drivers/nvme/host/core.c

   1 /*
   2  * NVM Express device driver
   3  * Copyright (c) 2011-2014, Intel Corporation.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  */
  14
  15 #include <linux/blkdev.h>
  16 #include <linux/blk-mq.h>
  17 #include <linux/delay.h>
  18 #include <linux/errno.h>
  19 #include <linux/hdreg.h>
  20 #include <linux/kernel.h>
  21 #include <linux/module.h>
  22 #include <linux/list_sort.h>
  23 #include <linux/slab.h>
  24 #include <linux/types.h>
  25 #include <linux/pr.h>
  26 #include <linux/ptrace.h>
  27 #include <linux/nvme_ioctl.h>
  28 #include <linux/t10-pi.h>
  29 #include <linux/pm_qos.h>
  30 #include <asm/unaligned.h>
  31
  32 #define CREATE_TRACE_POINTS
  33 #include "trace.h"
  34
  35 #include "nvme.h"
  36 #include "fabrics.h"
  37
  38 #define NVME_MINORS             (1U << MINORBITS)
  39
  40 unsigned int admin_timeout = 60;
  41 module_param(admin_timeout, uint, 0644);
  42 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  43 EXPORT_SYMBOL_GPL(admin_timeout);
  44
  45 unsigned int nvme_io_timeout = 30;
  46 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  47 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  48 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  49
  50 static unsigned char shutdown_timeout = 5;
  51 module_param(shutdown_timeout, byte, 0644);
  52 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  53
  54 static u8 nvme_max_retries = 5;
  55 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  56 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  57
  58 static unsigned long default_ps_max_latency_us = 100000;
  59 module_param(default_ps_max_latency_us, ulong, 0644);
  60 MODULE_PARM_DESC(default_ps_max_latency_us,
  61                  "max power saving latency for new devices; use PM QOS to change per device");
  62
  63 static bool force_apst;
  64 module_param(force_apst, bool, 0644);
  65 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  66
  67 static bool streams;
  68 module_param(streams, bool, 0644);
  69 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  70
  71 /*
  72  * nvme_wq - hosts nvme related works that are not reset or delete
  73  * nvme_reset_wq - hosts nvme reset works
  74  * nvme_delete_wq - hosts nvme delete works
  75  *
  76  * nvme_wq will host works such are scan, aen handling, fw activation,
  77  * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
  78  * runs reset works which also flush works hosted on nvme_wq for
  79  * serialization purposes. nvme_delete_wq host controller deletion
  80  * works which flush reset works for serialization.
  81  */
  82 struct workqueue_struct *nvme_wq;
  83 EXPORT_SYMBOL_GPL(nvme_wq);
  84
  85 struct workqueue_struct *nvme_reset_wq;
  86 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  87
  88 struct workqueue_struct *nvme_delete_wq;
  89 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  90
  91 static DEFINE_IDA(nvme_subsystems_ida);
  92 static LIST_HEAD(nvme_subsystems);
  93 static DEFINE_MUTEX(nvme_subsystems_lock);
  94
  95 static DEFINE_IDA(nvme_instance_ida);
  96 static dev_t nvme_chr_devt;
  97 static struct class *nvme_class;
  98 static struct class *nvme_subsys_class;
  99
 100 static int nvme_revalidate_disk(struct gendisk *disk);
 101 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
 102 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
 103                                            unsigned nsid);
 104
 105 static void nvme_set_queue_dying(struct nvme_ns *ns)
 106 {
 107         /*
 108          * Revalidating a dead namespace sets capacity to 0. This will end
 109          * buffered writers dirtying pages that can't be synced.
 110          */
 111         if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 112                 return;
 113         revalidate_disk(ns->disk);
 114         blk_set_queue_dying(ns->queue);
 115         /* Forcibly unquiesce queues to avoid blocking dispatch */
 116         blk_mq_unquiesce_queue(ns->queue);
 117 }
 118
 119 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 120 {
 121         /*
 122          * Only new queue scan work when admin and IO queues are both alive
 123          */
 124         if (ctrl->state == NVME_CTRL_LIVE)
 125                 queue_work(nvme_wq, &ctrl->scan_work);
 126 }
 127
 128 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 129 {
 130         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 131                 return -EBUSY;
 132         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 133                 return -EBUSY;
 134         return 0;
 135 }
 136 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 137
 138 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 139 {
 140         int ret;
 141
 142         ret = nvme_reset_ctrl(ctrl);
 143         if (!ret) {
 144                 flush_work(&ctrl->reset_work);
 145                 if (ctrl->state != NVME_CTRL_LIVE &&
 146                     ctrl->state != NVME_CTRL_ADMIN_ONLY)
 147                         ret = -ENETRESET;
 148         }
 149
 150         return ret;
 151 }
 152 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 153
 154 static void nvme_delete_ctrl_work(struct work_struct *work)
 155 {
 156         struct nvme_ctrl *ctrl =
 157                 container_of(work, struct nvme_ctrl, delete_work);
 158
 159         dev_info(ctrl->device,
 160                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 161
 162         flush_work(&ctrl->reset_work);
 163         nvme_stop_ctrl(ctrl);
 164         nvme_remove_namespaces(ctrl);
 165         ctrl->ops->delete_ctrl(ctrl);
 166         nvme_uninit_ctrl(ctrl);
 167         nvme_put_ctrl(ctrl);
 168 }
 169
 170 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 171 {
 172         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 173                 return -EBUSY;
 174         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 175                 return -EBUSY;
 176         return 0;
 177 }
 178 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 179
 180 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 181 {
 182         int ret = 0;
 183
 184         /*
 185          * Keep a reference until the work is flushed since ->delete_ctrl
 186          * can free the controller.
 187          */
 188         nvme_get_ctrl(ctrl);
 189         ret = nvme_delete_ctrl(ctrl);
 190         if (!ret)
 191                 flush_work(&ctrl->delete_work);
 192         nvme_put_ctrl(ctrl);
 193         return ret;
 194 }
 195 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
 196
 197 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 198 {
 199         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 200 }
 201
 202 static blk_status_t nvme_error_status(struct request *req)
 203 {
 204         switch (nvme_req(req)->status & 0x7ff) {
 205         case NVME_SC_SUCCESS:
 206                 return BLK_STS_OK;
 207         case NVME_SC_CAP_EXCEEDED:
 208                 return BLK_STS_NOSPC;
 209         case NVME_SC_LBA_RANGE:
 210                 return BLK_STS_TARGET;
 211         case NVME_SC_BAD_ATTRIBUTES:
 212         case NVME_SC_ONCS_NOT_SUPPORTED:
 213         case NVME_SC_INVALID_OPCODE:
 214         case NVME_SC_INVALID_FIELD:
 215         case NVME_SC_INVALID_NS:
 216                 return BLK_STS_NOTSUPP;
 217         case NVME_SC_WRITE_FAULT:
 218         case NVME_SC_READ_ERROR:
 219         case NVME_SC_UNWRITTEN_BLOCK:
 220         case NVME_SC_ACCESS_DENIED:
 221         case NVME_SC_READ_ONLY:
 222         case NVME_SC_COMPARE_FAILED:
 223                 return BLK_STS_MEDIUM;
 224         case NVME_SC_GUARD_CHECK:
 225         case NVME_SC_APPTAG_CHECK:
 226         case NVME_SC_REFTAG_CHECK:
 227         case NVME_SC_INVALID_PI:
 228                 return BLK_STS_PROTECTION;
 229         case NVME_SC_RESERVATION_CONFLICT:
 230                 return BLK_STS_NEXUS;
 231         default:
 232                 return BLK_STS_IOERR;
 233         }
 234 }
 235
 236 static inline bool nvme_req_needs_retry(struct request *req)
 237 {
 238         if (blk_noretry_request(req))
 239                 return false;
 240         if (nvme_req(req)->status & NVME_SC_DNR)
 241                 return false;
 242         if (nvme_req(req)->retries >= nvme_max_retries)
 243                 return false;
 244         return true;
 245 }
 246
 247 static void nvme_retry_req(struct request *req)
 248 {
 249         struct nvme_ns *ns = req->q->queuedata;
 250         unsigned long delay = 0;
 251         u16 crd;
 252
 253         /* The mask and shift result must be <= 3 */
 254         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 255         if (ns && crd)
 256                 delay = ns->ctrl->crdt[crd - 1] * 100;
 257
 258         nvme_req(req)->retries++;
 259         blk_mq_requeue_request(req, false);
 260         blk_mq_delay_kick_requeue_list(req->q, delay);
 261 }
 262
 263 void nvme_complete_rq(struct request *req)
 264 {
 265         blk_status_t status = nvme_error_status(req);
 266
 267         trace_nvme_complete_rq(req);
 268
 269         if (nvme_req(req)->ctrl->kas)
 270                 nvme_req(req)->ctrl->comp_seen = true;
 271
 272         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 273                 if ((req->cmd_flags & REQ_NVME_MPATH) &&
 274                     blk_path_error(status)) {
 275                         nvme_failover_req(req);
 276                         return;
 277                 }
 278
 279                 if (!blk_queue_dying(req->q)) {
 280                         nvme_retry_req(req);
 281                         return;
 282                 }
 283         }
 284         blk_mq_end_request(req, status);
 285 }
 286 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 287
 288 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 289 {
 290         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 291                                 "Cancelling I/O %d", req->tag);
 292
 293         nvme_req(req)->status = NVME_SC_ABORT_REQ;
 294         blk_mq_complete_request(req);
 295         return true;
 296 }
 297 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 298
 299 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 300                 enum nvme_ctrl_state new_state)
 301 {
 302         enum nvme_ctrl_state old_state;
 303         unsigned long flags;
 304         bool changed = false;
 305
 306         spin_lock_irqsave(&ctrl->lock, flags);
 307
 308         old_state = ctrl->state;
 309         switch (new_state) {
 310         case NVME_CTRL_ADMIN_ONLY:
 311                 switch (old_state) {
 312                 case NVME_CTRL_CONNECTING:
 313                         changed = true;
 314                         /* FALLTHRU */
 315                 default:
 316                         break;
 317                 }
 318                 break;
 319         case NVME_CTRL_LIVE:
 320                 switch (old_state) {
 321                 case NVME_CTRL_NEW:
 322                 case NVME_CTRL_RESETTING:
 323                 case NVME_CTRL_CONNECTING:
 324                         changed = true;
 325                         /* FALLTHRU */
 326                 default:
 327                         break;
 328                 }
 329                 break;
 330         case NVME_CTRL_RESETTING:
 331                 switch (old_state) {
 332                 case NVME_CTRL_NEW:
 333                 case NVME_CTRL_LIVE:
 334                 case NVME_CTRL_ADMIN_ONLY:
 335                         changed = true;
 336                         /* FALLTHRU */
 337                 default:
 338                         break;
 339                 }
 340                 break;
 341         case NVME_CTRL_CONNECTING:
 342                 switch (old_state) {
 343                 case NVME_CTRL_NEW:
 344                 case NVME_CTRL_RESETTING:
 345                         changed = true;
 346                         /* FALLTHRU */
 347                 default:
 348                         break;
 349                 }
 350                 break;
 351         case NVME_CTRL_DELETING:
 352                 switch (old_state) {
 353                 case NVME_CTRL_LIVE:
 354                 case NVME_CTRL_ADMIN_ONLY:
 355                 case NVME_CTRL_RESETTING:
 356                 case NVME_CTRL_CONNECTING:
 357                         changed = true;
 358                         /* FALLTHRU */
 359                 default:
 360                         break;
 361                 }
 362                 break;
 363         case NVME_CTRL_DEAD:
 364                 switch (old_state) {
 365                 case NVME_CTRL_DELETING:
 366                         changed = true;
 367                         /* FALLTHRU */
 368                 default:
 369                         break;
 370                 }
 371                 break;
 372         default:
 373                 break;
 374         }
 375
 376         if (changed)
 377                 ctrl->state = new_state;
 378
 379         spin_unlock_irqrestore(&ctrl->lock, flags);
 380         if (changed && ctrl->state == NVME_CTRL_LIVE)
 381                 nvme_kick_requeue_lists(ctrl);
 382         return changed;
 383 }
 384 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 385
 386 static void nvme_free_ns_head(struct kref *ref)
 387 {
 388         struct nvme_ns_head *head =
 389                 container_of(ref, struct nvme_ns_head, ref);
 390
 391         nvme_mpath_remove_disk(head);
 392         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 393         list_del_init(&head->entry);
 394         cleanup_srcu_struct_quiesced(&head->srcu);
 395         nvme_put_subsystem(head->subsys);
 396         kfree(head);
 397 }
 398
 399 static void nvme_put_ns_head(struct nvme_ns_head *head)
 400 {
 401         kref_put(&head->ref, nvme_free_ns_head);
 402 }
 403
 404 static void nvme_free_ns(struct kref *kref)
 405 {
 406         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 407
 408         if (ns->ndev)
 409                 nvme_nvm_unregister(ns);
 410
 411         put_disk(ns->disk);
 412         nvme_put_ns_head(ns->head);
 413         nvme_put_ctrl(ns->ctrl);
 414         kfree(ns);
 415 }
 416
 417 static void nvme_put_ns(struct nvme_ns *ns)
 418 {
 419         kref_put(&ns->kref, nvme_free_ns);
 420 }
 421
 422 static inline void nvme_clear_nvme_request(struct request *req)
 423 {
 424         if (!(req->rq_flags & RQF_DONTPREP)) {
 425                 nvme_req(req)->retries = 0;
 426                 nvme_req(req)->flags = 0;
 427                 req->rq_flags |= RQF_DONTPREP;
 428         }
 429 }
 430
 431 struct request *nvme_alloc_request(struct request_queue *q,
 432                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 433 {
 434         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 435         struct request *req;
 436
 437         if (qid == NVME_QID_ANY) {
 438                 req = blk_mq_alloc_request(q, op, flags);
 439         } else {
 440                 req = blk_mq_alloc_request_hctx(q, op, flags,
 441                                 qid ? qid - 1 : 0);
 442         }
 443         if (IS_ERR(req))
 444                 return req;
 445
 446         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 447         nvme_clear_nvme_request(req);
 448         nvme_req(req)->cmd = cmd;
 449
 450         return req;
 451 }
 452 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 453
 454 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 455 {
 456         struct nvme_command c;
 457
 458         memset(&c, 0, sizeof(c));
 459
 460         c.directive.opcode = nvme_admin_directive_send;
 461         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 462         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 463         c.directive.dtype = NVME_DIR_IDENTIFY;
 464         c.directive.tdtype = NVME_DIR_STREAMS;
 465         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 466
 467         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 468 }
 469
 470 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 471 {
 472         return nvme_toggle_streams(ctrl, false);
 473 }
 474
 475 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 476 {
 477         return nvme_toggle_streams(ctrl, true);
 478 }
 479
 480 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 481                                   struct streams_directive_params *s, u32 nsid)
 482 {
 483         struct nvme_command c;
 484
 485         memset(&c, 0, sizeof(c));
 486         memset(s, 0, sizeof(*s));
 487
 488         c.directive.opcode = nvme_admin_directive_recv;
 489         c.directive.nsid = cpu_to_le32(nsid);
 490         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
 491         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 492         c.directive.dtype = NVME_DIR_STREAMS;
 493
 494         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 495 }
 496
 497 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 498 {
 499         struct streams_directive_params s;
 500         int ret;
 501
 502         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 503                 return 0;
 504         if (!streams)
 505                 return 0;
 506
 507         ret = nvme_enable_streams(ctrl);
 508         if (ret)
 509                 return ret;
 510
 511         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 512         if (ret)
 513                 return ret;
 514
 515         ctrl->nssa = le16_to_cpu(s.nssa);
 516         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 517                 dev_info(ctrl->device, "too few streams (%u) available\n",
 518                                         ctrl->nssa);
 519                 nvme_disable_streams(ctrl);
 520                 return 0;
 521         }
 522
 523         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 524         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 525         return 0;
 526 }
 527
 528 /*
 529  * Check if 'req' has a write hint associated with it. If it does, assign
 530  * a valid namespace stream to the write.
 531  */
 532 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 533                                      struct request *req, u16 *control,
 534                                      u32 *dsmgmt)
 535 {
 536         enum rw_hint streamid = req->write_hint;
 537
 538         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 539                 streamid = 0;
 540         else {
 541                 streamid--;
 542                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 543                         return;
 544
 545                 *control |= NVME_RW_DTYPE_STREAMS;
 546                 *dsmgmt |= streamid << 16;
 547         }
 548
 549         if (streamid < ARRAY_SIZE(req->q->write_hints))
 550                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 551 }
 552
 553 static inline void nvme_setup_flush(struct nvme_ns *ns,
 554                 struct nvme_command *cmnd)
 555 {
 556         cmnd->common.opcode = nvme_cmd_flush;
 557         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 558 }
 559
 560 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 561                 struct nvme_command *cmnd)
 562 {
 563         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 564         struct nvme_dsm_range *range;
 565         struct bio *bio;
 566
 567         range = kmalloc_array(segments, sizeof(*range),
 568                                 GFP_ATOMIC | __GFP_NOWARN);
 569         if (!range) {
 570                 /*
 571                  * If we fail allocation our range, fallback to the controller
 572                  * discard page. If that's also busy, it's safe to return
 573                  * busy, as we know we can make progress once that's freed.
 574                  */
 575                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 576                         return BLK_STS_RESOURCE;
 577
 578                 range = page_address(ns->ctrl->discard_page);
 579         }
 580
 581         __rq_for_each_bio(bio, req) {
 582                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 583                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 584
 585                 if (n < segments) {
 586                         range[n].cattr = cpu_to_le32(0);
 587                         range[n].nlb = cpu_to_le32(nlb);
 588                         range[n].slba = cpu_to_le64(slba);
 589                 }
 590                 n++;
 591         }
 592
 593         if (WARN_ON_ONCE(n != segments)) {
 594                 if (virt_to_page(range) == ns->ctrl->discard_page)
 595                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 596                 else
 597                         kfree(range);
 598                 return BLK_STS_IOERR;
 599         }
 600
 601         cmnd->dsm.opcode = nvme_cmd_dsm;
 602         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 603         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 604         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 605
 606         req->special_vec.bv_page = virt_to_page(range);
 607         req->special_vec.bv_offset = offset_in_page(range);
 608         req->special_vec.bv_len = sizeof(*range) * segments;
 609         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 610
 611         return BLK_STS_OK;
 612 }
 613
 614 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 615                 struct request *req, struct nvme_command *cmnd)
 616 {
 617         struct nvme_ctrl *ctrl = ns->ctrl;
 618         u16 control = 0;
 619         u32 dsmgmt = 0;
 620
 621         if (req->cmd_flags & REQ_FUA)
 622                 control |= NVME_RW_FUA;
 623         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 624                 control |= NVME_RW_LR;
 625
 626         if (req->cmd_flags & REQ_RAHEAD)
 627                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 628
 629         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 630         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 631         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 632         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 633
 634         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 635                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 636
 637         if (ns->ms) {
 638                 /*
 639                  * If formated with metadata, the block layer always provides a
 640                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 641                  * we enable the PRACT bit for protection information or set the
 642                  * namespace capacity to zero to prevent any I/O.
 643                  */
 644                 if (!blk_integrity_rq(req)) {
 645                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 646                                 return BLK_STS_NOTSUPP;
 647                         control |= NVME_RW_PRINFO_PRACT;
 648                 } else if (req_op(req) == REQ_OP_WRITE) {
 649                         t10_pi_prepare(req, ns->pi_type);
 650                 }
 651
 652                 switch (ns->pi_type) {
 653                 case NVME_NS_DPS_PI_TYPE3:
 654                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 655                         break;
 656                 case NVME_NS_DPS_PI_TYPE1:
 657                 case NVME_NS_DPS_PI_TYPE2:
 658                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 659                                         NVME_RW_PRINFO_PRCHK_REF;
 660                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 661                         break;
 662                 }
 663         }
 664
 665         cmnd->rw.control = cpu_to_le16(control);
 666         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 667         return 0;
 668 }
 669
 670 void nvme_cleanup_cmd(struct request *req)
 671 {
 672         if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
 673             nvme_req(req)->status == 0) {
 674                 struct nvme_ns *ns = req->rq_disk->private_data;
 675
 676                 t10_pi_complete(req, ns->pi_type,
 677                                 blk_rq_bytes(req) >> ns->lba_shift);
 678         }
 679         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 680                 struct nvme_ns *ns = req->rq_disk->private_data;
 681                 struct page *page = req->special_vec.bv_page;
 682
 683                 if (page == ns->ctrl->discard_page)
 684                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 685                 else
 686                         kfree(page_address(page) + req->special_vec.bv_offset);
 687         }
 688 }
 689 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 690
 691 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 692                 struct nvme_command *cmd)
 693 {
 694         blk_status_t ret = BLK_STS_OK;
 695
 696         nvme_clear_nvme_request(req);
 697
 698         memset(cmd, 0, sizeof(*cmd));
 699         switch (req_op(req)) {
 700         case REQ_OP_DRV_IN:
 701         case REQ_OP_DRV_OUT:
 702                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 703                 break;
 704         case REQ_OP_FLUSH:
 705                 nvme_setup_flush(ns, cmd);
 706                 break;
 707         case REQ_OP_WRITE_ZEROES:
 708                 /* currently only aliased to deallocate for a few ctrls: */
 709         case REQ_OP_DISCARD:
 710                 ret = nvme_setup_discard(ns, req, cmd);
 711                 break;
 712         case REQ_OP_READ:
 713         case REQ_OP_WRITE:
 714                 ret = nvme_setup_rw(ns, req, cmd);
 715                 break;
 716         default:
 717                 WARN_ON_ONCE(1);
 718                 return BLK_STS_IOERR;
 719         }
 720
 721         cmd->common.command_id = req->tag;
 722         trace_nvme_setup_cmd(req, cmd);
 723         return ret;
 724 }
 725 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 726
 727 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 728 {
 729         struct completion *waiting = rq->end_io_data;
 730
 731         rq->end_io_data = NULL;
 732         complete(waiting);
 733 }
 734
 735 static void nvme_execute_rq_polled(struct request_queue *q,
 736                 struct gendisk *bd_disk, struct request *rq, int at_head)
 737 {
 738         DECLARE_COMPLETION_ONSTACK(wait);
 739
 740         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
 741
 742         rq->cmd_flags |= REQ_HIPRI;
 743         rq->end_io_data = &wait;
 744         blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
 745
 746         while (!completion_done(&wait)) {
 747                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
 748                 cond_resched();
 749         }
 750 }
 751
 752 /*
 753  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 754  * if the result is positive, it's an NVM Express status code
 755  */
 756 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 757                 union nvme_result *result, void *buffer, unsigned bufflen,
 758                 unsigned timeout, int qid, int at_head,
 759                 blk_mq_req_flags_t flags, bool poll)
 760 {
 761         struct request *req;
 762         int ret;
 763
 764         req = nvme_alloc_request(q, cmd, flags, qid);
 765         if (IS_ERR(req))
 766                 return PTR_ERR(req);
 767
 768         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 769
 770         if (buffer && bufflen) {
 771                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 772                 if (ret)
 773                         goto out;
 774         }
 775
 776         if (poll)
 777                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
 778         else
 779                 blk_execute_rq(req->q, NULL, req, at_head);
 780         if (result)
 781                 *result = nvme_req(req)->result;
 782         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 783                 ret = -EINTR;
 784         else
 785                 ret = nvme_req(req)->status;
 786  out:
 787         blk_mq_free_request(req);
 788         return ret;
 789 }
 790 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 791
 792 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 793                 void *buffer, unsigned bufflen)
 794 {
 795         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 796                         NVME_QID_ANY, 0, 0, false);
 797 }
 798 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 799
 800 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 801                 unsigned len, u32 seed, bool write)
 802 {
 803         struct bio_integrity_payload *bip;
 804         int ret = -ENOMEM;
 805         void *buf;
 806
 807         buf = kmalloc(len, GFP_KERNEL);
 808         if (!buf)
 809                 goto out;
 810
 811         ret = -EFAULT;
 812         if (write && copy_from_user(buf, ubuf, len))
 813                 goto out_free_meta;
 814
 815         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 816         if (IS_ERR(bip)) {
 817                 ret = PTR_ERR(bip);
 818                 goto out_free_meta;
 819         }
 820
 821         bip->bip_iter.bi_size = len;
 822         bip->bip_iter.bi_sector = seed;
 823         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 824                         offset_in_page(buf));
 825         if (ret == len)
 826                 return buf;
 827         ret = -ENOMEM;
 828 out_free_meta:
 829         kfree(buf);
 830 out:
 831         return ERR_PTR(ret);
 832 }
 833
 834 static int nvme_submit_user_cmd(struct request_queue *q,
 835                 struct nvme_command *cmd, void __user *ubuffer,
 836                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 837                 u32 meta_seed, u32 *result, unsigned timeout)
 838 {
 839         bool write = nvme_is_write(cmd);
 840         struct nvme_ns *ns = q->queuedata;
 841         struct gendisk *disk = ns ? ns->disk : NULL;
 842         struct request *req;
 843         struct bio *bio = NULL;
 844         void *meta = NULL;
 845         int ret;
 846
 847         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 848         if (IS_ERR(req))
 849                 return PTR_ERR(req);
 850
 851         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 852         nvme_req(req)->flags |= NVME_REQ_USERCMD;
 853
 854         if (ubuffer && bufflen) {
 855                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 856                                 GFP_KERNEL);
 857                 if (ret)
 858                         goto out;
 859                 bio = req->bio;
 860                 bio->bi_disk = disk;
 861                 if (disk && meta_buffer && meta_len) {
 862                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 863                                         meta_seed, write);
 864                         if (IS_ERR(meta)) {
 865                                 ret = PTR_ERR(meta);
 866                                 goto out_unmap;
 867                         }
 868                         req->cmd_flags |= REQ_INTEGRITY;
 869                 }
 870         }
 871
 872         blk_execute_rq(req->q, disk, req, 0);
 873         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 874                 ret = -EINTR;
 875         else
 876                 ret = nvme_req(req)->status;
 877         if (result)
 878                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 879         if (meta && !ret && !write) {
 880                 if (copy_to_user(meta_buffer, meta, meta_len))
 881                         ret = -EFAULT;
 882         }
 883         kfree(meta);
 884  out_unmap:
 885         if (bio)
 886                 blk_rq_unmap_user(bio);
 887  out:
 888         blk_mq_free_request(req);
 889         return ret;
 890 }
 891
 892 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 893 {
 894         struct nvme_ctrl *ctrl = rq->end_io_data;
 895         unsigned long flags;
 896         bool startka = false;
 897
 898         blk_mq_free_request(rq);
 899
 900         if (status) {
 901                 dev_err(ctrl->device,
 902                         "failed nvme_keep_alive_end_io error=%d\n",
 903                                 status);
 904                 return;
 905         }
 906
 907         ctrl->comp_seen = false;
 908         spin_lock_irqsave(&ctrl->lock, flags);
 909         if (ctrl->state == NVME_CTRL_LIVE ||
 910             ctrl->state == NVME_CTRL_CONNECTING)
 911                 startka = true;
 912         spin_unlock_irqrestore(&ctrl->lock, flags);
 913         if (startka)
 914                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 915 }
 916
 917 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 918 {
 919         struct request *rq;
 920
 921         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
 922                         NVME_QID_ANY);
 923         if (IS_ERR(rq))
 924                 return PTR_ERR(rq);
 925
 926         rq->timeout = ctrl->kato * HZ;
 927         rq->end_io_data = ctrl;
 928
 929         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 930
 931         return 0;
 932 }
 933
 934 static void nvme_keep_alive_work(struct work_struct *work)
 935 {
 936         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 937                         struct nvme_ctrl, ka_work);
 938         bool comp_seen = ctrl->comp_seen;
 939
 940         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
 941                 dev_dbg(ctrl->device,
 942                         "reschedule traffic based keep-alive timer\n");
 943                 ctrl->comp_seen = false;
 944                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 945                 return;
 946         }
 947
 948         if (nvme_keep_alive(ctrl)) {
 949                 /* allocation failure, reset the controller */
 950                 dev_err(ctrl->device, "keep-alive failed\n");
 951                 nvme_reset_ctrl(ctrl);
 952                 return;
 953         }
 954 }
 955
 956 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 957 {
 958         if (unlikely(ctrl->kato == 0))
 959                 return;
 960
 961         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 962 }
 963
 964 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 965 {
 966         if (unlikely(ctrl->kato == 0))
 967                 return;
 968
 969         cancel_delayed_work_sync(&ctrl->ka_work);
 970 }
 971 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 972
 973 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 974 {
 975         struct nvme_command c = { };
 976         int error;
 977
 978         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 979         c.identify.opcode = nvme_admin_identify;
 980         c.identify.cns = NVME_ID_CNS_CTRL;
 981
 982         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 983         if (!*id)
 984                 return -ENOMEM;
 985
 986         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 987                         sizeof(struct nvme_id_ctrl));
 988         if (error)
 989                 kfree(*id);
 990         return error;
 991 }
 992
 993 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
 994                 struct nvme_ns_ids *ids)
 995 {
 996         struct nvme_command c = { };
 997         int status;
 998         void *data;
 999         int pos;
1000         int len;
1001
1002         c.identify.opcode = nvme_admin_identify;
1003         c.identify.nsid = cpu_to_le32(nsid);
1004         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1005
1006         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1007         if (!data)
1008                 return -ENOMEM;
1009
1010         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1011                                       NVME_IDENTIFY_DATA_SIZE);
1012         if (status)
1013                 goto free_data;
1014
1015         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1016                 struct nvme_ns_id_desc *cur = data + pos;
1017
1018                 if (cur->nidl == 0)
1019                         break;
1020
1021                 switch (cur->nidt) {
1022                 case NVME_NIDT_EUI64:
1023                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1024                                 dev_warn(ctrl->device,
1025                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1026                                          cur->nidl);
1027                                 goto free_data;
1028                         }
1029                         len = NVME_NIDT_EUI64_LEN;
1030                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1031                         break;
1032                 case NVME_NIDT_NGUID:
1033                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1034                                 dev_warn(ctrl->device,
1035                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1036                                          cur->nidl);
1037                                 goto free_data;
1038                         }
1039                         len = NVME_NIDT_NGUID_LEN;
1040                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1041                         break;
1042                 case NVME_NIDT_UUID:
1043                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
1044                                 dev_warn(ctrl->device,
1045                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1046                                          cur->nidl);
1047                                 goto free_data;
1048                         }
1049                         len = NVME_NIDT_UUID_LEN;
1050                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1051                         break;
1052                 default:
1053                         /* Skip unknown types */
1054                         len = cur->nidl;
1055                         break;
1056                 }
1057
1058                 len += sizeof(*cur);
1059         }
1060 free_data:
1061         kfree(data);
1062         return status;
1063 }
1064
1065 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1066 {
1067         struct nvme_command c = { };
1068
1069         c.identify.opcode = nvme_admin_identify;
1070         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1071         c.identify.nsid = cpu_to_le32(nsid);
1072         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1073                                     NVME_IDENTIFY_DATA_SIZE);
1074 }
1075
1076 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
1077                 unsigned nsid)
1078 {
1079         struct nvme_id_ns *id;
1080         struct nvme_command c = { };
1081         int error;
1082
1083         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1084         c.identify.opcode = nvme_admin_identify;
1085         c.identify.nsid = cpu_to_le32(nsid);
1086         c.identify.cns = NVME_ID_CNS_NS;
1087
1088         id = kmalloc(sizeof(*id), GFP_KERNEL);
1089         if (!id)
1090                 return NULL;
1091
1092         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1093         if (error) {
1094                 dev_warn(ctrl->device, "Identify namespace failed\n");
1095                 kfree(id);
1096                 return NULL;
1097         }
1098
1099         return id;
1100 }
1101
1102 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
1103                       void *buffer, size_t buflen, u32 *result)
1104 {
1105         struct nvme_command c;
1106         union nvme_result res;
1107         int ret;
1108
1109         memset(&c, 0, sizeof(c));
1110         c.features.opcode = nvme_admin_set_features;
1111         c.features.fid = cpu_to_le32(fid);
1112         c.features.dword11 = cpu_to_le32(dword11);
1113
1114         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1115                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1116         if (ret >= 0 && result)
1117                 *result = le32_to_cpu(res.u32);
1118         return ret;
1119 }
1120
1121 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1122 {
1123         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1124         u32 result;
1125         int status, nr_io_queues;
1126
1127         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1128                         &result);
1129         if (status < 0)
1130                 return status;
1131
1132         /*
1133          * Degraded controllers might return an error when setting the queue
1134          * count.  We still want to be able to bring them online and offer
1135          * access to the admin queue, as that might be only way to fix them up.
1136          */
1137         if (status > 0) {
1138                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1139                 *count = 0;
1140         } else {
1141                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1142                 *count = min(*count, nr_io_queues);
1143         }
1144
1145         return 0;
1146 }
1147 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1148
1149 #define NVME_AEN_SUPPORTED \
1150         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1151
1152 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1153 {
1154         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1155         int status;
1156
1157         if (!supported_aens)
1158                 return;
1159
1160         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1161                         NULL, 0, &result);
1162         if (status)
1163                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1164                          supported_aens);
1165 }
1166
1167 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1168 {
1169         struct nvme_user_io io;
1170         struct nvme_command c;
1171         unsigned length, meta_len;
1172         void __user *metadata;
1173
1174         if (copy_from_user(&io, uio, sizeof(io)))
1175                 return -EFAULT;
1176         if (io.flags)
1177                 return -EINVAL;
1178
1179         switch (io.opcode) {
1180         case nvme_cmd_write:
1181         case nvme_cmd_read:
1182         case nvme_cmd_compare:
1183                 break;
1184         default:
1185                 return -EINVAL;
1186         }
1187
1188         length = (io.nblocks + 1) << ns->lba_shift;
1189         meta_len = (io.nblocks + 1) * ns->ms;
1190         metadata = (void __user *)(uintptr_t)io.metadata;
1191
1192         if (ns->ext) {
1193                 length += meta_len;
1194                 meta_len = 0;
1195         } else if (meta_len) {
1196                 if ((io.metadata & 3) || !io.metadata)
1197                         return -EINVAL;
1198         }
1199
1200         memset(&c, 0, sizeof(c));
1201         c.rw.opcode = io.opcode;
1202         c.rw.flags = io.flags;
1203         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1204         c.rw.slba = cpu_to_le64(io.slba);
1205         c.rw.length = cpu_to_le16(io.nblocks);
1206         c.rw.control = cpu_to_le16(io.control);
1207         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1208         c.rw.reftag = cpu_to_le32(io.reftag);
1209         c.rw.apptag = cpu_to_le16(io.apptag);
1210         c.rw.appmask = cpu_to_le16(io.appmask);
1211
1212         return nvme_submit_user_cmd(ns->queue, &c,
1213                         (void __user *)(uintptr_t)io.addr, length,
1214                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1215 }
1216
1217 static u32 nvme_known_admin_effects(u8 opcode)
1218 {
1219         switch (opcode) {
1220         case nvme_admin_format_nvm:
1221                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1222                                         NVME_CMD_EFFECTS_CSE_MASK;
1223         case nvme_admin_sanitize_nvm:
1224                 return NVME_CMD_EFFECTS_CSE_MASK;
1225         default:
1226                 break;
1227         }
1228         return 0;
1229 }
1230
1231 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1232                                                                 u8 opcode)
1233 {
1234         u32 effects = 0;
1235
1236         if (ns) {
1237                 if (ctrl->effects)
1238                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1239                 if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1240                         dev_warn(ctrl->device,
1241                                  "IO command:%02x has unhandled effects:%08x\n",
1242                                  opcode, effects);
1243                 return 0;
1244         }
1245
1246         if (ctrl->effects)
1247                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1248         else
1249                 effects = nvme_known_admin_effects(opcode);
1250
1251         /*
1252          * For simplicity, IO to all namespaces is quiesced even if the command
1253          * effects say only one namespace is affected.
1254          */
1255         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1256                 mutex_lock(&ctrl->scan_lock);
1257                 nvme_start_freeze(ctrl);
1258                 nvme_wait_freeze(ctrl);
1259         }
1260         return effects;
1261 }
1262
1263 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1264 {
1265         struct nvme_ns *ns;
1266
1267         down_read(&ctrl->namespaces_rwsem);
1268         list_for_each_entry(ns, &ctrl->namespaces, list)
1269                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1270                         nvme_set_queue_dying(ns);
1271         up_read(&ctrl->namespaces_rwsem);
1272
1273         nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1274 }
1275
1276 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1277 {
1278         /*
1279          * Revalidate LBA changes prior to unfreezing. This is necessary to
1280          * prevent memory corruption if a logical block size was changed by
1281          * this command.
1282          */
1283         if (effects & NVME_CMD_EFFECTS_LBCC)
1284                 nvme_update_formats(ctrl);
1285         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1286                 nvme_unfreeze(ctrl);
1287                 mutex_unlock(&ctrl->scan_lock);
1288         }
1289         if (effects & NVME_CMD_EFFECTS_CCC)
1290                 nvme_init_identify(ctrl);
1291         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1292                 nvme_queue_scan(ctrl);
1293 }
1294
1295 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1296                         struct nvme_passthru_cmd __user *ucmd)
1297 {
1298         struct nvme_passthru_cmd cmd;
1299         struct nvme_command c;
1300         unsigned timeout = 0;
1301         u32 effects;
1302         int status;
1303
1304         if (!capable(CAP_SYS_ADMIN))
1305                 return -EACCES;
1306         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1307                 return -EFAULT;
1308         if (cmd.flags)
1309                 return -EINVAL;
1310
1311         memset(&c, 0, sizeof(c));
1312         c.common.opcode = cmd.opcode;
1313         c.common.flags = cmd.flags;
1314         c.common.nsid = cpu_to_le32(cmd.nsid);
1315         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1316         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1317         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1318         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1319         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1320         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1321         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1322         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1323
1324         if (cmd.timeout_ms)
1325                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1326
1327         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1328         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1329                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1330                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1331                         0, &cmd.result, timeout);
1332         nvme_passthru_end(ctrl, effects);
1333
1334         if (status >= 0) {
1335                 if (put_user(cmd.result, &ucmd->result))
1336                         return -EFAULT;
1337         }
1338
1339         return status;
1340 }
1341
1342 /*
1343  * Issue ioctl requests on the first available path.  Note that unlike normal
1344  * block layer requests we will not retry failed request on another controller.
1345  */
1346 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1347                 struct nvme_ns_head **head, int *srcu_idx)
1348 {
1349 #ifdef CONFIG_NVME_MULTIPATH
1350         if (disk->fops == &nvme_ns_head_ops) {
1351                 *head = disk->private_data;
1352                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1353                 return nvme_find_path(*head);
1354         }
1355 #endif
1356         *head = NULL;
1357         *srcu_idx = -1;
1358         return disk->private_data;
1359 }
1360
1361 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1362 {
1363         if (head)
1364                 srcu_read_unlock(&head->srcu, idx);
1365 }
1366
1367 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1368 {
1369         switch (cmd) {
1370         case NVME_IOCTL_ID:
1371                 force_successful_syscall_return();
1372                 return ns->head->ns_id;
1373         case NVME_IOCTL_ADMIN_CMD:
1374                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1375         case NVME_IOCTL_IO_CMD:
1376                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1377         case NVME_IOCTL_SUBMIT_IO:
1378                 return nvme_submit_io(ns, (void __user *)arg);
1379         default:
1380 #ifdef CONFIG_NVM
1381                 if (ns->ndev)
1382                         return nvme_nvm_ioctl(ns, cmd, arg);
1383 #endif
1384                 if (is_sed_ioctl(cmd))
1385                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
1386                                          (void __user *) arg);
1387                 return -ENOTTY;
1388         }
1389 }
1390
1391 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1392                 unsigned int cmd, unsigned long arg)
1393 {
1394         struct nvme_ns_head *head = NULL;
1395         struct nvme_ns *ns;
1396         int srcu_idx, ret;
1397
1398         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1399         if (unlikely(!ns))
1400                 ret = -EWOULDBLOCK;
1401         else
1402                 ret = nvme_ns_ioctl(ns, cmd, arg);
1403         nvme_put_ns_from_disk(head, srcu_idx);
1404         return ret;
1405 }
1406
1407 static int nvme_open(struct block_device *bdev, fmode_t mode)
1408 {
1409         struct nvme_ns *ns = bdev->bd_disk->private_data;
1410
1411 #ifdef CONFIG_NVME_MULTIPATH
1412         /* should never be called due to GENHD_FL_HIDDEN */
1413         if (WARN_ON_ONCE(ns->head->disk))
1414                 goto fail;
1415 #endif
1416         if (!kref_get_unless_zero(&ns->kref))
1417                 goto fail;
1418         if (!try_module_get(ns->ctrl->ops->module))
1419                 goto fail_put_ns;
1420
1421         return 0;
1422
1423 fail_put_ns:
1424         nvme_put_ns(ns);
1425 fail:
1426         return -ENXIO;
1427 }
1428
1429 static void nvme_release(struct gendisk *disk, fmode_t mode)
1430 {
1431         struct nvme_ns *ns = disk->private_data;
1432
1433         module_put(ns->ctrl->ops->module);
1434         nvme_put_ns(ns);
1435 }
1436
1437 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1438 {
1439         /* some standard values */
1440         geo->heads = 1 << 6;
1441         geo->sectors = 1 << 5;
1442         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1443         return 0;
1444 }
1445
1446 #ifdef CONFIG_BLK_DEV_INTEGRITY
1447 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1448 {
1449         struct blk_integrity integrity;
1450
1451         memset(&integrity, 0, sizeof(integrity));
1452         switch (pi_type) {
1453         case NVME_NS_DPS_PI_TYPE3:
1454                 integrity.profile = &t10_pi_type3_crc;
1455                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1456                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1457                 break;
1458         case NVME_NS_DPS_PI_TYPE1:
1459         case NVME_NS_DPS_PI_TYPE2:
1460                 integrity.profile = &t10_pi_type1_crc;
1461                 integrity.tag_size = sizeof(u16);
1462                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1463                 break;
1464         default:
1465                 integrity.profile = NULL;
1466                 break;
1467         }
1468         integrity.tuple_size = ms;
1469         blk_integrity_register(disk, &integrity);
1470         blk_queue_max_integrity_segments(disk->queue, 1);
1471 }
1472 #else
1473 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1474 {
1475 }
1476 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1477
1478 static void nvme_set_chunk_size(struct nvme_ns *ns)
1479 {
1480         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1481         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1482 }
1483
1484 static void nvme_config_discard(struct nvme_ns *ns)
1485 {
1486         struct nvme_ctrl *ctrl = ns->ctrl;
1487         struct request_queue *queue = ns->queue;
1488         u32 size = queue_logical_block_size(queue);
1489
1490         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1491                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1492                 return;
1493         }
1494
1495         if (ctrl->nr_streams && ns->sws && ns->sgs)
1496                 size *= ns->sws * ns->sgs;
1497
1498         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1499                         NVME_DSM_MAX_RANGES);
1500
1501         queue->limits.discard_alignment = 0;
1502         queue->limits.discard_granularity = size;
1503
1504         /* If discard is already enabled, don't reset queue limits */
1505         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1506                 return;
1507
1508         blk_queue_max_discard_sectors(queue, UINT_MAX);
1509         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1510
1511         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1512                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1513 }
1514
1515 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1516                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1517 {
1518         memset(ids, 0, sizeof(*ids));
1519
1520         if (ctrl->vs >= NVME_VS(1, 1, 0))
1521                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1522         if (ctrl->vs >= NVME_VS(1, 2, 0))
1523                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1524         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1525                  /* Don't treat error as fatal we potentially
1526                   * already have a NGUID or EUI-64
1527                   */
1528                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1529                         dev_warn(ctrl->device,
1530                                  "%s: Identify Descriptors failed\n", __func__);
1531         }
1532 }
1533
1534 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1535 {
1536         return !uuid_is_null(&ids->uuid) ||
1537                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1538                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1539 }
1540
1541 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1542 {
1543         return uuid_equal(&a->uuid, &b->uuid) &&
1544                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1545                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1546 }
1547
1548 static void nvme_update_disk_info(struct gendisk *disk,
1549                 struct nvme_ns *ns, struct nvme_id_ns *id)
1550 {
1551         sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1552         unsigned short bs = 1 << ns->lba_shift;
1553
1554         blk_mq_freeze_queue(disk->queue);
1555         blk_integrity_unregister(disk);
1556
1557         blk_queue_logical_block_size(disk->queue, bs);
1558         blk_queue_physical_block_size(disk->queue, bs);
1559         blk_queue_io_min(disk->queue, bs);
1560
1561         if (ns->ms && !ns->ext &&
1562             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1563                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1564         if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1565                 capacity = 0;
1566
1567         set_capacity(disk, capacity);
1568         nvme_config_discard(ns);
1569
1570         if (id->nsattr & (1 << 0))
1571                 set_disk_ro(disk, true);
1572         else
1573                 set_disk_ro(disk, false);
1574
1575         blk_mq_unfreeze_queue(disk->queue);
1576 }
1577
1578 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1579 {
1580         struct nvme_ns *ns = disk->private_data;
1581
1582         /*
1583          * If identify namespace failed, use default 512 byte block size so
1584          * block layer can use before failing read/write for 0 capacity.
1585          */
1586         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1587         if (ns->lba_shift == 0)
1588                 ns->lba_shift = 9;
1589         ns->noiob = le16_to_cpu(id->noiob);
1590         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1591         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1592         /* the PI implementation requires metadata equal t10 pi tuple size */
1593         if (ns->ms == sizeof(struct t10_pi_tuple))
1594                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1595         else
1596                 ns->pi_type = 0;
1597
1598         if (ns->noiob)
1599                 nvme_set_chunk_size(ns);
1600         nvme_update_disk_info(disk, ns, id);
1601 #ifdef CONFIG_NVME_MULTIPATH
1602         if (ns->head->disk) {
1603                 nvme_update_disk_info(ns->head->disk, ns, id);
1604                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1605         }
1606 #endif
1607 }
1608
1609 static int nvme_revalidate_disk(struct gendisk *disk)
1610 {
1611         struct nvme_ns *ns = disk->private_data;
1612         struct nvme_ctrl *ctrl = ns->ctrl;
1613         struct nvme_id_ns *id;
1614         struct nvme_ns_ids ids;
1615         int ret = 0;
1616
1617         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1618                 set_capacity(disk, 0);
1619                 return -ENODEV;
1620         }
1621
1622         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1623         if (!id)
1624                 return -ENODEV;
1625
1626         if (id->ncap == 0) {
1627                 ret = -ENODEV;
1628                 goto out;
1629         }
1630
1631         __nvme_revalidate_disk(disk, id);
1632         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1633         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1634                 dev_err(ctrl->device,
1635                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1636                 ret = -ENODEV;
1637         }
1638
1639 out:
1640         kfree(id);
1641         return ret;
1642 }
1643
1644 static char nvme_pr_type(enum pr_type type)
1645 {
1646         switch (type) {
1647         case PR_WRITE_EXCLUSIVE:
1648                 return 1;
1649         case PR_EXCLUSIVE_ACCESS:
1650                 return 2;
1651         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1652                 return 3;
1653         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1654                 return 4;
1655         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1656                 return 5;
1657         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1658                 return 6;
1659         default:
1660                 return 0;
1661         }
1662 };
1663
1664 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1665                                 u64 key, u64 sa_key, u8 op)
1666 {
1667         struct nvme_ns_head *head = NULL;
1668         struct nvme_ns *ns;
1669         struct nvme_command c;
1670         int srcu_idx, ret;
1671         u8 data[16] = { 0, };
1672
1673         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1674         if (unlikely(!ns))
1675                 return -EWOULDBLOCK;
1676
1677         put_unaligned_le64(key, &data[0]);
1678         put_unaligned_le64(sa_key, &data[8]);
1679
1680         memset(&c, 0, sizeof(c));
1681         c.common.opcode = op;
1682         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1683         c.common.cdw10 = cpu_to_le32(cdw10);
1684
1685         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1686         nvme_put_ns_from_disk(head, srcu_idx);
1687         return ret;
1688 }
1689
1690 static int nvme_pr_register(struct block_device *bdev, u64 old,
1691                 u64 new, unsigned flags)
1692 {
1693         u32 cdw10;
1694
1695         if (flags & ~PR_FL_IGNORE_KEY)
1696                 return -EOPNOTSUPP;
1697
1698         cdw10 = old ? 2 : 0;
1699         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1700         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1701         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1702 }
1703
1704 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1705                 enum pr_type type, unsigned flags)
1706 {
1707         u32 cdw10;
1708
1709         if (flags & ~PR_FL_IGNORE_KEY)
1710                 return -EOPNOTSUPP;
1711
1712         cdw10 = nvme_pr_type(type) << 8;
1713         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1714         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1715 }
1716
1717 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1718                 enum pr_type type, bool abort)
1719 {
1720         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1721         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1722 }
1723
1724 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1725 {
1726         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1727         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1728 }
1729
1730 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1731 {
1732         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1733         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1734 }
1735
1736 static const struct pr_ops nvme_pr_ops = {
1737         .pr_register    = nvme_pr_register,
1738         .pr_reserve     = nvme_pr_reserve,
1739         .pr_release     = nvme_pr_release,
1740         .pr_preempt     = nvme_pr_preempt,
1741         .pr_clear       = nvme_pr_clear,
1742 };
1743
1744 #ifdef CONFIG_BLK_SED_OPAL
1745 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1746                 bool send)
1747 {
1748         struct nvme_ctrl *ctrl = data;
1749         struct nvme_command cmd;
1750
1751         memset(&cmd, 0, sizeof(cmd));
1752         if (send)
1753                 cmd.common.opcode = nvme_admin_security_send;
1754         else
1755                 cmd.common.opcode = nvme_admin_security_recv;
1756         cmd.common.nsid = 0;
1757         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1758         cmd.common.cdw11 = cpu_to_le32(len);
1759
1760         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1761                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1762 }
1763 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1764 #endif /* CONFIG_BLK_SED_OPAL */
1765
1766 static const struct block_device_operations nvme_fops = {
1767         .owner          = THIS_MODULE,
1768         .ioctl          = nvme_ioctl,
1769         .compat_ioctl   = nvme_ioctl,
1770         .open           = nvme_open,
1771         .release        = nvme_release,
1772         .getgeo         = nvme_getgeo,
1773         .revalidate_disk= nvme_revalidate_disk,
1774         .pr_ops         = &nvme_pr_ops,
1775 };
1776
1777 #ifdef CONFIG_NVME_MULTIPATH
1778 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1779 {
1780         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1781
1782         if (!kref_get_unless_zero(&head->ref))
1783                 return -ENXIO;
1784         return 0;
1785 }
1786
1787 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1788 {
1789         nvme_put_ns_head(disk->private_data);
1790 }
1791
1792 const struct block_device_operations nvme_ns_head_ops = {
1793         .owner          = THIS_MODULE,
1794         .open           = nvme_ns_head_open,
1795         .release        = nvme_ns_head_release,
1796         .ioctl          = nvme_ioctl,
1797         .compat_ioctl   = nvme_ioctl,
1798         .getgeo         = nvme_getgeo,
1799         .pr_ops         = &nvme_pr_ops,
1800 };
1801 #endif /* CONFIG_NVME_MULTIPATH */
1802
1803 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1804 {
1805         unsigned long timeout =
1806                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1807         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1808         int ret;
1809
1810         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1811                 if (csts == ~0)
1812                         return -ENODEV;
1813                 if ((csts & NVME_CSTS_RDY) == bit)
1814                         break;
1815
1816                 msleep(100);
1817                 if (fatal_signal_pending(current))
1818                         return -EINTR;
1819                 if (time_after(jiffies, timeout)) {
1820                         dev_err(ctrl->device,
1821                                 "Device not ready; aborting %s\n", enabled ?
1822                                                 "initialisation" : "reset");
1823                         return -ENODEV;
1824                 }
1825         }
1826
1827         return ret;
1828 }
1829
1830 /*
1831  * If the device has been passed off to us in an enabled state, just clear
1832  * the enabled bit.  The spec says we should set the 'shutdown notification
1833  * bits', but doing so may cause the device to complete commands to the
1834  * admin queue ... and we don't know what memory that might be pointing at!
1835  */
1836 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1837 {
1838         int ret;
1839
1840         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1841         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1842
1843         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1844         if (ret)
1845                 return ret;
1846
1847         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1848                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1849
1850         return nvme_wait_ready(ctrl, cap, false);
1851 }
1852 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1853
1854 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1855 {
1856         /*
1857          * Default to a 4K page size, with the intention to update this
1858          * path in the future to accomodate architectures with differing
1859          * kernel and IO page sizes.
1860          */
1861         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1862         int ret;
1863
1864         if (page_shift < dev_page_min) {
1865                 dev_err(ctrl->device,
1866                         "Minimum device page size %u too large for host (%u)\n",
1867                         1 << dev_page_min, 1 << page_shift);
1868                 return -ENODEV;
1869         }
1870
1871         ctrl->page_size = 1 << page_shift;
1872
1873         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1874         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1875         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1876         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1877         ctrl->ctrl_config |= NVME_CC_ENABLE;
1878
1879         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1880         if (ret)
1881                 return ret;
1882         return nvme_wait_ready(ctrl, cap, true);
1883 }
1884 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1885
1886 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1887 {
1888         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1889         u32 csts;
1890         int ret;
1891
1892         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1893         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1894
1895         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1896         if (ret)
1897                 return ret;
1898
1899         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1900                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1901                         break;
1902
1903                 msleep(100);
1904                 if (fatal_signal_pending(current))
1905                         return -EINTR;
1906                 if (time_after(jiffies, timeout)) {
1907                         dev_err(ctrl->device,
1908                                 "Device shutdown incomplete; abort shutdown\n");
1909                         return -ENODEV;
1910                 }
1911         }
1912
1913         return ret;
1914 }
1915 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1916
1917 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1918                 struct request_queue *q)
1919 {
1920         bool vwc = false;
1921
1922         if (ctrl->max_hw_sectors) {
1923                 u32 max_segments =
1924                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1925
1926                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1927                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1928                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1929         }
1930         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1931             is_power_of_2(ctrl->max_hw_sectors))
1932                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1933         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1934         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1935                 vwc = true;
1936         blk_queue_write_cache(q, vwc, vwc);
1937 }
1938
1939 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1940 {
1941         __le64 ts;
1942         int ret;
1943
1944         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1945                 return 0;
1946
1947         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1948         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1949                         NULL);
1950         if (ret)
1951                 dev_warn_once(ctrl->device,
1952                         "could not set timestamp (%d)\n", ret);
1953         return ret;
1954 }
1955
1956 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
1957 {
1958         struct nvme_feat_host_behavior *host;
1959         int ret;
1960
1961         /* Don't bother enabling the feature if retry delay is not reported */
1962         if (!ctrl->crdt[0])
1963                 return 0;
1964
1965         host = kzalloc(sizeof(*host), GFP_KERNEL);
1966         if (!host)
1967                 return 0;
1968
1969         host->acre = NVME_ENABLE_ACRE;
1970         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
1971                                 host, sizeof(*host), NULL);
1972         kfree(host);
1973         return ret;
1974 }
1975
1976 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1977 {
1978         /*
1979          * APST (Autonomous Power State Transition) lets us program a
1980          * table of power state transitions that the controller will
1981          * perform automatically.  We configure it with a simple
1982          * heuristic: we are willing to spend at most 2% of the time
1983          * transitioning between power states.  Therefore, when running
1984          * in any given state, we will enter the next lower-power
1985          * non-operational state after waiting 50 * (enlat + exlat)
1986          * microseconds, as long as that state's exit latency is under
1987          * the requested maximum latency.
1988          *
1989          * We will not autonomously enter any non-operational state for
1990          * which the total latency exceeds ps_max_latency_us.  Users
1991          * can set ps_max_latency_us to zero to turn off APST.
1992          */
1993
1994         unsigned apste;
1995         struct nvme_feat_auto_pst *table;
1996         u64 max_lat_us = 0;
1997         int max_ps = -1;
1998         int ret;
1999
2000         /*
2001          * If APST isn't supported or if we haven't been initialized yet,
2002          * then don't do anything.
2003          */
2004         if (!ctrl->apsta)
2005                 return 0;
2006
2007         if (ctrl->npss > 31) {
2008                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2009                 return 0;
2010         }
2011
2012         table = kzalloc(sizeof(*table), GFP_KERNEL);
2013         if (!table)
2014                 return 0;
2015
2016         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2017                 /* Turn off APST. */
2018                 apste = 0;
2019                 dev_dbg(ctrl->device, "APST disabled\n");
2020         } else {
2021                 __le64 target = cpu_to_le64(0);
2022                 int state;
2023
2024                 /*
2025                  * Walk through all states from lowest- to highest-power.
2026                  * According to the spec, lower-numbered states use more
2027                  * power.  NPSS, despite the name, is the index of the
2028                  * lowest-power state, not the number of states.
2029                  */
2030                 for (state = (int)ctrl->npss; state >= 0; state--) {
2031                         u64 total_latency_us, exit_latency_us, transition_ms;
2032
2033                         if (target)
2034                                 table->entries[state] = target;
2035
2036                         /*
2037                          * Don't allow transitions to the deepest state
2038                          * if it's quirked off.
2039                          */
2040                         if (state == ctrl->npss &&
2041                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2042                                 continue;
2043
2044                         /*
2045                          * Is this state a useful non-operational state for
2046                          * higher-power states to autonomously transition to?
2047                          */
2048                         if (!(ctrl->psd[state].flags &
2049                               NVME_PS_FLAGS_NON_OP_STATE))
2050                                 continue;
2051
2052                         exit_latency_us =
2053                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2054                         if (exit_latency_us > ctrl->ps_max_latency_us)
2055                                 continue;
2056
2057                         total_latency_us =
2058                                 exit_latency_us +
2059                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2060
2061                         /*
2062                          * This state is good.  Use it as the APST idle
2063                          * target for higher power states.
2064                          */
2065                         transition_ms = total_latency_us + 19;
2066                         do_div(transition_ms, 20);
2067                         if (transition_ms > (1 << 24) - 1)
2068                                 transition_ms = (1 << 24) - 1;
2069
2070                         target = cpu_to_le64((state << 3) |
2071                                              (transition_ms << 8));
2072
2073                         if (max_ps == -1)
2074                                 max_ps = state;
2075
2076                         if (total_latency_us > max_lat_us)
2077                                 max_lat_us = total_latency_us;
2078                 }
2079
2080                 apste = 1;
2081
2082                 if (max_ps == -1) {
2083                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2084                 } else {
2085                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2086                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2087                 }
2088         }
2089
2090         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2091                                 table, sizeof(*table), NULL);
2092         if (ret)
2093                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2094
2095         kfree(table);
2096         return ret;
2097 }
2098
2099 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2100 {
2101         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2102         u64 latency;
2103
2104         switch (val) {
2105         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2106         case PM_QOS_LATENCY_ANY:
2107                 latency = U64_MAX;
2108                 break;
2109
2110         default:
2111                 latency = val;
2112         }
2113
2114         if (ctrl->ps_max_latency_us != latency) {
2115                 ctrl->ps_max_latency_us = latency;
2116                 nvme_configure_apst(ctrl);
2117         }
2118 }
2119
2120 struct nvme_core_quirk_entry {
2121         /*
2122          * NVMe model and firmware strings are padded with spaces.  For
2123          * simplicity, strings in the quirk table are padded with NULLs
2124          * instead.
2125          */
2126         u16 vid;
2127         const char *mn;
2128         const char *fr;
2129         unsigned long quirks;
2130 };
2131
2132 static const struct nvme_core_quirk_entry core_quirks[] = {
2133         {
2134                 /*
2135                  * This Toshiba device seems to die using any APST states.  See:
2136                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2137                  */
2138                 .vid = 0x1179,
2139                 .mn = "THNSF5256GPUK TOSHIBA",
2140                 .quirks = NVME_QUIRK_NO_APST,
2141         }
2142 };
2143
2144 /* match is null-terminated but idstr is space-padded. */
2145 static bool string_matches(const char *idstr, const char *match, size_t len)
2146 {
2147         size_t matchlen;
2148
2149         if (!match)
2150                 return true;
2151
2152         matchlen = strlen(match);
2153         WARN_ON_ONCE(matchlen > len);
2154
2155         if (memcmp(idstr, match, matchlen))
2156                 return false;
2157
2158         for (; matchlen < len; matchlen++)
2159                 if (idstr[matchlen] != ' ')
2160                         return false;
2161
2162         return true;
2163 }
2164
2165 static bool quirk_matches(const struct nvme_id_ctrl *id,
2166                           const struct nvme_core_quirk_entry *q)
2167 {
2168         return q->vid == le16_to_cpu(id->vid) &&
2169                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2170                 string_matches(id->fr, q->fr, sizeof(id->fr));
2171 }
2172
2173 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2174                 struct nvme_id_ctrl *id)
2175 {
2176         size_t nqnlen;
2177         int off;
2178
2179         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2180                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2181                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2182                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2183                         return;
2184                 }
2185
2186                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2187                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2188         }
2189
2190         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2191         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2192                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2193                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2194         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2195         off += sizeof(id->sn);
2196         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2197         off += sizeof(id->mn);
2198         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2199 }
2200
2201 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2202 {
2203         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2204         kfree(subsys);
2205 }
2206
2207 static void nvme_release_subsystem(struct device *dev)
2208 {
2209         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2210 }
2211
2212 static void nvme_destroy_subsystem(struct kref *ref)
2213 {
2214         struct nvme_subsystem *subsys =
2215                         container_of(ref, struct nvme_subsystem, ref);
2216
2217         mutex_lock(&nvme_subsystems_lock);
2218         list_del(&subsys->entry);
2219         mutex_unlock(&nvme_subsystems_lock);
2220
2221         ida_destroy(&subsys->ns_ida);
2222         device_del(&subsys->dev);
2223         put_device(&subsys->dev);
2224 }
2225
2226 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2227 {
2228         kref_put(&subsys->ref, nvme_destroy_subsystem);
2229 }
2230
2231 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2232 {
2233         struct nvme_subsystem *subsys;
2234
2235         lockdep_assert_held(&nvme_subsystems_lock);
2236
2237         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2238                 if (strcmp(subsys->subnqn, subsysnqn))
2239                         continue;
2240                 if (!kref_get_unless_zero(&subsys->ref))
2241                         continue;
2242                 return subsys;
2243         }
2244
2245         return NULL;
2246 }
2247
2248 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2249         struct device_attribute subsys_attr_##_name = \
2250                 __ATTR(_name, _mode, _show, NULL)
2251
2252 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2253                                     struct device_attribute *attr,
2254                                     char *buf)
2255 {
2256         struct nvme_subsystem *subsys =
2257                 container_of(dev, struct nvme_subsystem, dev);
2258
2259         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2260 }
2261 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2262
2263 #define nvme_subsys_show_str_function(field)                            \
2264 static ssize_t subsys_##field##_show(struct device *dev,                \
2265                             struct device_attribute *attr, char *buf)   \
2266 {                                                                       \
2267         struct nvme_subsystem *subsys =                                 \
2268                 container_of(dev, struct nvme_subsystem, dev);          \
2269         return sprintf(buf, "%.*s\n",                                   \
2270                        (int)sizeof(subsys->field), subsys->field);      \
2271 }                                                                       \
2272 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2273
2274 nvme_subsys_show_str_function(model);
2275 nvme_subsys_show_str_function(serial);
2276 nvme_subsys_show_str_function(firmware_rev);
2277
2278 static struct attribute *nvme_subsys_attrs[] = {
2279         &subsys_attr_model.attr,
2280         &subsys_attr_serial.attr,
2281         &subsys_attr_firmware_rev.attr,
2282         &subsys_attr_subsysnqn.attr,
2283         NULL,
2284 };
2285
2286 static struct attribute_group nvme_subsys_attrs_group = {
2287         .attrs = nvme_subsys_attrs,
2288 };
2289
2290 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2291         &nvme_subsys_attrs_group,
2292         NULL,
2293 };
2294
2295 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2296 {
2297         int count = 0;
2298         struct nvme_ctrl *ctrl;
2299
2300         mutex_lock(&subsys->lock);
2301         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2302                 if (ctrl->state != NVME_CTRL_DELETING &&
2303                     ctrl->state != NVME_CTRL_DEAD)
2304                         count++;
2305         }
2306         mutex_unlock(&subsys->lock);
2307
2308         return count;
2309 }
2310
2311 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2312 {
2313         struct nvme_subsystem *subsys, *found;
2314         int ret;
2315
2316         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2317         if (!subsys)
2318                 return -ENOMEM;
2319         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2320         if (ret < 0) {
2321                 kfree(subsys);
2322                 return ret;
2323         }
2324         subsys->instance = ret;
2325         mutex_init(&subsys->lock);
2326         kref_init(&subsys->ref);
2327         INIT_LIST_HEAD(&subsys->ctrls);
2328         INIT_LIST_HEAD(&subsys->nsheads);
2329         nvme_init_subnqn(subsys, ctrl, id);
2330         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2331         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2332         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2333         subsys->vendor_id = le16_to_cpu(id->vid);
2334         subsys->cmic = id->cmic;
2335
2336         subsys->dev.class = nvme_subsys_class;
2337         subsys->dev.release = nvme_release_subsystem;
2338         subsys->dev.groups = nvme_subsys_attrs_groups;
2339         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2340         device_initialize(&subsys->dev);
2341
2342         mutex_lock(&nvme_subsystems_lock);
2343         found = __nvme_find_get_subsystem(subsys->subnqn);
2344         if (found) {
2345                 /*
2346                  * Verify that the subsystem actually supports multiple
2347                  * controllers, else bail out.
2348                  */
2349                 if (!(ctrl->opts && ctrl->opts->discovery_nqn) &&
2350                     nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2351                         dev_err(ctrl->device,
2352                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2353                                 found->subnqn);
2354                         nvme_put_subsystem(found);
2355                         ret = -EINVAL;
2356                         goto out_unlock;
2357                 }
2358
2359                 __nvme_release_subsystem(subsys);
2360                 subsys = found;
2361         } else {
2362                 ret = device_add(&subsys->dev);
2363                 if (ret) {
2364                         dev_err(ctrl->device,
2365                                 "failed to register subsystem device.\n");
2366                         goto out_unlock;
2367                 }
2368                 ida_init(&subsys->ns_ida);
2369                 list_add_tail(&subsys->entry, &nvme_subsystems);
2370         }
2371
2372         ctrl->subsys = subsys;
2373         mutex_unlock(&nvme_subsystems_lock);
2374
2375         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2376                         dev_name(ctrl->device))) {
2377                 dev_err(ctrl->device,
2378                         "failed to create sysfs link from subsystem.\n");
2379                 /* the transport driver will eventually put the subsystem */
2380                 return -EINVAL;
2381         }
2382
2383         mutex_lock(&subsys->lock);
2384         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2385         mutex_unlock(&subsys->lock);
2386
2387         return 0;
2388
2389 out_unlock:
2390         mutex_unlock(&nvme_subsystems_lock);
2391         put_device(&subsys->dev);
2392         return ret;
2393 }
2394
2395 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2396                 void *log, size_t size, u64 offset)
2397 {
2398         struct nvme_command c = { };
2399         unsigned long dwlen = size / 4 - 1;
2400
2401         c.get_log_page.opcode = nvme_admin_get_log_page;
2402         c.get_log_page.nsid = cpu_to_le32(nsid);
2403         c.get_log_page.lid = log_page;
2404         c.get_log_page.lsp = lsp;
2405         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2406         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2407         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2408         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2409
2410         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2411 }
2412
2413 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2414 {
2415         int ret;
2416
2417         if (!ctrl->effects)
2418                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2419
2420         if (!ctrl->effects)
2421                 return 0;
2422
2423         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2424                         ctrl->effects, sizeof(*ctrl->effects), 0);
2425         if (ret) {
2426                 kfree(ctrl->effects);
2427                 ctrl->effects = NULL;
2428         }
2429         return ret;
2430 }
2431
2432 /*
2433  * Initialize the cached copies of the Identify data and various controller
2434  * register in our nvme_ctrl structure.  This should be called as soon as
2435  * the admin queue is fully up and running.
2436  */
2437 int nvme_init_identify(struct nvme_ctrl *ctrl)
2438 {
2439         struct nvme_id_ctrl *id;
2440         u64 cap;
2441         int ret, page_shift;
2442         u32 max_hw_sectors;
2443         bool prev_apst_enabled;
2444
2445         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2446         if (ret) {
2447                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2448                 return ret;
2449         }
2450
2451         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2452         if (ret) {
2453                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2454                 return ret;
2455         }
2456         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2457
2458         if (ctrl->vs >= NVME_VS(1, 1, 0))
2459                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2460
2461         ret = nvme_identify_ctrl(ctrl, &id);
2462         if (ret) {
2463                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2464                 return -EIO;
2465         }
2466
2467         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2468                 ret = nvme_get_effects_log(ctrl);
2469                 if (ret < 0)
2470                         goto out_free;
2471         }
2472
2473         if (!ctrl->identified) {
2474                 int i;
2475
2476                 ret = nvme_init_subsystem(ctrl, id);
2477                 if (ret)
2478                         goto out_free;
2479
2480                 /*
2481                  * Check for quirks.  Quirk can depend on firmware version,
2482                  * so, in principle, the set of quirks present can change
2483                  * across a reset.  As a possible future enhancement, we
2484                  * could re-scan for quirks every time we reinitialize
2485                  * the device, but we'd have to make sure that the driver
2486                  * behaves intelligently if the quirks change.
2487                  */
2488                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2489                         if (quirk_matches(id, &core_quirks[i]))
2490                                 ctrl->quirks |= core_quirks[i].quirks;
2491                 }
2492         }
2493
2494         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2495                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2496                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2497         }
2498
2499         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2500         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2501         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2502
2503         ctrl->oacs = le16_to_cpu(id->oacs);
2504         ctrl->oncs = le16_to_cpup(&id->oncs);
2505         ctrl->oaes = le32_to_cpu(id->oaes);
2506         atomic_set(&ctrl->abort_limit, id->acl + 1);
2507         ctrl->vwc = id->vwc;
2508         if (id->mdts)
2509                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2510         else
2511                 max_hw_sectors = UINT_MAX;
2512         ctrl->max_hw_sectors =
2513                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2514
2515         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2516         ctrl->sgls = le32_to_cpu(id->sgls);
2517         ctrl->kas = le16_to_cpu(id->kas);
2518         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2519         ctrl->ctratt = le32_to_cpu(id->ctratt);
2520
2521         if (id->rtd3e) {
2522                 /* us -> s */
2523                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2524
2525                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2526                                                  shutdown_timeout, 60);
2527
2528                 if (ctrl->shutdown_timeout != shutdown_timeout)
2529                         dev_info(ctrl->device,
2530                                  "Shutdown timeout set to %u seconds\n",
2531                                  ctrl->shutdown_timeout);
2532         } else
2533                 ctrl->shutdown_timeout = shutdown_timeout;
2534
2535         ctrl->npss = id->npss;
2536         ctrl->apsta = id->apsta;
2537         prev_apst_enabled = ctrl->apst_enabled;
2538         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2539                 if (force_apst && id->apsta) {
2540                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2541                         ctrl->apst_enabled = true;
2542                 } else {
2543                         ctrl->apst_enabled = false;
2544                 }
2545         } else {
2546                 ctrl->apst_enabled = id->apsta;
2547         }
2548         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2549
2550         if (ctrl->ops->flags & NVME_F_FABRICS) {
2551                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2552                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2553                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2554                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2555
2556                 /*
2557                  * In fabrics we need to verify the cntlid matches the
2558                  * admin connect
2559                  */
2560                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2561                         ret = -EINVAL;
2562                         goto out_free;
2563                 }
2564
2565                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2566                         dev_err(ctrl->device,
2567                                 "keep-alive support is mandatory for fabrics\n");
2568                         ret = -EINVAL;
2569                         goto out_free;
2570                 }
2571         } else {
2572                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2573                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2574                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2575                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2576                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2577         }
2578
2579         ret = nvme_mpath_init(ctrl, id);
2580         kfree(id);
2581
2582         if (ret < 0)
2583                 return ret;
2584
2585         if (ctrl->apst_enabled && !prev_apst_enabled)
2586                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2587         else if (!ctrl->apst_enabled && prev_apst_enabled)
2588                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2589
2590         ret = nvme_configure_apst(ctrl);
2591         if (ret < 0)
2592                 return ret;
2593
2594         ret = nvme_configure_timestamp(ctrl);
2595         if (ret < 0)
2596                 return ret;
2597
2598         ret = nvme_configure_directives(ctrl);
2599         if (ret < 0)
2600                 return ret;
2601
2602         ret = nvme_configure_acre(ctrl);
2603         if (ret < 0)
2604                 return ret;
2605
2606         ctrl->identified = true;
2607
2608         return 0;
2609
2610 out_free:
2611         kfree(id);
2612         return ret;
2613 }
2614 EXPORT_SYMBOL_GPL(nvme_init_identify);
2615
2616 static int nvme_dev_open(struct inode *inode, struct file *file)
2617 {
2618         struct nvme_ctrl *ctrl =
2619                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2620
2621         switch (ctrl->state) {
2622         case NVME_CTRL_LIVE:
2623         case NVME_CTRL_ADMIN_ONLY:
2624                 break;
2625         default:
2626                 return -EWOULDBLOCK;
2627         }
2628
2629         file->private_data = ctrl;
2630         return 0;
2631 }
2632
2633 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2634 {
2635         struct nvme_ns *ns;
2636         int ret;
2637
2638         down_read(&ctrl->namespaces_rwsem);
2639         if (list_empty(&ctrl->namespaces)) {
2640                 ret = -ENOTTY;
2641                 goto out_unlock;
2642         }
2643
2644         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2645         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2646                 dev_warn(ctrl->device,
2647                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2648                 ret = -EINVAL;
2649                 goto out_unlock;
2650         }
2651
2652         dev_warn(ctrl->device,
2653                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2654         kref_get(&ns->kref);
2655         up_read(&ctrl->namespaces_rwsem);
2656
2657         ret = nvme_user_cmd(ctrl, ns, argp);
2658         nvme_put_ns(ns);
2659         return ret;
2660
2661 out_unlock:
2662         up_read(&ctrl->namespaces_rwsem);
2663         return ret;
2664 }
2665
2666 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2667                 unsigned long arg)
2668 {
2669         struct nvme_ctrl *ctrl = file->private_data;
2670         void __user *argp = (void __user *)arg;
2671
2672         switch (cmd) {
2673         case NVME_IOCTL_ADMIN_CMD:
2674                 return nvme_user_cmd(ctrl, NULL, argp);
2675         case NVME_IOCTL_IO_CMD:
2676                 return nvme_dev_user_cmd(ctrl, argp);
2677         case NVME_IOCTL_RESET:
2678                 dev_warn(ctrl->device, "resetting controller\n");
2679                 return nvme_reset_ctrl_sync(ctrl);
2680         case NVME_IOCTL_SUBSYS_RESET:
2681                 return nvme_reset_subsystem(ctrl);
2682         case NVME_IOCTL_RESCAN:
2683                 nvme_queue_scan(ctrl);
2684                 return 0;
2685         default:
2686                 return -ENOTTY;
2687         }
2688 }
2689
2690 static const struct file_operations nvme_dev_fops = {
2691         .owner          = THIS_MODULE,
2692         .open           = nvme_dev_open,
2693         .unlocked_ioctl = nvme_dev_ioctl,
2694         .compat_ioctl   = nvme_dev_ioctl,
2695 };
2696
2697 static ssize_t nvme_sysfs_reset(struct device *dev,
2698                                 struct device_attribute *attr, const char *buf,
2699                                 size_t count)
2700 {
2701         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2702         int ret;
2703
2704         ret = nvme_reset_ctrl_sync(ctrl);
2705         if (ret < 0)
2706                 return ret;
2707         return count;
2708 }
2709 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2710
2711 static ssize_t nvme_sysfs_rescan(struct device *dev,
2712                                 struct device_attribute *attr, const char *buf,
2713                                 size_t count)
2714 {
2715         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2716
2717         nvme_queue_scan(ctrl);
2718         return count;
2719 }
2720 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2721
2722 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2723 {
2724         struct gendisk *disk = dev_to_disk(dev);
2725
2726         if (disk->fops == &nvme_fops)
2727                 return nvme_get_ns_from_dev(dev)->head;
2728         else
2729                 return disk->private_data;
2730 }
2731
2732 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2733                 char *buf)
2734 {
2735         struct nvme_ns_head *head = dev_to_ns_head(dev);
2736         struct nvme_ns_ids *ids = &head->ids;
2737         struct nvme_subsystem *subsys = head->subsys;
2738         int serial_len = sizeof(subsys->serial);
2739         int model_len = sizeof(subsys->model);
2740
2741         if (!uuid_is_null(&ids->uuid))
2742                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2743
2744         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2745                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2746
2747         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2748                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2749
2750         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2751                                   subsys->serial[serial_len - 1] == '\0'))
2752                 serial_len--;
2753         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2754                                  subsys->model[model_len - 1] == '\0'))
2755                 model_len--;
2756
2757         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2758                 serial_len, subsys->serial, model_len, subsys->model,
2759                 head->ns_id);
2760 }
2761 static DEVICE_ATTR_RO(wwid);
2762
2763 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2764                 char *buf)
2765 {
2766         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2767 }
2768 static DEVICE_ATTR_RO(nguid);
2769
2770 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2771                 char *buf)
2772 {
2773         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2774
2775         /* For backward compatibility expose the NGUID to userspace if
2776          * we have no UUID set
2777          */
2778         if (uuid_is_null(&ids->uuid)) {
2779                 printk_ratelimited(KERN_WARNING
2780                                    "No UUID available providing old NGUID\n");
2781                 return sprintf(buf, "%pU\n", ids->nguid);
2782         }
2783         return sprintf(buf, "%pU\n", &ids->uuid);
2784 }
2785 static DEVICE_ATTR_RO(uuid);
2786
2787 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2788                 char *buf)
2789 {
2790         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2791 }
2792 static DEVICE_ATTR_RO(eui);
2793
2794 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2795                 char *buf)
2796 {
2797         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2798 }
2799 static DEVICE_ATTR_RO(nsid);
2800
2801 static struct attribute *nvme_ns_id_attrs[] = {
2802         &dev_attr_wwid.attr,
2803         &dev_attr_uuid.attr,
2804         &dev_attr_nguid.attr,
2805         &dev_attr_eui.attr,
2806         &dev_attr_nsid.attr,
2807 #ifdef CONFIG_NVME_MULTIPATH
2808         &dev_attr_ana_grpid.attr,
2809         &dev_attr_ana_state.attr,
2810 #endif
2811         NULL,
2812 };
2813
2814 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2815                 struct attribute *a, int n)
2816 {
2817         struct device *dev = container_of(kobj, struct device, kobj);
2818         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2819
2820         if (a == &dev_attr_uuid.attr) {
2821                 if (uuid_is_null(&ids->uuid) &&
2822                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2823                         return 0;
2824         }
2825         if (a == &dev_attr_nguid.attr) {
2826                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2827                         return 0;
2828         }
2829         if (a == &dev_attr_eui.attr) {
2830                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2831                         return 0;
2832         }
2833 #ifdef CONFIG_NVME_MULTIPATH
2834         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2835                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2836                         return 0;
2837                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2838                         return 0;
2839         }
2840 #endif
2841         return a->mode;
2842 }
2843
2844 static const struct attribute_group nvme_ns_id_attr_group = {
2845         .attrs          = nvme_ns_id_attrs,
2846         .is_visible     = nvme_ns_id_attrs_are_visible,
2847 };
2848
2849 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2850         &nvme_ns_id_attr_group,
2851 #ifdef CONFIG_NVM
2852         &nvme_nvm_attr_group,
2853 #endif
2854         NULL,
2855 };
2856
2857 #define nvme_show_str_function(field)                                           \
2858 static ssize_t  field##_show(struct device *dev,                                \
2859                             struct device_attribute *attr, char *buf)           \
2860 {                                                                               \
2861         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2862         return sprintf(buf, "%.*s\n",                                           \
2863                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2864 }                                                                               \
2865 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2866
2867 nvme_show_str_function(model);
2868 nvme_show_str_function(serial);
2869 nvme_show_str_function(firmware_rev);
2870
2871 #define nvme_show_int_function(field)                                           \
2872 static ssize_t  field##_show(struct device *dev,                                \
2873                             struct device_attribute *attr, char *buf)           \
2874 {                                                                               \
2875         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2876         return sprintf(buf, "%d\n", ctrl->field);       \
2877 }                                                                               \
2878 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2879
2880 nvme_show_int_function(cntlid);
2881 nvme_show_int_function(numa_node);
2882
2883 static ssize_t nvme_sysfs_delete(struct device *dev,
2884                                 struct device_attribute *attr, const char *buf,
2885                                 size_t count)
2886 {
2887         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2888
2889         if (device_remove_file_self(dev, attr))
2890                 nvme_delete_ctrl_sync(ctrl);
2891         return count;
2892 }
2893 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2894
2895 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2896                                          struct device_attribute *attr,
2897                                          char *buf)
2898 {
2899         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2900
2901         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2902 }
2903 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2904
2905 static ssize_t nvme_sysfs_show_state(struct device *dev,
2906                                      struct device_attribute *attr,
2907                                      char *buf)
2908 {
2909         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2910         static const char *const state_name[] = {
2911                 [NVME_CTRL_NEW]         = "new",
2912                 [NVME_CTRL_LIVE]        = "live",
2913                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2914                 [NVME_CTRL_RESETTING]   = "resetting",
2915                 [NVME_CTRL_CONNECTING]  = "connecting",
2916                 [NVME_CTRL_DELETING]    = "deleting",
2917                 [NVME_CTRL_DEAD]        = "dead",
2918         };
2919
2920         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2921             state_name[ctrl->state])
2922                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2923
2924         return sprintf(buf, "unknown state\n");
2925 }
2926
2927 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2928
2929 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2930                                          struct device_attribute *attr,
2931                                          char *buf)
2932 {
2933         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2934
2935         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2936 }
2937 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2938
2939 static ssize_t nvme_sysfs_show_address(struct device *dev,
2940                                          struct device_attribute *attr,
2941                                          char *buf)
2942 {
2943         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2944
2945         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2946 }
2947 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2948
2949 static struct attribute *nvme_dev_attrs[] = {
2950         &dev_attr_reset_controller.attr,
2951         &dev_attr_rescan_controller.attr,
2952         &dev_attr_model.attr,
2953         &dev_attr_serial.attr,
2954         &dev_attr_firmware_rev.attr,
2955         &dev_attr_cntlid.attr,
2956         &dev_attr_delete_controller.attr,
2957         &dev_attr_transport.attr,
2958         &dev_attr_subsysnqn.attr,
2959         &dev_attr_address.attr,
2960         &dev_attr_state.attr,
2961         &dev_attr_numa_node.attr,
2962         NULL
2963 };
2964
2965 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2966                 struct attribute *a, int n)
2967 {
2968         struct device *dev = container_of(kobj, struct device, kobj);
2969         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2970
2971         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2972                 return 0;
2973         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2974                 return 0;
2975
2976         return a->mode;
2977 }
2978
2979 static struct attribute_group nvme_dev_attrs_group = {
2980         .attrs          = nvme_dev_attrs,
2981         .is_visible     = nvme_dev_attrs_are_visible,
2982 };
2983
2984 static const struct attribute_group *nvme_dev_attr_groups[] = {
2985         &nvme_dev_attrs_group,
2986         NULL,
2987 };
2988
2989 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2990                 unsigned nsid)
2991 {
2992         struct nvme_ns_head *h;
2993
2994         lockdep_assert_held(&subsys->lock);
2995
2996         list_for_each_entry(h, &subsys->nsheads, entry) {
2997                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2998                         return h;
2999         }
3000
3001         return NULL;
3002 }
3003
3004 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3005                 struct nvme_ns_head *new)
3006 {
3007         struct nvme_ns_head *h;
3008
3009         lockdep_assert_held(&subsys->lock);
3010
3011         list_for_each_entry(h, &subsys->nsheads, entry) {
3012                 if (nvme_ns_ids_valid(&new->ids) &&
3013                     !list_empty(&h->list) &&
3014                     nvme_ns_ids_equal(&new->ids, &h->ids))
3015                         return -EINVAL;
3016         }
3017
3018         return 0;
3019 }
3020
3021 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3022                 unsigned nsid, struct nvme_id_ns *id)
3023 {
3024         struct nvme_ns_head *head;
3025         size_t size = sizeof(*head);
3026         int ret = -ENOMEM;
3027
3028 #ifdef CONFIG_NVME_MULTIPATH
3029         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3030 #endif
3031
3032         head = kzalloc(size, GFP_KERNEL);
3033         if (!head)
3034                 goto out;
3035         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3036         if (ret < 0)
3037                 goto out_free_head;
3038         head->instance = ret;
3039         INIT_LIST_HEAD(&head->list);
3040         ret = init_srcu_struct(&head->srcu);
3041         if (ret)
3042                 goto out_ida_remove;
3043         head->subsys = ctrl->subsys;
3044         head->ns_id = nsid;
3045         kref_init(&head->ref);
3046
3047         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3048
3049         ret = __nvme_check_ids(ctrl->subsys, head);
3050         if (ret) {
3051                 dev_err(ctrl->device,
3052                         "duplicate IDs for nsid %d\n", nsid);
3053                 goto out_cleanup_srcu;
3054         }
3055
3056         ret = nvme_mpath_alloc_disk(ctrl, head);
3057         if (ret)
3058                 goto out_cleanup_srcu;
3059
3060         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3061
3062         kref_get(&ctrl->subsys->ref);
3063
3064         return head;
3065 out_cleanup_srcu:
3066         cleanup_srcu_struct(&head->srcu);
3067 out_ida_remove:
3068         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3069 out_free_head:
3070         kfree(head);
3071 out:
3072         return ERR_PTR(ret);
3073 }
3074
3075 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3076                 struct nvme_id_ns *id)
3077 {
3078         struct nvme_ctrl *ctrl = ns->ctrl;
3079         bool is_shared = id->nmic & (1 << 0);
3080         struct nvme_ns_head *head = NULL;
3081         int ret = 0;
3082
3083         mutex_lock(&ctrl->subsys->lock);
3084         if (is_shared)
3085                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3086         if (!head) {
3087                 head = nvme_alloc_ns_head(ctrl, nsid, id);
3088                 if (IS_ERR(head)) {
3089                         ret = PTR_ERR(head);
3090                         goto out_unlock;
3091                 }
3092         } else {
3093                 struct nvme_ns_ids ids;
3094
3095                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3096                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3097                         dev_err(ctrl->device,
3098                                 "IDs don't match for shared namespace %d\n",
3099                                         nsid);
3100                         ret = -EINVAL;
3101                         goto out_unlock;
3102                 }
3103         }
3104
3105         list_add_tail(&ns->siblings, &head->list);
3106         ns->head = head;
3107
3108 out_unlock:
3109         mutex_unlock(&ctrl->subsys->lock);
3110         return ret;
3111 }
3112
3113 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3114 {
3115         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3116         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3117
3118         return nsa->head->ns_id - nsb->head->ns_id;
3119 }
3120
3121 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3122 {
3123         struct nvme_ns *ns, *ret = NULL;
3124
3125         down_read(&ctrl->namespaces_rwsem);
3126         list_for_each_entry(ns, &ctrl->namespaces, list) {
3127                 if (ns->head->ns_id == nsid) {
3128                         if (!kref_get_unless_zero(&ns->kref))
3129                                 continue;
3130                         ret = ns;
3131                         break;
3132                 }
3133                 if (ns->head->ns_id > nsid)
3134                         break;
3135         }
3136         up_read(&ctrl->namespaces_rwsem);
3137         return ret;
3138 }
3139
3140 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3141 {
3142         struct streams_directive_params s;
3143         int ret;
3144
3145         if (!ctrl->nr_streams)
3146                 return 0;
3147
3148         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3149         if (ret)
3150                 return ret;
3151
3152         ns->sws = le32_to_cpu(s.sws);
3153         ns->sgs = le16_to_cpu(s.sgs);
3154
3155         if (ns->sws) {
3156                 unsigned int bs = 1 << ns->lba_shift;
3157
3158                 blk_queue_io_min(ns->queue, bs * ns->sws);
3159                 if (ns->sgs)
3160                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3161         }
3162
3163         return 0;
3164 }
3165
3166 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3167 {
3168         struct nvme_ns *ns;
3169         struct gendisk *disk;
3170         struct nvme_id_ns *id;
3171         char disk_name[DISK_NAME_LEN];
3172         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT;
3173
3174         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3175         if (!ns)
3176                 return;
3177
3178         ns->queue = blk_mq_init_queue(ctrl->tagset);
3179         if (IS_ERR(ns->queue))
3180                 goto out_free_ns;
3181
3182         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3183         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3184                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3185
3186         ns->queue->queuedata = ns;
3187         ns->ctrl = ctrl;
3188
3189         kref_init(&ns->kref);
3190         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3191
3192         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3193         nvme_set_queue_limits(ctrl, ns->queue);
3194
3195         id = nvme_identify_ns(ctrl, nsid);
3196         if (!id)
3197                 goto out_free_queue;
3198
3199         if (id->ncap == 0)
3200                 goto out_free_id;
3201
3202         if (nvme_init_ns_head(ns, nsid, id))
3203                 goto out_free_id;
3204         nvme_setup_streams_ns(ctrl, ns);
3205         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3206
3207         disk = alloc_disk_node(0, node);
3208         if (!disk)
3209                 goto out_unlink_ns;
3210
3211         disk->fops = &nvme_fops;
3212         disk->private_data = ns;
3213         disk->queue = ns->queue;
3214         disk->flags = flags;
3215         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3216         ns->disk = disk;
3217
3218         __nvme_revalidate_disk(disk, id);
3219
3220         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3221                 if (nvme_nvm_register(ns, disk_name, node)) {
3222                         dev_warn(ctrl->device, "LightNVM init failure\n");
3223                         goto out_put_disk;
3224                 }
3225         }
3226
3227         down_write(&ctrl->namespaces_rwsem);
3228         list_add_tail(&ns->list, &ctrl->namespaces);
3229         up_write(&ctrl->namespaces_rwsem);
3230
3231         nvme_get_ctrl(ctrl);
3232
3233         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3234
3235         nvme_mpath_add_disk(ns, id);
3236         nvme_fault_inject_init(ns);
3237         kfree(id);
3238
3239         return;
3240  out_put_disk:
3241         put_disk(ns->disk);
3242  out_unlink_ns:
3243         mutex_lock(&ctrl->subsys->lock);
3244         list_del_rcu(&ns->siblings);
3245         mutex_unlock(&ctrl->subsys->lock);
3246  out_free_id:
3247         kfree(id);
3248  out_free_queue:
3249         blk_cleanup_queue(ns->queue);
3250  out_free_ns:
3251         kfree(ns);
3252 }
3253
3254 static void nvme_ns_remove(struct nvme_ns *ns)
3255 {
3256         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3257                 return;
3258
3259         nvme_fault_inject_fini(ns);
3260         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3261                 del_gendisk(ns->disk);
3262                 blk_cleanup_queue(ns->queue);
3263                 if (blk_get_integrity(ns->disk))
3264                         blk_integrity_unregister(ns->disk);
3265         }
3266
3267         mutex_lock(&ns->ctrl->subsys->lock);
3268         list_del_rcu(&ns->siblings);
3269         nvme_mpath_clear_current_path(ns);
3270         mutex_unlock(&ns->ctrl->subsys->lock);
3271
3272         down_write(&ns->ctrl->namespaces_rwsem);
3273         list_del_init(&ns->list);
3274         up_write(&ns->ctrl->namespaces_rwsem);
3275
3276         synchronize_srcu(&ns->head->srcu);
3277         nvme_mpath_check_last_path(ns);
3278         nvme_put_ns(ns);
3279 }
3280
3281 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3282 {
3283         struct nvme_ns *ns;
3284
3285         ns = nvme_find_get_ns(ctrl, nsid);
3286         if (ns) {
3287                 if (ns->disk && revalidate_disk(ns->disk))
3288                         nvme_ns_remove(ns);
3289                 nvme_put_ns(ns);
3290         } else
3291                 nvme_alloc_ns(ctrl, nsid);
3292 }
3293
3294 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3295                                         unsigned nsid)
3296 {
3297         struct nvme_ns *ns, *next;
3298         LIST_HEAD(rm_list);
3299
3300         down_write(&ctrl->namespaces_rwsem);
3301         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3302                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3303                         list_move_tail(&ns->list, &rm_list);
3304         }
3305         up_write(&ctrl->namespaces_rwsem);
3306
3307         list_for_each_entry_safe(ns, next, &rm_list, list)
3308                 nvme_ns_remove(ns);
3309
3310 }
3311
3312 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3313 {
3314         struct nvme_ns *ns;
3315         __le32 *ns_list;
3316         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3317         int ret = 0;
3318
3319         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3320         if (!ns_list)
3321                 return -ENOMEM;
3322
3323         for (i = 0; i < num_lists; i++) {
3324                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3325                 if (ret)
3326                         goto free;
3327
3328                 for (j = 0; j < min(nn, 1024U); j++) {
3329                         nsid = le32_to_cpu(ns_list[j]);
3330                         if (!nsid)
3331                                 goto out;
3332
3333                         nvme_validate_ns(ctrl, nsid);
3334
3335                         while (++prev < nsid) {
3336                                 ns = nvme_find_get_ns(ctrl, prev);
3337                                 if (ns) {
3338                                         nvme_ns_remove(ns);
3339                                         nvme_put_ns(ns);
3340                                 }
3341                         }
3342                 }
3343                 nn -= j;
3344         }
3345  out:
3346         nvme_remove_invalid_namespaces(ctrl, prev);
3347  free:
3348         kfree(ns_list);
3349         return ret;
3350 }
3351
3352 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3353 {
3354         unsigned i;
3355
3356         for (i = 1; i <= nn; i++)
3357                 nvme_validate_ns(ctrl, i);
3358
3359         nvme_remove_invalid_namespaces(ctrl, nn);
3360 }
3361
3362 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3363 {
3364         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3365         __le32 *log;
3366         int error;
3367
3368         log = kzalloc(log_size, GFP_KERNEL);
3369         if (!log)
3370                 return;
3371
3372         /*
3373          * We need to read the log to clear the AEN, but we don't want to rely
3374          * on it for the changed namespace information as userspace could have
3375          * raced with us in reading the log page, which could cause us to miss
3376          * updates.
3377          */
3378         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3379                         log_size, 0);
3380         if (error)
3381                 dev_warn(ctrl->device,
3382                         "reading changed ns log failed: %d\n", error);
3383
3384         kfree(log);
3385 }
3386
3387 static void nvme_scan_work(struct work_struct *work)
3388 {
3389         struct nvme_ctrl *ctrl =
3390                 container_of(work, struct nvme_ctrl, scan_work);
3391         struct nvme_id_ctrl *id;
3392         unsigned nn;
3393
3394         if (ctrl->state != NVME_CTRL_LIVE)
3395                 return;
3396
3397         WARN_ON_ONCE(!ctrl->tagset);
3398
3399         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3400                 dev_info(ctrl->device, "rescanning namespaces.\n");
3401                 nvme_clear_changed_ns_log(ctrl);
3402         }
3403
3404         if (nvme_identify_ctrl(ctrl, &id))
3405                 return;
3406
3407         mutex_lock(&ctrl->scan_lock);
3408         nn = le32_to_cpu(id->nn);
3409         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3410             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3411                 if (!nvme_scan_ns_list(ctrl, nn))
3412                         goto out_free_id;
3413         }
3414         nvme_scan_ns_sequential(ctrl, nn);
3415 out_free_id:
3416         mutex_unlock(&ctrl->scan_lock);
3417         kfree(id);
3418         down_write(&ctrl->namespaces_rwsem);
3419         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3420         up_write(&ctrl->namespaces_rwsem);
3421 }
3422
3423 /*
3424  * This function iterates the namespace list unlocked to allow recovery from
3425  * controller failure. It is up to the caller to ensure the namespace list is
3426  * not modified by scan work while this function is executing.
3427  */
3428 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3429 {
3430         struct nvme_ns *ns, *next;
3431         LIST_HEAD(ns_list);
3432
3433         /* prevent racing with ns scanning */
3434         flush_work(&ctrl->scan_work);
3435
3436         /*
3437          * The dead states indicates the controller was not gracefully
3438          * disconnected. In that case, we won't be able to flush any data while
3439          * removing the namespaces' disks; fail all the queues now to avoid
3440          * potentially having to clean up the failed sync later.
3441          */
3442         if (ctrl->state == NVME_CTRL_DEAD)
3443                 nvme_kill_queues(ctrl);
3444
3445         down_write(&ctrl->namespaces_rwsem);
3446         list_splice_init(&ctrl->namespaces, &ns_list);
3447         up_write(&ctrl->namespaces_rwsem);
3448
3449         list_for_each_entry_safe(ns, next, &ns_list, list)
3450                 nvme_ns_remove(ns);
3451 }
3452 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3453
3454 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3455 {
3456         char *envp[2] = { NULL, NULL };
3457         u32 aen_result = ctrl->aen_result;
3458
3459         ctrl->aen_result = 0;
3460         if (!aen_result)
3461                 return;
3462
3463         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3464         if (!envp[0])
3465                 return;
3466         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3467         kfree(envp[0]);
3468 }
3469
3470 static void nvme_async_event_work(struct work_struct *work)
3471 {
3472         struct nvme_ctrl *ctrl =
3473                 container_of(work, struct nvme_ctrl, async_event_work);
3474
3475         nvme_aen_uevent(ctrl);
3476         ctrl->ops->submit_async_event(ctrl);
3477 }
3478
3479 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3480 {
3481
3482         u32 csts;
3483
3484         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3485                 return false;
3486
3487         if (csts == ~0)
3488                 return false;
3489
3490         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3491 }
3492
3493 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3494 {
3495         struct nvme_fw_slot_info_log *log;
3496
3497         log = kmalloc(sizeof(*log), GFP_KERNEL);
3498         if (!log)
3499                 return;
3500
3501         if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3502                         sizeof(*log), 0))
3503                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3504         kfree(log);
3505 }
3506
3507 static void nvme_fw_act_work(struct work_struct *work)
3508 {
3509         struct nvme_ctrl *ctrl = container_of(work,
3510                                 struct nvme_ctrl, fw_act_work);
3511         unsigned long fw_act_timeout;
3512
3513         if (ctrl->mtfa)
3514                 fw_act_timeout = jiffies +
3515                                 msecs_to_jiffies(ctrl->mtfa * 100);
3516         else
3517                 fw_act_timeout = jiffies +
3518                                 msecs_to_jiffies(admin_timeout * 1000);
3519
3520         nvme_stop_queues(ctrl);
3521         while (nvme_ctrl_pp_status(ctrl)) {
3522                 if (time_after(jiffies, fw_act_timeout)) {
3523                         dev_warn(ctrl->device,
3524                                 "Fw activation timeout, reset controller\n");
3525                         nvme_reset_ctrl(ctrl);
3526                         break;
3527                 }
3528                 msleep(100);
3529         }
3530
3531         if (ctrl->state != NVME_CTRL_LIVE)
3532                 return;
3533
3534         nvme_start_queues(ctrl);
3535         /* read FW slot information to clear the AER */
3536         nvme_get_fw_slot_info(ctrl);
3537 }
3538
3539 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3540 {
3541         u32 aer_notice_type = (result & 0xff00) >> 8;
3542
3543         switch (aer_notice_type) {
3544         case NVME_AER_NOTICE_NS_CHANGED:
3545                 trace_nvme_async_event(ctrl, aer_notice_type);
3546                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3547                 nvme_queue_scan(ctrl);
3548                 break;
3549         case NVME_AER_NOTICE_FW_ACT_STARTING:
3550                 trace_nvme_async_event(ctrl, aer_notice_type);
3551                 queue_work(nvme_wq, &ctrl->fw_act_work);
3552                 break;
3553 #ifdef CONFIG_NVME_MULTIPATH
3554         case NVME_AER_NOTICE_ANA:
3555                 trace_nvme_async_event(ctrl, aer_notice_type);
3556                 if (!ctrl->ana_log_buf)
3557                         break;
3558                 queue_work(nvme_wq, &ctrl->ana_work);
3559                 break;
3560 #endif
3561         default:
3562                 dev_warn(ctrl->device, "async event result %08x\n", result);
3563         }
3564 }
3565
3566 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3567                 volatile union nvme_result *res)
3568 {
3569         u32 result = le32_to_cpu(res->u32);
3570         u32 aer_type = result & 0x07;
3571
3572         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3573                 return;
3574
3575         switch (aer_type) {
3576         case NVME_AER_NOTICE:
3577                 nvme_handle_aen_notice(ctrl, result);
3578                 break;
3579         case NVME_AER_ERROR:
3580         case NVME_AER_SMART:
3581         case NVME_AER_CSS:
3582         case NVME_AER_VS:
3583                 trace_nvme_async_event(ctrl, aer_type);
3584                 ctrl->aen_result = result;
3585                 break;
3586         default:
3587                 break;
3588         }
3589         queue_work(nvme_wq, &ctrl->async_event_work);
3590 }
3591 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3592
3593 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3594 {
3595         nvme_mpath_stop(ctrl);
3596         nvme_stop_keep_alive(ctrl);
3597         flush_work(&ctrl->async_event_work);
3598         cancel_work_sync(&ctrl->fw_act_work);
3599         if (ctrl->ops->stop_ctrl)
3600                 ctrl->ops->stop_ctrl(ctrl);
3601 }
3602 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3603
3604 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3605 {
3606         if (ctrl->kato)
3607                 nvme_start_keep_alive(ctrl);
3608
3609         if (ctrl->queue_count > 1) {
3610                 nvme_queue_scan(ctrl);
3611                 nvme_enable_aen(ctrl);
3612                 queue_work(nvme_wq, &ctrl->async_event_work);
3613                 nvme_start_queues(ctrl);
3614         }
3615 }
3616 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3617
3618 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3619 {
3620         cdev_device_del(&ctrl->cdev, ctrl->device);
3621 }
3622 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3623
3624 static void nvme_free_ctrl(struct device *dev)
3625 {
3626         struct nvme_ctrl *ctrl =
3627                 container_of(dev, struct nvme_ctrl, ctrl_device);
3628         struct nvme_subsystem *subsys = ctrl->subsys;
3629
3630         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3631         kfree(ctrl->effects);
3632         nvme_mpath_uninit(ctrl);
3633         __free_page(ctrl->discard_page);
3634
3635         if (subsys) {
3636                 mutex_lock(&subsys->lock);
3637                 list_del(&ctrl->subsys_entry);
3638                 mutex_unlock(&subsys->lock);
3639                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3640         }
3641
3642         ctrl->ops->free_ctrl(ctrl);
3643
3644         if (subsys)
3645                 nvme_put_subsystem(subsys);
3646 }
3647
3648 /*
3649  * Initialize a NVMe controller structures.  This needs to be called during
3650  * earliest initialization so that we have the initialized structured around
3651  * during probing.
3652  */
3653 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3654                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3655 {
3656         int ret;
3657
3658         ctrl->state = NVME_CTRL_NEW;
3659         spin_lock_init(&ctrl->lock);
3660         mutex_init(&ctrl->scan_lock);
3661         INIT_LIST_HEAD(&ctrl->namespaces);
3662         init_rwsem(&ctrl->namespaces_rwsem);
3663         ctrl->dev = dev;
3664         ctrl->ops = ops;
3665         ctrl->quirks = quirks;
3666         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3667         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3668         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3669         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3670
3671         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3672         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3673         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3674
3675         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3676                         PAGE_SIZE);
3677         ctrl->discard_page = alloc_page(GFP_KERNEL);
3678         if (!ctrl->discard_page) {
3679                 ret = -ENOMEM;
3680                 goto out;
3681         }
3682
3683         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3684         if (ret < 0)
3685                 goto out;
3686         ctrl->instance = ret;
3687
3688         device_initialize(&ctrl->ctrl_device);
3689         ctrl->device = &ctrl->ctrl_device;
3690         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3691         ctrl->device->class = nvme_class;
3692         ctrl->device->parent = ctrl->dev;
3693         ctrl->device->groups = nvme_dev_attr_groups;
3694         ctrl->device->release = nvme_free_ctrl;
3695         dev_set_drvdata(ctrl->device, ctrl);
3696         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3697         if (ret)
3698                 goto out_release_instance;
3699
3700         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3701         ctrl->cdev.owner = ops->module;
3702         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3703         if (ret)
3704                 goto out_free_name;
3705
3706         /*
3707          * Initialize latency tolerance controls.  The sysfs files won't
3708          * be visible to userspace unless the device actually supports APST.
3709          */
3710         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3711         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3712                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3713
3714         return 0;
3715 out_free_name:
3716         kfree_const(ctrl->device->kobj.name);
3717 out_release_instance:
3718         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3719 out:
3720         if (ctrl->discard_page)
3721                 __free_page(ctrl->discard_page);
3722         return ret;
3723 }
3724 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3725
3726 /**
3727  * nvme_kill_queues(): Ends all namespace queues
3728  * @ctrl: the dead controller that needs to end
3729  *
3730  * Call this function when the driver determines it is unable to get the
3731  * controller in a state capable of servicing IO.
3732  */
3733 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3734 {
3735         struct nvme_ns *ns;
3736
3737         down_read(&ctrl->namespaces_rwsem);
3738
3739         /* Forcibly unquiesce queues to avoid blocking dispatch */
3740         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3741                 blk_mq_unquiesce_queue(ctrl->admin_q);
3742
3743         list_for_each_entry(ns, &ctrl->namespaces, list)
3744                 nvme_set_queue_dying(ns);
3745
3746         up_read(&ctrl->namespaces_rwsem);
3747 }
3748 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3749
3750 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3751 {
3752         struct nvme_ns *ns;
3753
3754         down_read(&ctrl->namespaces_rwsem);
3755         list_for_each_entry(ns, &ctrl->namespaces, list)
3756                 blk_mq_unfreeze_queue(ns->queue);
3757         up_read(&ctrl->namespaces_rwsem);
3758 }
3759 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3760
3761 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3762 {
3763         struct nvme_ns *ns;
3764
3765         down_read(&ctrl->namespaces_rwsem);
3766         list_for_each_entry(ns, &ctrl->namespaces, list) {
3767                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3768                 if (timeout <= 0)
3769                         break;
3770         }
3771         up_read(&ctrl->namespaces_rwsem);
3772 }
3773 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3774
3775 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3776 {
3777         struct nvme_ns *ns;
3778
3779         down_read(&ctrl->namespaces_rwsem);
3780         list_for_each_entry(ns, &ctrl->namespaces, list)
3781                 blk_mq_freeze_queue_wait(ns->queue);
3782         up_read(&ctrl->namespaces_rwsem);
3783 }
3784 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3785
3786 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3787 {
3788         struct nvme_ns *ns;
3789
3790         down_read(&ctrl->namespaces_rwsem);
3791         list_for_each_entry(ns, &ctrl->namespaces, list)
3792                 blk_freeze_queue_start(ns->queue);
3793         up_read(&ctrl->namespaces_rwsem);
3794 }
3795 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3796
3797 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3798 {
3799         struct nvme_ns *ns;
3800
3801         down_read(&ctrl->namespaces_rwsem);
3802         list_for_each_entry(ns, &ctrl->namespaces, list)
3803                 blk_mq_quiesce_queue(ns->queue);
3804         up_read(&ctrl->namespaces_rwsem);
3805 }
3806 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3807
3808 void nvme_start_queues(struct nvme_ctrl *ctrl)
3809 {
3810         struct nvme_ns *ns;
3811
3812         down_read(&ctrl->namespaces_rwsem);
3813         list_for_each_entry(ns, &ctrl->namespaces, list)
3814                 blk_mq_unquiesce_queue(ns->queue);
3815         up_read(&ctrl->namespaces_rwsem);
3816 }
3817 EXPORT_SYMBOL_GPL(nvme_start_queues);
3818
3819 int __init nvme_core_init(void)
3820 {
3821         int result = -ENOMEM;
3822
3823         nvme_wq = alloc_workqueue("nvme-wq",
3824                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3825         if (!nvme_wq)
3826                 goto out;
3827
3828         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3829                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3830         if (!nvme_reset_wq)
3831                 goto destroy_wq;
3832
3833         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3834                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3835         if (!nvme_delete_wq)
3836                 goto destroy_reset_wq;
3837
3838         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3839         if (result < 0)
3840                 goto destroy_delete_wq;
3841
3842         nvme_class = class_create(THIS_MODULE, "nvme");
3843         if (IS_ERR(nvme_class)) {
3844                 result = PTR_ERR(nvme_class);
3845                 goto unregister_chrdev;
3846         }
3847
3848         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3849         if (IS_ERR(nvme_subsys_class)) {
3850                 result = PTR_ERR(nvme_subsys_class);
3851                 goto destroy_class;
3852         }
3853         return 0;
3854
3855 destroy_class:
3856         class_destroy(nvme_class);
3857 unregister_chrdev:
3858         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3859 destroy_delete_wq:
3860         destroy_workqueue(nvme_delete_wq);
3861 destroy_reset_wq:
3862         destroy_workqueue(nvme_reset_wq);
3863 destroy_wq:
3864         destroy_workqueue(nvme_wq);
3865 out:
3866         return result;
3867 }
3868
3869 void __exit nvme_core_exit(void)
3870 {
3871         ida_destroy(&nvme_subsystems_ida);
3872         class_destroy(nvme_subsys_class);
3873         class_destroy(nvme_class);
3874         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3875         destroy_workqueue(nvme_delete_wq);
3876         destroy_workqueue(nvme_reset_wq);
3877         destroy_workqueue(nvme_wq);
3878 }
3879
3880 MODULE_LICENSE("GPL");
3881 MODULE_VERSION("1.0");
3882 module_init(nvme_core_init);
3883 module_exit(nvme_core_exit);