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                 nvme_start_freeze(ctrl);
1257                 nvme_wait_freeze(ctrl);
1258         }
1259         return effects;
1260 }
1261
1262 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1263 {
1264         struct nvme_ns *ns;
1265
1266         down_read(&ctrl->namespaces_rwsem);
1267         list_for_each_entry(ns, &ctrl->namespaces, list)
1268                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1269                         nvme_set_queue_dying(ns);
1270         up_read(&ctrl->namespaces_rwsem);
1271
1272         nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1273 }
1274
1275 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1276 {
1277         /*
1278          * Revalidate LBA changes prior to unfreezing. This is necessary to
1279          * prevent memory corruption if a logical block size was changed by
1280          * this command.
1281          */
1282         if (effects & NVME_CMD_EFFECTS_LBCC)
1283                 nvme_update_formats(ctrl);
1284         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK))
1285                 nvme_unfreeze(ctrl);
1286         if (effects & NVME_CMD_EFFECTS_CCC)
1287                 nvme_init_identify(ctrl);
1288         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1289                 nvme_queue_scan(ctrl);
1290 }
1291
1292 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1293                         struct nvme_passthru_cmd __user *ucmd)
1294 {
1295         struct nvme_passthru_cmd cmd;
1296         struct nvme_command c;
1297         unsigned timeout = 0;
1298         u32 effects;
1299         int status;
1300
1301         if (!capable(CAP_SYS_ADMIN))
1302                 return -EACCES;
1303         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1304                 return -EFAULT;
1305         if (cmd.flags)
1306                 return -EINVAL;
1307
1308         memset(&c, 0, sizeof(c));
1309         c.common.opcode = cmd.opcode;
1310         c.common.flags = cmd.flags;
1311         c.common.nsid = cpu_to_le32(cmd.nsid);
1312         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1313         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1314         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1315         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1316         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1317         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1318         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1319         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1320
1321         if (cmd.timeout_ms)
1322                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1323
1324         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1325         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1326                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1327                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1328                         0, &cmd.result, timeout);
1329         nvme_passthru_end(ctrl, effects);
1330
1331         if (status >= 0) {
1332                 if (put_user(cmd.result, &ucmd->result))
1333                         return -EFAULT;
1334         }
1335
1336         return status;
1337 }
1338
1339 /*
1340  * Issue ioctl requests on the first available path.  Note that unlike normal
1341  * block layer requests we will not retry failed request on another controller.
1342  */
1343 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1344                 struct nvme_ns_head **head, int *srcu_idx)
1345 {
1346 #ifdef CONFIG_NVME_MULTIPATH
1347         if (disk->fops == &nvme_ns_head_ops) {
1348                 *head = disk->private_data;
1349                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1350                 return nvme_find_path(*head);
1351         }
1352 #endif
1353         *head = NULL;
1354         *srcu_idx = -1;
1355         return disk->private_data;
1356 }
1357
1358 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1359 {
1360         if (head)
1361                 srcu_read_unlock(&head->srcu, idx);
1362 }
1363
1364 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1365 {
1366         switch (cmd) {
1367         case NVME_IOCTL_ID:
1368                 force_successful_syscall_return();
1369                 return ns->head->ns_id;
1370         case NVME_IOCTL_ADMIN_CMD:
1371                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1372         case NVME_IOCTL_IO_CMD:
1373                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1374         case NVME_IOCTL_SUBMIT_IO:
1375                 return nvme_submit_io(ns, (void __user *)arg);
1376         default:
1377 #ifdef CONFIG_NVM
1378                 if (ns->ndev)
1379                         return nvme_nvm_ioctl(ns, cmd, arg);
1380 #endif
1381                 if (is_sed_ioctl(cmd))
1382                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
1383                                          (void __user *) arg);
1384                 return -ENOTTY;
1385         }
1386 }
1387
1388 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1389                 unsigned int cmd, unsigned long arg)
1390 {
1391         struct nvme_ns_head *head = NULL;
1392         struct nvme_ns *ns;
1393         int srcu_idx, ret;
1394
1395         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1396         if (unlikely(!ns))
1397                 ret = -EWOULDBLOCK;
1398         else
1399                 ret = nvme_ns_ioctl(ns, cmd, arg);
1400         nvme_put_ns_from_disk(head, srcu_idx);
1401         return ret;
1402 }
1403
1404 static int nvme_open(struct block_device *bdev, fmode_t mode)
1405 {
1406         struct nvme_ns *ns = bdev->bd_disk->private_data;
1407
1408 #ifdef CONFIG_NVME_MULTIPATH
1409         /* should never be called due to GENHD_FL_HIDDEN */
1410         if (WARN_ON_ONCE(ns->head->disk))
1411                 goto fail;
1412 #endif
1413         if (!kref_get_unless_zero(&ns->kref))
1414                 goto fail;
1415         if (!try_module_get(ns->ctrl->ops->module))
1416                 goto fail_put_ns;
1417
1418         return 0;
1419
1420 fail_put_ns:
1421         nvme_put_ns(ns);
1422 fail:
1423         return -ENXIO;
1424 }
1425
1426 static void nvme_release(struct gendisk *disk, fmode_t mode)
1427 {
1428         struct nvme_ns *ns = disk->private_data;
1429
1430         module_put(ns->ctrl->ops->module);
1431         nvme_put_ns(ns);
1432 }
1433
1434 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1435 {
1436         /* some standard values */
1437         geo->heads = 1 << 6;
1438         geo->sectors = 1 << 5;
1439         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1440         return 0;
1441 }
1442
1443 #ifdef CONFIG_BLK_DEV_INTEGRITY
1444 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1445 {
1446         struct blk_integrity integrity;
1447
1448         memset(&integrity, 0, sizeof(integrity));
1449         switch (pi_type) {
1450         case NVME_NS_DPS_PI_TYPE3:
1451                 integrity.profile = &t10_pi_type3_crc;
1452                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1453                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1454                 break;
1455         case NVME_NS_DPS_PI_TYPE1:
1456         case NVME_NS_DPS_PI_TYPE2:
1457                 integrity.profile = &t10_pi_type1_crc;
1458                 integrity.tag_size = sizeof(u16);
1459                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1460                 break;
1461         default:
1462                 integrity.profile = NULL;
1463                 break;
1464         }
1465         integrity.tuple_size = ms;
1466         blk_integrity_register(disk, &integrity);
1467         blk_queue_max_integrity_segments(disk->queue, 1);
1468 }
1469 #else
1470 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1471 {
1472 }
1473 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1474
1475 static void nvme_set_chunk_size(struct nvme_ns *ns)
1476 {
1477         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1478         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1479 }
1480
1481 static void nvme_config_discard(struct nvme_ns *ns)
1482 {
1483         struct nvme_ctrl *ctrl = ns->ctrl;
1484         struct request_queue *queue = ns->queue;
1485         u32 size = queue_logical_block_size(queue);
1486
1487         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1488                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1489                 return;
1490         }
1491
1492         if (ctrl->nr_streams && ns->sws && ns->sgs)
1493                 size *= ns->sws * ns->sgs;
1494
1495         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1496                         NVME_DSM_MAX_RANGES);
1497
1498         queue->limits.discard_alignment = 0;
1499         queue->limits.discard_granularity = size;
1500
1501         /* If discard is already enabled, don't reset queue limits */
1502         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1503                 return;
1504
1505         blk_queue_max_discard_sectors(queue, UINT_MAX);
1506         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1507
1508         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1509                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1510 }
1511
1512 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1513                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1514 {
1515         memset(ids, 0, sizeof(*ids));
1516
1517         if (ctrl->vs >= NVME_VS(1, 1, 0))
1518                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1519         if (ctrl->vs >= NVME_VS(1, 2, 0))
1520                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1521         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1522                  /* Don't treat error as fatal we potentially
1523                   * already have a NGUID or EUI-64
1524                   */
1525                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1526                         dev_warn(ctrl->device,
1527                                  "%s: Identify Descriptors failed\n", __func__);
1528         }
1529 }
1530
1531 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1532 {
1533         return !uuid_is_null(&ids->uuid) ||
1534                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1535                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1536 }
1537
1538 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1539 {
1540         return uuid_equal(&a->uuid, &b->uuid) &&
1541                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1542                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1543 }
1544
1545 static void nvme_update_disk_info(struct gendisk *disk,
1546                 struct nvme_ns *ns, struct nvme_id_ns *id)
1547 {
1548         sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1549         unsigned short bs = 1 << ns->lba_shift;
1550
1551         blk_mq_freeze_queue(disk->queue);
1552         blk_integrity_unregister(disk);
1553
1554         blk_queue_logical_block_size(disk->queue, bs);
1555         blk_queue_physical_block_size(disk->queue, bs);
1556         blk_queue_io_min(disk->queue, bs);
1557
1558         if (ns->ms && !ns->ext &&
1559             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1560                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1561         if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1562                 capacity = 0;
1563
1564         set_capacity(disk, capacity);
1565         nvme_config_discard(ns);
1566
1567         if (id->nsattr & (1 << 0))
1568                 set_disk_ro(disk, true);
1569         else
1570                 set_disk_ro(disk, false);
1571
1572         blk_mq_unfreeze_queue(disk->queue);
1573 }
1574
1575 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1576 {
1577         struct nvme_ns *ns = disk->private_data;
1578
1579         /*
1580          * If identify namespace failed, use default 512 byte block size so
1581          * block layer can use before failing read/write for 0 capacity.
1582          */
1583         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1584         if (ns->lba_shift == 0)
1585                 ns->lba_shift = 9;
1586         ns->noiob = le16_to_cpu(id->noiob);
1587         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1588         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1589         /* the PI implementation requires metadata equal t10 pi tuple size */
1590         if (ns->ms == sizeof(struct t10_pi_tuple))
1591                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1592         else
1593                 ns->pi_type = 0;
1594
1595         if (ns->noiob)
1596                 nvme_set_chunk_size(ns);
1597         nvme_update_disk_info(disk, ns, id);
1598 #ifdef CONFIG_NVME_MULTIPATH
1599         if (ns->head->disk) {
1600                 nvme_update_disk_info(ns->head->disk, ns, id);
1601                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1602         }
1603 #endif
1604 }
1605
1606 static int nvme_revalidate_disk(struct gendisk *disk)
1607 {
1608         struct nvme_ns *ns = disk->private_data;
1609         struct nvme_ctrl *ctrl = ns->ctrl;
1610         struct nvme_id_ns *id;
1611         struct nvme_ns_ids ids;
1612         int ret = 0;
1613
1614         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1615                 set_capacity(disk, 0);
1616                 return -ENODEV;
1617         }
1618
1619         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1620         if (!id)
1621                 return -ENODEV;
1622
1623         if (id->ncap == 0) {
1624                 ret = -ENODEV;
1625                 goto out;
1626         }
1627
1628         __nvme_revalidate_disk(disk, id);
1629         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1630         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1631                 dev_err(ctrl->device,
1632                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1633                 ret = -ENODEV;
1634         }
1635
1636 out:
1637         kfree(id);
1638         return ret;
1639 }
1640
1641 static char nvme_pr_type(enum pr_type type)
1642 {
1643         switch (type) {
1644         case PR_WRITE_EXCLUSIVE:
1645                 return 1;
1646         case PR_EXCLUSIVE_ACCESS:
1647                 return 2;
1648         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1649                 return 3;
1650         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1651                 return 4;
1652         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1653                 return 5;
1654         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1655                 return 6;
1656         default:
1657                 return 0;
1658         }
1659 };
1660
1661 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1662                                 u64 key, u64 sa_key, u8 op)
1663 {
1664         struct nvme_ns_head *head = NULL;
1665         struct nvme_ns *ns;
1666         struct nvme_command c;
1667         int srcu_idx, ret;
1668         u8 data[16] = { 0, };
1669
1670         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1671         if (unlikely(!ns))
1672                 return -EWOULDBLOCK;
1673
1674         put_unaligned_le64(key, &data[0]);
1675         put_unaligned_le64(sa_key, &data[8]);
1676
1677         memset(&c, 0, sizeof(c));
1678         c.common.opcode = op;
1679         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1680         c.common.cdw10 = cpu_to_le32(cdw10);
1681
1682         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1683         nvme_put_ns_from_disk(head, srcu_idx);
1684         return ret;
1685 }
1686
1687 static int nvme_pr_register(struct block_device *bdev, u64 old,
1688                 u64 new, unsigned flags)
1689 {
1690         u32 cdw10;
1691
1692         if (flags & ~PR_FL_IGNORE_KEY)
1693                 return -EOPNOTSUPP;
1694
1695         cdw10 = old ? 2 : 0;
1696         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1697         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1698         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1699 }
1700
1701 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1702                 enum pr_type type, unsigned flags)
1703 {
1704         u32 cdw10;
1705
1706         if (flags & ~PR_FL_IGNORE_KEY)
1707                 return -EOPNOTSUPP;
1708
1709         cdw10 = nvme_pr_type(type) << 8;
1710         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1711         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1712 }
1713
1714 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1715                 enum pr_type type, bool abort)
1716 {
1717         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1718         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1719 }
1720
1721 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1722 {
1723         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1724         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1725 }
1726
1727 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1728 {
1729         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1730         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1731 }
1732
1733 static const struct pr_ops nvme_pr_ops = {
1734         .pr_register    = nvme_pr_register,
1735         .pr_reserve     = nvme_pr_reserve,
1736         .pr_release     = nvme_pr_release,
1737         .pr_preempt     = nvme_pr_preempt,
1738         .pr_clear       = nvme_pr_clear,
1739 };
1740
1741 #ifdef CONFIG_BLK_SED_OPAL
1742 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1743                 bool send)
1744 {
1745         struct nvme_ctrl *ctrl = data;
1746         struct nvme_command cmd;
1747
1748         memset(&cmd, 0, sizeof(cmd));
1749         if (send)
1750                 cmd.common.opcode = nvme_admin_security_send;
1751         else
1752                 cmd.common.opcode = nvme_admin_security_recv;
1753         cmd.common.nsid = 0;
1754         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1755         cmd.common.cdw11 = cpu_to_le32(len);
1756
1757         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1758                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1759 }
1760 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1761 #endif /* CONFIG_BLK_SED_OPAL */
1762
1763 static const struct block_device_operations nvme_fops = {
1764         .owner          = THIS_MODULE,
1765         .ioctl          = nvme_ioctl,
1766         .compat_ioctl   = nvme_ioctl,
1767         .open           = nvme_open,
1768         .release        = nvme_release,
1769         .getgeo         = nvme_getgeo,
1770         .revalidate_disk= nvme_revalidate_disk,
1771         .pr_ops         = &nvme_pr_ops,
1772 };
1773
1774 #ifdef CONFIG_NVME_MULTIPATH
1775 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1776 {
1777         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1778
1779         if (!kref_get_unless_zero(&head->ref))
1780                 return -ENXIO;
1781         return 0;
1782 }
1783
1784 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1785 {
1786         nvme_put_ns_head(disk->private_data);
1787 }
1788
1789 const struct block_device_operations nvme_ns_head_ops = {
1790         .owner          = THIS_MODULE,
1791         .open           = nvme_ns_head_open,
1792         .release        = nvme_ns_head_release,
1793         .ioctl          = nvme_ioctl,
1794         .compat_ioctl   = nvme_ioctl,
1795         .getgeo         = nvme_getgeo,
1796         .pr_ops         = &nvme_pr_ops,
1797 };
1798 #endif /* CONFIG_NVME_MULTIPATH */
1799
1800 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1801 {
1802         unsigned long timeout =
1803                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1804         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1805         int ret;
1806
1807         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1808                 if (csts == ~0)
1809                         return -ENODEV;
1810                 if ((csts & NVME_CSTS_RDY) == bit)
1811                         break;
1812
1813                 msleep(100);
1814                 if (fatal_signal_pending(current))
1815                         return -EINTR;
1816                 if (time_after(jiffies, timeout)) {
1817                         dev_err(ctrl->device,
1818                                 "Device not ready; aborting %s\n", enabled ?
1819                                                 "initialisation" : "reset");
1820                         return -ENODEV;
1821                 }
1822         }
1823
1824         return ret;
1825 }
1826
1827 /*
1828  * If the device has been passed off to us in an enabled state, just clear
1829  * the enabled bit.  The spec says we should set the 'shutdown notification
1830  * bits', but doing so may cause the device to complete commands to the
1831  * admin queue ... and we don't know what memory that might be pointing at!
1832  */
1833 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1834 {
1835         int ret;
1836
1837         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1838         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1839
1840         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1841         if (ret)
1842                 return ret;
1843
1844         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1845                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1846
1847         return nvme_wait_ready(ctrl, cap, false);
1848 }
1849 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1850
1851 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1852 {
1853         /*
1854          * Default to a 4K page size, with the intention to update this
1855          * path in the future to accomodate architectures with differing
1856          * kernel and IO page sizes.
1857          */
1858         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1859         int ret;
1860
1861         if (page_shift < dev_page_min) {
1862                 dev_err(ctrl->device,
1863                         "Minimum device page size %u too large for host (%u)\n",
1864                         1 << dev_page_min, 1 << page_shift);
1865                 return -ENODEV;
1866         }
1867
1868         ctrl->page_size = 1 << page_shift;
1869
1870         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1871         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1872         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1873         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1874         ctrl->ctrl_config |= NVME_CC_ENABLE;
1875
1876         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1877         if (ret)
1878                 return ret;
1879         return nvme_wait_ready(ctrl, cap, true);
1880 }
1881 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1882
1883 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1884 {
1885         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1886         u32 csts;
1887         int ret;
1888
1889         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1890         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1891
1892         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1893         if (ret)
1894                 return ret;
1895
1896         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1897                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1898                         break;
1899
1900                 msleep(100);
1901                 if (fatal_signal_pending(current))
1902                         return -EINTR;
1903                 if (time_after(jiffies, timeout)) {
1904                         dev_err(ctrl->device,
1905                                 "Device shutdown incomplete; abort shutdown\n");
1906                         return -ENODEV;
1907                 }
1908         }
1909
1910         return ret;
1911 }
1912 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1913
1914 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1915                 struct request_queue *q)
1916 {
1917         bool vwc = false;
1918
1919         if (ctrl->max_hw_sectors) {
1920                 u32 max_segments =
1921                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1922
1923                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1924                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1925                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1926         }
1927         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1928             is_power_of_2(ctrl->max_hw_sectors))
1929                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1930         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1931         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1932                 vwc = true;
1933         blk_queue_write_cache(q, vwc, vwc);
1934 }
1935
1936 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1937 {
1938         __le64 ts;
1939         int ret;
1940
1941         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1942                 return 0;
1943
1944         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1945         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1946                         NULL);
1947         if (ret)
1948                 dev_warn_once(ctrl->device,
1949                         "could not set timestamp (%d)\n", ret);
1950         return ret;
1951 }
1952
1953 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
1954 {
1955         struct nvme_feat_host_behavior *host;
1956         int ret;
1957
1958         /* Don't bother enabling the feature if retry delay is not reported */
1959         if (!ctrl->crdt[0])
1960                 return 0;
1961
1962         host = kzalloc(sizeof(*host), GFP_KERNEL);
1963         if (!host)
1964                 return 0;
1965
1966         host->acre = NVME_ENABLE_ACRE;
1967         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
1968                                 host, sizeof(*host), NULL);
1969         kfree(host);
1970         return ret;
1971 }
1972
1973 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1974 {
1975         /*
1976          * APST (Autonomous Power State Transition) lets us program a
1977          * table of power state transitions that the controller will
1978          * perform automatically.  We configure it with a simple
1979          * heuristic: we are willing to spend at most 2% of the time
1980          * transitioning between power states.  Therefore, when running
1981          * in any given state, we will enter the next lower-power
1982          * non-operational state after waiting 50 * (enlat + exlat)
1983          * microseconds, as long as that state's exit latency is under
1984          * the requested maximum latency.
1985          *
1986          * We will not autonomously enter any non-operational state for
1987          * which the total latency exceeds ps_max_latency_us.  Users
1988          * can set ps_max_latency_us to zero to turn off APST.
1989          */
1990
1991         unsigned apste;
1992         struct nvme_feat_auto_pst *table;
1993         u64 max_lat_us = 0;
1994         int max_ps = -1;
1995         int ret;
1996
1997         /*
1998          * If APST isn't supported or if we haven't been initialized yet,
1999          * then don't do anything.
2000          */
2001         if (!ctrl->apsta)
2002                 return 0;
2003
2004         if (ctrl->npss > 31) {
2005                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2006                 return 0;
2007         }
2008
2009         table = kzalloc(sizeof(*table), GFP_KERNEL);
2010         if (!table)
2011                 return 0;
2012
2013         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2014                 /* Turn off APST. */
2015                 apste = 0;
2016                 dev_dbg(ctrl->device, "APST disabled\n");
2017         } else {
2018                 __le64 target = cpu_to_le64(0);
2019                 int state;
2020
2021                 /*
2022                  * Walk through all states from lowest- to highest-power.
2023                  * According to the spec, lower-numbered states use more
2024                  * power.  NPSS, despite the name, is the index of the
2025                  * lowest-power state, not the number of states.
2026                  */
2027                 for (state = (int)ctrl->npss; state >= 0; state--) {
2028                         u64 total_latency_us, exit_latency_us, transition_ms;
2029
2030                         if (target)
2031                                 table->entries[state] = target;
2032
2033                         /*
2034                          * Don't allow transitions to the deepest state
2035                          * if it's quirked off.
2036                          */
2037                         if (state == ctrl->npss &&
2038                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2039                                 continue;
2040
2041                         /*
2042                          * Is this state a useful non-operational state for
2043                          * higher-power states to autonomously transition to?
2044                          */
2045                         if (!(ctrl->psd[state].flags &
2046                               NVME_PS_FLAGS_NON_OP_STATE))
2047                                 continue;
2048
2049                         exit_latency_us =
2050                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2051                         if (exit_latency_us > ctrl->ps_max_latency_us)
2052                                 continue;
2053
2054                         total_latency_us =
2055                                 exit_latency_us +
2056                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2057
2058                         /*
2059                          * This state is good.  Use it as the APST idle
2060                          * target for higher power states.
2061                          */
2062                         transition_ms = total_latency_us + 19;
2063                         do_div(transition_ms, 20);
2064                         if (transition_ms > (1 << 24) - 1)
2065                                 transition_ms = (1 << 24) - 1;
2066
2067                         target = cpu_to_le64((state << 3) |
2068                                              (transition_ms << 8));
2069
2070                         if (max_ps == -1)
2071                                 max_ps = state;
2072
2073                         if (total_latency_us > max_lat_us)
2074                                 max_lat_us = total_latency_us;
2075                 }
2076
2077                 apste = 1;
2078
2079                 if (max_ps == -1) {
2080                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2081                 } else {
2082                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2083                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2084                 }
2085         }
2086
2087         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2088                                 table, sizeof(*table), NULL);
2089         if (ret)
2090                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2091
2092         kfree(table);
2093         return ret;
2094 }
2095
2096 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2097 {
2098         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2099         u64 latency;
2100
2101         switch (val) {
2102         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2103         case PM_QOS_LATENCY_ANY:
2104                 latency = U64_MAX;
2105                 break;
2106
2107         default:
2108                 latency = val;
2109         }
2110
2111         if (ctrl->ps_max_latency_us != latency) {
2112                 ctrl->ps_max_latency_us = latency;
2113                 nvme_configure_apst(ctrl);
2114         }
2115 }
2116
2117 struct nvme_core_quirk_entry {
2118         /*
2119          * NVMe model and firmware strings are padded with spaces.  For
2120          * simplicity, strings in the quirk table are padded with NULLs
2121          * instead.
2122          */
2123         u16 vid;
2124         const char *mn;
2125         const char *fr;
2126         unsigned long quirks;
2127 };
2128
2129 static const struct nvme_core_quirk_entry core_quirks[] = {
2130         {
2131                 /*
2132                  * This Toshiba device seems to die using any APST states.  See:
2133                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2134                  */
2135                 .vid = 0x1179,
2136                 .mn = "THNSF5256GPUK TOSHIBA",
2137                 .quirks = NVME_QUIRK_NO_APST,
2138         }
2139 };
2140
2141 /* match is null-terminated but idstr is space-padded. */
2142 static bool string_matches(const char *idstr, const char *match, size_t len)
2143 {
2144         size_t matchlen;
2145
2146         if (!match)
2147                 return true;
2148
2149         matchlen = strlen(match);
2150         WARN_ON_ONCE(matchlen > len);
2151
2152         if (memcmp(idstr, match, matchlen))
2153                 return false;
2154
2155         for (; matchlen < len; matchlen++)
2156                 if (idstr[matchlen] != ' ')
2157                         return false;
2158
2159         return true;
2160 }
2161
2162 static bool quirk_matches(const struct nvme_id_ctrl *id,
2163                           const struct nvme_core_quirk_entry *q)
2164 {
2165         return q->vid == le16_to_cpu(id->vid) &&
2166                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2167                 string_matches(id->fr, q->fr, sizeof(id->fr));
2168 }
2169
2170 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2171                 struct nvme_id_ctrl *id)
2172 {
2173         size_t nqnlen;
2174         int off;
2175
2176         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2177                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2178                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2179                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2180                         return;
2181                 }
2182
2183                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2184                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2185         }
2186
2187         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2188         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2189                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2190                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2191         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2192         off += sizeof(id->sn);
2193         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2194         off += sizeof(id->mn);
2195         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2196 }
2197
2198 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2199 {
2200         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2201         kfree(subsys);
2202 }
2203
2204 static void nvme_release_subsystem(struct device *dev)
2205 {
2206         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2207 }
2208
2209 static void nvme_destroy_subsystem(struct kref *ref)
2210 {
2211         struct nvme_subsystem *subsys =
2212                         container_of(ref, struct nvme_subsystem, ref);
2213
2214         mutex_lock(&nvme_subsystems_lock);
2215         list_del(&subsys->entry);
2216         mutex_unlock(&nvme_subsystems_lock);
2217
2218         ida_destroy(&subsys->ns_ida);
2219         device_del(&subsys->dev);
2220         put_device(&subsys->dev);
2221 }
2222
2223 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2224 {
2225         kref_put(&subsys->ref, nvme_destroy_subsystem);
2226 }
2227
2228 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2229 {
2230         struct nvme_subsystem *subsys;
2231
2232         lockdep_assert_held(&nvme_subsystems_lock);
2233
2234         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2235                 if (strcmp(subsys->subnqn, subsysnqn))
2236                         continue;
2237                 if (!kref_get_unless_zero(&subsys->ref))
2238                         continue;
2239                 return subsys;
2240         }
2241
2242         return NULL;
2243 }
2244
2245 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2246         struct device_attribute subsys_attr_##_name = \
2247                 __ATTR(_name, _mode, _show, NULL)
2248
2249 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2250                                     struct device_attribute *attr,
2251                                     char *buf)
2252 {
2253         struct nvme_subsystem *subsys =
2254                 container_of(dev, struct nvme_subsystem, dev);
2255
2256         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2257 }
2258 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2259
2260 #define nvme_subsys_show_str_function(field)                            \
2261 static ssize_t subsys_##field##_show(struct device *dev,                \
2262                             struct device_attribute *attr, char *buf)   \
2263 {                                                                       \
2264         struct nvme_subsystem *subsys =                                 \
2265                 container_of(dev, struct nvme_subsystem, dev);          \
2266         return sprintf(buf, "%.*s\n",                                   \
2267                        (int)sizeof(subsys->field), subsys->field);      \
2268 }                                                                       \
2269 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2270
2271 nvme_subsys_show_str_function(model);
2272 nvme_subsys_show_str_function(serial);
2273 nvme_subsys_show_str_function(firmware_rev);
2274
2275 static struct attribute *nvme_subsys_attrs[] = {
2276         &subsys_attr_model.attr,
2277         &subsys_attr_serial.attr,
2278         &subsys_attr_firmware_rev.attr,
2279         &subsys_attr_subsysnqn.attr,
2280         NULL,
2281 };
2282
2283 static struct attribute_group nvme_subsys_attrs_group = {
2284         .attrs = nvme_subsys_attrs,
2285 };
2286
2287 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2288         &nvme_subsys_attrs_group,
2289         NULL,
2290 };
2291
2292 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2293 {
2294         int count = 0;
2295         struct nvme_ctrl *ctrl;
2296
2297         mutex_lock(&subsys->lock);
2298         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2299                 if (ctrl->state != NVME_CTRL_DELETING &&
2300                     ctrl->state != NVME_CTRL_DEAD)
2301                         count++;
2302         }
2303         mutex_unlock(&subsys->lock);
2304
2305         return count;
2306 }
2307
2308 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2309 {
2310         struct nvme_subsystem *subsys, *found;
2311         int ret;
2312
2313         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2314         if (!subsys)
2315                 return -ENOMEM;
2316         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2317         if (ret < 0) {
2318                 kfree(subsys);
2319                 return ret;
2320         }
2321         subsys->instance = ret;
2322         mutex_init(&subsys->lock);
2323         kref_init(&subsys->ref);
2324         INIT_LIST_HEAD(&subsys->ctrls);
2325         INIT_LIST_HEAD(&subsys->nsheads);
2326         nvme_init_subnqn(subsys, ctrl, id);
2327         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2328         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2329         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2330         subsys->vendor_id = le16_to_cpu(id->vid);
2331         subsys->cmic = id->cmic;
2332
2333         subsys->dev.class = nvme_subsys_class;
2334         subsys->dev.release = nvme_release_subsystem;
2335         subsys->dev.groups = nvme_subsys_attrs_groups;
2336         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2337         device_initialize(&subsys->dev);
2338
2339         mutex_lock(&nvme_subsystems_lock);
2340         found = __nvme_find_get_subsystem(subsys->subnqn);
2341         if (found) {
2342                 /*
2343                  * Verify that the subsystem actually supports multiple
2344                  * controllers, else bail out.
2345                  */
2346                 if (!(ctrl->opts && ctrl->opts->discovery_nqn) &&
2347                     nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2348                         dev_err(ctrl->device,
2349                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2350                                 found->subnqn);
2351                         nvme_put_subsystem(found);
2352                         ret = -EINVAL;
2353                         goto out_unlock;
2354                 }
2355
2356                 __nvme_release_subsystem(subsys);
2357                 subsys = found;
2358         } else {
2359                 ret = device_add(&subsys->dev);
2360                 if (ret) {
2361                         dev_err(ctrl->device,
2362                                 "failed to register subsystem device.\n");
2363                         goto out_unlock;
2364                 }
2365                 ida_init(&subsys->ns_ida);
2366                 list_add_tail(&subsys->entry, &nvme_subsystems);
2367         }
2368
2369         ctrl->subsys = subsys;
2370         mutex_unlock(&nvme_subsystems_lock);
2371
2372         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2373                         dev_name(ctrl->device))) {
2374                 dev_err(ctrl->device,
2375                         "failed to create sysfs link from subsystem.\n");
2376                 /* the transport driver will eventually put the subsystem */
2377                 return -EINVAL;
2378         }
2379
2380         mutex_lock(&subsys->lock);
2381         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2382         mutex_unlock(&subsys->lock);
2383
2384         return 0;
2385
2386 out_unlock:
2387         mutex_unlock(&nvme_subsystems_lock);
2388         put_device(&subsys->dev);
2389         return ret;
2390 }
2391
2392 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2393                 void *log, size_t size, u64 offset)
2394 {
2395         struct nvme_command c = { };
2396         unsigned long dwlen = size / 4 - 1;
2397
2398         c.get_log_page.opcode = nvme_admin_get_log_page;
2399         c.get_log_page.nsid = cpu_to_le32(nsid);
2400         c.get_log_page.lid = log_page;
2401         c.get_log_page.lsp = lsp;
2402         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2403         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2404         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2405         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2406
2407         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2408 }
2409
2410 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2411 {
2412         int ret;
2413
2414         if (!ctrl->effects)
2415                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2416
2417         if (!ctrl->effects)
2418                 return 0;
2419
2420         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2421                         ctrl->effects, sizeof(*ctrl->effects), 0);
2422         if (ret) {
2423                 kfree(ctrl->effects);
2424                 ctrl->effects = NULL;
2425         }
2426         return ret;
2427 }
2428
2429 /*
2430  * Initialize the cached copies of the Identify data and various controller
2431  * register in our nvme_ctrl structure.  This should be called as soon as
2432  * the admin queue is fully up and running.
2433  */
2434 int nvme_init_identify(struct nvme_ctrl *ctrl)
2435 {
2436         struct nvme_id_ctrl *id;
2437         u64 cap;
2438         int ret, page_shift;
2439         u32 max_hw_sectors;
2440         bool prev_apst_enabled;
2441
2442         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2443         if (ret) {
2444                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2445                 return ret;
2446         }
2447
2448         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2449         if (ret) {
2450                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2451                 return ret;
2452         }
2453         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2454
2455         if (ctrl->vs >= NVME_VS(1, 1, 0))
2456                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2457
2458         ret = nvme_identify_ctrl(ctrl, &id);
2459         if (ret) {
2460                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2461                 return -EIO;
2462         }
2463
2464         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2465                 ret = nvme_get_effects_log(ctrl);
2466                 if (ret < 0)
2467                         goto out_free;
2468         }
2469
2470         if (!ctrl->identified) {
2471                 int i;
2472
2473                 ret = nvme_init_subsystem(ctrl, id);
2474                 if (ret)
2475                         goto out_free;
2476
2477                 /*
2478                  * Check for quirks.  Quirk can depend on firmware version,
2479                  * so, in principle, the set of quirks present can change
2480                  * across a reset.  As a possible future enhancement, we
2481                  * could re-scan for quirks every time we reinitialize
2482                  * the device, but we'd have to make sure that the driver
2483                  * behaves intelligently if the quirks change.
2484                  */
2485                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2486                         if (quirk_matches(id, &core_quirks[i]))
2487                                 ctrl->quirks |= core_quirks[i].quirks;
2488                 }
2489         }
2490
2491         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2492                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2493                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2494         }
2495
2496         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2497         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2498         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2499
2500         ctrl->oacs = le16_to_cpu(id->oacs);
2501         ctrl->oncs = le16_to_cpup(&id->oncs);
2502         ctrl->oaes = le32_to_cpu(id->oaes);
2503         atomic_set(&ctrl->abort_limit, id->acl + 1);
2504         ctrl->vwc = id->vwc;
2505         if (id->mdts)
2506                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2507         else
2508                 max_hw_sectors = UINT_MAX;
2509         ctrl->max_hw_sectors =
2510                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2511
2512         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2513         ctrl->sgls = le32_to_cpu(id->sgls);
2514         ctrl->kas = le16_to_cpu(id->kas);
2515         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2516         ctrl->ctratt = le32_to_cpu(id->ctratt);
2517
2518         if (id->rtd3e) {
2519                 /* us -> s */
2520                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2521
2522                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2523                                                  shutdown_timeout, 60);
2524
2525                 if (ctrl->shutdown_timeout != shutdown_timeout)
2526                         dev_info(ctrl->device,
2527                                  "Shutdown timeout set to %u seconds\n",
2528                                  ctrl->shutdown_timeout);
2529         } else
2530                 ctrl->shutdown_timeout = shutdown_timeout;
2531
2532         ctrl->npss = id->npss;
2533         ctrl->apsta = id->apsta;
2534         prev_apst_enabled = ctrl->apst_enabled;
2535         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2536                 if (force_apst && id->apsta) {
2537                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2538                         ctrl->apst_enabled = true;
2539                 } else {
2540                         ctrl->apst_enabled = false;
2541                 }
2542         } else {
2543                 ctrl->apst_enabled = id->apsta;
2544         }
2545         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2546
2547         if (ctrl->ops->flags & NVME_F_FABRICS) {
2548                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2549                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2550                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2551                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2552
2553                 /*
2554                  * In fabrics we need to verify the cntlid matches the
2555                  * admin connect
2556                  */
2557                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2558                         ret = -EINVAL;
2559                         goto out_free;
2560                 }
2561
2562                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2563                         dev_err(ctrl->device,
2564                                 "keep-alive support is mandatory for fabrics\n");
2565                         ret = -EINVAL;
2566                         goto out_free;
2567                 }
2568         } else {
2569                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2570                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2571                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2572                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2573                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2574         }
2575
2576         ret = nvme_mpath_init(ctrl, id);
2577         kfree(id);
2578
2579         if (ret < 0)
2580                 return ret;
2581
2582         if (ctrl->apst_enabled && !prev_apst_enabled)
2583                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2584         else if (!ctrl->apst_enabled && prev_apst_enabled)
2585                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2586
2587         ret = nvme_configure_apst(ctrl);
2588         if (ret < 0)
2589                 return ret;
2590
2591         ret = nvme_configure_timestamp(ctrl);
2592         if (ret < 0)
2593                 return ret;
2594
2595         ret = nvme_configure_directives(ctrl);
2596         if (ret < 0)
2597                 return ret;
2598
2599         ret = nvme_configure_acre(ctrl);
2600         if (ret < 0)
2601                 return ret;
2602
2603         ctrl->identified = true;
2604
2605         return 0;
2606
2607 out_free:
2608         kfree(id);
2609         return ret;
2610 }
2611 EXPORT_SYMBOL_GPL(nvme_init_identify);
2612
2613 static int nvme_dev_open(struct inode *inode, struct file *file)
2614 {
2615         struct nvme_ctrl *ctrl =
2616                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2617
2618         switch (ctrl->state) {
2619         case NVME_CTRL_LIVE:
2620         case NVME_CTRL_ADMIN_ONLY:
2621                 break;
2622         default:
2623                 return -EWOULDBLOCK;
2624         }
2625
2626         file->private_data = ctrl;
2627         return 0;
2628 }
2629
2630 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2631 {
2632         struct nvme_ns *ns;
2633         int ret;
2634
2635         down_read(&ctrl->namespaces_rwsem);
2636         if (list_empty(&ctrl->namespaces)) {
2637                 ret = -ENOTTY;
2638                 goto out_unlock;
2639         }
2640
2641         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2642         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2643                 dev_warn(ctrl->device,
2644                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2645                 ret = -EINVAL;
2646                 goto out_unlock;
2647         }
2648
2649         dev_warn(ctrl->device,
2650                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2651         kref_get(&ns->kref);
2652         up_read(&ctrl->namespaces_rwsem);
2653
2654         ret = nvme_user_cmd(ctrl, ns, argp);
2655         nvme_put_ns(ns);
2656         return ret;
2657
2658 out_unlock:
2659         up_read(&ctrl->namespaces_rwsem);
2660         return ret;
2661 }
2662
2663 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2664                 unsigned long arg)
2665 {
2666         struct nvme_ctrl *ctrl = file->private_data;
2667         void __user *argp = (void __user *)arg;
2668
2669         switch (cmd) {
2670         case NVME_IOCTL_ADMIN_CMD:
2671                 return nvme_user_cmd(ctrl, NULL, argp);
2672         case NVME_IOCTL_IO_CMD:
2673                 return nvme_dev_user_cmd(ctrl, argp);
2674         case NVME_IOCTL_RESET:
2675                 dev_warn(ctrl->device, "resetting controller\n");
2676                 return nvme_reset_ctrl_sync(ctrl);
2677         case NVME_IOCTL_SUBSYS_RESET:
2678                 return nvme_reset_subsystem(ctrl);
2679         case NVME_IOCTL_RESCAN:
2680                 nvme_queue_scan(ctrl);
2681                 return 0;
2682         default:
2683                 return -ENOTTY;
2684         }
2685 }
2686
2687 static const struct file_operations nvme_dev_fops = {
2688         .owner          = THIS_MODULE,
2689         .open           = nvme_dev_open,
2690         .unlocked_ioctl = nvme_dev_ioctl,
2691         .compat_ioctl   = nvme_dev_ioctl,
2692 };
2693
2694 static ssize_t nvme_sysfs_reset(struct device *dev,
2695                                 struct device_attribute *attr, const char *buf,
2696                                 size_t count)
2697 {
2698         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2699         int ret;
2700
2701         ret = nvme_reset_ctrl_sync(ctrl);
2702         if (ret < 0)
2703                 return ret;
2704         return count;
2705 }
2706 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2707
2708 static ssize_t nvme_sysfs_rescan(struct device *dev,
2709                                 struct device_attribute *attr, const char *buf,
2710                                 size_t count)
2711 {
2712         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2713
2714         nvme_queue_scan(ctrl);
2715         return count;
2716 }
2717 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2718
2719 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2720 {
2721         struct gendisk *disk = dev_to_disk(dev);
2722
2723         if (disk->fops == &nvme_fops)
2724                 return nvme_get_ns_from_dev(dev)->head;
2725         else
2726                 return disk->private_data;
2727 }
2728
2729 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2730                 char *buf)
2731 {
2732         struct nvme_ns_head *head = dev_to_ns_head(dev);
2733         struct nvme_ns_ids *ids = &head->ids;
2734         struct nvme_subsystem *subsys = head->subsys;
2735         int serial_len = sizeof(subsys->serial);
2736         int model_len = sizeof(subsys->model);
2737
2738         if (!uuid_is_null(&ids->uuid))
2739                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2740
2741         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2742                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2743
2744         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2745                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2746
2747         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2748                                   subsys->serial[serial_len - 1] == '\0'))
2749                 serial_len--;
2750         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2751                                  subsys->model[model_len - 1] == '\0'))
2752                 model_len--;
2753
2754         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2755                 serial_len, subsys->serial, model_len, subsys->model,
2756                 head->ns_id);
2757 }
2758 static DEVICE_ATTR_RO(wwid);
2759
2760 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2761                 char *buf)
2762 {
2763         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2764 }
2765 static DEVICE_ATTR_RO(nguid);
2766
2767 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2768                 char *buf)
2769 {
2770         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2771
2772         /* For backward compatibility expose the NGUID to userspace if
2773          * we have no UUID set
2774          */
2775         if (uuid_is_null(&ids->uuid)) {
2776                 printk_ratelimited(KERN_WARNING
2777                                    "No UUID available providing old NGUID\n");
2778                 return sprintf(buf, "%pU\n", ids->nguid);
2779         }
2780         return sprintf(buf, "%pU\n", &ids->uuid);
2781 }
2782 static DEVICE_ATTR_RO(uuid);
2783
2784 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2785                 char *buf)
2786 {
2787         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2788 }
2789 static DEVICE_ATTR_RO(eui);
2790
2791 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2792                 char *buf)
2793 {
2794         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2795 }
2796 static DEVICE_ATTR_RO(nsid);
2797
2798 static struct attribute *nvme_ns_id_attrs[] = {
2799         &dev_attr_wwid.attr,
2800         &dev_attr_uuid.attr,
2801         &dev_attr_nguid.attr,
2802         &dev_attr_eui.attr,
2803         &dev_attr_nsid.attr,
2804 #ifdef CONFIG_NVME_MULTIPATH
2805         &dev_attr_ana_grpid.attr,
2806         &dev_attr_ana_state.attr,
2807 #endif
2808         NULL,
2809 };
2810
2811 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2812                 struct attribute *a, int n)
2813 {
2814         struct device *dev = container_of(kobj, struct device, kobj);
2815         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2816
2817         if (a == &dev_attr_uuid.attr) {
2818                 if (uuid_is_null(&ids->uuid) &&
2819                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2820                         return 0;
2821         }
2822         if (a == &dev_attr_nguid.attr) {
2823                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2824                         return 0;
2825         }
2826         if (a == &dev_attr_eui.attr) {
2827                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2828                         return 0;
2829         }
2830 #ifdef CONFIG_NVME_MULTIPATH
2831         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2832                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2833                         return 0;
2834                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2835                         return 0;
2836         }
2837 #endif
2838         return a->mode;
2839 }
2840
2841 static const struct attribute_group nvme_ns_id_attr_group = {
2842         .attrs          = nvme_ns_id_attrs,
2843         .is_visible     = nvme_ns_id_attrs_are_visible,
2844 };
2845
2846 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2847         &nvme_ns_id_attr_group,
2848 #ifdef CONFIG_NVM
2849         &nvme_nvm_attr_group,
2850 #endif
2851         NULL,
2852 };
2853
2854 #define nvme_show_str_function(field)                                           \
2855 static ssize_t  field##_show(struct device *dev,                                \
2856                             struct device_attribute *attr, char *buf)           \
2857 {                                                                               \
2858         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2859         return sprintf(buf, "%.*s\n",                                           \
2860                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2861 }                                                                               \
2862 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2863
2864 nvme_show_str_function(model);
2865 nvme_show_str_function(serial);
2866 nvme_show_str_function(firmware_rev);
2867
2868 #define nvme_show_int_function(field)                                           \
2869 static ssize_t  field##_show(struct device *dev,                                \
2870                             struct device_attribute *attr, char *buf)           \
2871 {                                                                               \
2872         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2873         return sprintf(buf, "%d\n", ctrl->field);       \
2874 }                                                                               \
2875 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2876
2877 nvme_show_int_function(cntlid);
2878 nvme_show_int_function(numa_node);
2879
2880 static ssize_t nvme_sysfs_delete(struct device *dev,
2881                                 struct device_attribute *attr, const char *buf,
2882                                 size_t count)
2883 {
2884         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2885
2886         if (device_remove_file_self(dev, attr))
2887                 nvme_delete_ctrl_sync(ctrl);
2888         return count;
2889 }
2890 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2891
2892 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2893                                          struct device_attribute *attr,
2894                                          char *buf)
2895 {
2896         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2897
2898         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2899 }
2900 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2901
2902 static ssize_t nvme_sysfs_show_state(struct device *dev,
2903                                      struct device_attribute *attr,
2904                                      char *buf)
2905 {
2906         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2907         static const char *const state_name[] = {
2908                 [NVME_CTRL_NEW]         = "new",
2909                 [NVME_CTRL_LIVE]        = "live",
2910                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2911                 [NVME_CTRL_RESETTING]   = "resetting",
2912                 [NVME_CTRL_CONNECTING]  = "connecting",
2913                 [NVME_CTRL_DELETING]    = "deleting",
2914                 [NVME_CTRL_DEAD]        = "dead",
2915         };
2916
2917         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2918             state_name[ctrl->state])
2919                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2920
2921         return sprintf(buf, "unknown state\n");
2922 }
2923
2924 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2925
2926 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2927                                          struct device_attribute *attr,
2928                                          char *buf)
2929 {
2930         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2931
2932         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2933 }
2934 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2935
2936 static ssize_t nvme_sysfs_show_address(struct device *dev,
2937                                          struct device_attribute *attr,
2938                                          char *buf)
2939 {
2940         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2941
2942         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2943 }
2944 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2945
2946 static struct attribute *nvme_dev_attrs[] = {
2947         &dev_attr_reset_controller.attr,
2948         &dev_attr_rescan_controller.attr,
2949         &dev_attr_model.attr,
2950         &dev_attr_serial.attr,
2951         &dev_attr_firmware_rev.attr,
2952         &dev_attr_cntlid.attr,
2953         &dev_attr_delete_controller.attr,
2954         &dev_attr_transport.attr,
2955         &dev_attr_subsysnqn.attr,
2956         &dev_attr_address.attr,
2957         &dev_attr_state.attr,
2958         &dev_attr_numa_node.attr,
2959         NULL
2960 };
2961
2962 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2963                 struct attribute *a, int n)
2964 {
2965         struct device *dev = container_of(kobj, struct device, kobj);
2966         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2967
2968         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2969                 return 0;
2970         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2971                 return 0;
2972
2973         return a->mode;
2974 }
2975
2976 static struct attribute_group nvme_dev_attrs_group = {
2977         .attrs          = nvme_dev_attrs,
2978         .is_visible     = nvme_dev_attrs_are_visible,
2979 };
2980
2981 static const struct attribute_group *nvme_dev_attr_groups[] = {
2982         &nvme_dev_attrs_group,
2983         NULL,
2984 };
2985
2986 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2987                 unsigned nsid)
2988 {
2989         struct nvme_ns_head *h;
2990
2991         lockdep_assert_held(&subsys->lock);
2992
2993         list_for_each_entry(h, &subsys->nsheads, entry) {
2994                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2995                         return h;
2996         }
2997
2998         return NULL;
2999 }
3000
3001 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3002                 struct nvme_ns_head *new)
3003 {
3004         struct nvme_ns_head *h;
3005
3006         lockdep_assert_held(&subsys->lock);
3007
3008         list_for_each_entry(h, &subsys->nsheads, entry) {
3009                 if (nvme_ns_ids_valid(&new->ids) &&
3010                     !list_empty(&h->list) &&
3011                     nvme_ns_ids_equal(&new->ids, &h->ids))
3012                         return -EINVAL;
3013         }
3014
3015         return 0;
3016 }
3017
3018 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3019                 unsigned nsid, struct nvme_id_ns *id)
3020 {
3021         struct nvme_ns_head *head;
3022         size_t size = sizeof(*head);
3023         int ret = -ENOMEM;
3024
3025 #ifdef CONFIG_NVME_MULTIPATH
3026         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3027 #endif
3028
3029         head = kzalloc(size, GFP_KERNEL);
3030         if (!head)
3031                 goto out;
3032         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3033         if (ret < 0)
3034                 goto out_free_head;
3035         head->instance = ret;
3036         INIT_LIST_HEAD(&head->list);
3037         ret = init_srcu_struct(&head->srcu);
3038         if (ret)
3039                 goto out_ida_remove;
3040         head->subsys = ctrl->subsys;
3041         head->ns_id = nsid;
3042         kref_init(&head->ref);
3043
3044         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3045
3046         ret = __nvme_check_ids(ctrl->subsys, head);
3047         if (ret) {
3048                 dev_err(ctrl->device,
3049                         "duplicate IDs for nsid %d\n", nsid);
3050                 goto out_cleanup_srcu;
3051         }
3052
3053         ret = nvme_mpath_alloc_disk(ctrl, head);
3054         if (ret)
3055                 goto out_cleanup_srcu;
3056
3057         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3058
3059         kref_get(&ctrl->subsys->ref);
3060
3061         return head;
3062 out_cleanup_srcu:
3063         cleanup_srcu_struct(&head->srcu);
3064 out_ida_remove:
3065         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3066 out_free_head:
3067         kfree(head);
3068 out:
3069         return ERR_PTR(ret);
3070 }
3071
3072 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3073                 struct nvme_id_ns *id)
3074 {
3075         struct nvme_ctrl *ctrl = ns->ctrl;
3076         bool is_shared = id->nmic & (1 << 0);
3077         struct nvme_ns_head *head = NULL;
3078         int ret = 0;
3079
3080         mutex_lock(&ctrl->subsys->lock);
3081         if (is_shared)
3082                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3083         if (!head) {
3084                 head = nvme_alloc_ns_head(ctrl, nsid, id);
3085                 if (IS_ERR(head)) {
3086                         ret = PTR_ERR(head);
3087                         goto out_unlock;
3088                 }
3089         } else {
3090                 struct nvme_ns_ids ids;
3091
3092                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3093                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3094                         dev_err(ctrl->device,
3095                                 "IDs don't match for shared namespace %d\n",
3096                                         nsid);
3097                         ret = -EINVAL;
3098                         goto out_unlock;
3099                 }
3100         }
3101
3102         list_add_tail(&ns->siblings, &head->list);
3103         ns->head = head;
3104
3105 out_unlock:
3106         mutex_unlock(&ctrl->subsys->lock);
3107         return ret;
3108 }
3109
3110 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3111 {
3112         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3113         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3114
3115         return nsa->head->ns_id - nsb->head->ns_id;
3116 }
3117
3118 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3119 {
3120         struct nvme_ns *ns, *ret = NULL;
3121
3122         down_read(&ctrl->namespaces_rwsem);
3123         list_for_each_entry(ns, &ctrl->namespaces, list) {
3124                 if (ns->head->ns_id == nsid) {
3125                         if (!kref_get_unless_zero(&ns->kref))
3126                                 continue;
3127                         ret = ns;
3128                         break;
3129                 }
3130                 if (ns->head->ns_id > nsid)
3131                         break;
3132         }
3133         up_read(&ctrl->namespaces_rwsem);
3134         return ret;
3135 }
3136
3137 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3138 {
3139         struct streams_directive_params s;
3140         int ret;
3141
3142         if (!ctrl->nr_streams)
3143                 return 0;
3144
3145         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3146         if (ret)
3147                 return ret;
3148
3149         ns->sws = le32_to_cpu(s.sws);
3150         ns->sgs = le16_to_cpu(s.sgs);
3151
3152         if (ns->sws) {
3153                 unsigned int bs = 1 << ns->lba_shift;
3154
3155                 blk_queue_io_min(ns->queue, bs * ns->sws);
3156                 if (ns->sgs)
3157                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3158         }
3159
3160         return 0;
3161 }
3162
3163 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3164 {
3165         struct nvme_ns *ns;
3166         struct gendisk *disk;
3167         struct nvme_id_ns *id;
3168         char disk_name[DISK_NAME_LEN];
3169         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT;
3170
3171         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3172         if (!ns)
3173                 return;
3174
3175         ns->queue = blk_mq_init_queue(ctrl->tagset);
3176         if (IS_ERR(ns->queue))
3177                 goto out_free_ns;
3178
3179         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3180         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3181                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3182
3183         ns->queue->queuedata = ns;
3184         ns->ctrl = ctrl;
3185
3186         kref_init(&ns->kref);
3187         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3188
3189         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3190         nvme_set_queue_limits(ctrl, ns->queue);
3191
3192         id = nvme_identify_ns(ctrl, nsid);
3193         if (!id)
3194                 goto out_free_queue;
3195
3196         if (id->ncap == 0)
3197                 goto out_free_id;
3198
3199         if (nvme_init_ns_head(ns, nsid, id))
3200                 goto out_free_id;
3201         nvme_setup_streams_ns(ctrl, ns);
3202         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3203
3204         disk = alloc_disk_node(0, node);
3205         if (!disk)
3206                 goto out_unlink_ns;
3207
3208         disk->fops = &nvme_fops;
3209         disk->private_data = ns;
3210         disk->queue = ns->queue;
3211         disk->flags = flags;
3212         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3213         ns->disk = disk;
3214
3215         __nvme_revalidate_disk(disk, id);
3216
3217         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3218                 if (nvme_nvm_register(ns, disk_name, node)) {
3219                         dev_warn(ctrl->device, "LightNVM init failure\n");
3220                         goto out_put_disk;
3221                 }
3222         }
3223
3224         down_write(&ctrl->namespaces_rwsem);
3225         list_add_tail(&ns->list, &ctrl->namespaces);
3226         up_write(&ctrl->namespaces_rwsem);
3227
3228         nvme_get_ctrl(ctrl);
3229
3230         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3231
3232         nvme_mpath_add_disk(ns, id);
3233         nvme_fault_inject_init(ns);
3234         kfree(id);
3235
3236         return;
3237  out_put_disk:
3238         put_disk(ns->disk);
3239  out_unlink_ns:
3240         mutex_lock(&ctrl->subsys->lock);
3241         list_del_rcu(&ns->siblings);
3242         mutex_unlock(&ctrl->subsys->lock);
3243  out_free_id:
3244         kfree(id);
3245  out_free_queue:
3246         blk_cleanup_queue(ns->queue);
3247  out_free_ns:
3248         kfree(ns);
3249 }
3250
3251 static void nvme_ns_remove(struct nvme_ns *ns)
3252 {
3253         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3254                 return;
3255
3256         nvme_fault_inject_fini(ns);
3257         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3258                 del_gendisk(ns->disk);
3259                 blk_cleanup_queue(ns->queue);
3260                 if (blk_get_integrity(ns->disk))
3261                         blk_integrity_unregister(ns->disk);
3262         }
3263
3264         mutex_lock(&ns->ctrl->subsys->lock);
3265         list_del_rcu(&ns->siblings);
3266         nvme_mpath_clear_current_path(ns);
3267         mutex_unlock(&ns->ctrl->subsys->lock);
3268
3269         down_write(&ns->ctrl->namespaces_rwsem);
3270         list_del_init(&ns->list);
3271         up_write(&ns->ctrl->namespaces_rwsem);
3272
3273         synchronize_srcu(&ns->head->srcu);
3274         nvme_mpath_check_last_path(ns);
3275         nvme_put_ns(ns);
3276 }
3277
3278 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3279 {
3280         struct nvme_ns *ns;
3281
3282         ns = nvme_find_get_ns(ctrl, nsid);
3283         if (ns) {
3284                 if (ns->disk && revalidate_disk(ns->disk))
3285                         nvme_ns_remove(ns);
3286                 nvme_put_ns(ns);
3287         } else
3288                 nvme_alloc_ns(ctrl, nsid);
3289 }
3290
3291 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3292                                         unsigned nsid)
3293 {
3294         struct nvme_ns *ns, *next;
3295         LIST_HEAD(rm_list);
3296
3297         down_write(&ctrl->namespaces_rwsem);
3298         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3299                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3300                         list_move_tail(&ns->list, &rm_list);
3301         }
3302         up_write(&ctrl->namespaces_rwsem);
3303
3304         list_for_each_entry_safe(ns, next, &rm_list, list)
3305                 nvme_ns_remove(ns);
3306
3307 }
3308
3309 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3310 {
3311         struct nvme_ns *ns;
3312         __le32 *ns_list;
3313         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3314         int ret = 0;
3315
3316         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3317         if (!ns_list)
3318                 return -ENOMEM;
3319
3320         for (i = 0; i < num_lists; i++) {
3321                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3322                 if (ret)
3323                         goto free;
3324
3325                 for (j = 0; j < min(nn, 1024U); j++) {
3326                         nsid = le32_to_cpu(ns_list[j]);
3327                         if (!nsid)
3328                                 goto out;
3329
3330                         nvme_validate_ns(ctrl, nsid);
3331
3332                         while (++prev < nsid) {
3333                                 ns = nvme_find_get_ns(ctrl, prev);
3334                                 if (ns) {
3335                                         nvme_ns_remove(ns);
3336                                         nvme_put_ns(ns);
3337                                 }
3338                         }
3339                 }
3340                 nn -= j;
3341         }
3342  out:
3343         nvme_remove_invalid_namespaces(ctrl, prev);
3344  free:
3345         kfree(ns_list);
3346         return ret;
3347 }
3348
3349 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3350 {
3351         unsigned i;
3352
3353         for (i = 1; i <= nn; i++)
3354                 nvme_validate_ns(ctrl, i);
3355
3356         nvme_remove_invalid_namespaces(ctrl, nn);
3357 }
3358
3359 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3360 {
3361         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3362         __le32 *log;
3363         int error;
3364
3365         log = kzalloc(log_size, GFP_KERNEL);
3366         if (!log)
3367                 return;
3368
3369         /*
3370          * We need to read the log to clear the AEN, but we don't want to rely
3371          * on it for the changed namespace information as userspace could have
3372          * raced with us in reading the log page, which could cause us to miss
3373          * updates.
3374          */
3375         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3376                         log_size, 0);
3377         if (error)
3378                 dev_warn(ctrl->device,
3379                         "reading changed ns log failed: %d\n", error);
3380
3381         kfree(log);
3382 }
3383
3384 static void nvme_scan_work(struct work_struct *work)
3385 {
3386         struct nvme_ctrl *ctrl =
3387                 container_of(work, struct nvme_ctrl, scan_work);
3388         struct nvme_id_ctrl *id;
3389         unsigned nn;
3390
3391         if (ctrl->state != NVME_CTRL_LIVE)
3392                 return;
3393
3394         WARN_ON_ONCE(!ctrl->tagset);
3395
3396         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3397                 dev_info(ctrl->device, "rescanning namespaces.\n");
3398                 nvme_clear_changed_ns_log(ctrl);
3399         }
3400
3401         if (nvme_identify_ctrl(ctrl, &id))
3402                 return;
3403
3404         nn = le32_to_cpu(id->nn);
3405         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3406             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3407                 if (!nvme_scan_ns_list(ctrl, nn))
3408                         goto out_free_id;
3409         }
3410         nvme_scan_ns_sequential(ctrl, nn);
3411 out_free_id:
3412         kfree(id);
3413         down_write(&ctrl->namespaces_rwsem);
3414         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3415         up_write(&ctrl->namespaces_rwsem);
3416 }
3417
3418 /*
3419  * This function iterates the namespace list unlocked to allow recovery from
3420  * controller failure. It is up to the caller to ensure the namespace list is
3421  * not modified by scan work while this function is executing.
3422  */
3423 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3424 {
3425         struct nvme_ns *ns, *next;
3426         LIST_HEAD(ns_list);
3427
3428         /* prevent racing with ns scanning */
3429         flush_work(&ctrl->scan_work);
3430
3431         /*
3432          * The dead states indicates the controller was not gracefully
3433          * disconnected. In that case, we won't be able to flush any data while
3434          * removing the namespaces' disks; fail all the queues now to avoid
3435          * potentially having to clean up the failed sync later.
3436          */
3437         if (ctrl->state == NVME_CTRL_DEAD)
3438                 nvme_kill_queues(ctrl);
3439
3440         down_write(&ctrl->namespaces_rwsem);
3441         list_splice_init(&ctrl->namespaces, &ns_list);
3442         up_write(&ctrl->namespaces_rwsem);
3443
3444         list_for_each_entry_safe(ns, next, &ns_list, list)
3445                 nvme_ns_remove(ns);
3446 }
3447 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3448
3449 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3450 {
3451         char *envp[2] = { NULL, NULL };
3452         u32 aen_result = ctrl->aen_result;
3453
3454         ctrl->aen_result = 0;
3455         if (!aen_result)
3456                 return;
3457
3458         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3459         if (!envp[0])
3460                 return;
3461         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3462         kfree(envp[0]);
3463 }
3464
3465 static void nvme_async_event_work(struct work_struct *work)
3466 {
3467         struct nvme_ctrl *ctrl =
3468                 container_of(work, struct nvme_ctrl, async_event_work);
3469
3470         nvme_aen_uevent(ctrl);
3471         ctrl->ops->submit_async_event(ctrl);
3472 }
3473
3474 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3475 {
3476
3477         u32 csts;
3478
3479         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3480                 return false;
3481
3482         if (csts == ~0)
3483                 return false;
3484
3485         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3486 }
3487
3488 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3489 {
3490         struct nvme_fw_slot_info_log *log;
3491
3492         log = kmalloc(sizeof(*log), GFP_KERNEL);
3493         if (!log)
3494                 return;
3495
3496         if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3497                         sizeof(*log), 0))
3498                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3499         kfree(log);
3500 }
3501
3502 static void nvme_fw_act_work(struct work_struct *work)
3503 {
3504         struct nvme_ctrl *ctrl = container_of(work,
3505                                 struct nvme_ctrl, fw_act_work);
3506         unsigned long fw_act_timeout;
3507
3508         if (ctrl->mtfa)
3509                 fw_act_timeout = jiffies +
3510                                 msecs_to_jiffies(ctrl->mtfa * 100);
3511         else
3512                 fw_act_timeout = jiffies +
3513                                 msecs_to_jiffies(admin_timeout * 1000);
3514
3515         nvme_stop_queues(ctrl);
3516         while (nvme_ctrl_pp_status(ctrl)) {
3517                 if (time_after(jiffies, fw_act_timeout)) {
3518                         dev_warn(ctrl->device,
3519                                 "Fw activation timeout, reset controller\n");
3520                         nvme_reset_ctrl(ctrl);
3521                         break;
3522                 }
3523                 msleep(100);
3524         }
3525
3526         if (ctrl->state != NVME_CTRL_LIVE)
3527                 return;
3528
3529         nvme_start_queues(ctrl);
3530         /* read FW slot information to clear the AER */
3531         nvme_get_fw_slot_info(ctrl);
3532 }
3533
3534 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3535 {
3536         u32 aer_notice_type = (result & 0xff00) >> 8;
3537
3538         switch (aer_notice_type) {
3539         case NVME_AER_NOTICE_NS_CHANGED:
3540                 trace_nvme_async_event(ctrl, aer_notice_type);
3541                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3542                 nvme_queue_scan(ctrl);
3543                 break;
3544         case NVME_AER_NOTICE_FW_ACT_STARTING:
3545                 trace_nvme_async_event(ctrl, aer_notice_type);
3546                 queue_work(nvme_wq, &ctrl->fw_act_work);
3547                 break;
3548 #ifdef CONFIG_NVME_MULTIPATH
3549         case NVME_AER_NOTICE_ANA:
3550                 trace_nvme_async_event(ctrl, aer_notice_type);
3551                 if (!ctrl->ana_log_buf)
3552                         break;
3553                 queue_work(nvme_wq, &ctrl->ana_work);
3554                 break;
3555 #endif
3556         default:
3557                 dev_warn(ctrl->device, "async event result %08x\n", result);
3558         }
3559 }
3560
3561 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3562                 volatile union nvme_result *res)
3563 {
3564         u32 result = le32_to_cpu(res->u32);
3565         u32 aer_type = result & 0x07;
3566
3567         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3568                 return;
3569
3570         switch (aer_type) {
3571         case NVME_AER_NOTICE:
3572                 nvme_handle_aen_notice(ctrl, result);
3573                 break;
3574         case NVME_AER_ERROR:
3575         case NVME_AER_SMART:
3576         case NVME_AER_CSS:
3577         case NVME_AER_VS:
3578                 trace_nvme_async_event(ctrl, aer_type);
3579                 ctrl->aen_result = result;
3580                 break;
3581         default:
3582                 break;
3583         }
3584         queue_work(nvme_wq, &ctrl->async_event_work);
3585 }
3586 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3587
3588 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3589 {
3590         nvme_mpath_stop(ctrl);
3591         nvme_stop_keep_alive(ctrl);
3592         flush_work(&ctrl->async_event_work);
3593         cancel_work_sync(&ctrl->fw_act_work);
3594         if (ctrl->ops->stop_ctrl)
3595                 ctrl->ops->stop_ctrl(ctrl);
3596 }
3597 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3598
3599 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3600 {
3601         if (ctrl->kato)
3602                 nvme_start_keep_alive(ctrl);
3603
3604         if (ctrl->queue_count > 1) {
3605                 nvme_queue_scan(ctrl);
3606                 nvme_enable_aen(ctrl);
3607                 queue_work(nvme_wq, &ctrl->async_event_work);
3608                 nvme_start_queues(ctrl);
3609         }
3610 }
3611 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3612
3613 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3614 {
3615         cdev_device_del(&ctrl->cdev, ctrl->device);
3616 }
3617 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3618
3619 static void nvme_free_ctrl(struct device *dev)
3620 {
3621         struct nvme_ctrl *ctrl =
3622                 container_of(dev, struct nvme_ctrl, ctrl_device);
3623         struct nvme_subsystem *subsys = ctrl->subsys;
3624
3625         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3626         kfree(ctrl->effects);
3627         nvme_mpath_uninit(ctrl);
3628         __free_page(ctrl->discard_page);
3629
3630         if (subsys) {
3631                 mutex_lock(&subsys->lock);
3632                 list_del(&ctrl->subsys_entry);
3633                 mutex_unlock(&subsys->lock);
3634                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3635         }
3636
3637         ctrl->ops->free_ctrl(ctrl);
3638
3639         if (subsys)
3640                 nvme_put_subsystem(subsys);
3641 }
3642
3643 /*
3644  * Initialize a NVMe controller structures.  This needs to be called during
3645  * earliest initialization so that we have the initialized structured around
3646  * during probing.
3647  */
3648 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3649                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3650 {
3651         int ret;
3652
3653         ctrl->state = NVME_CTRL_NEW;
3654         spin_lock_init(&ctrl->lock);
3655         INIT_LIST_HEAD(&ctrl->namespaces);
3656         init_rwsem(&ctrl->namespaces_rwsem);
3657         ctrl->dev = dev;
3658         ctrl->ops = ops;
3659         ctrl->quirks = quirks;
3660         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3661         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3662         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3663         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3664
3665         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3666         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3667         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3668
3669         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3670                         PAGE_SIZE);
3671         ctrl->discard_page = alloc_page(GFP_KERNEL);
3672         if (!ctrl->discard_page) {
3673                 ret = -ENOMEM;
3674                 goto out;
3675         }
3676
3677         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3678         if (ret < 0)
3679                 goto out;
3680         ctrl->instance = ret;
3681
3682         device_initialize(&ctrl->ctrl_device);
3683         ctrl->device = &ctrl->ctrl_device;
3684         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3685         ctrl->device->class = nvme_class;
3686         ctrl->device->parent = ctrl->dev;
3687         ctrl->device->groups = nvme_dev_attr_groups;
3688         ctrl->device->release = nvme_free_ctrl;
3689         dev_set_drvdata(ctrl->device, ctrl);
3690         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3691         if (ret)
3692                 goto out_release_instance;
3693
3694         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3695         ctrl->cdev.owner = ops->module;
3696         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3697         if (ret)
3698                 goto out_free_name;
3699
3700         /*
3701          * Initialize latency tolerance controls.  The sysfs files won't
3702          * be visible to userspace unless the device actually supports APST.
3703          */
3704         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3705         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3706                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3707
3708         return 0;
3709 out_free_name:
3710         kfree_const(ctrl->device->kobj.name);
3711 out_release_instance:
3712         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3713 out:
3714         if (ctrl->discard_page)
3715                 __free_page(ctrl->discard_page);
3716         return ret;
3717 }
3718 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3719
3720 /**
3721  * nvme_kill_queues(): Ends all namespace queues
3722  * @ctrl: the dead controller that needs to end
3723  *
3724  * Call this function when the driver determines it is unable to get the
3725  * controller in a state capable of servicing IO.
3726  */
3727 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3728 {
3729         struct nvme_ns *ns;
3730
3731         down_read(&ctrl->namespaces_rwsem);
3732
3733         /* Forcibly unquiesce queues to avoid blocking dispatch */
3734         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3735                 blk_mq_unquiesce_queue(ctrl->admin_q);
3736
3737         list_for_each_entry(ns, &ctrl->namespaces, list)
3738                 nvme_set_queue_dying(ns);
3739
3740         up_read(&ctrl->namespaces_rwsem);
3741 }
3742 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3743
3744 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3745 {
3746         struct nvme_ns *ns;
3747
3748         down_read(&ctrl->namespaces_rwsem);
3749         list_for_each_entry(ns, &ctrl->namespaces, list)
3750                 blk_mq_unfreeze_queue(ns->queue);
3751         up_read(&ctrl->namespaces_rwsem);
3752 }
3753 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3754
3755 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3756 {
3757         struct nvme_ns *ns;
3758
3759         down_read(&ctrl->namespaces_rwsem);
3760         list_for_each_entry(ns, &ctrl->namespaces, list) {
3761                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3762                 if (timeout <= 0)
3763                         break;
3764         }
3765         up_read(&ctrl->namespaces_rwsem);
3766 }
3767 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3768
3769 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3770 {
3771         struct nvme_ns *ns;
3772
3773         down_read(&ctrl->namespaces_rwsem);
3774         list_for_each_entry(ns, &ctrl->namespaces, list)
3775                 blk_mq_freeze_queue_wait(ns->queue);
3776         up_read(&ctrl->namespaces_rwsem);
3777 }
3778 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3779
3780 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3781 {
3782         struct nvme_ns *ns;
3783
3784         down_read(&ctrl->namespaces_rwsem);
3785         list_for_each_entry(ns, &ctrl->namespaces, list)
3786                 blk_freeze_queue_start(ns->queue);
3787         up_read(&ctrl->namespaces_rwsem);
3788 }
3789 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3790
3791 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3792 {
3793         struct nvme_ns *ns;
3794
3795         down_read(&ctrl->namespaces_rwsem);
3796         list_for_each_entry(ns, &ctrl->namespaces, list)
3797                 blk_mq_quiesce_queue(ns->queue);
3798         up_read(&ctrl->namespaces_rwsem);
3799 }
3800 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3801
3802 void nvme_start_queues(struct nvme_ctrl *ctrl)
3803 {
3804         struct nvme_ns *ns;
3805
3806         down_read(&ctrl->namespaces_rwsem);
3807         list_for_each_entry(ns, &ctrl->namespaces, list)
3808                 blk_mq_unquiesce_queue(ns->queue);
3809         up_read(&ctrl->namespaces_rwsem);
3810 }
3811 EXPORT_SYMBOL_GPL(nvme_start_queues);
3812
3813 int __init nvme_core_init(void)
3814 {
3815         int result = -ENOMEM;
3816
3817         nvme_wq = alloc_workqueue("nvme-wq",
3818                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3819         if (!nvme_wq)
3820                 goto out;
3821
3822         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3823                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3824         if (!nvme_reset_wq)
3825                 goto destroy_wq;
3826
3827         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3828                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3829         if (!nvme_delete_wq)
3830                 goto destroy_reset_wq;
3831
3832         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3833         if (result < 0)
3834                 goto destroy_delete_wq;
3835
3836         nvme_class = class_create(THIS_MODULE, "nvme");
3837         if (IS_ERR(nvme_class)) {
3838                 result = PTR_ERR(nvme_class);
3839                 goto unregister_chrdev;
3840         }
3841
3842         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3843         if (IS_ERR(nvme_subsys_class)) {
3844                 result = PTR_ERR(nvme_subsys_class);
3845                 goto destroy_class;
3846         }
3847         return 0;
3848
3849 destroy_class:
3850         class_destroy(nvme_class);
3851 unregister_chrdev:
3852         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3853 destroy_delete_wq:
3854         destroy_workqueue(nvme_delete_wq);
3855 destroy_reset_wq:
3856         destroy_workqueue(nvme_reset_wq);
3857 destroy_wq:
3858         destroy_workqueue(nvme_wq);
3859 out:
3860         return result;
3861 }
3862
3863 void __exit nvme_core_exit(void)
3864 {
3865         ida_destroy(&nvme_subsystems_ida);
3866         class_destroy(nvme_subsys_class);
3867         class_destroy(nvme_class);
3868         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3869         destroy_workqueue(nvme_delete_wq);
3870         destroy_workqueue(nvme_reset_wq);
3871         destroy_workqueue(nvme_wq);
3872 }
3873
3874 MODULE_LICENSE("GPL");
3875 MODULE_VERSION("1.0");
3876 module_init(nvme_core_init);
3877 module_exit(nvme_core_exit);