 return BLK_STS_TARGET;
 223         case NVME_SC_BAD_ATTRIBUTES:
 224         case NVME_SC_ONCS_NOT_SUPPORTED:
 225         case NVME_SC_INVALID_OPCODE:
 226         case NVME_SC_INVALID_FIELD:
 227         case NVME_SC_INVALID_NS:
 228                 return BLK_STS_NOTSUPP;
 229         case NVME_SC_WRITE_FAULT:
 230         case NVME_SC_READ_ERROR:
 231         case NVME_SC_UNWRITTEN_BLOCK:
 232         case NVME_SC_ACCESS_DENIED:
 233         case NVME_SC_READ_ONLY:
 234         case NVME_SC_COMPARE_FAILED:
 235                 return BLK_STS_MEDIUM;
 236         case NVME_SC_GUARD_CHECK:
 237         case NVME_SC_APPTAG_CHECK:
 238         case NVME_SC_REFTAG_CHECK:
 239         case NVME_SC_INVALID_PI:
 240                 return BLK_STS_PROTECTION;
 241         case NVME_SC_RESERVATION_CONFLICT:
 242                 return BLK_STS_NEXUS;
 243         case NVME_SC_HOST_PATH_ERROR:
 244                 return BLK_STS_TRANSPORT;
 245         default:
 246                 return BLK_STS_IOERR;
 247         }
 248 }
 249
 250 static inline bool nvme_req_needs_retry(struct request *req)
 251 {
 252         if (blk_noretry_request(req))
 253                 return false;
 254         if (nvme_req(req)->status & NVME_SC_DNR)
 255                 return false;
 256         if (nvme_req(req)->retries >= nvme_max_retries)
 257                 return false;
 258         return true;
 259 }
 260
 261 static void nvme_retry_req(struct request *req)
 262 {
 263         struct nvme_ns *ns = req->q->queuedata;
 264         unsigned long delay = 0;
 265         u16 crd;
 266
 267         /* The mask and shift result must be <= 3 */
 268         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 269         if (ns && crd)
 270                 delay = ns->ctrl->crdt[crd - 1] * 100;
 271
 272         nvme_req(req)->retries++;
 273         blk_mq_requeue_request(req, false);
 274         blk_mq_delay_kick_requeue_list(req->q, delay);
 275 }
 276
 277 void nvme_complete_rq(struct request *req)
 278 {
 279         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 280
 281         trace_nvme_complete_rq(req);
 282
 283         nvme_cleanup_cmd(req);
 284
 285         if (nvme_req(req)->ctrl->kas)
 286                 nvme_req(req)->ctrl->comp_seen = true;
 287
 288         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 289                 if ((req->cmd_flags & REQ_NVME_MPATH) && nvme_failover_req(req))
 290                         return;
 291
 292                 if (!blk_queue_dying(req->q)) {
 293                         nvme_retry_req(req);
 294                         return;
 295                 }
 296         }
 297
 298         nvme_trace_bio_complete(req, status);
 299         blk_mq_end_request(req, status);
 300 }
 301 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 302
 303 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 304 {
 305         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 306                                 "Cancelling I/O %d", req->tag);
 307
 308         /* don't abort one completed request */
 309         if (blk_mq_request_completed(req))
 310                 return true;
 311
 312         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 313         blk_mq_complete_request(req);
 314         return true;
 315 }
 316 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 317
 318 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 319                 enum nvme_ctrl_state new_state)
 320 {
 321         enum nvme_ctrl_state old_state;
 322         unsigned long flags;
 323         bool changed = false;
 324
 325         spin_lock_irqsave(&ctrl->lock, flags);
 326
 327         old_state = ctrl->state;
 328         switch (new_state) {
 329         case NVME_CTRL_LIVE:
 330                 switch (old_state) {
 331                 case NVME_CTRL_NEW:
 332                 case NVME_CTRL_RESETTING:
 333                 case NVME_CTRL_CONNECTING:
 334                         changed = true;
 335                         /* FALLTHRU */
 336                 default:
 337                         break;
 338                 }
 339                 break;
 340         case NVME_CTRL_RESETTING:
 341                 switch (old_state) {
 342                 case NVME_CTRL_NEW:
 343                 case NVME_CTRL_LIVE:
 344                         changed = true;
 345                         /* FALLTHRU */
 346                 default:
 347                         break;
 348                 }
 349                 break;
 350         case NVME_CTRL_CONNECTING:
 351                 switch (old_state) {
 352                 case NVME_CTRL_NEW:
 353                 case NVME_CTRL_RESETTING:
 354                         changed = true;
 355                         /* FALLTHRU */
 356                 default:
 357                         break;
 358                 }
 359                 break;
 360         case NVME_CTRL_DELETING:
 361                 switch (old_state) {
 362                 case NVME_CTRL_LIVE:
 363                 case NVME_CTRL_RESETTING:
 364                 case NVME_CTRL_CONNECTING:
 365                         changed = true;
 366                         /* FALLTHRU */
 367                 default:
 368                         break;
 369                 }
 370                 break;
 371         case NVME_CTRL_DEAD:
 372                 switch (old_state) {
 373                 case NVME_CTRL_DELETING:
 374                         changed = true;
 375                         /* FALLTHRU */
 376                 default:
 377                         break;
 378                 }
 379                 break;
 380         default:
 381                 break;
 382         }
 383
 384         if (changed) {
 385                 ctrl->state = new_state;
 386                 wake_up_all(&ctrl->state_wq);
 387         }
 388
 389         spin_unlock_irqrestore(&ctrl->lock, flags);
 390         if (changed && ctrl->state == NVME_CTRL_LIVE)
 391                 nvme_kick_requeue_lists(ctrl);
 392         return changed;
 393 }
 394 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 395
 396 /*
 397  * Returns true for sink states that can't ever transition back to live.
 398  */
 399 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 400 {
 401         switch (ctrl->state) {
 402         case NVME_CTRL_NEW:
 403         case NVME_CTRL_LIVE:
 404         case NVME_CTRL_RESETTING:
 405         case NVME_CTRL_CONNECTING:
 406                 return false;
 407         case NVME_CTRL_DELETING:
 408         case NVME_CTRL_DEAD:
 409                 return true;
 410         default:
 411                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 412                 return true;
 413         }
 414 }
 415
 416 /*
 417  * Waits for the controller state to be resetting, or returns false if it is
 418  * not possible to ever transition to that state.
 419  */
 420 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 421 {
 422         wait_event(ctrl->state_wq,
 423                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 424                    nvme_state_terminal(ctrl));
 425         return ctrl->state == NVME_CTRL_RESETTING;
 426 }
 427 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 428
 429 static void nvme_free_ns_head(struct kref *ref)
 430 {
 431         struct nvme_ns_head *head =
 432                 container_of(ref, struct nvme_ns_head, ref);
 433
 434         nvme_mpath_remove_disk(head);
 435         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 436         cleanup_srcu_struct(&head->srcu);
 437         nvme_put_subsystem(head->subsys);
 438         kfree(head);
 439 }
 440
 441 static void nvme_put_ns_head(struct nvme_ns_head *head)
 442 {
 443         kref_put(&head->ref, nvme_free_ns_head);
 444 }
 445
 446 static void nvme_free_ns(struct kref *kref)
 447 {
 448         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 449
 450         if (ns->ndev)
 451                 nvme_nvm_unregister(ns);
 452
 453         put_disk(ns->disk);
 454         nvme_put_ns_head(ns->head);
 455         nvme_put_ctrl(ns->ctrl);
 456         kfree(ns);
 457 }
 458
 459 static void nvme_put_ns(struct nvme_ns *ns)
 460 {
 461         kref_put(&ns->kref, nvme_free_ns);
 462 }
 463
 464 static inline void nvme_clear_nvme_request(struct request *req)
 465 {
 466         if (!(req->rq_flags & RQF_DONTPREP)) {
 467                 nvme_req(req)->retries = 0;
 468                 nvme_req(req)->flags = 0;
 469                 req->rq_flags |= RQF_DONTPREP;
 470         }
 471 }
 472
 473 struct request *nvme_alloc_request(struct request_queue *q,
 474                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 475 {
 476         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 477         struct request *req;
 478
 479         if (qid == NVME_QID_ANY) {
 480                 req = blk_mq_alloc_request(q, op, flags);
 481         } else {
 482                 req = blk_mq_alloc_request_hctx(q, op, flags,
 483                                 qid ? qid - 1 : 0);
 484         }
 485         if (IS_ERR(req))
 486                 return req;
 487
 488         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 489         nvme_clear_nvme_request(req);
 490         nvme_req(req)->cmd = cmd;
 491
 492         return req;
 493 }
 494 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 495
 496 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 497 {
 498         struct nvme_command c;
 499
 500         memset(&c, 0, sizeof(c));
 501
 502         c.directive.opcode = nvme_admin_directive_send;
 503         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 504         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 505         c.directive.dtype = NVME_DIR_IDENTIFY;
 506         c.directive.tdtype = NVME_DIR_STREAMS;
 507         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 508
 509         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 510 }
 511
 512 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 513 {
 514         return nvme_toggle_streams(ctrl, false);
 515 }
 516
 517 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 518 {
 519         return nvme_toggle_streams(ctrl, true);
 520 }
 521
 522 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 523                                   struct streams_directive_params *s, u32 nsid)
 524 {
 525         struct nvme_command c;
 526
 527         memset(&c, 0, sizeof(c));
 528         memset(s, 0, sizeof(*s));
 529
 530         c.directive.opcode = nvme_admin_directive_recv;
 531         c.directive.nsid = cpu_to_le32(nsid);
 532         c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
 533         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 534         c.directive.dtype = NVME_DIR_STREAMS;
 535
 536         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 537 }
 538
 539 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 540 {
 541         struct streams_directive_params s;
 542         int ret;
 543
 544         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 545                 return 0;
 546         if (!streams)
 547                 return 0;
 548
 549         ret = nvme_enable_streams(ctrl);
 550         if (ret)
 551                 return ret;
 552
 553         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 554         if (ret)
 555                 return ret;
 556
 557         ctrl->nssa = le16_to_cpu(s.nssa);
 558         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 559                 dev_info(ctrl->device, "too few streams (%u) available\n",
 560                                         ctrl->nssa);
 561                 nvme_disable_streams(ctrl);
 562                 return 0;
 563         }
 564
 565         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 566         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 567         return 0;
 568 }
 569
 570 /*
 571  * Check if 'req' has a write hint associated with it. If it does, assign
 572  * a valid namespace stream to the write.
 573  */
 574 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 575                                      struct request *req, u16 *control,
 576                                      u32 *dsmgmt)
 577 {
 578         enum rw_hint streamid = req->write_hint;
 579
 580         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 581                 streamid = 0;
 582         else {
 583                 streamid--;
 584                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 585                         return;
 586
 587                 *control |= NVME_RW_DTYPE_STREAMS;
 588                 *dsmgmt |= streamid << 16;
 589         }
 590
 591         if (streamid < ARRAY_SIZE(req->q->write_hints))
 592                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 593 }
 594
 595 static inline void nvme_setup_flush(struct nvme_ns *ns,
 596                 struct nvme_command *cmnd)
 597 {
 598         cmnd->common.opcode = nvme_cmd_flush;
 599         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 600 }
 601
 602 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 603                 struct nvme_command *cmnd)
 604 {
 605         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 606         struct nvme_dsm_range *range;
 607         struct bio *bio;
 608
 609         /*
 610          * Some devices do not consider the DSM 'Number of Ranges' field when
 611          * determining how much data to DMA. Always allocate memory for maximum
 612          * number of segments to prevent device reading beyond end of buffer.
 613          */
 614         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 615
 616         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 617         if (!range) {
 618                 /*
 619                  * If we fail allocation our range, fallback to the controller
 620                  * discard page. If that's also busy, it's safe to return
 621                  * busy, as we know we can make progress once that's freed.
 622                  */
 623                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 624                         return BLK_STS_RESOURCE;
 625
 626                 range = page_address(ns->ctrl->discard_page);
 627         }
 628
 629         __rq_for_each_bio(bio, req) {
 630                 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 631                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 632
 633                 if (n < segments) {
 634                         range[n].cattr = cpu_to_le32(0);
 635                         range[n].nlb = cpu_to_le32(nlb);
 636                         range[n].slba = cpu_to_le64(slba);
 637                 }
 638                 n++;
 639         }
 640
 641         if (WARN_ON_ONCE(n != segments)) {
 642                 if (virt_to_page(range) == ns->ctrl->discard_page)
 643                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 644                 else
 645                         kfree(range);
 646                 return BLK_STS_IOERR;
 647         }
 648
 649         cmnd->dsm.opcode = nvme_cmd_dsm;
 650         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 651         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 652         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 653
 654         req->special_vec.bv_page = virt_to_page(range);
 655         req->special_vec.bv_offset = offset_in_page(range);
 656         req->special_vec.bv_len = alloc_size;
 657         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 658
 659         return BLK_STS_OK;
 660 }
 661
 662 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 663                 struct request *req, struct nvme_command *cmnd)
 664 {
 665         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 666                 return nvme_setup_discard(ns, req, cmnd);
 667
 668         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 669         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 670         cmnd->write_zeroes.slba =
 671                 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 672         cmnd->write_zeroes.length =
 673                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 674         cmnd->write_zeroes.control = 0;
 675         return BLK_STS_OK;
 676 }
 677
 678 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 679                 struct request *req, struct nvme_command *cmnd)
 680 {
 681         struct nvme_ctrl *ctrl = ns->ctrl;
 682         u16 control = 0;
 683         u32 dsmgmt = 0;
 684
 685         if (req->cmd_flags & REQ_FUA)
 686                 control |= NVME_RW_FUA;
 687         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 688                 control |= NVME_RW_LR;
 689
 690         if (req->cmd_flags & REQ_RAHEAD)
 691                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 692
 693         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 694         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 695         cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 696         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 697
 698         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 699                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 700
 701         if (ns->ms) {
 702                 /*
 703                  * If formated with metadata, the block layer always provides a
 704                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 705                  * we enable the PRACT bit for protection information or set the
 706                  * namespace capacity to zero to prevent any I/O.
 707                  */
 708                 if (!blk_integrity_rq(req)) {
 709                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 710                                 return BLK_STS_NOTSUPP;
 711                         control |= NVME_RW_PRINFO_PRACT;
 712                 }
 713
 714                 switch (ns->pi_type) {
 715                 case NVME_NS_DPS_PI_TYPE3:
 716                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 717                         break;
 718                 case NVME_NS_DPS_PI_TYPE1:
 719                 case NVME_NS_DPS_PI_TYPE2:
 720                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 721                                         NVME_RW_PRINFO_PRCHK_REF;
 722                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 723                         break;
 724                 }
 725         }
 726
 727         cmnd->rw.control = cpu_to_le16(control);
 728         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 729         return 0;
 730 }
 731
 732 void nvme_cleanup_cmd(struct request *req)
 733 {
 734         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 735                 struct nvme_ns *ns = req->rq_disk->private_data;
 736                 struct page *page = req->special_vec.bv_page;
 737
 738                 if (page == ns->ctrl->discard_page)
 739                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 740                 else
 741                         kfree(page_address(page) + req->special_vec.bv_offset);
 742         }
 743 }
 744 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 745
 746 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 747                 struct nvme_command *cmd)
 748 {
 749         blk_status_t ret = BLK_STS_OK;
 750
 751         nvme_clear_nvme_request(req);
 752
 753         memset(cmd, 0, sizeof(*cmd));
 754         switch (req_op(req)) {
 755         case REQ_OP_DRV_IN:
 756         case REQ_OP_DRV_OUT:
 757                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 758                 break;
 759         case REQ_OP_FLUSH:
 760                 nvme_setup_flush(ns, cmd);
 761                 break;
 762         case REQ_OP_WRITE_ZEROES:
 763                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 764                 break;
 765         case REQ_OP_DISCARD:
 766                 ret = nvme_setup_discard(ns, req, cmd);
 767                 break;
 768         case REQ_OP_READ:
 769         case REQ_OP_WRITE:
 770                 ret = nvme_setup_rw(ns, req, cmd);
 771                 break;
 772         default:
 773                 WARN_ON_ONCE(1);
 774                 return BLK_STS_IOERR;
 775         }
 776
 777         cmd->common.command_id = req->tag;
 778         trace_nvme_setup_cmd(req, cmd);
 779         return ret;
 780 }
 781 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 782
 783 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 784 {
 785         struct completion *waiting = rq->end_io_data;
 786
 787         rq->end_io_data = NULL;
 788         complete(waiting);
 789 }
 790
 791 static void nvme_execute_rq_polled(struct request_queue *q,
 792                 struct gendisk *bd_disk, struct request *rq, int at_head)
 793 {
 794         DECLARE_COMPLETION_ONSTACK(wait);
 795
 796         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
 797
 798         rq->cmd_flags |= REQ_HIPRI;
 799         rq->end_io_data = &wait;
 800         blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
 801
 802         while (!completion_done(&wait)) {
 803                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
 804                 cond_resched();
 805         }
 806 }
 807
 808 /*
 809  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 810  * if the result is positive, it's an NVM Express status code
 811  */
 812 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 813                 union nvme_result *result, void *buffer, unsigned bufflen,
 814                 unsigned timeout, int qid, int at_head,
 815                 blk_mq_req_flags_t flags, bool poll)
 816 {
 817         struct request *req;
 818         int ret;
 819
 820         req = nvme_alloc_request(q, cmd, flags, qid);
 821         if (IS_ERR(req))
 822                 return PTR_ERR(req);
 823
 824         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 825
 826         if (buffer && bufflen) {
 827                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 828                 if (ret)
 829                         goto out;
 830         }
 831
 832         if (poll)
 833                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
 834         else
 835                 blk_execute_rq(req->q, NULL, req, at_head);
 836         if (result)
 837                 *result = nvme_req(req)->result;
 838         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 839                 ret = -EINTR;
 840         else
 841                 ret = nvme_req(req)->status;
 842  out:
 843         blk_mq_free_request(req);
 844         return ret;
 845 }
 846 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 847
 848 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 849                 void *buffer, unsigned bufflen)
 850 {
 851         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 852                         NVME_QID_ANY, 0, 0, false);
 853 }
 854 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 855
 856 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 857                 unsigned len, u32 seed, bool write)
 858 {
 859         struct bio_integrity_payload *bip;
 860         int ret = -ENOMEM;
 861         void *buf;
 862
 863         buf = kmalloc(len, GFP_KERNEL);
 864         if (!buf)
 865                 goto out;
 866
 867         ret = -EFAULT;
 868         if (write && copy_from_user(buf, ubuf, len))
 869                 goto out_free_meta;
 870
 871         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 872         if (IS_ERR(bip)) {
 873                 ret = PTR_ERR(bip);
 874                 goto out_free_meta;
 875         }
 876
 877         bip->bip_iter.bi_size = len;
 878         bip->bip_iter.bi_sector = seed;
 879         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 880                         offset_in_page(buf));
 881         if (ret == len)
 882                 return buf;
 883         ret = -ENOMEM;
 884 out_free_meta:
 885         kfree(buf);
 886 out:
 887         return ERR_PTR(ret);
 888 }
 889
 890 static int nvme_submit_user_cmd(struct request_queue *q,
 891                 struct nvme_command *cmd, void __user *ubuffer,
 892                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 893                 u32 meta_seed, u64 *result, unsigned timeout)
 894 {
 895         bool write = nvme_is_write(cmd);
 896         struct nvme_ns *ns = q->queuedata;
 897         struct gendisk *disk = ns ? ns->disk : NULL;
 898         struct request *req;
 899         struct bio *bio = NULL;
 900         void *meta = NULL;
 901         int ret;
 902
 903         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 904         if (IS_ERR(req))
 905                 return PTR_ERR(req);
 906
 907         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 908         nvme_req(req)->flags |= NVME_REQ_USERCMD;
 909
 910         if (ubuffer && bufflen) {
 911                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 912                                 GFP_KERNEL);
 913                 if (ret)
 914                         goto out;
 915                 bio = req->bio;
 916                 bio->bi_disk = disk;
 917                 if (disk && meta_buffer && meta_len) {
 918                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 919                                         meta_seed, write);
 920                         if (IS_ERR(meta)) {
 921                                 ret = PTR_ERR(meta);
 922                                 goto out_unmap;
 923                         }
 924                         req->cmd_flags |= REQ_INTEGRITY;
 925                 }
 926         }
 927
 928         blk_execute_rq(req->q, disk, req, 0);
 929         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 930                 ret = -EINTR;
 931         else
 932                 ret = nvme_req(req)->status;
 933         if (result)
 934                 *result = le64_to_cpu(nvme_req(req)->result.u64);
 935         if (meta && !ret && !write) {
 936                 if (copy_to_user(meta_buffer, meta, meta_len))
 937                         ret = -EFAULT;
 938         }
 939         kfree(meta);
 940  out_unmap:
 941         if (bio)
 942                 blk_rq_unmap_user(bio);
 943  out:
 944         blk_mq_free_request(req);
 945         return ret;
 946 }
 947
 948 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 949 {
 950         struct nvme_ctrl *ctrl = rq->end_io_data;
 951         unsigned long flags;
 952         bool startka = false;
 953
 954         blk_mq_free_request(rq);
 955
 956         if (status) {
 957                 dev_err(ctrl->device,
 958                         "failed nvme_keep_alive_end_io error=%d\n",
 959                                 status);
 960                 return;
 961         }
 962
 963         ctrl->comp_seen = false;
 964         spin_lock_irqsave(&ctrl->lock, flags);
 965         if (ctrl->state == NVME_CTRL_LIVE ||
 966             ctrl->state == NVME_CTRL_CONNECTING)
 967                 startka = true;
 968         spin_unlock_irqrestore(&ctrl->lock, flags);
 969         if (startka)
 970                 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
 971 }
 972
 973 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 974 {
 975         struct request *rq;
 976
 977         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
 978                         NVME_QID_ANY);
 979         if (IS_ERR(rq))
 980                 return PTR_ERR(rq);
 981
 982         rq->timeout = ctrl->kato * HZ;
 983         rq->end_io_data = ctrl;
 984
 985         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 986
 987         return 0;
 988 }
 989
 990 static void nvme_keep_alive_work(struct work_struct *work)
 991 {
 992         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 993                         struct nvme_ctrl, ka_work);
 994         bool comp_seen = ctrl->comp_seen;
 995
 996         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
 997                 dev_dbg(ctrl->device,
 998                         "reschedule traffic based keep-alive timer\n");
 999                 ctrl->comp_seen = false;
1000                 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1001                 return;
1002         }
1003
1004         if (nvme_keep_alive(ctrl)) {
1005                 /* allocation failure, reset the controller */
1006                 dev_err(ctrl->device, "keep-alive failed\n");
1007                 nvme_reset_ctrl(ctrl);
1008                 return;
1009         }
1010 }
1011
1012 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1013 {
1014         if (unlikely(ctrl->kato == 0))
1015                 return;
1016
1017         queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1018 }
1019
1020 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1021 {
1022         if (unlikely(ctrl->kato == 0))
1023                 return;
1024
1025         cancel_delayed_work_sync(&ctrl->ka_work);
1026 }
1027 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1028
1029 /*
1030  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1031  * flag, thus sending any new CNS opcodes has a big chance of not working.
1032  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1033  * (but not for any later version).
1034  */
1035 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1036 {
1037         if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1038                 return ctrl->vs < NVME_VS(1, 2, 0);
1039         return ctrl->vs < NVME_VS(1, 1, 0);
1040 }
1041
1042 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1043 {
1044         struct nvme_command c = { };
1045         int error;
1046
1047         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1048         c.identify.opcode = nvme_admin_identify;
1049         c.identify.cns = NVME_ID_CNS_CTRL;
1050
1051         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1052         if (!*id)
1053                 return -ENOMEM;
1054
1055         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1056                         sizeof(struct nvme_id_ctrl));
1057         if (error)
1058                 kfree(*id);
1059         return error;
1060 }
1061
1062 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1063                 struct nvme_ns_id_desc *cur)
1064 {
1065         const char *warn_str = "ctrl returned bogus length:";
1066         void *data = cur;
1067
1068         switch (cur->nidt) {
1069         case NVME_NIDT_EUI64:
1070                 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1071                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1072                                  warn_str, cur->nidl);
1073                         return -1;
1074                 }
1075                 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1076                 return NVME_NIDT_EUI64_LEN;
1077         case NVME_NIDT_NGUID:
1078                 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1079                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1080                                  warn_str, cur->nidl);
1081                         return -1;
1082                 }
1083                 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1084                 return NVME_NIDT_NGUID_LEN;
1085         case NVME_NIDT_UUID:
1086                 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1087                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1088                                  warn_str, cur->nidl);
1089                         return -1;
1090                 }
1091                 uuid_copy(&ids->uuid, data + sizeof(*cur));
1092                 return NVME_NIDT_UUID_LEN;
1093         default:
1094                 /* Skip unknown types */
1095                 return cur->nidl;
1096         }
1097 }
1098
1099 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1100                 struct nvme_ns_ids *ids)
1101 {
1102         struct nvme_command c = { };
1103         int status;
1104         void *data;
1105         int pos;
1106         int len;
1107
1108         c.identify.opcode = nvme_admin_identify;
1109         c.identify.nsid = cpu_to_le32(nsid);
1110         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1111
1112         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1113         if (!data)
1114                 return -ENOMEM;
1115
1116         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1117                                       NVME_IDENTIFY_DATA_SIZE);
1118         if (status) {
1119                 dev_warn(ctrl->device,
1120                         "Identify Descriptors failed (%d)\n", status);
1121                  /*
1122                   * Don't treat an error as fatal, as we potentially already
1123                   * have a NGUID or EUI-64.
1124                   */
1125                 if (status > 0 && !(status & NVME_SC_DNR))
1126                         status = 0;
1127                 goto free_data;
1128         }
1129
1130         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1131                 struct nvme_ns_id_desc *cur = data + pos;
1132
1133                 if (cur->nidl == 0)
1134                         break;
1135
1136                 len = nvme_process_ns_desc(ctrl, ids, cur);
1137                 if (len < 0)
1138                         goto free_data;
1139
1140                 len += sizeof(*cur);
1141         }
1142 free_data:
1143         kfree(data);
1144         return status;
1145 }
1146
1147 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1148 {
1149         struct nvme_command c = { };
1150
1151         c.identify.opcode = nvme_admin_identify;
1152         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1153         c.identify.nsid = cpu_to_le32(nsid);
1154         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1155                                     NVME_IDENTIFY_DATA_SIZE);
1156 }
1157
1158 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1159                 unsigned nsid, struct nvme_id_ns **id)
1160 {
1161         struct nvme_command c = { };
1162         int error;
1163
1164         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1165         c.identify.opcode = nvme_admin_identify;
1166         c.identify.nsid = cpu_to_le32(nsid);
1167         c.identify.cns = NVME_ID_CNS_NS;
1168
1169         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1170         if (!*id)
1171                 return -ENOMEM;
1172
1173         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1174         if (error) {
1175                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1176                 kfree(*id);
1177         }
1178
1179         return error;
1180 }
1181
1182 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1183                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1184 {
1185         union nvme_result res = { 0 };
1186         struct nvme_command c;
1187         int ret;
1188
1189         memset(&c, 0, sizeof(c));
1190         c.features.opcode = op;
1191         c.features.fid = cpu_to_le32(fid);
1192         c.features.dword11 = cpu_to_le32(dword11);
1193
1194         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1195                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1196         if (ret >= 0 && result)
1197                 *result = le32_to_cpu(res.u32);
1198         return ret;
1199 }
1200
1201 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1202                       unsigned int dword11, void *buffer, size_t buflen,
1203                       u32 *result)
1204 {
1205         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1206                              buflen, result);
1207 }
1208 EXPORT_SYMBOL_GPL(nvme_set_features);
1209
1210 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1211                       unsigned int dword11, void *buffer, size_t buflen,
1212                       u32 *result)
1213 {
1214         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1215                              buflen, result);
1216 }
1217 EXPORT_SYMBOL_GPL(nvme_get_features);
1218
1219 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1220 {
1221         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1222         u32 result;
1223         int status, nr_io_queues;
1224
1225         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1226                         &result);
1227         if (status < 0)
1228                 return status;
1229
1230         /*
1231          * Degraded controllers might return an error when setting the queue
1232          * count.  We still want to be able to bring them online and offer
1233          * access to the admin queue, as that might be only way to fix them up.
1234          */
1235         if (status > 0) {
1236                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1237                 *count = 0;
1238         } else {
1239                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1240                 *count = min(*count, nr_io_queues);
1241         }
1242
1243         return 0;
1244 }
1245 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1246
1247 #define NVME_AEN_SUPPORTED \
1248         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1249          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1250
1251 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1252 {
1253         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1254         int status;
1255
1256         if (!supported_aens)
1257                 return;
1258
1259         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1260                         NULL, 0, &result);
1261         if (status)
1262                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1263                          supported_aens);
1264
1265         queue_work(nvme_wq, &ctrl->async_event_work);
1266 }
1267
1268 /*
1269  * Convert integer values from ioctl structures to user pointers, silently
1270  * ignoring the upper bits in the compat case to match behaviour of 32-bit
1271  * kernels.
1272  */
1273 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1274 {
1275         if (in_compat_syscall())
1276                 ptrval = (compat_uptr_t)ptrval;
1277         return (void __user *)ptrval;
1278 }
1279
1280 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1281 {
1282         struct nvme_user_io io;
1283         struct nvme_command c;
1284         unsigned length, meta_len;
1285         void __user *metadata;
1286
1287         if (copy_from_user(&io, uio, sizeof(io)))
1288                 return -EFAULT;
1289         if (io.flags)
1290                 return -EINVAL;
1291
1292         switch (io.opcode) {
1293         case nvme_cmd_write:
1294         case nvme_cmd_read:
1295         case nvme_cmd_compare:
1296                 break;
1297         default:
1298                 return -EINVAL;
1299         }
1300
1301         length = (io.nblocks + 1) << ns->lba_shift;
1302         meta_len = (io.nblocks + 1) * ns->ms;
1303         metadata = nvme_to_user_ptr(io.metadata);
1304
1305         if (ns->ext) {
1306                 length += meta_len;
1307                 meta_len = 0;
1308         } else if (meta_len) {
1309                 if ((io.metadata & 3) || !io.metadata)
1310                         return -EINVAL;
1311         }
1312
1313         memset(&c, 0, sizeof(c));
1314         c.rw.opcode = io.opcode;
1315         c.rw.flags = io.flags;
1316         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1317         c.rw.slba = cpu_to_le64(io.slba);
1318         c.rw.length = cpu_to_le16(io.nblocks);
1319         c.rw.control = cpu_to_le16(io.control);
1320         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1321         c.rw.reftag = cpu_to_le32(io.reftag);
1322         c.rw.apptag = cpu_to_le16(io.apptag);
1323         c.rw.appmask = cpu_to_le16(io.appmask);
1324
1325         return nvme_submit_user_cmd(ns->queue, &c,
1326                         nvme_to_user_ptr(io.addr), length,
1327                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1328 }
1329
1330 static u32 nvme_known_admin_effects(u8 opcode)
1331 {
1332         switch (opcode) {
1333         case nvme_admin_format_nvm:
1334                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1335                                         NVME_CMD_EFFECTS_CSE_MASK;
1336         case nvme_admin_sanitize_nvm:
1337                 return NVME_CMD_EFFECTS_CSE_MASK;
1338         default:
1339                 break;
1340         }
1341         return 0;
1342 }
1343
1344 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1345                                                                 u8 opcode)
1346 {
1347         u32 effects = 0;
1348
1349         if (ns) {
1350                 if (ctrl->effects)
1351                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1352                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1353                         dev_warn(ctrl->device,
1354                                  "IO command:%02x has unhandled effects:%08x\n",
1355                                  opcode, effects);
1356                 return 0;
1357         }
1358
1359         if (ctrl->effects)
1360                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1361         effects |= nvme_known_admin_effects(opcode);
1362
1363         /*
1364          * For simplicity, IO to all namespaces is quiesced even if the command
1365          * effects say only one namespace is affected.
1366          */
1367         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1368                 mutex_lock(&ctrl->scan_lock);
1369                 mutex_lock(&ctrl->subsys->lock);
1370                 nvme_mpath_start_freeze(ctrl->subsys);
1371                 nvme_mpath_wait_freeze(ctrl->subsys);
1372                 nvme_start_freeze(ctrl);
1373                 nvme_wait_freeze(ctrl);
1374         }
1375         return effects;
1376 }
1377
1378 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1379 {
1380         struct nvme_ns *ns;
1381
1382         down_read(&ctrl->namespaces_rwsem);
1383         list_for_each_entry(ns, &ctrl->namespaces, list)
1384                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1385                         nvme_set_queue_dying(ns);
1386         up_read(&ctrl->namespaces_rwsem);
1387 }
1388
1389 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1390 {
1391         /*
1392          * Revalidate LBA changes prior to unfreezing. This is necessary to
1393          * prevent memory corruption if a logical block size was changed by
1394          * this command.
1395          */
1396         if (effects & NVME_CMD_EFFECTS_LBCC)
1397                 nvme_update_formats(ctrl);
1398         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1399                 nvme_unfreeze(ctrl);
1400                 nvme_mpath_unfreeze(ctrl->subsys);
1401                 mutex_unlock(&ctrl->subsys->lock);
1402                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1403                 mutex_unlock(&ctrl->scan_lock);
1404         }
1405         if (effects & NVME_CMD_EFFECTS_CCC)
1406                 nvme_init_identify(ctrl);
1407         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1408                 nvme_queue_scan(ctrl);
1409 }
1410
1411 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1412                         struct nvme_passthru_cmd __user *ucmd)
1413 {
1414         struct nvme_passthru_cmd cmd;
1415         struct nvme_command c;
1416         unsigned timeout = 0;
1417         u32 effects;
1418         u64 result;
1419         int status;
1420
1421         if (!capable(CAP_SYS_ADMIN))
1422                 return -EACCES;
1423         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1424                 return -EFAULT;
1425         if (cmd.flags)
1426                 return -EINVAL;
1427
1428         memset(&c, 0, sizeof(c));
1429         c.common.opcode = cmd.opcode;
1430         c.common.flags = cmd.flags;
1431         c.common.nsid = cpu_to_le32(cmd.nsid);
1432         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1433         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1434         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1435         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1436         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1437         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1438         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1439         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1440
1441         if (cmd.timeout_ms)
1442                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1443
1444         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1445         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1446                         nvme_to_user_ptr(cmd.addr), cmd.data_len,
1447                         nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1448                         0, &result, timeout);
1449         nvme_passthru_end(ctrl, effects);
1450
1451         if (status >= 0) {
1452                 if (put_user(result, &ucmd->result))
1453                         return -EFAULT;
1454         }
1455
1456         return status;
1457 }
1458
1459 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1460                         struct nvme_passthru_cmd64 __user *ucmd)
1461 {
1462         struct nvme_passthru_cmd64 cmd;
1463         struct nvme_command c;
1464         unsigned timeout = 0;
1465         u32 effects;
1466         int status;
1467
1468         if (!capable(CAP_SYS_ADMIN))
1469                 return -EACCES;
1470         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1471                 return -EFAULT;
1472         if (cmd.flags)
1473                 return -EINVAL;
1474
1475         memset(&c, 0, sizeof(c));
1476         c.common.opcode = cmd.opcode;
1477         c.common.flags = cmd.flags;
1478         c.common.nsid = cpu_to_le32(cmd.nsid);
1479         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1480         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1481         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1482         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1483         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1484         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1485         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1486         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1487
1488         if (cmd.timeout_ms)
1489                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1490
1491         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1492         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1493                         nvme_to_user_ptr(cmd.addr), cmd.data_len,
1494                         nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1495                         0, &cmd.result, timeout);
1496         nvme_passthru_end(ctrl, effects);
1497
1498         if (status >= 0) {
1499                 if (put_user(cmd.result, &ucmd->result))
1500                         return -EFAULT;
1501         }
1502
1503         return status;
1504 }
1505
1506 /*
1507  * Issue ioctl requests on the first available path.  Note that unlike normal
1508  * block layer requests we will not retry failed request on another controller.
1509  */
1510 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1511                 struct nvme_ns_head **head, int *srcu_idx)
1512 {
1513 #ifdef CONFIG_NVME_MULTIPATH
1514         if (disk->fops == &nvme_ns_head_ops) {
1515                 struct nvme_ns *ns;
1516
1517                 *head = disk->private_data;
1518                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1519                 ns = nvme_find_path(*head);
1520                 if (!ns)
1521                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1522                 return ns;
1523         }
1524 #endif
1525         *head = NULL;
1526         *srcu_idx = -1;
1527         return disk->private_data;
1528 }
1529
1530 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1531 {
1532         if (head)
1533                 srcu_read_unlock(&head->srcu, idx);
1534 }
1535
1536 static bool is_ctrl_ioctl(unsigned int cmd)
1537 {
1538         if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1539                 return true;
1540         if (is_sed_ioctl(cmd))
1541                 return true;
1542         return false;
1543 }
1544
1545 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1546                                   void __user *argp,
1547                                   struct nvme_ns_head *head,
1548                                   int srcu_idx)
1549 {
1550         struct nvme_ctrl *ctrl = ns->ctrl;
1551         int ret;
1552
1553         nvme_get_ctrl(ns->ctrl);
1554         nvme_put_ns_from_disk(head, srcu_idx);
1555
1556         switch (cmd) {
1557         case NVME_IOCTL_ADMIN_CMD:
1558                 ret = nvme_user_cmd(ctrl, NULL, argp);
1559                 break;
1560         case NVME_IOCTL_ADMIN64_CMD:
1561                 ret = nvme_user_cmd64(ctrl, NULL, argp);
1562                 break;
1563         default:
1564                 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1565                 break;
1566         }
1567         nvme_put_ctrl(ctrl);
1568         return ret;
1569 }
1570
1571 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1572                 unsigned int cmd, unsigned long arg)
1573 {
1574         struct nvme_ns_head *head = NULL;
1575         void __user *argp = (void __user *)arg;
1576         struct nvme_ns *ns;
1577         int srcu_idx, ret;
1578
1579         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1580         if (unlikely(!ns))
1581                 return -EWOULDBLOCK;
1582
1583         /*
1584          * Handle ioctls that apply to the controller instead of the namespace
1585          * seperately and drop the ns SRCU reference early.  This avoids a
1586          * deadlock when deleting namespaces using the passthrough interface.
1587          */
1588         if (is_ctrl_ioctl(cmd))
1589                 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1590
1591         switch (cmd) {
1592         case NVME_IOCTL_ID:
1593                 force_successful_syscall_return();
1594                 ret = ns->head->ns_id;
1595                 break;
1596         case NVME_IOCTL_IO_CMD:
1597                 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1598                 break;
1599         case NVME_IOCTL_SUBMIT_IO:
1600                 ret = nvme_submit_io(ns, argp);
1601                 break;
1602         case NVME_IOCTL_IO64_CMD:
1603                 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1604                 break;
1605         default:
1606                 if (ns->ndev)
1607                         ret = nvme_nvm_ioctl(ns, cmd, arg);
1608                 else
1609                         ret = -ENOTTY;
1610         }
1611
1612         nvme_put_ns_from_disk(head, srcu_idx);
1613         return ret;
1614 }
1615
1616 #ifdef CONFIG_COMPAT
1617 struct nvme_user_io32 {
1618         __u8    opcode;
1619         __u8    flags;
1620         __u16   control;
1621         __u16   nblocks;
1622         __u16   rsvd;
1623         __u64   metadata;
1624         __u64   addr;
1625         __u64   slba;
1626         __u32   dsmgmt;
1627         __u32   reftag;
1628         __u16   apptag;
1629         __u16   appmask;
1630 } __attribute__((__packed__));
1631
1632 #define NVME_IOCTL_SUBMIT_IO32  _IOW('N', 0x42, struct nvme_user_io32)
1633
1634 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1635                 unsigned int cmd, unsigned long arg)
1636 {
1637         /*
1638          * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1639          * between 32 bit programs and 64 bit kernel.
1640          * The cause is that the results of sizeof(struct nvme_user_io),
1641          * which is used to define NVME_IOCTL_SUBMIT_IO,
1642          * are not same between 32 bit compiler and 64 bit compiler.
1643          * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1644          * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1645          * Other IOCTL numbers are same between 32 bit and 64 bit.
1646          * So there is nothing to do regarding to other IOCTL numbers.
1647          */
1648         if (cmd == NVME_IOCTL_SUBMIT_IO32)
1649                 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1650
1651         return nvme_ioctl(bdev, mode, cmd, arg);
1652 }
1653 #else
1654 #define nvme_compat_ioctl       NULL
1655 #endif /* CONFIG_COMPAT */
1656
1657 static int nvme_open(struct block_device *bdev, fmode_t mode)
1658 {
1659         struct nvme_ns *ns = bdev->bd_disk->private_data;
1660
1661 #ifdef CONFIG_NVME_MULTIPATH
1662         /* should never be called due to GENHD_FL_HIDDEN */
1663         if (WARN_ON_ONCE(ns->head->disk))
1664                 goto fail;
1665 #endif
1666         if (!kref_get_unless_zero(&ns->kref))
1667                 goto fail;
1668         if (!try_module_get(ns->ctrl->ops->module))
1669                 goto fail_put_ns;
1670
1671         return 0;
1672
1673 fail_put_ns:
1674         nvme_put_ns(ns);
1675 fail:
1676         return -ENXIO;
1677 }
1678
1679 static void nvme_release(struct gendisk *disk, fmode_t mode)
1680 {
1681         struct nvme_ns *ns = disk->private_data;
1682
1683         module_put(ns->ctrl->ops->module);
1684         nvme_put_ns(ns);
1685 }
1686
1687 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1688 {
1689         /* some standard values */
1690         geo->heads = 1 << 6;
1691         geo->sectors = 1 << 5;
1692         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1693         return 0;
1694 }
1695
1696 #ifdef CONFIG_BLK_DEV_INTEGRITY
1697 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1698 {
1699         struct blk_integrity integrity;
1700
1701         memset(&integrity, 0, sizeof(integrity));
1702         switch (pi_type) {
1703         case NVME_NS_DPS_PI_TYPE3:
1704                 integrity.profile = &t10_pi_type3_crc;
1705                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1706                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1707                 break;
1708         case NVME_NS_DPS_PI_TYPE1:
1709         case NVME_NS_DPS_PI_TYPE2:
1710                 integrity.profile = &t10_pi_type1_crc;
1711                 integrity.tag_size = sizeof(u16);
1712                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1713                 break;
1714         default:
1715                 integrity.profile = NULL;
1716                 break;
1717         }
1718         integrity.tuple_size = ms;
1719         blk_integrity_register(disk, &integrity);
1720         blk_queue_max_integrity_segments(disk->queue, 1);
1721 }
1722 #else
1723 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1724 {
1725 }
1726 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1727
1728 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1729 {
1730         struct nvme_ctrl *ctrl = ns->ctrl;
1731         struct request_queue *queue = disk->queue;
1732         u32 size = queue_logical_block_size(queue);
1733
1734         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1735                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1736                 return;
1737         }
1738
1739         if (ctrl->nr_streams && ns->sws && ns->sgs)
1740                 size *= ns->sws * ns->sgs;
1741
1742         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1743                         NVME_DSM_MAX_RANGES);
1744
1745         queue->limits.discard_alignment = 0;
1746         queue->limits.discard_granularity = size;
1747
1748         /* If discard is already enabled, don't reset queue limits */
1749         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1750                 return;
1751
1752         blk_queue_max_discard_sectors(queue, UINT_MAX);
1753         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1754
1755         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1756                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1757 }
1758
1759 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1760 {
1761         u64 max_blocks;
1762
1763         if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1764             (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1765                 return;
1766         /*
1767          * Even though NVMe spec explicitly states that MDTS is not
1768          * applicable to the write-zeroes:- "The restriction does not apply to
1769          * commands that do not transfer data between the host and the
1770          * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1771          * In order to be more cautious use controller's max_hw_sectors value
1772          * to configure the maximum sectors for the write-zeroes which is
1773          * configured based on the controller's MDTS field in the
1774          * nvme_init_identify() if available.
1775          */
1776         if (ns->ctrl->max_hw_sectors == UINT_MAX)
1777                 max_blocks = (u64)USHRT_MAX + 1;
1778         else
1779                 max_blocks = ns->ctrl->max_hw_sectors + 1;
1780
1781         blk_queue_max_write_zeroes_sectors(disk->queue,
1782                                            nvme_lba_to_sect(ns, max_blocks));
1783 }
1784
1785 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1786                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1787 {
1788         memset(ids, 0, sizeof(*ids));
1789
1790         if (ctrl->vs >= NVME_VS(1, 1, 0))
1791                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1792         if (ctrl->vs >= NVME_VS(1, 2, 0))
1793                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1794         if (ctrl->vs >= NVME_VS(1, 3, 0))
1795                 return nvme_identify_ns_descs(ctrl, nsid, ids);
1796         return 0;
1797 }
1798
1799 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1800 {
1801         return !uuid_is_null(&ids->uuid) ||
1802                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1803                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1804 }
1805
1806 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1807 {
1808         return uuid_equal(&a->uuid, &b->uuid) &&
1809                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1810                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1811 }
1812
1813 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1814                                  u32 *phys_bs, u32 *io_opt)
1815 {
1816         struct streams_directive_params s;
1817         int ret;
1818
1819         if (!ctrl->nr_streams)
1820                 return 0;
1821
1822         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1823         if (ret)
1824                 return ret;
1825
1826         ns->sws = le32_to_cpu(s.sws);
1827         ns->sgs = le16_to_cpu(s.sgs);
1828
1829         if (ns->sws) {
1830                 *phys_bs = ns->sws * (1 << ns->lba_shift);
1831                 if (ns->sgs)
1832                         *io_opt = *phys_bs * ns->sgs;
1833         }
1834
1835         return 0;
1836 }
1837
1838 static void nvme_update_disk_info(struct gendisk *disk,
1839                 struct nvme_ns *ns, struct nvme_id_ns *id)
1840 {
1841         sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1842         unsigned short bs = 1 << ns->lba_shift;
1843         u32 atomic_bs, phys_bs, io_opt;
1844
1845         if (ns->lba_shift > PAGE_SHIFT) {
1846                 /* unsupported block size, set capacity to 0 later */
1847                 bs = (1 << 9);
1848         }
1849         blk_mq_freeze_queue(disk->queue);
1850         blk_integrity_unregister(disk);
1851
1852         atomic_bs = phys_bs = io_opt = bs;
1853         nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1854         if (id->nabo == 0) {
1855                 /*
1856                  * Bit 1 indicates whether NAWUPF is defined for this namespace
1857                  * and whether it should be used instead of AWUPF. If NAWUPF ==
1858                  * 0 then AWUPF must be used instead.
1859                  */
1860                 if (id->nsfeat & (1 << 1) && id->nawupf)
1861                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1862                 else
1863                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1864         }
1865
1866         if (id->nsfeat & (1 << 4)) {
1867                 /* NPWG = Namespace Preferred Write Granularity */
1868                 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1869                 /* NOWS = Namespace Optimal Write Size */
1870                 io_opt = bs * (1 + le16_to_cpu(id->nows));
1871         }
1872
1873         blk_queue_logical_block_size(disk->queue, bs);
1874         /*
1875          * Linux filesystems assume writing a single physical block is
1876          * an atomic operation. Hence limit the physical block size to the
1877          * value of the Atomic Write Unit Power Fail parameter.
1878          */
1879         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1880         blk_queue_io_min(disk->queue, phys_bs);
1881         blk_queue_io_opt(disk->queue, io_opt);
1882
1883         if (ns->ms && !ns->ext &&
1884             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1885                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1886         if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1887             ns->lba_shift > PAGE_SHIFT)
1888                 capacity = 0;
1889
1890         set_capacity_revalidate_and_notify(disk, capacity, false);
1891
1892         nvme_config_discard(disk, ns);
1893         nvme_config_write_zeroes(disk, ns);
1894
1895         if (id->nsattr & (1 << 0))
1896                 set_disk_ro(disk, true);
1897         else
1898                 set_disk_ro(disk, false);
1899
1900         blk_mq_unfreeze_queue(disk->queue);
1901 }
1902
1903 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1904 {
1905         struct nvme_ns *ns = disk->private_data;
1906         u32 iob;
1907
1908         /*
1909          * If identify namespace failed, use default 512 byte block size so
1910          * block layer can use before failing read/write for 0 capacity.
1911          */
1912         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1913         if (ns->lba_shift == 0)
1914                 ns->lba_shift = 9;
1915
1916         if ((ns->ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1917             is_power_of_2(ns->ctrl->max_hw_sectors))
1918                 iob = ns->ctrl->max_hw_sectors;
1919         else
1920                 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1921
1922         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1923         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1924         /* the PI implementation requires metadata equal t10 pi tuple size */
1925         if (ns->ms == sizeof(struct t10_pi_tuple))
1926                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1927         else
1928                 ns->pi_type = 0;
1929
1930         if (iob)
1931                 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(iob));
1932         nvme_update_disk_info(disk, ns, id);
1933 #ifdef CONFIG_NVME_MULTIPATH
1934         if (ns->head->disk) {
1935                 nvme_update_disk_info(ns->head->disk, ns, id);
1936                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1937                 revalidate_disk(ns->head->disk);
1938         }
1939 #endif
1940 }
1941
1942 static int nvme_revalidate_disk(struct gendisk *disk)
1943 {
1944         struct nvme_ns *ns = disk->private_data;
1945         struct nvme_ctrl *ctrl = ns->ctrl;
1946         struct nvme_id_ns *id;
1947         struct nvme_ns_ids ids;
1948         int ret = 0;
1949
1950         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1951                 set_capacity(disk, 0);
1952                 return -ENODEV;
1953         }
1954
1955         ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1956         if (ret)
1957                 goto out;
1958
1959         if (id->ncap == 0) {
1960                 ret = -ENODEV;
1961                 goto free_id;
1962         }
1963
1964         ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1965         if (ret)
1966                 goto free_id;
1967
1968         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1969                 dev_err(ctrl->device,
1970                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1971                 ret = -ENODEV;
1972                 goto free_id;
1973         }
1974
1975         __nvme_revalidate_disk(disk, id);
1976 free_id:
1977         kfree(id);
1978 out:
1979         /*
1980          * Only fail the function if we got a fatal error back from the
1981          * device, otherwise ignore the error and just move on.
1982          */
1983         if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1984                 ret = 0;
1985         else if (ret > 0)
1986                 ret = blk_status_to_errno(nvme_error_status(ret));
1987         return ret;
1988 }
1989
1990 static char nvme_pr_type(enum pr_type type)
1991 {
1992         switch (type) {
1993         case PR_WRITE_EXCLUSIVE:
1994                 return 1;
1995         case PR_EXCLUSIVE_ACCESS:
1996                 return 2;
1997         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1998                 return 3;
1999         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2000                 return 4;
2001         case PR_WRITE_EXCLUSIVE_ALL_REGS:
2002                 return 5;
2003         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2004                 return 6;
2005         default:
2006                 return 0;
2007         }
2008 };
2009
2010 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2011                                 u64 key, u64 sa_key, u8 op)
2012 {
2013         struct nvme_ns_head *head = NULL;
2014         struct nvme_ns *ns;
2015         struct nvme_command c;
2016         int srcu_idx, ret;
2017         u8 data[16] = { 0, };
2018
2019         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2020         if (unlikely(!ns))
2021                 return -EWOULDBLOCK;
2022
2023         put_unaligned_le64(key, &data[0]);
2024         put_unaligned_le64(sa_key, &data[8]);
2025
2026         memset(&c, 0, sizeof(c));
2027         c.common.opcode = op;
2028         c.common.nsid = cpu_to_le32(ns->head->ns_id);
2029         c.common.cdw10 = cpu_to_le32(cdw10);
2030
2031         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2032         nvme_put_ns_from_disk(head, srcu_idx);
2033         return ret;
2034 }
2035
2036 static int nvme_pr_register(struct block_device *bdev, u64 old,
2037                 u64 new, unsigned flags)
2038 {
2039         u32 cdw10;
2040
2041         if (flags & ~PR_FL_IGNORE_KEY)
2042                 return -EOPNOTSUPP;
2043
2044         cdw10 = old ? 2 : 0;
2045         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2046         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2047         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2048 }
2049
2050 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2051                 enum pr_type type, unsigned flags)
2052 {
2053         u32 cdw10;
2054
2055         if (flags & ~PR_FL_IGNORE_KEY)
2056                 return -EOPNOTSUPP;
2057
2058         cdw10 = nvme_pr_type(type) << 8;
2059         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2060         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2061 }
2062
2063 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2064                 enum pr_type type, bool abort)
2065 {
2066         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2067         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2068 }
2069
2070 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2071 {
2072         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2073         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2074 }
2075
2076 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2077 {
2078         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2079         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2080 }
2081
2082 static const struct pr_ops nvme_pr_ops = {
2083         .pr_register    = nvme_pr_register,
2084         .pr_reserve     = nvme_pr_reserve,
2085         .pr_release     = nvme_pr_release,
2086         .pr_preempt     = nvme_pr_preempt,
2087         .pr_clear       = nvme_pr_clear,
2088 };
2089
2090 #ifdef CONFIG_BLK_SED_OPAL
2091 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2092                 bool send)
2093 {
2094         struct nvme_ctrl *ctrl = data;
2095         struct nvme_command cmd;
2096
2097         memset(&cmd, 0, sizeof(cmd));
2098         if (send)
2099                 cmd.common.opcode = nvme_admin_security_send;
2100         else
2101                 cmd.common.opcode = nvme_admin_security_recv;
2102         cmd.common.nsid = 0;
2103         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2104         cmd.common.cdw11 = cpu_to_le32(len);
2105
2106         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2107                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2108 }
2109 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2110 #endif /* CONFIG_BLK_SED_OPAL */
2111
2112 static const struct block_device_operations nvme_fops = {
2113         .owner          = THIS_MODULE,
2114         .ioctl          = nvme_ioctl,
2115         .compat_ioctl   = nvme_compat_ioctl,
2116         .open           = nvme_open,
2117         .release        = nvme_release,
2118         .getgeo         = nvme_getgeo,
2119         .revalidate_disk= nvme_revalidate_disk,
2120         .pr_ops         = &nvme_pr_ops,
2121 };
2122
2123 #ifdef CONFIG_NVME_MULTIPATH
2124 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2125 {
2126         struct nvme_ns_head *head = bdev->bd_disk->private_data;
2127
2128         if (!kref_get_unless_zero(&head->ref))
2129                 return -ENXIO;
2130         return 0;
2131 }
2132
2133 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2134 {
2135         nvme_put_ns_head(disk->private_data);
2136 }
2137
2138 const struct block_device_operations nvme_ns_head_ops = {
2139         .owner          = THIS_MODULE,
2140         .open           = nvme_ns_head_open,
2141         .release        = nvme_ns_head_release,
2142         .ioctl          = nvme_ioctl,
2143         .compat_ioctl   = nvme_compat_ioctl,
2144         .getgeo         = nvme_getgeo,
2145         .pr_ops         = &nvme_pr_ops,
2146 };
2147 #endif /* CONFIG_NVME_MULTIPATH */
2148
2149 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2150 {
2151         unsigned long timeout =
2152                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2153         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2154         int ret;
2155
2156         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2157                 if (csts == ~0)
2158                         return -ENODEV;
2159                 if ((csts & NVME_CSTS_RDY) == bit)
2160                         break;
2161
2162                 usleep_range(1000, 2000);
2163                 if (fatal_signal_pending(current))
2164                         return -EINTR;
2165                 if (time_after(jiffies, timeout)) {
2166                         dev_err(ctrl->device,
2167                                 "Device not ready; aborting %s, CSTS=0x%x\n",
2168                                 enabled ? "initialisation" : "reset", csts);
2169                         return -ENODEV;
2170                 }
2171         }
2172
2173         return ret;
2174 }
2175
2176 /*
2177  * If the device has been passed off to us in an enabled state, just clear
2178  * the enabled bit.  The spec says we should set the 'shutdown notification
2179  * bits', but doing so may cause the device to complete commands to the
2180  * admin queue ... and we don't know what memory that might be pointing at!
2181  */
2182 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2183 {
2184         int ret;
2185
2186         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2187         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2188
2189         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2190         if (ret)
2191                 return ret;
2192
2193         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2194                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2195
2196         return nvme_wait_ready(ctrl, ctrl->cap, false);
2197 }
2198 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2199
2200 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2201 {
2202         /*
2203          * Default to a 4K page size, with the intention to update this
2204          * path in the future to accomodate architectures with differing
2205          * kernel and IO page sizes.
2206          */
2207         unsigned dev_page_min, page_shift = 12;
2208         int ret;
2209
2210         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2211         if (ret) {
2212                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2213                 return ret;
2214         }
2215         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2216
2217         if (page_shift < dev_page_min) {
2218                 dev_err(ctrl->device,
2219                         "Minimum device page size %u too large for host (%u)\n",
2220                         1 << dev_page_min, 1 << page_shift);
2221                 return -ENODEV;
2222         }
2223
2224         ctrl->page_size = 1 << page_shift;
2225
2226         ctrl->ctrl_config = NVME_CC_CSS_NVM;
2227         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2228         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2229         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2230         ctrl->ctrl_config |= NVME_CC_ENABLE;
2231
2232         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2233         if (ret)
2234                 return ret;
2235         return nvme_wait_ready(ctrl, ctrl->cap, true);
2236 }
2237 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2238
2239 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2240 {
2241         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2242         u32 csts;
2243         int ret;
2244
2245         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2246         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2247
2248         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2249         if (ret)
2250                 return ret;
2251
2252         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2253                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2254                         break;
2255
2256                 msleep(100);
2257                 if (fatal_signal_pending(current))
2258                         return -EINTR;
2259                 if (time_after(jiffies, timeout)) {
2260                         dev_err(ctrl->device,
2261                                 "Device shutdown incomplete; abort shutdown\n");
2262                         return -ENODEV;
2263                 }
2264         }
2265
2266         return ret;
2267 }
2268 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2269
2270 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2271                 struct request_queue *q)
2272 {
2273         bool vwc = false;
2274
2275         if (ctrl->max_hw_sectors) {
2276                 u32 max_segments =
2277                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2278
2279                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2280                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2281                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2282         }
2283         blk_queue_virt_boundary(q, ctrl->page_size - 1);
2284         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2285                 vwc = true;
2286         blk_queue_write_cache(q, vwc, vwc);
2287 }
2288
2289 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2290 {
2291         __le64 ts;
2292         int ret;
2293
2294         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2295                 return 0;
2296
2297         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2298         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2299                         NULL);
2300         if (ret)
2301                 dev_warn_once(ctrl->device,
2302                         "could not set timestamp (%d)\n", ret);
2303         return ret;
2304 }
2305
2306 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2307 {
2308         struct nvme_feat_host_behavior *host;
2309         int ret;
2310
2311         /* Don't bother enabling the feature if retry delay is not reported */
2312         if (!ctrl->crdt[0])
2313                 return 0;
2314
2315         host = kzalloc(sizeof(*host), GFP_KERNEL);
2316         if (!host)
2317                 return 0;
2318
2319         host->acre = NVME_ENABLE_ACRE;
2320         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2321                                 host, sizeof(*host), NULL);
2322         kfree(host);
2323         return ret;
2324 }
2325
2326 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2327 {
2328         /*
2329          * APST (Autonomous Power State Transition) lets us program a
2330          * table of power state transitions that the controller will
2331          * perform automatically.  We configure it with a simple
2332          * heuristic: we are willing to spend at most 2% of the time
2333          * transitioning between power states.  Therefore, when running
2334          * in any given state, we will enter the next lower-power
2335          * non-operational state after waiting 50 * (enlat + exlat)
2336          * microseconds, as long as that state's exit latency is under
2337          * the requested maximum latency.
2338          *
2339          * We will not autonomously enter any non-operational state for
2340          * which the total latency exceeds ps_max_latency_us.  Users
2341          * can set ps_max_latency_us to zero to turn off APST.
2342          */
2343
2344         unsigned apste;
2345         struct nvme_feat_auto_pst *table;
2346         u64 max_lat_us = 0;
2347         int max_ps = -1;
2348         int ret;
2349
2350         /*
2351          * If APST isn't supported or if we haven't been initialized yet,
2352          * then don't do anything.
2353          */
2354         if (!ctrl->apsta)
2355                 return 0;
2356
2357         if (ctrl->npss > 31) {
2358                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2359                 return 0;
2360         }
2361
2362         table = kzalloc(sizeof(*table), GFP_KERNEL);
2363         if (!table)
2364                 return 0;
2365
2366         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2367                 /* Turn off APST. */
2368                 apste = 0;
2369                 dev_dbg(ctrl->device, "APST disabled\n");
2370         } else {
2371                 __le64 target = cpu_to_le64(0);
2372                 int state;
2373
2374                 /*
2375                  * Walk through all states from lowest- to highest-power.
2376                  * According to the spec, lower-numbered states use more
2377                  * power.  NPSS, despite the name, is the index of the
2378                  * lowest-power state, not the number of states.
2379                  */
2380                 for (state = (int)ctrl->npss; state >= 0; state--) {
2381                         u64 total_latency_us, exit_latency_us, transition_ms;
2382
2383                         if (target)
2384                                 table->entries[state] = target;
2385
2386                         /*
2387                          * Don't allow transitions to the deepest state
2388                          * if it's quirked off.
2389                          */
2390                         if (state == ctrl->npss &&
2391                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2392                                 continue;
2393
2394                         /*
2395                          * Is this state a useful non-operational state for
2396                          * higher-power states to autonomously transition to?
2397                          */
2398                         if (!(ctrl->psd[state].flags &
2399                               NVME_PS_FLAGS_NON_OP_STATE))
2400                                 continue;
2401
2402                         exit_latency_us =
2403                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2404                         if (exit_latency_us > ctrl->ps_max_latency_us)
2405                                 continue;
2406
2407                         total_latency_us =
2408                                 exit_latency_us +
2409                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2410
2411                         /*
2412                          * This state is good.  Use it as the APST idle
2413                          * target for higher power states.
2414                          */
2415                         transition_ms = total_latency_us + 19;
2416                         do_div(transition_ms, 20);
2417                         if (transition_ms > (1 << 24) - 1)
2418                                 transition_ms = (1 << 24) - 1;
2419
2420                         target = cpu_to_le64((state << 3) |
2421                                              (transition_ms << 8));
2422
2423                         if (max_ps == -1)
2424                                 max_ps = state;
2425
2426                         if (total_latency_us > max_lat_us)
2427                                 max_lat_us = total_latency_us;
2428                 }
2429
2430                 apste = 1;
2431
2432                 if (max_ps == -1) {
2433                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2434                 } else {
2435                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2436                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2437                 }
2438         }
2439
2440         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2441                                 table, sizeof(*table), NULL);
2442         if (ret)
2443                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2444
2445         kfree(table);
2446         return ret;
2447 }
2448
2449 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2450 {
2451         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2452         u64 latency;
2453
2454         switch (val) {
2455         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2456         case PM_QOS_LATENCY_ANY:
2457                 latency = U64_MAX;
2458                 break;
2459
2460         default:
2461                 latency = val;
2462         }
2463
2464         if (ctrl->ps_max_latency_us != latency) {
2465                 ctrl->ps_max_latency_us = latency;
2466                 nvme_configure_apst(ctrl);
2467         }
2468 }
2469
2470 struct nvme_core_quirk_entry {
2471         /*
2472          * NVMe model and firmware strings are padded with spaces.  For
2473          * simplicity, strings in the quirk table are padded with NULLs
2474          * instead.
2475          */
2476         u16 vid;
2477         const char *mn;
2478         const char *fr;
2479         unsigned long quirks;
2480 };
2481
2482 static const struct nvme_core_quirk_entry core_quirks[] = {
2483         {
2484                 /*
2485                  * This Toshiba device seems to die using any APST states.  See:
2486                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2487                  */
2488                 .vid = 0x1179,
2489                 .mn = "THNSF5256GPUK TOSHIBA",
2490                 .quirks = NVME_QUIRK_NO_APST,
2491         },
2492         {
2493                 /*
2494                  * This LiteON CL1-3D*-Q11 firmware version has a race
2495                  * condition associated with actions related to suspend to idle
2496                  * LiteON has resolved the problem in future firmware
2497                  */
2498                 .vid = 0x14a4,
2499                 .fr = "22301111",
2500                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2501         }
2502 };
2503
2504 /* match is null-terminated but idstr is space-padded. */
2505 static bool string_matches(const char *idstr, const char *match, size_t len)
2506 {
2507         size_t matchlen;
2508
2509         if (!match)
2510                 return true;
2511
2512         matchlen = strlen(match);
2513         WARN_ON_ONCE(matchlen > len);
2514
2515         if (memcmp(idstr, match, matchlen))
2516                 return false;
2517
2518         for (; matchlen < len; matchlen++)
2519                 if (idstr[matchlen] != ' ')
2520                         return false;
2521
2522         return true;
2523 }
2524
2525 static bool quirk_matches(const struct nvme_id_ctrl *id,
2526                           const struct nvme_core_quirk_entry *q)
2527 {
2528         return q->vid == le16_to_cpu(id->vid) &&
2529                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2530                 string_matches(id->fr, q->fr, sizeof(id->fr));
2531 }
2532
2533 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2534                 struct nvme_id_ctrl *id)
2535 {
2536         size_t nqnlen;
2537         int off;
2538
2539         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2540                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2541                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2542                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2543                         return;
2544                 }
2545
2546                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2547                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2548         }
2549
2550         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2551         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2552                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2553                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2554         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2555         off += sizeof(id->sn);
2556         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2557         off += sizeof(id->mn);
2558         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2559 }
2560
2561 static void nvme_release_subsystem(struct device *dev)
2562 {
2563         struct nvme_subsystem *subsys =
2564                 container_of(dev, struct nvme_subsystem, dev);
2565
2566         if (subsys->instance >= 0)
2567                 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2568         kfree(subsys);
2569 }
2570
2571 static void nvme_destroy_subsystem(struct kref *ref)
2572 {
2573         struct nvme_subsystem *subsys =
2574                         container_of(ref, struct nvme_subsystem, ref);
2575
2576         mutex_lock(&nvme_subsystems_lock);
2577         list_del(&subsys->entry);
2578         mutex_unlock(&nvme_subsystems_lock);
2579
2580         ida_destroy(&subsys->ns_ida);
2581         device_del(&subsys->dev);
2582         put_device(&subsys->dev);
2583 }
2584
2585 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2586 {
2587         kref_put(&subsys->ref, nvme_destroy_subsystem);
2588 }
2589
2590 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2591 {
2592         struct nvme_subsystem *subsys;
2593
2594         lockdep_assert_held(&nvme_subsystems_lock);
2595
2596         /*
2597          * Fail matches for discovery subsystems. This results
2598          * in each discovery controller bound to a unique subsystem.
2599          * This avoids issues with validating controller values
2600          * that can only be true when there is a single unique subsystem.
2601          * There may be multiple and completely independent entities
2602          * that provide discovery controllers.
2603          */
2604         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2605                 return NULL;
2606
2607         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2608                 if (strcmp(subsys->subnqn, subsysnqn))
2609                         continue;
2610                 if (!kref_get_unless_zero(&subsys->ref))
2611                         continue;
2612                 return subsys;
2613         }
2614
2615         return NULL;
2616 }
2617
2618 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2619         struct device_attribute subsys_attr_##_name = \
2620                 __ATTR(_name, _mode, _show, NULL)
2621
2622 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2623                                     struct device_attribute *attr,
2624                                     char *buf)
2625 {
2626         struct nvme_subsystem *subsys =
2627                 container_of(dev, struct nvme_subsystem, dev);
2628
2629         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2630 }
2631 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2632
2633 #define nvme_subsys_show_str_function(field)                            \
2634 static ssize_t subsys_##field##_show(struct device *dev,                \
2635                             struct device_attribute *attr, char *buf)   \
2636 {                                                                       \
2637         struct nvme_subsystem *subsys =                                 \
2638                 container_of(dev, struct nvme_subsystem, dev);          \
2639         return sprintf(buf, "%.*s\n",                                   \
2640                        (int)sizeof(subsys->field), subsys->field);      \
2641 }                                                                       \
2642 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2643
2644 nvme_subsys_show_str_function(model);
2645 nvme_subsys_show_str_function(serial);
2646 nvme_subsys_show_str_function(firmware_rev);
2647
2648 static struct attribute *nvme_subsys_attrs[] = {
2649         &subsys_attr_model.attr,
2650         &subsys_attr_serial.attr,
2651         &subsys_attr_firmware_rev.attr,
2652         &subsys_attr_subsysnqn.attr,
2653 #ifdef CONFIG_NVME_MULTIPATH
2654         &subsys_attr_iopolicy.attr,
2655 #endif
2656         NULL,
2657 };
2658
2659 static struct attribute_group nvme_subsys_attrs_group = {
2660         .attrs = nvme_subsys_attrs,
2661 };
2662
2663 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2664         &nvme_subsys_attrs_group,
2665         NULL,
2666 };
2667
2668 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2669                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2670 {
2671         struct nvme_ctrl *tmp;
2672
2673         lockdep_assert_held(&nvme_subsystems_lock);
2674
2675         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2676                 if (nvme_state_terminal(tmp))
2677                         continue;
2678
2679                 if (tmp->cntlid == ctrl->cntlid) {
2680                         dev_err(ctrl->device,
2681                                 "Duplicate cntlid %u with %s, rejecting\n",
2682                                 ctrl->cntlid, dev_name(tmp->device));
2683                         return false;
2684                 }
2685
2686                 if ((id->cmic & (1 << 1)) ||
2687                     (ctrl->opts && ctrl->opts->discovery_nqn))
2688                         continue;
2689
2690                 dev_err(ctrl->device,
2691                         "Subsystem does not support multiple controllers\n");
2692                 return false;
2693         }
2694
2695         return true;
2696 }
2697
2698 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2699 {
2700         struct nvme_subsystem *subsys, *found;
2701         int ret;
2702
2703         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2704         if (!subsys)
2705                 return -ENOMEM;
2706
2707         subsys->instance = -1;
2708         mutex_init(&subsys->lock);
2709         kref_init(&subsys->ref);
2710         INIT_LIST_HEAD(&subsys->ctrls);
2711         INIT_LIST_HEAD(&subsys->nsheads);
2712         nvme_init_subnqn(subsys, ctrl, id);
2713         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2714         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2715         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2716         subsys->vendor_id = le16_to_cpu(id->vid);
2717         subsys->cmic = id->cmic;
2718         subsys->awupf = le16_to_cpu(id->awupf);
2719 #ifdef CONFIG_NVME_MULTIPATH
2720         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2721 #endif
2722
2723         subsys->dev.class = nvme_subsys_class;
2724         subsys->dev.release = nvme_release_subsystem;
2725         subsys->dev.groups = nvme_subsys_attrs_groups;
2726         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2727         device_initialize(&subsys->dev);
2728
2729         mutex_lock(&nvme_subsystems_lock);
2730         found = __nvme_find_get_subsystem(subsys->subnqn);
2731         if (found) {
2732                 put_device(&subsys->dev);
2733                 subsys = found;
2734
2735                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2736                         ret = -EINVAL;
2737                         goto out_put_subsystem;
2738                 }
2739         } else {
2740                 ret = device_add(&subsys->dev);
2741                 if (ret) {
2742                         dev_err(ctrl->device,
2743                                 "failed to register subsystem device.\n");
2744                         put_device(&subsys->dev);
2745                         goto out_unlock;
2746                 }
2747                 ida_init(&subsys->ns_ida);
2748                 list_add_tail(&subsys->entry, &nvme_subsystems);
2749         }
2750
2751         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2752                                 dev_name(ctrl->device));
2753         if (ret) {
2754                 dev_err(ctrl->device,
2755                         "failed to create sysfs link from subsystem.\n");
2756                 goto out_put_subsystem;
2757         }
2758
2759         if (!found)
2760                 subsys->instance = ctrl->instance;
2761         ctrl->subsys = subsys;
2762         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2763         mutex_unlock(&nvme_subsystems_lock);
2764         return 0;
2765
2766 out_put_subsystem:
2767         nvme_put_subsystem(subsys);
2768 out_unlock:
2769         mutex_unlock(&nvme_subsystems_lock);
2770         return ret;
2771 }
2772
2773 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2774                 void *log, size_t size, u64 offset)
2775 {
2776         struct nvme_command c = { };
2777         u32 dwlen = nvme_bytes_to_numd(size);
2778
2779         c.get_log_page.opcode = nvme_admin_get_log_page;
2780         c.get_log_page.nsid = cpu_to_le32(nsid);
2781         c.get_log_page.lid = log_page;
2782         c.get_log_page.lsp = lsp;
2783         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2784         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2785         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2786         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2787
2788         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2789 }
2790
2791 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2792 {
2793         int ret;
2794
2795         if (!ctrl->effects)
2796                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2797
2798         if (!ctrl->effects)
2799                 return 0;
2800
2801         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2802                         ctrl->effects, sizeof(*ctrl->effects), 0);
2803         if (ret) {
2804                 kfree(ctrl->effects);
2805                 ctrl->effects = NULL;
2806         }
2807         return ret;
2808 }
2809
2810 /*
2811  * Initialize the cached copies of the Identify data and various controller
2812  * register in our nvme_ctrl structure.  This should be called as soon as
2813  * the admin queue is fully up and running.
2814  */
2815 int nvme_init_identify(struct nvme_ctrl *ctrl)
2816 {
2817         struct nvme_id_ctrl *id;
2818         int ret, page_shift;
2819         u32 max_hw_sectors;
2820         bool prev_apst_enabled;
2821
2822         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2823         if (ret) {
2824                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2825                 return ret;
2826         }
2827         page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2828         ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2829
2830         if (ctrl->vs >= NVME_VS(1, 1, 0))
2831                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2832
2833         ret = nvme_identify_ctrl(ctrl, &id);
2834         if (ret) {
2835                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2836                 return -EIO;
2837         }
2838
2839         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2840                 ret = nvme_get_effects_log(ctrl);
2841                 if (ret < 0)
2842                         goto out_free;
2843         }
2844
2845         if (!(ctrl->ops->flags & NVME_F_FABRICS))
2846                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2847
2848         if (!ctrl->identified) {
2849                 int i;
2850
2851                 ret = nvme_init_subsystem(ctrl, id);
2852                 if (ret)
2853                         goto out_free;
2854
2855                 /*
2856                  * Check for quirks.  Quirk can depend on firmware version,
2857                  * so, in principle, the set of quirks present can change
2858                  * across a reset.  As a possible future enhancement, we
2859                  * could re-scan for quirks every time we reinitialize
2860                  * the device, but we'd have to make sure that the driver
2861                  * behaves intelligently if the quirks change.
2862                  */
2863                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2864                         if (quirk_matches(id, &core_quirks[i]))
2865                                 ctrl->quirks |= core_quirks[i].quirks;
2866                 }
2867         }
2868
2869         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2870                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2871                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2872         }
2873
2874         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2875         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2876         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2877
2878         ctrl->oacs = le16_to_cpu(id->oacs);
2879         ctrl->oncs = le16_to_cpu(id->oncs);
2880         ctrl->mtfa = le16_to_cpu(id->mtfa);
2881         ctrl->oaes = le32_to_cpu(id->oaes);
2882         ctrl->wctemp = le16_to_cpu(id->wctemp);
2883         ctrl->cctemp = le16_to_cpu(id->cctemp);
2884
2885         atomic_set(&ctrl->abort_limit, id->acl + 1);
2886         ctrl->vwc = id->vwc;
2887         if (id->mdts)
2888                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2889         else
2890                 max_hw_sectors = UINT_MAX;
2891         ctrl->max_hw_sectors =
2892                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2893
2894         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2895         ctrl->sgls = le32_to_cpu(id->sgls);
2896         ctrl->kas = le16_to_cpu(id->kas);
2897         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2898         ctrl->ctratt = le32_to_cpu(id->ctratt);
2899
2900         if (id->rtd3e) {
2901                 /* us -> s */
2902                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2903
2904                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2905                                                  shutdown_timeout, 60);
2906
2907                 if (ctrl->shutdown_timeout != shutdown_timeout)
2908                         dev_info(ctrl->device,
2909                                  "Shutdown timeout set to %u seconds\n",
2910                                  ctrl->shutdown_timeout);
2911         } else
2912                 ctrl->shutdown_timeout = shutdown_timeout;
2913
2914         ctrl->npss = id->npss;
2915         ctrl->apsta = id->apsta;
2916         prev_apst_enabled = ctrl->apst_enabled;
2917         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2918                 if (force_apst && id->apsta) {
2919                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2920                         ctrl->apst_enabled = true;
2921                 } else {
2922                         ctrl->apst_enabled = false;
2923                 }
2924         } else {
2925                 ctrl->apst_enabled = id->apsta;
2926         }
2927         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2928
2929         if (ctrl->ops->flags & NVME_F_FABRICS) {
2930                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2931                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2932                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2933                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2934
2935                 /*
2936                  * In fabrics we need to verify the cntlid matches the
2937                  * admin connect
2938                  */
2939                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2940                         dev_err(ctrl->device,
2941                                 "Mismatching cntlid: Connect %u vs Identify "
2942                                 "%u, rejecting\n",
2943                                 ctrl->cntlid, le16_to_cpu(id->cntlid));
2944                         ret = -EINVAL;
2945                         goto out_free;
2946                 }
2947
2948                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2949                         dev_err(ctrl->device,
2950                                 "keep-alive support is mandatory for fabrics\n");
2951                         ret = -EINVAL;
2952                         goto out_free;
2953                 }
2954         } else {
2955                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2956                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2957                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2958                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2959         }
2960
2961         ret = nvme_mpath_init(ctrl, id);
2962         kfree(id);
2963
2964         if (ret < 0)
2965                 return ret;
2966
2967         if (ctrl->apst_enabled && !prev_apst_enabled)
2968                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2969         else if (!ctrl->apst_enabled && prev_apst_enabled)
2970                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2971
2972         ret = nvme_configure_apst(ctrl);
2973         if (ret < 0)
2974                 return ret;
2975
2976         ret = nvme_configure_timestamp(ctrl);
2977         if (ret < 0)
2978                 return ret;
2979
2980         ret = nvme_configure_directives(ctrl);
2981         if (ret < 0)
2982                 return ret;
2983
2984         ret = nvme_configure_acre(ctrl);
2985         if (ret < 0)
2986                 return ret;
2987
2988         if (!ctrl->identified)
2989                 nvme_hwmon_init(ctrl);
2990
2991         ctrl->identified = true;
2992
2993         return 0;
2994
2995 out_free:
2996         kfree(id);
2997         return ret;
2998 }
2999 EXPORT_SYMBOL_GPL(nvme_init_identify);
3000
3001 static int nvme_dev_open(struct inode *inode, struct file *file)
3002 {
3003         struct nvme_ctrl *ctrl =
3004                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3005
3006         switch (ctrl->state) {
3007         case NVME_CTRL_LIVE:
3008                 break;
3009         default:
3010                 return -EWOULDBLOCK;
3011         }
3012
3013         file->private_data = ctrl;
3014         return 0;
3015 }
3016
3017 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3018 {
3019         struct nvme_ns *ns;
3020         int ret;
3021
3022         down_read(&ctrl->namespaces_rwsem);
3023         if (list_empty(&ctrl->namespaces)) {
3024                 ret = -ENOTTY;
3025                 goto out_unlock;
3026         }
3027
3028         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3029         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3030                 dev_warn(ctrl->device,
3031                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3032                 ret = -EINVAL;
3033                 goto out_unlock;
3034         }
3035
3036         dev_warn(ctrl->device,
3037                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3038         kref_get(&ns->kref);
3039         up_read(&ctrl->namespaces_rwsem);
3040
3041         ret = nvme_user_cmd(ctrl, ns, argp);
3042         nvme_put_ns(ns);
3043         return ret;
3044
3045 out_unlock:
3046         up_read(&ctrl->namespaces_rwsem);
3047         return ret;
3048 }
3049
3050 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3051                 unsigned long arg)
3052 {
3053         struct nvme_ctrl *ctrl = file->private_data;
3054         void __user *argp = (void __user *)arg;
3055
3056         switch (cmd) {
3057         case NVME_IOCTL_ADMIN_CMD:
3058                 return nvme_user_cmd(ctrl, NULL, argp);
3059         case NVME_IOCTL_ADMIN64_CMD:
3060                 return nvme_user_cmd64(ctrl, NULL, argp);
3061         case NVME_IOCTL_IO_CMD:
3062                 return nvme_dev_user_cmd(ctrl, argp);
3063         case NVME_IOCTL_RESET:
3064                 dev_warn(ctrl->device, "resetting controller\n");
3065                 return nvme_reset_ctrl_sync(ctrl);
3066         case NVME_IOCTL_SUBSYS_RESET:
3067                 return nvme_reset_subsystem(ctrl);
3068         case NVME_IOCTL_RESCAN:
3069                 nvme_queue_scan(ctrl);
3070                 return 0;
3071         default:
3072                 return -ENOTTY;
3073         }
3074 }
3075
3076 static const struct file_operations nvme_dev_fops = {
3077         .owner          = THIS_MODULE,
3078         .open           = nvme_dev_open,
3079         .unlocked_ioctl = nvme_dev_ioctl,
3080         .compat_ioctl   = compat_ptr_ioctl,
3081 };
3082
3083 static ssize_t nvme_sysfs_reset(struct device *dev,
3084                                 struct device_attribute *attr, const char *buf,
3085                                 size_t count)
3086 {
3087         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3088         int ret;
3089
3090         ret = nvme_reset_ctrl_sync(ctrl);
3091         if (ret < 0)
3092                 return ret;
3093         return count;
3094 }
3095 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3096
3097 static ssize_t nvme_sysfs_rescan(struct device *dev,
3098                                 struct device_attribute *attr, const char *buf,
3099                                 size_t count)
3100 {
3101         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3102
3103         nvme_queue_scan(ctrl);
3104         return count;
3105 }
3106 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3107
3108 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3109 {
3110         struct gendisk *disk = dev_to_disk(dev);
3111
3112         if (disk->fops == &nvme_fops)
3113                 return nvme_get_ns_from_dev(dev)->head;
3114         else
3115                 return disk->private_data;
3116 }
3117
3118 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3119                 char *buf)
3120 {
3121         struct nvme_ns_head *head = dev_to_ns_head(dev);
3122         struct nvme_ns_ids *ids = &head->ids;
3123         struct nvme_subsystem *subsys = head->subsys;
3124         int serial_len = sizeof(subsys->serial);
3125         int model_len = sizeof(subsys->model);
3126
3127         if (!uuid_is_null(&ids->uuid))
3128                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3129
3130         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3131                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3132
3133         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3134                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3135
3136         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3137                                   subsys->serial[serial_len - 1] == '\0'))
3138                 serial_len--;
3139         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3140                                  subsys->model[model_len - 1] == '\0'))
3141                 model_len--;
3142
3143         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3144                 serial_len, subsys->serial, model_len, subsys->model,
3145                 head->ns_id);
3146 }
3147 static DEVICE_ATTR_RO(wwid);
3148
3149 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3150                 char *buf)
3151 {
3152         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3153 }
3154 static DEVICE_ATTR_RO(nguid);
3155
3156 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3157                 char *buf)
3158 {
3159         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3160
3161         /* For backward compatibility expose the NGUID to userspace if
3162          * we have no UUID set
3163          */
3164         if (uuid_is_null(&ids->uuid)) {
3165                 printk_ratelimited(KERN_WARNING
3166                                    "No UUID available providing old NGUID\n");
3167                 return sprintf(buf, "%pU\n", ids->nguid);
3168         }
3169         return sprintf(buf, "%pU\n", &ids->uuid);
3170 }
3171 static DEVICE_ATTR_RO(uuid);
3172
3173 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3174                 char *buf)
3175 {
3176         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3177 }
3178 static DEVICE_ATTR_RO(eui);
3179
3180 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3181                 char *buf)
3182 {
3183         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3184 }
3185 static DEVICE_ATTR_RO(nsid);
3186
3187 static struct attribute *nvme_ns_id_attrs[] = {
3188         &dev_attr_wwid.attr,
3189         &dev_attr_uuid.attr,
3190         &dev_attr_nguid.attr,
3191         &dev_attr_eui.attr,
3192         &dev_attr_nsid.attr,
3193 #ifdef CONFIG_NVME_MULTIPATH
3194         &dev_attr_ana_grpid.attr,
3195         &dev_attr_ana_state.attr,
3196 #endif
3197         NULL,
3198 };
3199
3200 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3201                 struct attribute *a, int n)
3202 {
3203         struct device *dev = container_of(kobj, struct device, kobj);
3204         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3205
3206         if (a == &dev_attr_uuid.attr) {
3207                 if (uuid_is_null(&ids->uuid) &&
3208                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3209                         return 0;
3210         }
3211         if (a == &dev_attr_nguid.attr) {
3212                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3213                         return 0;
3214         }
3215         if (a == &dev_attr_eui.attr) {
3216                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3217                         return 0;
3218         }
3219 #ifdef CONFIG_NVME_MULTIPATH
3220         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3221                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3222                         return 0;
3223                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3224                         return 0;
3225         }
3226 #endif
3227         return a->mode;
3228 }
3229
3230 static const struct attribute_group nvme_ns_id_attr_group = {
3231         .attrs          = nvme_ns_id_attrs,
3232         .is_visible     = nvme_ns_id_attrs_are_visible,
3233 };
3234
3235 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3236         &nvme_ns_id_attr_group,
3237 #ifdef CONFIG_NVM
3238         &nvme_nvm_attr_group,
3239 #endif
3240         NULL,
3241 };
3242
3243 #define nvme_show_str_function(field)                                           \
3244 static ssize_t  field##_show(struct device *dev,                                \
3245                             struct device_attribute *attr, char *buf)           \
3246 {                                                                               \
3247         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3248         return sprintf(buf, "%.*s\n",                                           \
3249                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3250 }                                                                               \
3251 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3252
3253 nvme_show_str_function(model);
3254 nvme_show_str_function(serial);
3255 nvme_show_str_function(firmware_rev);
3256
3257 #define nvme_show_int_function(field)                                           \
3258 static ssize_t  field##_show(struct device *dev,                                \
3259                             struct device_attribute *attr, char *buf)           \
3260 {                                                                               \
3261         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3262         return sprintf(buf, "%d\n", ctrl->field);       \
3263 }                                                                               \
3264 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3265
3266 nvme_show_int_function(cntlid);
3267 nvme_show_int_function(numa_node);
3268 nvme_show_int_function(queue_count);
3269 nvme_show_int_function(sqsize);
3270
3271 static ssize_t nvme_sysfs_delete(struct device *dev,
3272                                 struct device_attribute *attr, const char *buf,
3273                                 size_t count)
3274 {
3275         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3276
3277         /* Can't delete non-created controllers */
3278         if (!ctrl->created)
3279                 return -EBUSY;
3280
3281         if (device_remove_file_self(dev, attr))
3282                 nvme_delete_ctrl_sync(ctrl);
3283         return count;
3284 }
3285 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3286
3287 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3288                                          struct device_attribute *attr,
3289                                          char *buf)
3290 {
3291         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3292
3293         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3294 }
3295 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3296
3297 static ssize_t nvme_sysfs_show_state(struct device *dev,
3298                                      struct device_attribute *attr,
3299                                      char *buf)
3300 {
3301         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3302         static const char *const state_name[] = {
3303                 [NVME_CTRL_NEW]         = "new",
3304                 [NVME_CTRL_LIVE]        = "live",
3305                 [NVME_CTRL_RESETTING]   = "resetting",
3306                 [NVME_CTRL_CONNECTING]  = "connecting",
3307                 [NVME_CTRL_DELETING]    = "deleting",
3308                 [NVME_CTRL_DEAD]        = "dead",
3309         };
3310
3311         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3312             state_name[ctrl->state])
3313                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3314
3315         return sprintf(buf, "unknown state\n");
3316 }
3317
3318 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3319
3320 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3321                                          struct device_attribute *attr,
3322                                          char *buf)
3323 {
3324         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3325
3326         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3327 }
3328 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3329
3330 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3331                                         struct device_attribute *attr,
3332                                         char *buf)
3333 {
3334         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3335
3336         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3337 }
3338 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3339
3340 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3341                                         struct device_attribute *attr,
3342                                         char *buf)
3343 {
3344         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3345
3346         return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3347 }
3348 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3349
3350 static ssize_t nvme_sysfs_show_address(struct device *dev,
3351                                          struct device_attribute *attr,
3352                                          char *buf)
3353 {
3354         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3355
3356         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3357 }
3358 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3359
3360 static struct attribute *nvme_dev_attrs[] = {
3361         &dev_attr_reset_controller.attr,
3362         &dev_attr_rescan_controller.attr,
3363         &dev_attr_model.attr,
3364         &dev_attr_serial.attr,
3365         &dev_attr_firmware_rev.attr,
3366         &dev_attr_cntlid.attr,
3367         &dev_attr_delete_controller.attr,
3368         &dev_attr_transport.attr,
3369         &dev_attr_subsysnqn.attr,
3370         &dev_attr_address.attr,
3371         &dev_attr_state.attr,
3372         &dev_attr_numa_node.attr,
3373         &dev_attr_queue_count.attr,
3374         &dev_attr_sqsize.attr,
3375         &dev_attr_hostnqn.attr,
3376         &dev_attr_hostid.attr,
3377         NULL
3378 };
3379
3380 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3381                 struct attribute *a, int n)
3382 {
3383         struct device *dev = container_of(kobj, struct device, kobj);
3384         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3385
3386         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3387                 return 0;
3388         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3389                 return 0;
3390         if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3391                 return 0;
3392         if (a == &dev_attr_hostid.attr && !ctrl->opts)
3393                 return 0;
3394
3395         return a->mode;
3396 }
3397
3398 static struct attribute_group nvme_dev_attrs_group = {
3399         .attrs          = nvme_dev_attrs,
3400         .is_visible     = nvme_dev_attrs_are_visible,
3401 };
3402
3403 static const struct attribute_group *nvme_dev_attr_groups[] = {
3404         &nvme_dev_attrs_group,
3405         NULL,
3406 };
3407
3408 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3409                 unsigned nsid)
3410 {
3411         struct nvme_ns_head *h;
3412
3413         lockdep_assert_held(&subsys->lock);
3414
3415         list_for_each_entry(h, &subsys->nsheads, entry) {
3416                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3417                         return h;
3418         }
3419
3420         return NULL;
3421 }
3422
3423 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3424                 struct nvme_ns_head *new)
3425 {
3426         struct nvme_ns_head *h;
3427
3428         lockdep_assert_held(&subsys->lock);
3429
3430         list_for_each_entry(h, &subsys->nsheads, entry) {
3431                 if (nvme_ns_ids_valid(&new->ids) &&
3432                     nvme_ns_ids_equal(&new->ids, &h->ids))
3433                         return -EINVAL;
3434         }
3435
3436         return 0;
3437 }
3438
3439 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3440                 unsigned nsid, struct nvme_ns_ids *ids)
3441 {
3442         struct nvme_ns_head *head;
3443         size_t size = sizeof(*head);
3444         int ret = -ENOMEM;
3445
3446 #ifdef CONFIG_NVME_MULTIPATH
3447         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3448 #endif
3449
3450         head = kzalloc(size, GFP_KERNEL);
3451         if (!head)
3452                 goto out;
3453         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3454         if (ret < 0)
3455                 goto out_free_head;
3456         head->instance = ret;
3457         INIT_LIST_HEAD(&head->list);
3458         ret = init_srcu_struct(&head->srcu);
3459         if (ret)
3460                 goto out_ida_remove;
3461         head->subsys = ctrl->subsys;
3462         head->ns_id = nsid;
3463         head->ids = *ids;
3464         kref_init(&head->ref);
3465
3466         ret = __nvme_check_ids(ctrl->subsys, head);
3467         if (ret) {
3468                 dev_err(ctrl->device,
3469                         "duplicate IDs for nsid %d\n", nsid);
3470                 goto out_cleanup_srcu;
3471         }
3472
3473         ret = nvme_mpath_alloc_disk(ctrl, head);
3474         if (ret)
3475                 goto out_cleanup_srcu;
3476
3477         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3478
3479         kref_get(&ctrl->subsys->ref);
3480
3481         return head;
3482 out_cleanup_srcu:
3483         cleanup_srcu_struct(&head->srcu);
3484 out_ida_remove:
3485         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3486 out_free_head:
3487         kfree(head);
3488 out:
3489         if (ret > 0)
3490                 ret = blk_status_to_errno(nvme_error_status(ret));
3491         return ERR_PTR(ret);
3492 }
3493
3494 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3495                 struct nvme_id_ns *id)
3496 {
3497         struct nvme_ctrl *ctrl = ns->ctrl;
3498         bool is_shared = id->nmic & (1 << 0);
3499         struct nvme_ns_head *head = NULL;
3500         struct nvme_ns_ids ids;
3501         int ret = 0;
3502
3503         ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3504         if (ret)
3505                 goto out;
3506
3507         mutex_lock(&ctrl->subsys->lock);
3508         head = nvme_find_ns_head(ctrl->subsys, nsid);
3509         if (!head) {
3510                 head = nvme_alloc_ns_head(ctrl, nsid, &ids);
3511                 if (IS_ERR(head)) {
3512                         ret = PTR_ERR(head);
3513                         goto out_unlock;
3514                 }
3515                 head->shared = is_shared;
3516         } else {
3517                 if (!is_shared || !head->shared) {
3518                         dev_err(ctrl->device,
3519                                 "Duplicate unshared namespace %d\n",
3520                                         nsid);
3521                         ret = -EINVAL;
3522                         nvme_put_ns_head(head);
3523                         goto out_unlock;
3524                 }
3525                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3526                         dev_err(ctrl->device,
3527                                 "IDs don't match for shared namespace %d\n",
3528                                         nsid);
3529                         ret = -EINVAL;
3530                         nvme_put_ns_head(head);
3531                         goto out_unlock;
3532                 }
3533         }
3534
3535         list_add_tail(&ns->siblings, &head->list);
3536         ns->head = head;
3537
3538 out_unlock:
3539         mutex_unlock(&ctrl->subsys->lock);
3540 out:
3541         if (ret > 0)
3542                 ret = blk_status_to_errno(nvme_error_status(ret));
3543         return ret;
3544 }
3545
3546 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3547 {
3548         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3549         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3550
3551         return nsa->head->ns_id - nsb->head->ns_id;
3552 }
3553
3554 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3555 {
3556         struct nvme_ns *ns, *ret = NULL;
3557
3558         down_read(&ctrl->namespaces_rwsem);
3559         list_for_each_entry(ns, &ctrl->namespaces, list) {
3560                 if (ns->head->ns_id == nsid) {
3561                         if (!kref_get_unless_zero(&ns->kref))
3562                                 continue;
3563                         ret = ns;
3564                         break;
3565                 }
3566                 if (ns->head->ns_id > nsid)
3567                         break;
3568         }
3569         up_read(&ctrl->namespaces_rwsem);
3570         return ret;
3571 }
3572
3573 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3574 {
3575         struct nvme_ns *ns;
3576         struct gendisk *disk;
3577         struct nvme_id_ns *id;
3578         char disk_name[DISK_NAME_LEN];
3579         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3580
3581         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3582         if (!ns)
3583                 return;
3584
3585         ns->queue = blk_mq_init_queue(ctrl->tagset);
3586         if (IS_ERR(ns->queue))
3587                 goto out_free_ns;
3588
3589         if (ctrl->opts && ctrl->opts->data_digest)
3590                 ns->queue->backing_dev_info->capabilities
3591                         |= BDI_CAP_STABLE_WRITES;
3592
3593         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3594         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3595                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3596
3597         ns->queue->queuedata = ns;
3598         ns->ctrl = ctrl;
3599
3600         kref_init(&ns->kref);
3601         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3602
3603         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3604         nvme_set_queue_limits(ctrl, ns->queue);
3605
3606         ret = nvme_identify_ns(ctrl, nsid, &id);
3607         if (ret)
3608                 goto out_free_queue;
3609
3610         if (id->ncap == 0)      /* no namespace (legacy quirk) */
3611                 goto out_free_id;
3612
3613         ret = nvme_init_ns_head(ns, nsid, id);
3614         if (ret)
3615                 goto out_free_id;
3616         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3617
3618         disk = alloc_disk_node(0, node);
3619         if (!disk)
3620                 goto out_unlink_ns;
3621
3622         disk->fops = &nvme_fops;
3623         disk->private_data = ns;
3624         disk->queue = ns->queue;
3625         disk->flags = flags;
3626         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3627         ns->disk = disk;
3628
3629         __nvme_revalidate_disk(disk, id);
3630
3631         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3632                 ret = nvme_nvm_register(ns, disk_name, node);
3633                 if (ret) {
3634                         dev_warn(ctrl->device, "LightNVM init failure\n");
3635                         goto out_put_disk;
3636                 }
3637         }
3638
3639         down_write(&ctrl->namespaces_rwsem);
3640         list_add_tail(&ns->list, &ctrl->namespaces);
3641         up_write(&ctrl->namespaces_rwsem);
3642
3643         nvme_get_ctrl(ctrl);
3644
3645         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3646
3647         nvme_mpath_add_disk(ns, id);
3648         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3649         kfree(id);
3650
3651         return;
3652  out_put_disk:
3653         /* prevent double queue cleanup */
3654         ns->disk->queue = NULL;
3655         put_disk(ns->disk);
3656  out_unlink_ns:
3657         mutex_lock(&ctrl->subsys->lock);
3658         list_del_rcu(&ns->siblings);
3659         if (list_empty(&ns->head->list))
3660                 list_del_init(&ns->head->entry);
3661         mutex_unlock(&ctrl->subsys->lock);
3662         nvme_put_ns_head(ns->head);
3663  out_free_id:
3664         kfree(id);
3665  out_free_queue:
3666         blk_cleanup_queue(ns->queue);
3667  out_free_ns:
3668         kfree(ns);
3669 }
3670
3671 static void nvme_ns_remove(struct nvme_ns *ns)
3672 {
3673         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3674                 return;
3675
3676         nvme_fault_inject_fini(&ns->fault_inject);
3677
3678         mutex_lock(&ns->ctrl->subsys->lock);
3679         list_del_rcu(&ns->siblings);
3680         if (list_empty(&ns->head->list))
3681                 list_del_init(&ns->head->entry);
3682         mutex_unlock(&ns->ctrl->subsys->lock);
3683
3684         synchronize_rcu(); /* guarantee not available in head->list */
3685         nvme_mpath_clear_current_path(ns);
3686         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3687
3688         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3689                 del_gendisk(ns->disk);
3690                 blk_cleanup_queue(ns->queue);
3691                 if (blk_get_integrity(ns->disk))
3692                         blk_integrity_unregister(ns->disk);
3693         }
3694
3695         down_write(&ns->ctrl->namespaces_rwsem);
3696         list_del_init(&ns->list);
3697         up_write(&ns->ctrl->namespaces_rwsem);
3698
3699         nvme_mpath_check_last_path(ns);
3700         nvme_put_ns(ns);
3701 }
3702
3703 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3704 {
3705         struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3706
3707         if (ns) {
3708                 nvme_ns_remove(ns);
3709                 nvme_put_ns(ns);
3710         }
3711 }
3712
3713 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3714 {
3715         struct nvme_ns *ns;
3716
3717         ns = nvme_find_get_ns(ctrl, nsid);
3718         if (ns) {
3719                 if (ns->disk && revalidate_disk(ns->disk))
3720                         nvme_ns_remove(ns);
3721                 nvme_put_ns(ns);
3722         } else
3723                 nvme_alloc_ns(ctrl, nsid);
3724 }
3725
3726 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3727                                         unsigned nsid)
3728 {
3729         struct nvme_ns *ns, *next;
3730         LIST_HEAD(rm_list);
3731
3732         down_write(&ctrl->namespaces_rwsem);
3733         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3734                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3735                         list_move_tail(&ns->list, &rm_list);
3736         }
3737         up_write(&ctrl->namespaces_rwsem);
3738
3739         list_for_each_entry_safe(ns, next, &rm_list, list)
3740                 nvme_ns_remove(ns);
3741
3742 }
3743
3744 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3745 {
3746         const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3747         __le32 *ns_list;
3748         u32 prev = 0;
3749         int ret = 0, i;
3750
3751         if (nvme_ctrl_limited_cns(ctrl))
3752                 return -EOPNOTSUPP;
3753
3754         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3755         if (!ns_list)
3756                 return -ENOMEM;
3757
3758         for (;;) {
3759                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3760                 if (ret)
3761                         goto free;
3762
3763                 for (i = 0; i < nr_entries; i++) {
3764                         u32 nsid = le32_to_cpu(ns_list[i]);
3765
3766                         if (!nsid)      /* end of the list? */
3767                                 goto out;
3768                         nvme_validate_ns(ctrl, nsid);
3769                         while (++prev < nsid)
3770                                 nvme_ns_remove_by_nsid(ctrl, prev);
3771                 }
3772         }
3773  out:
3774         nvme_remove_invalid_namespaces(ctrl, prev);
3775  free:
3776         kfree(ns_list);
3777         return ret;
3778 }
3779
3780 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3781 {
3782         struct nvme_id_ctrl *id;
3783         u32 nn, i;
3784
3785         if (nvme_identify_ctrl(ctrl, &id))
3786                 return;
3787         nn = le32_to_cpu(id->nn);
3788         kfree(id);
3789
3790         for (i = 1; i <= nn; i++)
3791                 nvme_validate_ns(ctrl, i);
3792
3793         nvme_remove_invalid_namespaces(ctrl, nn);
3794 }
3795
3796 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3797 {
3798         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3799         __le32 *log;
3800         int error;
3801
3802         log = kzalloc(log_size, GFP_KERNEL);
3803         if (!log)
3804                 return;
3805
3806         /*
3807          * We need to read the log to clear the AEN, but we don't want to rely
3808          * on it for the changed namespace information as userspace could have
3809          * raced with us in reading the log page, which could cause us to miss
3810          * updates.
3811          */
3812         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3813                         log_size, 0);
3814         if (error)
3815                 dev_warn(ctrl->device,
3816                         "reading changed ns log failed: %d\n", error);
3817
3818         kfree(log);
3819 }
3820
3821 static void nvme_scan_work(struct work_struct *work)
3822 {
3823         struct nvme_ctrl *ctrl =
3824                 container_of(work, struct nvme_ctrl, scan_work);
3825
3826         /* No tagset on a live ctrl means IO queues could not created */
3827         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3828                 return;
3829
3830         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3831                 dev_info(ctrl->device, "rescanning namespaces.\n");
3832                 nvme_clear_changed_ns_log(ctrl);
3833         }
3834
3835         mutex_lock(&ctrl->scan_lock);
3836         if (nvme_scan_ns_list(ctrl) != 0)
3837                 nvme_scan_ns_sequential(ctrl);
3838         mutex_unlock(&ctrl->scan_lock);
3839
3840         down_write(&ctrl->namespaces_rwsem);
3841         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3842         up_write(&ctrl->namespaces_rwsem);
3843 }
3844
3845 /*
3846  * This function iterates the namespace list unlocked to allow recovery from
3847  * controller failure. It is up to the caller to ensure the namespace list is
3848  * not modified by scan work while this function is executing.
3849  */
3850 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3851 {
3852         struct nvme_ns *ns, *next;
3853         LIST_HEAD(ns_list);
3854
3855         /*
3856          * make sure to requeue I/O to all namespaces as these
3857          * might result from the scan itself and must complete
3858          * for the scan_work to make progress
3859          */
3860         nvme_mpath_clear_ctrl_paths(ctrl);
3861
3862         /* prevent racing with ns scanning */
3863         flush_work(&ctrl->scan_work);
3864
3865         /*
3866          * The dead states indicates the controller was not gracefully
3867          * disconnected. In that case, we won't be able to flush any data while
3868          * removing the namespaces' disks; fail all the queues now to avoid
3869          * potentially having to clean up the failed sync later.
3870          */
3871         if (ctrl->state == NVME_CTRL_DEAD)
3872                 nvme_kill_queues(ctrl);
3873
3874         down_write(&ctrl->namespaces_rwsem);
3875         list_splice_init(&ctrl->namespaces, &ns_list);
3876         up_write(&ctrl->namespaces_rwsem);
3877
3878         list_for_each_entry_safe(ns, next, &ns_list, list)
3879                 nvme_ns_remove(ns);
3880 }
3881 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3882
3883 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3884 {
3885         struct nvme_ctrl *ctrl =
3886                 container_of(dev, struct nvme_ctrl, ctrl_device);
3887         struct nvmf_ctrl_options *opts = ctrl->opts;
3888         int ret;
3889
3890         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3891         if (ret)
3892                 return ret;
3893
3894         if (opts) {
3895                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3896                 if (ret)
3897                         return ret;
3898
3899                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3900                                 opts->trsvcid ?: "none");
3901                 if (ret)
3902                         return ret;
3903
3904                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3905                                 opts->host_traddr ?: "none");
3906         }
3907         return ret;
3908 }
3909
3910 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3911 {
3912         char *envp[2] = { NULL, NULL };
3913         u32 aen_result = ctrl->aen_result;
3914
3915         ctrl->aen_result = 0;
3916         if (!aen_result)
3917                 return;
3918
3919         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3920         if (!envp[0])
3921                 return;
3922         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3923         kfree(envp[0]);
3924 }
3925
3926 static void nvme_async_event_work(struct work_struct *work)
3927 {
3928         struct nvme_ctrl *ctrl =
3929                 container_of(work, struct nvme_ctrl, async_event_work);
3930
3931         nvme_aen_uevent(ctrl);
3932         ctrl->ops->submit_async_event(ctrl);
3933 }
3934
3935 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3936 {
3937
3938         u32 csts;
3939
3940         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3941                 return false;
3942
3943         if (csts == ~0)
3944                 return false;
3945
3946         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3947 }
3948
3949 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3950 {
3951         struct nvme_fw_slot_info_log *log;
3952
3953         log = kmalloc(sizeof(*log), GFP_KERNEL);
3954         if (!log)
3955                 return;
3956
3957         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, log,
3958                         sizeof(*log), 0))
3959                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3960         kfree(log);
3961 }
3962
3963 static void nvme_fw_act_work(struct work_struct *work)
3964 {
3965         struct nvme_ctrl *ctrl = container_of(work,
3966                                 struct nvme_ctrl, fw_act_work);
3967         unsigned long fw_act_timeout;
3968
3969         if (ctrl->mtfa)
3970                 fw_act_timeout = jiffies +
3971                                 msecs_to_jiffies(ctrl->mtfa * 100);
3972         else
3973                 fw_act_timeout = jiffies +
3974                                 msecs_to_jiffies(admin_timeout * 1000);
3975
3976         nvme_stop_queues(ctrl);
3977         while (nvme_ctrl_pp_status(ctrl)) {
3978                 if (time_after(jiffies, fw_act_timeout)) {
3979                         dev_warn(ctrl->device,
3980                                 "Fw activation timeout, reset controller\n");
3981                         nvme_try_sched_reset(ctrl);
3982                         return;
3983                 }
3984                 msleep(100);
3985         }
3986
3987         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
3988                 return;
3989
3990         nvme_start_queues(ctrl);
3991         /* read FW slot information to clear the AER */
3992         nvme_get_fw_slot_info(ctrl);
3993 }
3994
3995 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3996 {
3997         u32 aer_notice_type = (result & 0xff00) >> 8;
3998
3999         trace_nvme_async_event(ctrl, aer_notice_type);
4000
4001         switch (aer_notice_type) {
4002         case NVME_AER_NOTICE_NS_CHANGED:
4003                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4004                 nvme_queue_scan(ctrl);
4005                 break;
4006         case NVME_AER_NOTICE_FW_ACT_STARTING:
4007                 /*
4008                  * We are (ab)using the RESETTING state to prevent subsequent
4009                  * recovery actions from interfering with the controller's
4010                  * firmware activation.
4011                  */
4012                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4013                         queue_work(nvme_wq, &ctrl->fw_act_work);
4014                 break;
4015 #ifdef CONFIG_NVME_MULTIPATH
4016         case NVME_AER_NOTICE_ANA:
4017                 if (!ctrl->ana_log_buf)
4018                         break;
4019                 queue_work(nvme_wq, &ctrl->ana_work);
4020                 break;
4021 #endif
4022         case NVME_AER_NOTICE_DISC_CHANGED:
4023                 ctrl->aen_result = result;
4024                 break;
4025         default:
4026                 dev_warn(ctrl->device, "async event result %08x\n", result);
4027         }
4028 }
4029
4030 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4031                 volatile union nvme_result *res)
4032 {
4033         u32 result = le32_to_cpu(res->u32);
4034         u32 aer_type = result & 0x07;
4035
4036         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4037                 return;
4038
4039         switch (aer_type) {
4040         case NVME_AER_NOTICE:
4041                 nvme_handle_aen_notice(ctrl, result);
4042                 break;
4043         case NVME_AER_ERROR:
4044         case NVME_AER_SMART:
4045         case NVME_AER_CSS:
4046         case NVME_AER_VS:
4047                 trace_nvme_async_event(ctrl, aer_type);
4048                 ctrl->aen_result = result;
4049                 break;
4050         default:
4051                 break;
4052         }
4053         queue_work(nvme_wq, &ctrl->async_event_work);
4054 }
4055 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4056
4057 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4058 {
4059         nvme_mpath_stop(ctrl);
4060         nvme_stop_keep_alive(ctrl);
4061         flush_work(&ctrl->async_event_work);
4062         cancel_work_sync(&ctrl->fw_act_work);
4063 }
4064 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4065
4066 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4067 {
4068         if (ctrl->kato)
4069                 nvme_start_keep_alive(ctrl);
4070
4071         nvme_enable_aen(ctrl);
4072
4073         if (ctrl->queue_count > 1) {
4074                 nvme_queue_scan(ctrl);
4075                 nvme_start_queues(ctrl);
4076         }
4077         ctrl->created = true;
4078 }
4079 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4080
4081 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4082 {
4083         nvme_fault_inject_fini(&ctrl->fault_inject);
4084         dev_pm_qos_hide_latency_tolerance(ctrl->device);
4085         cdev_device_del(&ctrl->cdev, ctrl->device);
4086         nvme_put_ctrl(ctrl);
4087 }
4088 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4089
4090 static void nvme_free_ctrl(struct device *dev)
4091 {
4092         struct nvme_ctrl *ctrl =
4093                 container_of(dev, struct nvme_ctrl, ctrl_device);
4094         struct nvme_subsystem *subsys = ctrl->subsys;
4095
4096         if (subsys && ctrl->instance != subsys->instance)
4097                 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4098
4099         kfree(ctrl->effects);
4100         nvme_mpath_uninit(ctrl);
4101         __free_page(ctrl->discard_page);
4102
4103         if (subsys) {
4104                 mutex_lock(&nvme_subsystems_lock);
4105                 list_del(&ctrl->subsys_entry);
4106                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4107                 mutex_unlock(&nvme_subsystems_lock);
4108         }
4109
4110         ctrl->ops->free_ctrl(ctrl);
4111
4112         if (subsys)
4113                 nvme_put_subsystem(subsys);
4114 }
4115
4116 /*
4117  * Initialize a NVMe controller structures.  This needs to be called during
4118  * earliest initialization so that we have the initialized structured around
4119  * during probing.
4120  */
4121 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4122                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4123 {
4124         int ret;
4125
4126         ctrl->state = NVME_CTRL_NEW;
4127         spin_lock_init(&ctrl->lock);
4128         mutex_init(&ctrl->scan_lock);
4129         INIT_LIST_HEAD(&ctrl->namespaces);
4130         init_rwsem(&ctrl->namespaces_rwsem);
4131         ctrl->dev = dev;
4132         ctrl->ops = ops;
4133         ctrl->quirks = quirks;
4134         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4135         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4136         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4137         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4138         init_waitqueue_head(&ctrl->state_wq);
4139
4140         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4141         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4142         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4143
4144         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4145                         PAGE_SIZE);
4146         ctrl->discard_page = alloc_page(GFP_KERNEL);
4147         if (!ctrl->discard_page) {
4148                 ret = -ENOMEM;
4149                 goto out;
4150         }
4151
4152         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4153         if (ret < 0)
4154                 goto out;
4155         ctrl->instance = ret;
4156
4157         device_initialize(&ctrl->ctrl_device);
4158         ctrl->device = &ctrl->ctrl_device;
4159         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4160         ctrl->device->class = nvme_class;
4161         ctrl->device->parent = ctrl->dev;
4162         ctrl->device->groups = nvme_dev_attr_groups;
4163         ctrl->device->release = nvme_free_ctrl;
4164         dev_set_drvdata(ctrl->device, ctrl);
4165         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4166         if (ret)
4167                 goto out_release_instance;
4168
4169         nvme_get_ctrl(ctrl);
4170         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4171         ctrl->cdev.owner = ops->module;
4172         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4173         if (ret)
4174                 goto out_free_name;
4175
4176         /*
4177          * Initialize latency tolerance controls.  The sysfs files won't
4178          * be visible to userspace unless the device actually supports APST.
4179          */
4180         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4181         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4182                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4183
4184         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4185
4186         return 0;
4187 out_free_name:
4188         nvme_put_ctrl(ctrl);
4189         kfree_const(ctrl->device->kobj.name);
4190 out_release_instance:
4191         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4192 out:
4193         if (ctrl->discard_page)
4194                 __free_page(ctrl->discard_page);
4195         return ret;
4196 }
4197 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4198
4199 /**
4200  * nvme_kill_queues(): Ends all namespace queues
4201  * @ctrl: the dead controller that needs to end
4202  *
4203  * Call this function when the driver determines it is unable to get the
4204  * controller in a state capable of servicing IO.
4205  */
4206 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4207 {
4208         struct nvme_ns *ns;
4209
4210         down_read(&ctrl->namespaces_rwsem);
4211
4212         /* Forcibly unquiesce queues to avoid blocking dispatch */
4213         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4214                 blk_mq_unquiesce_queue(ctrl->admin_q);
4215
4216         list_for_each_entry(ns, &ctrl->namespaces, list)
4217                 nvme_set_queue_dying(ns);
4218
4219         up_read(&ctrl->namespaces_rwsem);
4220 }
4221 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4222
4223 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4224 {
4225         struct nvme_ns *ns;
4226
4227         down_read(&ctrl->namespaces_rwsem);
4228         list_for_each_entry(ns, &ctrl->namespaces, list)
4229                 blk_mq_unfreeze_queue(ns->queue);
4230         up_read(&ctrl->namespaces_rwsem);
4231 }
4232 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4233
4234 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4235 {
4236         struct nvme_ns *ns;
4237
4238         down_read(&ctrl->namespaces_rwsem);
4239         list_for_each_entry(ns, &ctrl->namespaces, list) {
4240                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4241                 if (timeout <= 0)
4242                         break;
4243         }
4244         up_read(&ctrl->namespaces_rwsem);
4245 }
4246 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4247
4248 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4249 {
4250         struct nvme_ns *ns;
4251
4252         down_read(&ctrl->namespaces_rwsem);
4253         list_for_each_entry(ns, &ctrl->namespaces, list)
4254                 blk_mq_freeze_queue_wait(ns->queue);
4255         up_read(&ctrl->namespaces_rwsem);
4256 }
4257 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4258
4259 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4260 {
4261         struct nvme_ns *ns;
4262
4263         down_read(&ctrl->namespaces_rwsem);
4264         list_for_each_entry(ns, &ctrl->namespaces, list)
4265                 blk_freeze_queue_start(ns->queue);
4266         up_read(&ctrl->namespaces_rwsem);
4267 }
4268 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4269
4270 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4271 {
4272         struct nvme_ns *ns;
4273
4274         down_read(&ctrl->namespaces_rwsem);
4275         list_for_each_entry(ns, &ctrl->namespaces, list)
4276                 blk_mq_quiesce_queue(ns->queue);
4277         up_read(&ctrl->namespaces_rwsem);
4278 }
4279 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4280
4281 void nvme_start_queues(struct nvme_ctrl *ctrl)
4282 {
4283         struct nvme_ns *ns;
4284
4285         down_read(&ctrl->namespaces_rwsem);
4286         list_for_each_entry(ns, &ctrl->namespaces, list)
4287                 blk_mq_unquiesce_queue(ns->queue);
4288         up_read(&ctrl->namespaces_rwsem);
4289 }
4290 EXPORT_SYMBOL_GPL(nvme_start_queues);
4291
4292
4293 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4294 {
4295         struct nvme_ns *ns;
4296
4297         down_read(&ctrl->namespaces_rwsem);
4298         list_for_each_entry(ns, &ctrl->namespaces, list)
4299                 blk_sync_queue(ns->queue);
4300         up_read(&ctrl->namespaces_rwsem);
4301
4302         if (ctrl->admin_q)
4303                 blk_sync_queue(ctrl->admin_q);
4304 }
4305 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4306
4307 /*
4308  * Check we didn't inadvertently grow the command structure sizes:
4309  */
4310 static inline void _nvme_check_size(void)
4311 {
4312         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4313         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4314         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4315         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4316         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4317         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4318         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4319         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4320         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4321         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4322         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4323         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4324         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4325         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4326         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4327         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4328         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4329 }
4330
4331
4332 static int __init nvme_core_init(void)
4333 {
4334         int result = -ENOMEM;
4335
4336         _nvme_check_size();
4337
4338         nvme_wq = alloc_workqueue("nvme-wq",
4339                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4340         if (!nvme_wq)
4341                 goto out;
4342
4343         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4344                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4345         if (!nvme_reset_wq)
4346                 goto destroy_wq;
4347
4348         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4349                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4350         if (!nvme_delete_wq)
4351                 goto destroy_reset_wq;
4352
4353         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4354         if (result < 0)
4355                 goto destroy_delete_wq;
4356
4357         nvme_class = class_create(THIS_MODULE, "nvme");
4358         if (IS_ERR(nvme_class)) {
4359                 result = PTR_ERR(nvme_class);
4360                 goto unregister_chrdev;
4361         }
4362         nvme_class->dev_uevent = nvme_class_uevent;
4363
4364         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4365         if (IS_ERR(nvme_subsys_class)) {
4366                 result = PTR_ERR(nvme_subsys_class);
4367                 goto destroy_class;
4368         }
4369         return 0;
4370
4371 destroy_class:
4372         class_destroy(nvme_class);
4373 unregister_chrdev:
4374         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4375 destroy_delete_wq:
4376         destroy_workqueue(nvme_delete_wq);
4377 destroy_reset_wq:
4378         destroy_workqueue(nvme_reset_wq);
4379 destroy_wq:
4380         destroy_workqueue(nvme_wq);
4381 out:
4382         return result;
4383 }
4384
4385 static void __exit nvme_core_exit(void)
4386 {
4387         class_destroy(nvme_subsys_class);
4388         class_destroy(nvme_class);
4389         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4390         destroy_workqueue(nvme_delete_wq);
4391         destroy_workqueue(nvme_reset_wq);
4392         destroy_workqueue(nvme_wq);
4393         ida_destroy(&nvme_instance_ida);
4394 }
4395
4396 MODULE_LICENSE("GPL");
4397 MODULE_VERSION("1.0");
4398 module_init(nvme_core_init);
4399 module_exit(nvme_core_exit);