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 void nvme_ns_remove(struct nvme_ns *ns);
 101 static int nvme_revalidate_disk(struct gendisk *disk);
 102 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
 103 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
 104                                            unsigned nsid);
 105
 106 static void nvme_set_queue_dying(struct nvme_ns *ns)
 107 {
 108         /*
 109          * Revalidating a dead namespace sets capacity to 0. This will end
 110          * buffered writers dirtying pages that can't be synced.
 111          */
 112         if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 113                 return;
 114         blk_set_queue_dying(ns->queue);
 115         /* Forcibly unquiesce queues to avoid blocking dispatch */
 116         blk_mq_unquiesce_queue(ns->queue);
 117         /*
 118          * Revalidate after unblocking dispatchers that may be holding bd_butex
 119          */
 120         revalidate_disk(ns->disk);
 121 }
 122
 123 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 124 {
 125         /*
 126          * Only new queue scan work when admin and IO queues are both alive
 127          */
 128         if (ctrl->state == NVME_CTRL_LIVE)
 129                 queue_work(nvme_wq, &ctrl->scan_work);
 130 }
 131
 132 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 133 {
 134         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 135                 return -EBUSY;
 136         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 137                 return -EBUSY;
 138         return 0;
 139 }
 140 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 141
 142 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 143 {
 144         int ret;
 145
 146         ret = nvme_reset_ctrl(ctrl);
 147         if (!ret) {
 148                 flush_work(&ctrl->reset_work);
 149                 if (ctrl->state != NVME_CTRL_LIVE &&
 150                     ctrl->state != NVME_CTRL_ADMIN_ONLY)
 151                         ret = -ENETRESET;
 152         }
 153
 154         return ret;
 155 }
 156 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 157
 158 static void nvme_delete_ctrl_work(struct work_struct *work)
 159 {
 160         struct nvme_ctrl *ctrl =
 161                 container_of(work, struct nvme_ctrl, delete_work);
 162
 163         dev_info(ctrl->device,
 164                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 165
 166         flush_work(&ctrl->reset_work);
 167         nvme_stop_ctrl(ctrl);
 168         nvme_remove_namespaces(ctrl);
 169         ctrl->ops->delete_ctrl(ctrl);
 170         nvme_uninit_ctrl(ctrl);
 171         nvme_put_ctrl(ctrl);
 172 }
 173
 174 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 175 {
 176         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 177                 return -EBUSY;
 178         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 179                 return -EBUSY;
 180         return 0;
 181 }
 182 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 183
 184 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 185 {
 186         int ret = 0;
 187
 188         /*
 189          * Keep a reference until the work is flushed since ->delete_ctrl
 190          * can free the controller.
 191          */
 192         nvme_get_ctrl(ctrl);
 193         ret = nvme_delete_ctrl(ctrl);
 194         if (!ret)
 195                 flush_work(&ctrl->delete_work);
 196         nvme_put_ctrl(ctrl);
 197         return ret;
 198 }
 199 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
 200
 201 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 202 {
 203         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 204 }
 205
 206 static blk_status_t nvme_error_status(struct request *req)
 207 {
 208         switch (nvme_req(req)->status & 0x7ff) {
 209         case NVME_SC_SUCCESS:
 210                 return BLK_STS_OK;
 211         case NVME_SC_CAP_EXCEEDED:
 212                 return BLK_STS_NOSPC;
 213         case NVME_SC_LBA_RANGE:
 214                 return BLK_STS_TARGET;
 215         case NVME_SC_BAD_ATTRIBUTES:
 216         case NVME_SC_ONCS_NOT_SUPPORTED:
 217         case NVME_SC_INVALID_OPCODE:
 218         case NVME_SC_INVALID_FIELD:
 219         case NVME_SC_INVALID_NS:
 220                 return BLK_STS_NOTSUPP;
 221         case NVME_SC_WRITE_FAULT:
 222         case NVME_SC_READ_ERROR:
 223         case NVME_SC_UNWRITTEN_BLOCK:
 224         case NVME_SC_ACCESS_DENIED:
 225         case NVME_SC_READ_ONLY:
 226         case NVME_SC_COMPARE_FAILED:
 227                 return BLK_STS_MEDIUM;
 228         case NVME_SC_GUARD_CHECK:
 229         case NVME_SC_APPTAG_CHECK:
 230         case NVME_SC_REFTAG_CHECK:
 231         case NVME_SC_INVALID_PI:
 232                 return BLK_STS_PROTECTION;
 233         case NVME_SC_RESERVATION_CONFLICT:
 234                 return BLK_STS_NEXUS;
 235         default:
 236                 return BLK_STS_IOERR;
 237         }
 238 }
 239
 240 static inline bool nvme_req_needs_retry(struct request *req)
 241 {
 242         if (blk_noretry_request(req))
 243                 return false;
 244         if (nvme_req(req)->status & NVME_SC_DNR)
 245                 return false;
 246         if (nvme_req(req)->retries >= nvme_max_retries)
 247                 return false;
 248         return true;
 249 }
 250
 251 void nvme_complete_rq(struct request *req)
 252 {
 253         blk_status_t status = nvme_error_status(req);
 254
 255         trace_nvme_complete_rq(req);
 256
 257         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 258                 if ((req->cmd_flags & REQ_NVME_MPATH) && nvme_failover_req(req))
 259                         return;
 260
 261                 if (!blk_queue_dying(req->q)) {
 262                         nvme_req(req)->retries++;
 263                         blk_mq_requeue_request(req, true);
 264                         return;
 265                 }
 266         }
 267         blk_mq_end_request(req, status);
 268 }
 269 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 270
 271 void nvme_cancel_request(struct request *req, void *data, bool reserved)
 272 {
 273         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 274                                 "Cancelling I/O %d", req->tag);
 275
 276         nvme_req(req)->status = NVME_SC_ABORT_REQ;
 277         blk_mq_complete_request(req);
 278
 279 }
 280 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 281
 282 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 283                 enum nvme_ctrl_state new_state)
 284 {
 285         enum nvme_ctrl_state old_state;
 286         unsigned long flags;
 287         bool changed = false;
 288
 289         spin_lock_irqsave(&ctrl->lock, flags);
 290
 291         old_state = ctrl->state;
 292         switch (new_state) {
 293         case NVME_CTRL_ADMIN_ONLY:
 294                 switch (old_state) {
 295                 case NVME_CTRL_CONNECTING:
 296                         changed = true;
 297                         /* FALLTHRU */
 298                 default:
 299                         break;
 300                 }
 301                 break;
 302         case NVME_CTRL_LIVE:
 303                 switch (old_state) {
 304                 case NVME_CTRL_NEW:
 305                 case NVME_CTRL_RESETTING:
 306                 case NVME_CTRL_CONNECTING:
 307                         changed = true;
 308                         /* FALLTHRU */
 309                 default:
 310                         break;
 311                 }
 312                 break;
 313         case NVME_CTRL_RESETTING:
 314                 switch (old_state) {
 315                 case NVME_CTRL_NEW:
 316                 case NVME_CTRL_LIVE:
 317                 case NVME_CTRL_ADMIN_ONLY:
 318                         changed = true;
 319                         /* FALLTHRU */
 320                 default:
 321                         break;
 322                 }
 323                 break;
 324         case NVME_CTRL_CONNECTING:
 325                 switch (old_state) {
 326                 case NVME_CTRL_NEW:
 327                 case NVME_CTRL_RESETTING:
 328                         changed = true;
 329                         /* FALLTHRU */
 330                 default:
 331                         break;
 332                 }
 333                 break;
 334         case NVME_CTRL_DELETING:
 335                 switch (old_state) {
 336                 case NVME_CTRL_LIVE:
 337                 case NVME_CTRL_ADMIN_ONLY:
 338                 case NVME_CTRL_RESETTING:
 339                 case NVME_CTRL_CONNECTING:
 340                         changed = true;
 341                         /* FALLTHRU */
 342                 default:
 343                         break;
 344                 }
 345                 break;
 346         case NVME_CTRL_DEAD:
 347                 switch (old_state) {
 348                 case NVME_CTRL_DELETING:
 349                         changed = true;
 350                         /* FALLTHRU */
 351                 default:
 352                         break;
 353                 }
 354                 break;
 355         default:
 356                 break;
 357         }
 358
 359         if (changed)
 360                 ctrl->state = new_state;
 361
 362         spin_unlock_irqrestore(&ctrl->lock, flags);
 363         if (changed && ctrl->state == NVME_CTRL_LIVE)
 364                 nvme_kick_requeue_lists(ctrl);
 365         return changed;
 366 }
 367 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 368
 369 static void nvme_free_ns_head(struct kref *ref)
 370 {
 371         struct nvme_ns_head *head =
 372                 container_of(ref, struct nvme_ns_head, ref);
 373
 374         nvme_mpath_remove_disk(head);
 375         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 376         list_del_init(&head->entry);
 377         cleanup_srcu_struct_quiesced(&head->srcu);
 378         nvme_put_subsystem(head->subsys);
 379         kfree(head);
 380 }
 381
 382 static void nvme_put_ns_head(struct nvme_ns_head *head)
 383 {
 384         kref_put(&head->ref, nvme_free_ns_head);
 385 }
 386
 387 static void nvme_free_ns(struct kref *kref)
 388 {
 389         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 390
 391         if (ns->ndev)
 392                 nvme_nvm_unregister(ns);
 393
 394         put_disk(ns->disk);
 395         nvme_put_ns_head(ns->head);
 396         nvme_put_ctrl(ns->ctrl);
 397         kfree(ns);
 398 }
 399
 400 static void nvme_put_ns(struct nvme_ns *ns)
 401 {
 402         kref_put(&ns->kref, nvme_free_ns);
 403 }
 404
 405 static inline void nvme_clear_nvme_request(struct request *req)
 406 {
 407         if (!(req->rq_flags & RQF_DONTPREP)) {
 408                 nvme_req(req)->retries = 0;
 409                 nvme_req(req)->flags = 0;
 410                 req->rq_flags |= RQF_DONTPREP;
 411         }
 412 }
 413
 414 struct request *nvme_alloc_request(struct request_queue *q,
 415                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 416 {
 417         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 418         struct request *req;
 419
 420         if (qid == NVME_QID_ANY) {
 421                 req = blk_mq_alloc_request(q, op, flags);
 422         } else {
 423                 req = blk_mq_alloc_request_hctx(q, op, flags,
 424                                 qid ? qid - 1 : 0);
 425         }
 426         if (IS_ERR(req))
 427                 return req;
 428
 429         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 430         nvme_clear_nvme_request(req);
 431         nvme_req(req)->cmd = cmd;
 432
 433         return req;
 434 }
 435 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 436
 437 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 438 {
 439         struct nvme_command c;
 440
 441         memset(&c, 0, sizeof(c));
 442
 443         c.directive.opcode = nvme_admin_directive_send;
 444         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 445         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 446         c.directive.dtype = NVME_DIR_IDENTIFY;
 447         c.directive.tdtype = NVME_DIR_STREAMS;
 448         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 449
 450         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 451 }
 452
 453 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 454 {
 455         return nvme_toggle_streams(ctrl, false);
 456 }
 457
 458 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 459 {
 460         return nvme_toggle_streams(ctrl, true);
 461 }
 462
 463 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 464                                   struct streams_directive_params *s, u32 nsid)
 465 {
 466         struct nvme_command c;
 467
 468         memset(&c, 0, sizeof(c));
 469         memset(s, 0, sizeof(*s));
 470
 471         c.directive.opcode = nvme_admin_directive_recv;
 472         c.directive.nsid = cpu_to_le32(nsid);
 473         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
 474         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 475         c.directive.dtype = NVME_DIR_STREAMS;
 476
 477         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 478 }
 479
 480 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 481 {
 482         struct streams_directive_params s;
 483         int ret;
 484
 485         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 486                 return 0;
 487         if (!streams)
 488                 return 0;
 489
 490         ret = nvme_enable_streams(ctrl);
 491         if (ret)
 492                 return ret;
 493
 494         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 495         if (ret)
 496                 return ret;
 497
 498         ctrl->nssa = le16_to_cpu(s.nssa);
 499         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 500                 dev_info(ctrl->device, "too few streams (%u) available\n",
 501                                         ctrl->nssa);
 502                 nvme_disable_streams(ctrl);
 503                 return 0;
 504         }
 505
 506         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 507         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 508         return 0;
 509 }
 510
 511 /*
 512  * Check if 'req' has a write hint associated with it. If it does, assign
 513  * a valid namespace stream to the write.
 514  */
 515 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 516                                      struct request *req, u16 *control,
 517                                      u32 *dsmgmt)
 518 {
 519         enum rw_hint streamid = req->write_hint;
 520
 521         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 522                 streamid = 0;
 523         else {
 524                 streamid--;
 525                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 526                         return;
 527
 528                 *control |= NVME_RW_DTYPE_STREAMS;
 529                 *dsmgmt |= streamid << 16;
 530         }
 531
 532         if (streamid < ARRAY_SIZE(req->q->write_hints))
 533                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 534 }
 535
 536 static inline void nvme_setup_flush(struct nvme_ns *ns,
 537                 struct nvme_command *cmnd)
 538 {
 539         memset(cmnd, 0, sizeof(*cmnd));
 540         cmnd->common.opcode = nvme_cmd_flush;
 541         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 542 }
 543
 544 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 545                 struct nvme_command *cmnd)
 546 {
 547         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 548         struct nvme_dsm_range *range;
 549         struct bio *bio;
 550
 551         /*
 552          * Some devices do not consider the DSM 'Number of Ranges' field when
 553          * determining how much data to DMA. Always allocate memory for maximum
 554          * number of segments to prevent device reading beyond end of buffer.
 555          */
 556         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 557
 558         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 559         if (!range) {
 560                 /*
 561                  * If we fail allocation our range, fallback to the controller
 562                  * discard page. If that's also busy, it's safe to return
 563                  * busy, as we know we can make progress once that's freed.
 564                  */
 565                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 566                         return BLK_STS_RESOURCE;
 567
 568                 range = page_address(ns->ctrl->discard_page);
 569         }
 570
 571         __rq_for_each_bio(bio, req) {
 572                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 573                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 574
 575                 if (n < segments) {
 576                         range[n].cattr = cpu_to_le32(0);
 577                         range[n].nlb = cpu_to_le32(nlb);
 578                         range[n].slba = cpu_to_le64(slba);
 579                 }
 580                 n++;
 581         }
 582
 583         if (WARN_ON_ONCE(n != segments)) {
 584                 if (virt_to_page(range) == ns->ctrl->discard_page)
 585                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 586                 else
 587                         kfree(range);
 588                 return BLK_STS_IOERR;
 589         }
 590
 591         memset(cmnd, 0, sizeof(*cmnd));
 592         cmnd->dsm.opcode = nvme_cmd_dsm;
 593         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 594         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 595         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 596
 597         req->special_vec.bv_page = virt_to_page(range);
 598         req->special_vec.bv_offset = offset_in_page(range);
 599         req->special_vec.bv_len = alloc_size;
 600         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 601
 602         return BLK_STS_OK;
 603 }
 604
 605 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 606                 struct request *req, struct nvme_command *cmnd)
 607 {
 608         struct nvme_ctrl *ctrl = ns->ctrl;
 609         u16 control = 0;
 610         u32 dsmgmt = 0;
 611
 612         if (req->cmd_flags & REQ_FUA)
 613                 control |= NVME_RW_FUA;
 614         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 615                 control |= NVME_RW_LR;
 616
 617         if (req->cmd_flags & REQ_RAHEAD)
 618                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 619
 620         memset(cmnd, 0, sizeof(*cmnd));
 621         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 622         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 623         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 624         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 625
 626         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 627                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 628
 629         if (ns->ms) {
 630                 /*
 631                  * If formated with metadata, the block layer always provides a
 632                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 633                  * we enable the PRACT bit for protection information or set the
 634                  * namespace capacity to zero to prevent any I/O.
 635                  */
 636                 if (!blk_integrity_rq(req)) {
 637                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 638                                 return BLK_STS_NOTSUPP;
 639                         control |= NVME_RW_PRINFO_PRACT;
 640                 } else if (req_op(req) == REQ_OP_WRITE) {
 641                         t10_pi_prepare(req, ns->pi_type);
 642                 }
 643
 644                 switch (ns->pi_type) {
 645                 case NVME_NS_DPS_PI_TYPE3:
 646                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 647                         break;
 648                 case NVME_NS_DPS_PI_TYPE1:
 649                 case NVME_NS_DPS_PI_TYPE2:
 650                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 651                                         NVME_RW_PRINFO_PRCHK_REF;
 652                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 653                         break;
 654                 }
 655         }
 656
 657         cmnd->rw.control = cpu_to_le16(control);
 658         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 659         return 0;
 660 }
 661
 662 void nvme_cleanup_cmd(struct request *req)
 663 {
 664         if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
 665             nvme_req(req)->status == 0) {
 666                 struct nvme_ns *ns = req->rq_disk->private_data;
 667
 668                 t10_pi_complete(req, ns->pi_type,
 669                                 blk_rq_bytes(req) >> ns->lba_shift);
 670         }
 671         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 672                 struct nvme_ns *ns = req->rq_disk->private_data;
 673                 struct page *page = req->special_vec.bv_page;
 674
 675                 if (page == ns->ctrl->discard_page)
 676                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 677                 else
 678                         kfree(page_address(page) + req->special_vec.bv_offset);
 679         }
 680 }
 681 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 682
 683 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 684                 struct nvme_command *cmd)
 685 {
 686         blk_status_t ret = BLK_STS_OK;
 687
 688         nvme_clear_nvme_request(req);
 689
 690         switch (req_op(req)) {
 691         case REQ_OP_DRV_IN:
 692         case REQ_OP_DRV_OUT:
 693                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 694                 break;
 695         case REQ_OP_FLUSH:
 696                 nvme_setup_flush(ns, cmd);
 697                 break;
 698         case REQ_OP_WRITE_ZEROES:
 699                 /* currently only aliased to deallocate for a few ctrls: */
 700         case REQ_OP_DISCARD:
 701                 ret = nvme_setup_discard(ns, req, cmd);
 702                 break;
 703         case REQ_OP_READ:
 704         case REQ_OP_WRITE:
 705                 ret = nvme_setup_rw(ns, req, cmd);
 706                 break;
 707         default:
 708                 WARN_ON_ONCE(1);
 709                 return BLK_STS_IOERR;
 710         }
 711
 712         cmd->common.command_id = req->tag;
 713         trace_nvme_setup_cmd(req, cmd);
 714         return ret;
 715 }
 716 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 717
 718 /*
 719  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 720  * if the result is positive, it's an NVM Express status code
 721  */
 722 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 723                 union nvme_result *result, void *buffer, unsigned bufflen,
 724                 unsigned timeout, int qid, int at_head,
 725                 blk_mq_req_flags_t flags)
 726 {
 727         struct request *req;
 728         int ret;
 729
 730         req = nvme_alloc_request(q, cmd, flags, qid);
 731         if (IS_ERR(req))
 732                 return PTR_ERR(req);
 733
 734         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 735
 736         if (buffer && bufflen) {
 737                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 738                 if (ret)
 739                         goto out;
 740         }
 741
 742         blk_execute_rq(req->q, NULL, req, at_head);
 743         if (result)
 744                 *result = nvme_req(req)->result;
 745         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 746                 ret = -EINTR;
 747         else
 748                 ret = nvme_req(req)->status;
 749  out:
 750         blk_mq_free_request(req);
 751         return ret;
 752 }
 753 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 754
 755 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 756                 void *buffer, unsigned bufflen)
 757 {
 758         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 759                         NVME_QID_ANY, 0, 0);
 760 }
 761 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 762
 763 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 764                 unsigned len, u32 seed, bool write)
 765 {
 766         struct bio_integrity_payload *bip;
 767         int ret = -ENOMEM;
 768         void *buf;
 769
 770         buf = kmalloc(len, GFP_KERNEL);
 771         if (!buf)
 772                 goto out;
 773
 774         ret = -EFAULT;
 775         if (write && copy_from_user(buf, ubuf, len))
 776                 goto out_free_meta;
 777
 778         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 779         if (IS_ERR(bip)) {
 780                 ret = PTR_ERR(bip);
 781                 goto out_free_meta;
 782         }
 783
 784         bip->bip_iter.bi_size = len;
 785         bip->bip_iter.bi_sector = seed;
 786         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 787                         offset_in_page(buf));
 788         if (ret == len)
 789                 return buf;
 790         ret = -ENOMEM;
 791 out_free_meta:
 792         kfree(buf);
 793 out:
 794         return ERR_PTR(ret);
 795 }
 796
 797 static int nvme_submit_user_cmd(struct request_queue *q,
 798                 struct nvme_command *cmd, void __user *ubuffer,
 799                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 800                 u32 meta_seed, u32 *result, unsigned timeout)
 801 {
 802         bool write = nvme_is_write(cmd);
 803         struct nvme_ns *ns = q->queuedata;
 804         struct gendisk *disk = ns ? ns->disk : NULL;
 805         struct request *req;
 806         struct bio *bio = NULL;
 807         void *meta = NULL;
 808         int ret;
 809
 810         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 811         if (IS_ERR(req))
 812                 return PTR_ERR(req);
 813
 814         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 815         nvme_req(req)->flags |= NVME_REQ_USERCMD;
 816
 817         if (ubuffer && bufflen) {
 818                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 819                                 GFP_KERNEL);
 820                 if (ret)
 821                         goto out;
 822                 bio = req->bio;
 823                 bio->bi_disk = disk;
 824                 if (disk && meta_buffer && meta_len) {
 825                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 826                                         meta_seed, write);
 827                         if (IS_ERR(meta)) {
 828                                 ret = PTR_ERR(meta);
 829                                 goto out_unmap;
 830                         }
 831                         req->cmd_flags |= REQ_INTEGRITY;
 832                 }
 833         }
 834
 835         blk_execute_rq(req->q, disk, req, 0);
 836         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 837                 ret = -EINTR;
 838         else
 839                 ret = nvme_req(req)->status;
 840         if (result)
 841                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 842         if (meta && !ret && !write) {
 843                 if (copy_to_user(meta_buffer, meta, meta_len))
 844                         ret = -EFAULT;
 845         }
 846         kfree(meta);
 847  out_unmap:
 848         if (bio)
 849                 blk_rq_unmap_user(bio);
 850  out:
 851         blk_mq_free_request(req);
 852         return ret;
 853 }
 854
 855 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 856 {
 857         struct nvme_ctrl *ctrl = rq->end_io_data;
 858         unsigned long flags;
 859         bool startka = false;
 860
 861         blk_mq_free_request(rq);
 862
 863         if (status) {
 864                 dev_err(ctrl->device,
 865                         "failed nvme_keep_alive_end_io error=%d\n",
 866                                 status);
 867                 return;
 868         }
 869
 870         spin_lock_irqsave(&ctrl->lock, flags);
 871         if (ctrl->state == NVME_CTRL_LIVE ||
 872             ctrl->state == NVME_CTRL_CONNECTING)
 873                 startka = true;
 874         spin_unlock_irqrestore(&ctrl->lock, flags);
 875         if (startka)
 876                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 877 }
 878
 879 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 880 {
 881         struct request *rq;
 882
 883         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
 884                         NVME_QID_ANY);
 885         if (IS_ERR(rq))
 886                 return PTR_ERR(rq);
 887
 888         rq->timeout = ctrl->kato * HZ;
 889         rq->end_io_data = ctrl;
 890
 891         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 892
 893         return 0;
 894 }
 895
 896 static void nvme_keep_alive_work(struct work_struct *work)
 897 {
 898         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 899                         struct nvme_ctrl, ka_work);
 900
 901         if (nvme_keep_alive(ctrl)) {
 902                 /* allocation failure, reset the controller */
 903                 dev_err(ctrl->device, "keep-alive failed\n");
 904                 nvme_reset_ctrl(ctrl);
 905                 return;
 906         }
 907 }
 908
 909 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 910 {
 911         if (unlikely(ctrl->kato == 0))
 912                 return;
 913
 914         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 915 }
 916
 917 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 918 {
 919         if (unlikely(ctrl->kato == 0))
 920                 return;
 921
 922         cancel_delayed_work_sync(&ctrl->ka_work);
 923 }
 924 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 925
 926 /*
 927  * In NVMe 1.0 the CNS field was just a binary controller or namespace
 928  * flag, thus sending any new CNS opcodes has a big chance of not working.
 929  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
 930  * (but not for any later version).
 931  */
 932 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
 933 {
 934         if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
 935                 return ctrl->vs < NVME_VS(1, 2, 0);
 936         return ctrl->vs < NVME_VS(1, 1, 0);
 937 }
 938
 939 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 940 {
 941         struct nvme_command c = { };
 942         int error;
 943
 944         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 945         c.identify.opcode = nvme_admin_identify;
 946         c.identify.cns = NVME_ID_CNS_CTRL;
 947
 948         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 949         if (!*id)
 950                 return -ENOMEM;
 951
 952         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 953                         sizeof(struct nvme_id_ctrl));
 954         if (error)
 955                 kfree(*id);
 956         return error;
 957 }
 958
 959 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
 960                 struct nvme_ns_ids *ids)
 961 {
 962         struct nvme_command c = { };
 963         int status;
 964         void *data;
 965         int pos;
 966         int len;
 967
 968         c.identify.opcode = nvme_admin_identify;
 969         c.identify.nsid = cpu_to_le32(nsid);
 970         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
 971
 972         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
 973         if (!data)
 974                 return -ENOMEM;
 975
 976         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
 977                                       NVME_IDENTIFY_DATA_SIZE);
 978         if (status)
 979                 goto free_data;
 980
 981         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
 982                 struct nvme_ns_id_desc *cur = data + pos;
 983
 984                 if (cur->nidl == 0)
 985                         break;
 986
 987                 switch (cur->nidt) {
 988                 case NVME_NIDT_EUI64:
 989                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
 990                                 dev_warn(ctrl->device,
 991                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
 992                                          cur->nidl);
 993                                 goto free_data;
 994                         }
 995                         len = NVME_NIDT_EUI64_LEN;
 996                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
 997                         break;
 998                 case NVME_NIDT_NGUID:
 999                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1000                                 dev_warn(ctrl->device,
1001                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1002                                          cur->nidl);
1003                                 goto free_data;
1004                         }
1005                         len = NVME_NIDT_NGUID_LEN;
1006                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1007                         break;
1008                 case NVME_NIDT_UUID:
1009                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
1010                                 dev_warn(ctrl->device,
1011                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1012                                          cur->nidl);
1013                                 goto free_data;
1014                         }
1015                         len = NVME_NIDT_UUID_LEN;
1016                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1017                         break;
1018                 default:
1019                         /* Skip unnkown types */
1020                         len = cur->nidl;
1021                         break;
1022                 }
1023
1024                 len += sizeof(*cur);
1025         }
1026 free_data:
1027         kfree(data);
1028         return status;
1029 }
1030
1031 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1032 {
1033         struct nvme_command c = { };
1034
1035         c.identify.opcode = nvme_admin_identify;
1036         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1037         c.identify.nsid = cpu_to_le32(nsid);
1038         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1039                                     NVME_IDENTIFY_DATA_SIZE);
1040 }
1041
1042 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
1043                 unsigned nsid)
1044 {
1045         struct nvme_id_ns *id;
1046         struct nvme_command c = { };
1047         int error;
1048
1049         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1050         c.identify.opcode = nvme_admin_identify;
1051         c.identify.nsid = cpu_to_le32(nsid);
1052         c.identify.cns = NVME_ID_CNS_NS;
1053
1054         id = kmalloc(sizeof(*id), GFP_KERNEL);
1055         if (!id)
1056                 return NULL;
1057
1058         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1059         if (error) {
1060                 dev_warn(ctrl->device, "Identify namespace failed\n");
1061                 kfree(id);
1062                 return NULL;
1063         }
1064
1065         return id;
1066 }
1067
1068 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
1069                       void *buffer, size_t buflen, u32 *result)
1070 {
1071         union nvme_result res = { 0 };
1072         struct nvme_command c;
1073         int ret;
1074
1075         memset(&c, 0, sizeof(c));
1076         c.features.opcode = nvme_admin_set_features;
1077         c.features.fid = cpu_to_le32(fid);
1078         c.features.dword11 = cpu_to_le32(dword11);
1079
1080         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1081                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1082         if (ret >= 0 && result)
1083                 *result = le32_to_cpu(res.u32);
1084         return ret;
1085 }
1086
1087 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1088 {
1089         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1090         u32 result;
1091         int status, nr_io_queues;
1092
1093         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1094                         &result);
1095         if (status < 0)
1096                 return status;
1097
1098         /*
1099          * Degraded controllers might return an error when setting the queue
1100          * count.  We still want to be able to bring them online and offer
1101          * access to the admin queue, as that might be only way to fix them up.
1102          */
1103         if (status > 0) {
1104                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1105                 *count = 0;
1106         } else {
1107                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1108                 *count = min(*count, nr_io_queues);
1109         }
1110
1111         return 0;
1112 }
1113 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1114
1115 #define NVME_AEN_SUPPORTED \
1116         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1117
1118 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1119 {
1120         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1121         int status;
1122
1123         if (!supported_aens)
1124                 return;
1125
1126         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1127                         NULL, 0, &result);
1128         if (status)
1129                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1130                          supported_aens);
1131 }
1132
1133 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1134 {
1135         struct nvme_user_io io;
1136         struct nvme_command c;
1137         unsigned length, meta_len;
1138         void __user *metadata;
1139
1140         if (copy_from_user(&io, uio, sizeof(io)))
1141                 return -EFAULT;
1142         if (io.flags)
1143                 return -EINVAL;
1144
1145         switch (io.opcode) {
1146         case nvme_cmd_write:
1147         case nvme_cmd_read:
1148         case nvme_cmd_compare:
1149                 break;
1150         default:
1151                 return -EINVAL;
1152         }
1153
1154         length = (io.nblocks + 1) << ns->lba_shift;
1155         meta_len = (io.nblocks + 1) * ns->ms;
1156         metadata = (void __user *)(uintptr_t)io.metadata;
1157
1158         if (ns->ext) {
1159                 length += meta_len;
1160                 meta_len = 0;
1161         } else if (meta_len) {
1162                 if ((io.metadata & 3) || !io.metadata)
1163                         return -EINVAL;
1164         }
1165
1166         memset(&c, 0, sizeof(c));
1167         c.rw.opcode = io.opcode;
1168         c.rw.flags = io.flags;
1169         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1170         c.rw.slba = cpu_to_le64(io.slba);
1171         c.rw.length = cpu_to_le16(io.nblocks);
1172         c.rw.control = cpu_to_le16(io.control);
1173         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1174         c.rw.reftag = cpu_to_le32(io.reftag);
1175         c.rw.apptag = cpu_to_le16(io.apptag);
1176         c.rw.appmask = cpu_to_le16(io.appmask);
1177
1178         return nvme_submit_user_cmd(ns->queue, &c,
1179                         (void __user *)(uintptr_t)io.addr, length,
1180                         metadata, meta_len, io.slba, NULL, 0);
1181 }
1182
1183 static u32 nvme_known_admin_effects(u8 opcode)
1184 {
1185         switch (opcode) {
1186         case nvme_admin_format_nvm:
1187                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1188                                         NVME_CMD_EFFECTS_CSE_MASK;
1189         case nvme_admin_sanitize_nvm:
1190                 return NVME_CMD_EFFECTS_CSE_MASK;
1191         default:
1192                 break;
1193         }
1194         return 0;
1195 }
1196
1197 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1198                                                                 u8 opcode)
1199 {
1200         u32 effects = 0;
1201
1202         if (ns) {
1203                 if (ctrl->effects)
1204                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1205                 if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1206                         dev_warn(ctrl->device,
1207                                  "IO command:%02x has unhandled effects:%08x\n",
1208                                  opcode, effects);
1209                 return 0;
1210         }
1211
1212         if (ctrl->effects)
1213                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1214         else
1215                 effects = nvme_known_admin_effects(opcode);
1216
1217         /*
1218          * For simplicity, IO to all namespaces is quiesced even if the command
1219          * effects say only one namespace is affected.
1220          */
1221         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1222                 mutex_lock(&ctrl->scan_lock);
1223                 mutex_lock(&ctrl->subsys->lock);
1224                 nvme_mpath_start_freeze(ctrl->subsys);
1225                 nvme_mpath_wait_freeze(ctrl->subsys);
1226                 nvme_start_freeze(ctrl);
1227                 nvme_wait_freeze(ctrl);
1228         }
1229         return effects;
1230 }
1231
1232 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1233 {
1234         struct nvme_ns *ns;
1235
1236         down_read(&ctrl->namespaces_rwsem);
1237         list_for_each_entry(ns, &ctrl->namespaces, list)
1238                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1239                         nvme_set_queue_dying(ns);
1240         up_read(&ctrl->namespaces_rwsem);
1241
1242         nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1243 }
1244
1245 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1246 {
1247         /*
1248          * Revalidate LBA changes prior to unfreezing. This is necessary to
1249          * prevent memory corruption if a logical block size was changed by
1250          * this command.
1251          */
1252         if (effects & NVME_CMD_EFFECTS_LBCC)
1253                 nvme_update_formats(ctrl);
1254         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1255                 nvme_unfreeze(ctrl);
1256                 nvme_mpath_unfreeze(ctrl->subsys);
1257                 mutex_unlock(&ctrl->subsys->lock);
1258                 mutex_unlock(&ctrl->scan_lock);
1259         }
1260         if (effects & NVME_CMD_EFFECTS_CCC)
1261                 nvme_init_identify(ctrl);
1262         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1263                 nvme_queue_scan(ctrl);
1264 }
1265
1266 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1267                         struct nvme_passthru_cmd __user *ucmd)
1268 {
1269         struct nvme_passthru_cmd cmd;
1270         struct nvme_command c;
1271         unsigned timeout = 0;
1272         u32 effects;
1273         int status;
1274
1275         if (!capable(CAP_SYS_ADMIN))
1276                 return -EACCES;
1277         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1278                 return -EFAULT;
1279         if (cmd.flags)
1280                 return -EINVAL;
1281
1282         memset(&c, 0, sizeof(c));
1283         c.common.opcode = cmd.opcode;
1284         c.common.flags = cmd.flags;
1285         c.common.nsid = cpu_to_le32(cmd.nsid);
1286         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1287         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1288         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1289         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1290         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1291         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1292         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1293         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1294
1295         if (cmd.timeout_ms)
1296                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1297
1298         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1299         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1300                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1301                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1302                         0, &cmd.result, timeout);
1303         nvme_passthru_end(ctrl, effects);
1304
1305         if (status >= 0) {
1306                 if (put_user(cmd.result, &ucmd->result))
1307                         return -EFAULT;
1308         }
1309
1310         return status;
1311 }
1312
1313 /*
1314  * Issue ioctl requests on the first available path.  Note that unlike normal
1315  * block layer requests we will not retry failed request on another controller.
1316  */
1317 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1318                 struct nvme_ns_head **head, int *srcu_idx)
1319 {
1320 #ifdef CONFIG_NVME_MULTIPATH
1321         if (disk->fops == &nvme_ns_head_ops) {
1322                 struct nvme_ns *ns;
1323
1324                 *head = disk->private_data;
1325                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1326                 ns = nvme_find_path(*head);
1327                 if (!ns)
1328                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1329                 return ns;
1330         }
1331 #endif
1332         *head = NULL;
1333         *srcu_idx = -1;
1334         return disk->private_data;
1335 }
1336
1337 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1338 {
1339         if (head)
1340                 srcu_read_unlock(&head->srcu, idx);
1341 }
1342
1343 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1344                 unsigned int cmd, unsigned long arg)
1345 {
1346         struct nvme_ns_head *head = NULL;
1347         void __user *argp = (void __user *)arg;
1348         struct nvme_ns *ns;
1349         int srcu_idx, ret;
1350
1351         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1352         if (unlikely(!ns))
1353                 return -EWOULDBLOCK;
1354
1355         /*
1356          * Handle ioctls that apply to the controller instead of the namespace
1357          * seperately and drop the ns SRCU reference early.  This avoids a
1358          * deadlock when deleting namespaces using the passthrough interface.
1359          */
1360         if (cmd == NVME_IOCTL_ADMIN_CMD || is_sed_ioctl(cmd)) {
1361                 struct nvme_ctrl *ctrl = ns->ctrl;
1362
1363                 nvme_get_ctrl(ns->ctrl);
1364                 nvme_put_ns_from_disk(head, srcu_idx);
1365
1366                 if (cmd == NVME_IOCTL_ADMIN_CMD)
1367                         ret = nvme_user_cmd(ctrl, NULL, argp);
1368                 else
1369                         ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1370
1371                 nvme_put_ctrl(ctrl);
1372                 return ret;
1373         }
1374
1375         switch (cmd) {
1376         case NVME_IOCTL_ID:
1377                 force_successful_syscall_return();
1378                 ret = ns->head->ns_id;
1379                 break;
1380         case NVME_IOCTL_IO_CMD:
1381                 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1382                 break;
1383         case NVME_IOCTL_SUBMIT_IO:
1384                 ret = nvme_submit_io(ns, argp);
1385                 break;
1386         default:
1387                 if (ns->ndev)
1388                         ret = nvme_nvm_ioctl(ns, cmd, arg);
1389                 else
1390                         ret = -ENOTTY;
1391         }
1392
1393         nvme_put_ns_from_disk(head, srcu_idx);
1394         return ret;
1395 }
1396
1397 static int nvme_open(struct block_device *bdev, fmode_t mode)
1398 {
1399         struct nvme_ns *ns = bdev->bd_disk->private_data;
1400
1401 #ifdef CONFIG_NVME_MULTIPATH
1402         /* should never be called due to GENHD_FL_HIDDEN */
1403         if (WARN_ON_ONCE(ns->head->disk))
1404                 goto fail;
1405 #endif
1406         if (!kref_get_unless_zero(&ns->kref))
1407                 goto fail;
1408         if (!try_module_get(ns->ctrl->ops->module))
1409                 goto fail_put_ns;
1410
1411         return 0;
1412
1413 fail_put_ns:
1414         nvme_put_ns(ns);
1415 fail:
1416         return -ENXIO;
1417 }
1418
1419 static void nvme_release(struct gendisk *disk, fmode_t mode)
1420 {
1421         struct nvme_ns *ns = disk->private_data;
1422
1423         module_put(ns->ctrl->ops->module);
1424         nvme_put_ns(ns);
1425 }
1426
1427 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1428 {
1429         /* some standard values */
1430         geo->heads = 1 << 6;
1431         geo->sectors = 1 << 5;
1432         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1433         return 0;
1434 }
1435
1436 #ifdef CONFIG_BLK_DEV_INTEGRITY
1437 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1438 {
1439         struct blk_integrity integrity;
1440
1441         memset(&integrity, 0, sizeof(integrity));
1442         switch (pi_type) {
1443         case NVME_NS_DPS_PI_TYPE3:
1444                 integrity.profile = &t10_pi_type3_crc;
1445                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1446                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1447                 break;
1448         case NVME_NS_DPS_PI_TYPE1:
1449         case NVME_NS_DPS_PI_TYPE2:
1450                 integrity.profile = &t10_pi_type1_crc;
1451                 integrity.tag_size = sizeof(u16);
1452                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1453                 break;
1454         default:
1455                 integrity.profile = NULL;
1456                 break;
1457         }
1458         integrity.tuple_size = ms;
1459         blk_integrity_register(disk, &integrity);
1460         blk_queue_max_integrity_segments(disk->queue, 1);
1461 }
1462 #else
1463 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1464 {
1465 }
1466 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1467
1468 static void nvme_set_chunk_size(struct nvme_ns *ns)
1469 {
1470         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1471         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1472 }
1473
1474 static void nvme_config_discard(struct nvme_ns *ns)
1475 {
1476         struct nvme_ctrl *ctrl = ns->ctrl;
1477         struct request_queue *queue = ns->queue;
1478         u32 size = queue_logical_block_size(queue);
1479
1480         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1481                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1482                 return;
1483         }
1484
1485         if (ctrl->nr_streams && ns->sws && ns->sgs)
1486                 size *= ns->sws * ns->sgs;
1487
1488         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1489                         NVME_DSM_MAX_RANGES);
1490
1491         queue->limits.discard_alignment = 0;
1492         queue->limits.discard_granularity = size;
1493
1494         /* If discard is already enabled, don't reset queue limits */
1495         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1496                 return;
1497
1498         blk_queue_max_discard_sectors(queue, UINT_MAX);
1499         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1500
1501         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1502                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1503 }
1504
1505 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1506                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1507 {
1508         memset(ids, 0, sizeof(*ids));
1509
1510         if (ctrl->vs >= NVME_VS(1, 1, 0))
1511                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1512         if (ctrl->vs >= NVME_VS(1, 2, 0))
1513                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1514         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1515                  /* Don't treat error as fatal we potentially
1516                   * already have a NGUID or EUI-64
1517                   */
1518                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1519                         dev_warn(ctrl->device,
1520                                  "%s: Identify Descriptors failed\n", __func__);
1521         }
1522 }
1523
1524 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1525 {
1526         return !uuid_is_null(&ids->uuid) ||
1527                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1528                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1529 }
1530
1531 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1532 {
1533         return uuid_equal(&a->uuid, &b->uuid) &&
1534                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1535                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1536 }
1537
1538 static void nvme_update_disk_info(struct gendisk *disk,
1539                 struct nvme_ns *ns, struct nvme_id_ns *id)
1540 {
1541         sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1542         unsigned short bs = 1 << ns->lba_shift;
1543
1544         if (ns->lba_shift > PAGE_SHIFT) {
1545                 /* unsupported block size, set capacity to 0 later */
1546                 bs = (1 << 9);
1547         }
1548         blk_mq_freeze_queue(disk->queue);
1549         blk_integrity_unregister(disk);
1550
1551         blk_queue_logical_block_size(disk->queue, bs);
1552         blk_queue_physical_block_size(disk->queue, bs);
1553         blk_queue_io_min(disk->queue, bs);
1554
1555         if (ns->ms && !ns->ext &&
1556             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1557                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1558         if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1559             ns->lba_shift > PAGE_SHIFT)
1560                 capacity = 0;
1561
1562         set_capacity(disk, capacity);
1563         nvme_config_discard(ns);
1564
1565         if (id->nsattr & (1 << 0))
1566                 set_disk_ro(disk, true);
1567         else
1568                 set_disk_ro(disk, false);
1569
1570         blk_mq_unfreeze_queue(disk->queue);
1571 }
1572
1573 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1574 {
1575         struct nvme_ns *ns = disk->private_data;
1576
1577         /*
1578          * If identify namespace failed, use default 512 byte block size so
1579          * block layer can use before failing read/write for 0 capacity.
1580          */
1581         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1582         if (ns->lba_shift == 0)
1583                 ns->lba_shift = 9;
1584         ns->noiob = le16_to_cpu(id->noiob);
1585         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1586         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1587         /* the PI implementation requires metadata equal t10 pi tuple size */
1588         if (ns->ms == sizeof(struct t10_pi_tuple))
1589                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1590         else
1591                 ns->pi_type = 0;
1592
1593         if (ns->noiob)
1594                 nvme_set_chunk_size(ns);
1595         nvme_update_disk_info(disk, ns, id);
1596         if (ns->ndev)
1597                 nvme_nvm_update_nvm_info(ns);
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                 nvme_mpath_update_disk_size(ns->head->disk);
1603         }
1604 #endif
1605 }
1606
1607 static int nvme_revalidate_disk(struct gendisk *disk)
1608 {
1609         struct nvme_ns *ns = disk->private_data;
1610         struct nvme_ctrl *ctrl = ns->ctrl;
1611         struct nvme_id_ns *id;
1612         struct nvme_ns_ids ids;
1613         int ret = 0;
1614
1615         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1616                 set_capacity(disk, 0);
1617                 return -ENODEV;
1618         }
1619
1620         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1621         if (!id)
1622                 return -ENODEV;
1623
1624         if (id->ncap == 0) {
1625                 ret = -ENODEV;
1626                 goto out;
1627         }
1628
1629         __nvme_revalidate_disk(disk, id);
1630         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1631         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1632                 dev_err(ctrl->device,
1633                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1634                 ret = -ENODEV;
1635         }
1636
1637 out:
1638         kfree(id);
1639         return ret;
1640 }
1641
1642 static char nvme_pr_type(enum pr_type type)
1643 {
1644         switch (type) {
1645         case PR_WRITE_EXCLUSIVE:
1646                 return 1;
1647         case PR_EXCLUSIVE_ACCESS:
1648                 return 2;
1649         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1650                 return 3;
1651         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1652                 return 4;
1653         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1654                 return 5;
1655         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1656                 return 6;
1657         default:
1658                 return 0;
1659         }
1660 };
1661
1662 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1663                                 u64 key, u64 sa_key, u8 op)
1664 {
1665         struct nvme_ns_head *head = NULL;
1666         struct nvme_ns *ns;
1667         struct nvme_command c;
1668         int srcu_idx, ret;
1669         u8 data[16] = { 0, };
1670
1671         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1672         if (unlikely(!ns))
1673                 return -EWOULDBLOCK;
1674
1675         put_unaligned_le64(key, &data[0]);
1676         put_unaligned_le64(sa_key, &data[8]);
1677
1678         memset(&c, 0, sizeof(c));
1679         c.common.opcode = op;
1680         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1681         c.common.cdw10[0] = cpu_to_le32(cdw10);
1682
1683         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1684         nvme_put_ns_from_disk(head, srcu_idx);
1685         return ret;
1686 }
1687
1688 static int nvme_pr_register(struct block_device *bdev, u64 old,
1689                 u64 new, unsigned flags)
1690 {
1691         u32 cdw10;
1692
1693         if (flags & ~PR_FL_IGNORE_KEY)
1694                 return -EOPNOTSUPP;
1695
1696         cdw10 = old ? 2 : 0;
1697         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1698         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1699         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1700 }
1701
1702 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1703                 enum pr_type type, unsigned flags)
1704 {
1705         u32 cdw10;
1706
1707         if (flags & ~PR_FL_IGNORE_KEY)
1708                 return -EOPNOTSUPP;
1709
1710         cdw10 = nvme_pr_type(type) << 8;
1711         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1712         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1713 }
1714
1715 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1716                 enum pr_type type, bool abort)
1717 {
1718         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1719
1720         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1721 }
1722
1723 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1724 {
1725         u32 cdw10 = 1 | (key ? 0 : 1 << 3);
1726
1727         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1728 }
1729
1730 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1731 {
1732         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 0 : 1 << 3);
1733
1734         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1735 }
1736
1737 static const struct pr_ops nvme_pr_ops = {
1738         .pr_register    = nvme_pr_register,
1739         .pr_reserve     = nvme_pr_reserve,
1740         .pr_release     = nvme_pr_release,
1741         .pr_preempt     = nvme_pr_preempt,
1742         .pr_clear       = nvme_pr_clear,
1743 };
1744
1745 #ifdef CONFIG_BLK_SED_OPAL
1746 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1747                 bool send)
1748 {
1749         struct nvme_ctrl *ctrl = data;
1750         struct nvme_command cmd;
1751
1752         memset(&cmd, 0, sizeof(cmd));
1753         if (send)
1754                 cmd.common.opcode = nvme_admin_security_send;
1755         else
1756                 cmd.common.opcode = nvme_admin_security_recv;
1757         cmd.common.nsid = 0;
1758         cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1759         cmd.common.cdw10[1] = cpu_to_le32(len);
1760
1761         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1762                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1763 }
1764 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1765 #endif /* CONFIG_BLK_SED_OPAL */
1766
1767 static const struct block_device_operations nvme_fops = {
1768         .owner          = THIS_MODULE,
1769         .ioctl          = nvme_ioctl,
1770         .compat_ioctl   = nvme_ioctl,
1771         .open           = nvme_open,
1772         .release        = nvme_release,
1773         .getgeo         = nvme_getgeo,
1774         .revalidate_disk= nvme_revalidate_disk,
1775         .pr_ops         = &nvme_pr_ops,
1776 };
1777
1778 #ifdef CONFIG_NVME_MULTIPATH
1779 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1780 {
1781         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1782
1783         if (!kref_get_unless_zero(&head->ref))
1784                 return -ENXIO;
1785         return 0;
1786 }
1787
1788 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1789 {
1790         nvme_put_ns_head(disk->private_data);
1791 }
1792
1793 const struct block_device_operations nvme_ns_head_ops = {
1794         .owner          = THIS_MODULE,
1795         .open           = nvme_ns_head_open,
1796         .release        = nvme_ns_head_release,
1797         .ioctl          = nvme_ioctl,
1798         .compat_ioctl   = nvme_ioctl,
1799         .getgeo         = nvme_getgeo,
1800         .pr_ops         = &nvme_pr_ops,
1801 };
1802 #endif /* CONFIG_NVME_MULTIPATH */
1803
1804 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1805 {
1806         unsigned long timeout =
1807                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1808         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1809         int ret;
1810
1811         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1812                 if (csts == ~0)
1813                         return -ENODEV;
1814                 if ((csts & NVME_CSTS_RDY) == bit)
1815                         break;
1816
1817                 msleep(100);
1818                 if (fatal_signal_pending(current))
1819                         return -EINTR;
1820                 if (time_after(jiffies, timeout)) {
1821                         dev_err(ctrl->device,
1822                                 "Device not ready; aborting %s\n", enabled ?
1823                                                 "initialisation" : "reset");
1824                         return -ENODEV;
1825                 }
1826         }
1827
1828         return ret;
1829 }
1830
1831 /*
1832  * If the device has been passed off to us in an enabled state, just clear
1833  * the enabled bit.  The spec says we should set the 'shutdown notification
1834  * bits', but doing so may cause the device to complete commands to the
1835  * admin queue ... and we don't know what memory that might be pointing at!
1836  */
1837 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1838 {
1839         int ret;
1840
1841         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1842         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1843
1844         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1845         if (ret)
1846                 return ret;
1847
1848         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1849                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1850
1851         return nvme_wait_ready(ctrl, cap, false);
1852 }
1853 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1854
1855 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1856 {
1857         /*
1858          * Default to a 4K page size, with the intention to update this
1859          * path in the future to accomodate architectures with differing
1860          * kernel and IO page sizes.
1861          */
1862         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1863         int ret;
1864
1865         if (page_shift < dev_page_min) {
1866                 dev_err(ctrl->device,
1867                         "Minimum device page size %u too large for host (%u)\n",
1868                         1 << dev_page_min, 1 << page_shift);
1869                 return -ENODEV;
1870         }
1871
1872         ctrl->page_size = 1 << page_shift;
1873
1874         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1875         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1876         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1877         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1878         ctrl->ctrl_config |= NVME_CC_ENABLE;
1879
1880         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1881         if (ret)
1882                 return ret;
1883         return nvme_wait_ready(ctrl, cap, true);
1884 }
1885 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1886
1887 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1888 {
1889         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1890         u32 csts;
1891         int ret;
1892
1893         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1894         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1895
1896         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1897         if (ret)
1898                 return ret;
1899
1900         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1901                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1902                         break;
1903
1904                 msleep(100);
1905                 if (fatal_signal_pending(current))
1906                         return -EINTR;
1907                 if (time_after(jiffies, timeout)) {
1908                         dev_err(ctrl->device,
1909                                 "Device shutdown incomplete; abort shutdown\n");
1910                         return -ENODEV;
1911                 }
1912         }
1913
1914         return ret;
1915 }
1916 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1917
1918 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1919                 struct request_queue *q)
1920 {
1921         bool vwc = false;
1922
1923         if (ctrl->max_hw_sectors) {
1924                 u32 max_segments =
1925                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1926
1927                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1928                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1929                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1930         }
1931         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1932             is_power_of_2(ctrl->max_hw_sectors))
1933                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1934         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1935         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1936                 vwc = true;
1937         blk_queue_write_cache(q, vwc, vwc);
1938 }
1939
1940 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1941 {
1942         __le64 ts;
1943         int ret;
1944
1945         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1946                 return 0;
1947
1948         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1949         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1950                         NULL);
1951         if (ret)
1952                 dev_warn_once(ctrl->device,
1953                         "could not set timestamp (%d)\n", ret);
1954         return ret;
1955 }
1956
1957 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1958 {
1959         /*
1960          * APST (Autonomous Power State Transition) lets us program a
1961          * table of power state transitions that the controller will
1962          * perform automatically.  We configure it with a simple
1963          * heuristic: we are willing to spend at most 2% of the time
1964          * transitioning between power states.  Therefore, when running
1965          * in any given state, we will enter the next lower-power
1966          * non-operational state after waiting 50 * (enlat + exlat)
1967          * microseconds, as long as that state's exit latency is under
1968          * the requested maximum latency.
1969          *
1970          * We will not autonomously enter any non-operational state for
1971          * which the total latency exceeds ps_max_latency_us.  Users
1972          * can set ps_max_latency_us to zero to turn off APST.
1973          */
1974
1975         unsigned apste;
1976         struct nvme_feat_auto_pst *table;
1977         u64 max_lat_us = 0;
1978         int max_ps = -1;
1979         int ret;
1980
1981         /*
1982          * If APST isn't supported or if we haven't been initialized yet,
1983          * then don't do anything.
1984          */
1985         if (!ctrl->apsta)
1986                 return 0;
1987
1988         if (ctrl->npss > 31) {
1989                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1990                 return 0;
1991         }
1992
1993         table = kzalloc(sizeof(*table), GFP_KERNEL);
1994         if (!table)
1995                 return 0;
1996
1997         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1998                 /* Turn off APST. */
1999                 apste = 0;
2000                 dev_dbg(ctrl->device, "APST disabled\n");
2001         } else {
2002                 __le64 target = cpu_to_le64(0);
2003                 int state;
2004
2005                 /*
2006                  * Walk through all states from lowest- to highest-power.
2007                  * According to the spec, lower-numbered states use more
2008                  * power.  NPSS, despite the name, is the index of the
2009                  * lowest-power state, not the number of states.
2010                  */
2011                 for (state = (int)ctrl->npss; state >= 0; state--) {
2012                         u64 total_latency_us, exit_latency_us, transition_ms;
2013
2014                         if (target)
2015                                 table->entries[state] = target;
2016
2017                         /*
2018                          * Don't allow transitions to the deepest state
2019                          * if it's quirked off.
2020                          */
2021                         if (state == ctrl->npss &&
2022                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2023                                 continue;
2024
2025                         /*
2026                          * Is this state a useful non-operational state for
2027                          * higher-power states to autonomously transition to?
2028                          */
2029                         if (!(ctrl->psd[state].flags &
2030                               NVME_PS_FLAGS_NON_OP_STATE))
2031                                 continue;
2032
2033                         exit_latency_us =
2034                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2035                         if (exit_latency_us > ctrl->ps_max_latency_us)
2036                                 continue;
2037
2038                         total_latency_us =
2039                                 exit_latency_us +
2040                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2041
2042                         /*
2043                          * This state is good.  Use it as the APST idle
2044                          * target for higher power states.
2045                          */
2046                         transition_ms = total_latency_us + 19;
2047                         do_div(transition_ms, 20);
2048                         if (transition_ms > (1 << 24) - 1)
2049                                 transition_ms = (1 << 24) - 1;
2050
2051                         target = cpu_to_le64((state << 3) |
2052                                              (transition_ms << 8));
2053
2054                         if (max_ps == -1)
2055                                 max_ps = state;
2056
2057                         if (total_latency_us > max_lat_us)
2058                                 max_lat_us = total_latency_us;
2059                 }
2060
2061                 apste = 1;
2062
2063                 if (max_ps == -1) {
2064                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2065                 } else {
2066                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2067                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2068                 }
2069         }
2070
2071         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2072                                 table, sizeof(*table), NULL);
2073         if (ret)
2074                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2075
2076         kfree(table);
2077         return ret;
2078 }
2079
2080 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2081 {
2082         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2083         u64 latency;
2084
2085         switch (val) {
2086         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2087         case PM_QOS_LATENCY_ANY:
2088                 latency = U64_MAX;
2089                 break;
2090
2091         default:
2092                 latency = val;
2093         }
2094
2095         if (ctrl->ps_max_latency_us != latency) {
2096                 ctrl->ps_max_latency_us = latency;
2097                 if (ctrl->state == NVME_CTRL_LIVE)
2098                         nvme_configure_apst(ctrl);
2099         }
2100 }
2101
2102 struct nvme_core_quirk_entry {
2103         /*
2104          * NVMe model and firmware strings are padded with spaces.  For
2105          * simplicity, strings in the quirk table are padded with NULLs
2106          * instead.
2107          */
2108         u16 vid;
2109         const char *mn;
2110         const char *fr;
2111         unsigned long quirks;
2112 };
2113
2114 static const struct nvme_core_quirk_entry core_quirks[] = {
2115         {
2116                 /*
2117                  * This Toshiba device seems to die using any APST states.  See:
2118                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2119                  */
2120                 .vid = 0x1179,
2121                 .mn = "THNSF5256GPUK TOSHIBA",
2122                 .quirks = NVME_QUIRK_NO_APST,
2123         }
2124 };
2125
2126 /* match is null-terminated but idstr is space-padded. */
2127 static bool string_matches(const char *idstr, const char *match, size_t len)
2128 {
2129         size_t matchlen;
2130
2131         if (!match)
2132                 return true;
2133
2134         matchlen = strlen(match);
2135         WARN_ON_ONCE(matchlen > len);
2136
2137         if (memcmp(idstr, match, matchlen))
2138                 return false;
2139
2140         for (; matchlen < len; matchlen++)
2141                 if (idstr[matchlen] != ' ')
2142                         return false;
2143
2144         return true;
2145 }
2146
2147 static bool quirk_matches(const struct nvme_id_ctrl *id,
2148                           const struct nvme_core_quirk_entry *q)
2149 {
2150         return q->vid == le16_to_cpu(id->vid) &&
2151                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2152                 string_matches(id->fr, q->fr, sizeof(id->fr));
2153 }
2154
2155 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2156                 struct nvme_id_ctrl *id)
2157 {
2158         size_t nqnlen;
2159         int off;
2160
2161         nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2162         if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2163                 strncpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2164                 return;
2165         }
2166
2167         if (ctrl->vs >= NVME_VS(1, 2, 1))
2168                 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2169
2170         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2171         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2172                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2173                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2174         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2175         off += sizeof(id->sn);
2176         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2177         off += sizeof(id->mn);
2178         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2179 }
2180
2181 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2182 {
2183         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2184         kfree(subsys);
2185 }
2186
2187 static void nvme_release_subsystem(struct device *dev)
2188 {
2189         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2190 }
2191
2192 static void nvme_destroy_subsystem(struct kref *ref)
2193 {
2194         struct nvme_subsystem *subsys =
2195                         container_of(ref, struct nvme_subsystem, ref);
2196
2197         mutex_lock(&nvme_subsystems_lock);
2198         list_del(&subsys->entry);
2199         mutex_unlock(&nvme_subsystems_lock);
2200
2201         ida_destroy(&subsys->ns_ida);
2202         device_del(&subsys->dev);
2203         put_device(&subsys->dev);
2204 }
2205
2206 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2207 {
2208         kref_put(&subsys->ref, nvme_destroy_subsystem);
2209 }
2210
2211 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2212 {
2213         struct nvme_subsystem *subsys;
2214
2215         lockdep_assert_held(&nvme_subsystems_lock);
2216
2217         /*
2218          * Fail matches for discovery subsystems. This results
2219          * in each discovery controller bound to a unique subsystem.
2220          * This avoids issues with validating controller values
2221          * that can only be true when there is a single unique subsystem.
2222          * There may be multiple and completely independent entities
2223          * that provide discovery controllers.
2224          */
2225         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2226                 return NULL;
2227
2228         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2229                 if (strcmp(subsys->subnqn, subsysnqn))
2230                         continue;
2231                 if (!kref_get_unless_zero(&subsys->ref))
2232                         continue;
2233                 return subsys;
2234         }
2235
2236         return NULL;
2237 }
2238
2239 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2240         struct device_attribute subsys_attr_##_name = \
2241                 __ATTR(_name, _mode, _show, NULL)
2242
2243 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2244                                     struct device_attribute *attr,
2245                                     char *buf)
2246 {
2247         struct nvme_subsystem *subsys =
2248                 container_of(dev, struct nvme_subsystem, dev);
2249
2250         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2251 }
2252 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2253
2254 #define nvme_subsys_show_str_function(field)                            \
2255 static ssize_t subsys_##field##_show(struct device *dev,                \
2256                             struct device_attribute *attr, char *buf)   \
2257 {                                                                       \
2258         struct nvme_subsystem *subsys =                                 \
2259                 container_of(dev, struct nvme_subsystem, dev);          \
2260         return sprintf(buf, "%.*s\n",                                   \
2261                        (int)sizeof(subsys->field), subsys->field);      \
2262 }                                                                       \
2263 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2264
2265 nvme_subsys_show_str_function(model);
2266 nvme_subsys_show_str_function(serial);
2267 nvme_subsys_show_str_function(firmware_rev);
2268
2269 static struct attribute *nvme_subsys_attrs[] = {
2270         &subsys_attr_model.attr,
2271         &subsys_attr_serial.attr,
2272         &subsys_attr_firmware_rev.attr,
2273         &subsys_attr_subsysnqn.attr,
2274         NULL,
2275 };
2276
2277 static struct attribute_group nvme_subsys_attrs_group = {
2278         .attrs = nvme_subsys_attrs,
2279 };
2280
2281 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2282         &nvme_subsys_attrs_group,
2283         NULL,
2284 };
2285
2286 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2287 {
2288         int count = 0;
2289         struct nvme_ctrl *ctrl;
2290
2291         mutex_lock(&subsys->lock);
2292         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2293                 if (ctrl->state != NVME_CTRL_DELETING &&
2294                     ctrl->state != NVME_CTRL_DEAD)
2295                         count++;
2296         }
2297         mutex_unlock(&subsys->lock);
2298
2299         return count;
2300 }
2301
2302 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2303 {
2304         struct nvme_subsystem *subsys, *found;
2305         int ret;
2306
2307         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2308         if (!subsys)
2309                 return -ENOMEM;
2310         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2311         if (ret < 0) {
2312                 kfree(subsys);
2313                 return ret;
2314         }
2315         subsys->instance = ret;
2316         mutex_init(&subsys->lock);
2317         kref_init(&subsys->ref);
2318         INIT_LIST_HEAD(&subsys->ctrls);
2319         INIT_LIST_HEAD(&subsys->nsheads);
2320         nvme_init_subnqn(subsys, ctrl, id);
2321         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2322         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2323         subsys->vendor_id = le16_to_cpu(id->vid);
2324         subsys->cmic = id->cmic;
2325
2326         subsys->dev.class = nvme_subsys_class;
2327         subsys->dev.release = nvme_release_subsystem;
2328         subsys->dev.groups = nvme_subsys_attrs_groups;
2329         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2330         device_initialize(&subsys->dev);
2331
2332         mutex_lock(&nvme_subsystems_lock);
2333         found = __nvme_find_get_subsystem(subsys->subnqn);
2334         if (found) {
2335                 /*
2336                  * Verify that the subsystem actually supports multiple
2337                  * controllers, else bail out.
2338                  */
2339                 if (!(ctrl->opts && ctrl->opts->discovery_nqn) &&
2340                     nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2341                         dev_err(ctrl->device,
2342                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2343                                 found->subnqn);
2344                         nvme_put_subsystem(found);
2345                         ret = -EINVAL;
2346                         goto out_unlock;
2347                 }
2348
2349                 __nvme_release_subsystem(subsys);
2350                 subsys = found;
2351         } else {
2352                 ret = device_add(&subsys->dev);
2353                 if (ret) {
2354                         dev_err(ctrl->device,
2355                                 "failed to register subsystem device.\n");
2356                         goto out_unlock;
2357                 }
2358                 ida_init(&subsys->ns_ida);
2359                 list_add_tail(&subsys->entry, &nvme_subsystems);
2360         }
2361
2362         ctrl->subsys = subsys;
2363         mutex_unlock(&nvme_subsystems_lock);
2364
2365         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2366                         dev_name(ctrl->device))) {
2367                 dev_err(ctrl->device,
2368                         "failed to create sysfs link from subsystem.\n");
2369                 /* the transport driver will eventually put the subsystem */
2370                 return -EINVAL;
2371         }
2372
2373         mutex_lock(&subsys->lock);
2374         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2375         mutex_unlock(&subsys->lock);
2376
2377         return 0;
2378
2379 out_unlock:
2380         mutex_unlock(&nvme_subsystems_lock);
2381         put_device(&subsys->dev);
2382         return ret;
2383 }
2384
2385 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2386                 void *log, size_t size, u64 offset)
2387 {
2388         struct nvme_command c = { };
2389         unsigned long dwlen = size / 4 - 1;
2390
2391         c.get_log_page.opcode = nvme_admin_get_log_page;
2392         c.get_log_page.nsid = cpu_to_le32(nsid);
2393         c.get_log_page.lid = log_page;
2394         c.get_log_page.lsp = lsp;
2395         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2396         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2397         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2398         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2399
2400         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2401 }
2402
2403 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2404 {
2405         int ret;
2406
2407         if (!ctrl->effects)
2408                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2409
2410         if (!ctrl->effects)
2411                 return 0;
2412
2413         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2414                         ctrl->effects, sizeof(*ctrl->effects), 0);
2415         if (ret) {
2416                 kfree(ctrl->effects);
2417                 ctrl->effects = NULL;
2418         }
2419         return ret;
2420 }
2421
2422 /*
2423  * Initialize the cached copies of the Identify data and various controller
2424  * register in our nvme_ctrl structure.  This should be called as soon as
2425  * the admin queue is fully up and running.
2426  */
2427 int nvme_init_identify(struct nvme_ctrl *ctrl)
2428 {
2429         struct nvme_id_ctrl *id;
2430         u64 cap;
2431         int ret, page_shift;
2432         u32 max_hw_sectors;
2433         bool prev_apst_enabled;
2434
2435         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2436         if (ret) {
2437                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2438                 return ret;
2439         }
2440
2441         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2442         if (ret) {
2443                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2444                 return ret;
2445         }
2446         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2447
2448         if (ctrl->vs >= NVME_VS(1, 1, 0))
2449                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2450
2451         ret = nvme_identify_ctrl(ctrl, &id);
2452         if (ret) {
2453                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2454                 return -EIO;
2455         }
2456
2457         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2458                 ret = nvme_get_effects_log(ctrl);
2459                 if (ret < 0)
2460                         goto out_free;
2461         }
2462
2463         if (!ctrl->identified) {
2464                 int i;
2465
2466                 ret = nvme_init_subsystem(ctrl, id);
2467                 if (ret)
2468                         goto out_free;
2469
2470                 /*
2471                  * Check for quirks.  Quirk can depend on firmware version,
2472                  * so, in principle, the set of quirks present can change
2473                  * across a reset.  As a possible future enhancement, we
2474                  * could re-scan for quirks every time we reinitialize
2475                  * the device, but we'd have to make sure that the driver
2476                  * behaves intelligently if the quirks change.
2477                  */
2478                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2479                         if (quirk_matches(id, &core_quirks[i]))
2480                                 ctrl->quirks |= core_quirks[i].quirks;
2481                 }
2482         }
2483         memcpy(ctrl->subsys->firmware_rev, id->fr,
2484                sizeof(ctrl->subsys->firmware_rev));
2485
2486         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2487                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2488                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2489         }
2490
2491         ctrl->oacs = le16_to_cpu(id->oacs);
2492         ctrl->oncs = le16_to_cpup(&id->oncs);
2493         ctrl->oaes = le32_to_cpu(id->oaes);
2494         atomic_set(&ctrl->abort_limit, id->acl + 1);
2495         ctrl->vwc = id->vwc;
2496         ctrl->cntlid = le16_to_cpup(&id->cntlid);
2497         if (id->mdts)
2498                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2499         else
2500                 max_hw_sectors = UINT_MAX;
2501         ctrl->max_hw_sectors =
2502                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2503
2504         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2505         ctrl->sgls = le32_to_cpu(id->sgls);
2506         ctrl->kas = le16_to_cpu(id->kas);
2507         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2508
2509         if (id->rtd3e) {
2510                 /* us -> s */
2511                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2512
2513                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2514                                                  shutdown_timeout, 60);
2515
2516                 if (ctrl->shutdown_timeout != shutdown_timeout)
2517                         dev_info(ctrl->device,
2518                                  "Shutdown timeout set to %u seconds\n",
2519                                  ctrl->shutdown_timeout);
2520         } else
2521                 ctrl->shutdown_timeout = shutdown_timeout;
2522
2523         ctrl->npss = id->npss;
2524         ctrl->apsta = id->apsta;
2525         prev_apst_enabled = ctrl->apst_enabled;
2526         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2527                 if (force_apst && id->apsta) {
2528                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2529                         ctrl->apst_enabled = true;
2530                 } else {
2531                         ctrl->apst_enabled = false;
2532                 }
2533         } else {
2534                 ctrl->apst_enabled = id->apsta;
2535         }
2536         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2537
2538         if (ctrl->ops->flags & NVME_F_FABRICS) {
2539                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2540                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2541                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2542                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2543
2544                 /*
2545                  * In fabrics we need to verify the cntlid matches the
2546                  * admin connect
2547                  */
2548                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2549                         ret = -EINVAL;
2550                         goto out_free;
2551                 }
2552
2553                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2554                         dev_err(ctrl->device,
2555                                 "keep-alive support is mandatory for fabrics\n");
2556                         ret = -EINVAL;
2557                         goto out_free;
2558                 }
2559         } else {
2560                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2561                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2562                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2563                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2564                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2565         }
2566
2567         ret = nvme_mpath_init(ctrl, id);
2568         kfree(id);
2569
2570         if (ret < 0)
2571                 return ret;
2572
2573         if (ctrl->apst_enabled && !prev_apst_enabled)
2574                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2575         else if (!ctrl->apst_enabled && prev_apst_enabled)
2576                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2577
2578         ret = nvme_configure_apst(ctrl);
2579         if (ret < 0)
2580                 return ret;
2581
2582         ret = nvme_configure_timestamp(ctrl);
2583         if (ret < 0)
2584                 return ret;
2585
2586         ret = nvme_configure_directives(ctrl);
2587         if (ret < 0)
2588                 return ret;
2589
2590         ctrl->identified = true;
2591
2592         return 0;
2593
2594 out_free:
2595         kfree(id);
2596         return ret;
2597 }
2598 EXPORT_SYMBOL_GPL(nvme_init_identify);
2599
2600 static int nvme_dev_open(struct inode *inode, struct file *file)
2601 {
2602         struct nvme_ctrl *ctrl =
2603                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2604
2605         switch (ctrl->state) {
2606         case NVME_CTRL_LIVE:
2607         case NVME_CTRL_ADMIN_ONLY:
2608                 break;
2609         default:
2610                 return -EWOULDBLOCK;
2611         }
2612
2613         nvme_get_ctrl(ctrl);
2614         if (!try_module_get(ctrl->ops->module)) {
2615                 nvme_put_ctrl(ctrl);
2616                 return -EINVAL;
2617         }
2618
2619         file->private_data = ctrl;
2620         return 0;
2621 }
2622
2623 static int nvme_dev_release(struct inode *inode, struct file *file)
2624 {
2625         struct nvme_ctrl *ctrl =
2626                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2627
2628         module_put(ctrl->ops->module);
2629         nvme_put_ctrl(ctrl);
2630         return 0;
2631 }
2632
2633 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2634 {
2635         struct nvme_ns *ns;
2636         int ret;
2637
2638         down_read(&ctrl->namespaces_rwsem);
2639         if (list_empty(&ctrl->namespaces)) {
2640                 ret = -ENOTTY;
2641                 goto out_unlock;
2642         }
2643
2644         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2645         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2646                 dev_warn(ctrl->device,
2647                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2648                 ret = -EINVAL;
2649                 goto out_unlock;
2650         }
2651
2652         dev_warn(ctrl->device,
2653                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2654         kref_get(&ns->kref);
2655         up_read(&ctrl->namespaces_rwsem);
2656
2657         ret = nvme_user_cmd(ctrl, ns, argp);
2658         nvme_put_ns(ns);
2659         return ret;
2660
2661 out_unlock:
2662         up_read(&ctrl->namespaces_rwsem);
2663         return ret;
2664 }
2665
2666 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2667                 unsigned long arg)
2668 {
2669         struct nvme_ctrl *ctrl = file->private_data;
2670         void __user *argp = (void __user *)arg;
2671
2672         switch (cmd) {
2673         case NVME_IOCTL_ADMIN_CMD:
2674                 return nvme_user_cmd(ctrl, NULL, argp);
2675         case NVME_IOCTL_IO_CMD:
2676                 return nvme_dev_user_cmd(ctrl, argp);
2677         case NVME_IOCTL_RESET:
2678                 dev_warn(ctrl->device, "resetting controller\n");
2679                 return nvme_reset_ctrl_sync(ctrl);
2680         case NVME_IOCTL_SUBSYS_RESET:
2681                 return nvme_reset_subsystem(ctrl);
2682         case NVME_IOCTL_RESCAN:
2683                 nvme_queue_scan(ctrl);
2684                 return 0;
2685         default:
2686                 return -ENOTTY;
2687         }
2688 }
2689
2690 static const struct file_operations nvme_dev_fops = {
2691         .owner          = THIS_MODULE,
2692         .open           = nvme_dev_open,
2693         .release        = nvme_dev_release,
2694         .unlocked_ioctl = nvme_dev_ioctl,
2695         .compat_ioctl   = nvme_dev_ioctl,
2696 };
2697
2698 static ssize_t nvme_sysfs_reset(struct device *dev,
2699                                 struct device_attribute *attr, const char *buf,
2700                                 size_t count)
2701 {
2702         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2703         int ret;
2704
2705         ret = nvme_reset_ctrl_sync(ctrl);
2706         if (ret < 0)
2707                 return ret;
2708         return count;
2709 }
2710 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2711
2712 static ssize_t nvme_sysfs_rescan(struct device *dev,
2713                                 struct device_attribute *attr, const char *buf,
2714                                 size_t count)
2715 {
2716         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2717
2718         nvme_queue_scan(ctrl);
2719         return count;
2720 }
2721 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2722
2723 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2724 {
2725         struct gendisk *disk = dev_to_disk(dev);
2726
2727         if (disk->fops == &nvme_fops)
2728                 return nvme_get_ns_from_dev(dev)->head;
2729         else
2730                 return disk->private_data;
2731 }
2732
2733 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2734                 char *buf)
2735 {
2736         struct nvme_ns_head *head = dev_to_ns_head(dev);
2737         struct nvme_ns_ids *ids = &head->ids;
2738         struct nvme_subsystem *subsys = head->subsys;
2739         int serial_len = sizeof(subsys->serial);
2740         int model_len = sizeof(subsys->model);
2741
2742         if (!uuid_is_null(&ids->uuid))
2743                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2744
2745         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2746                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2747
2748         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2749                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2750
2751         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2752                                   subsys->serial[serial_len - 1] == '\0'))
2753                 serial_len--;
2754         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2755                                  subsys->model[model_len - 1] == '\0'))
2756                 model_len--;
2757
2758         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2759                 serial_len, subsys->serial, model_len, subsys->model,
2760                 head->ns_id);
2761 }
2762 static DEVICE_ATTR_RO(wwid);
2763
2764 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2765                 char *buf)
2766 {
2767         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2768 }
2769 static DEVICE_ATTR_RO(nguid);
2770
2771 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2772                 char *buf)
2773 {
2774         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2775
2776         /* For backward compatibility expose the NGUID to userspace if
2777          * we have no UUID set
2778          */
2779         if (uuid_is_null(&ids->uuid)) {
2780                 printk_ratelimited(KERN_WARNING
2781                                    "No UUID available providing old NGUID\n");
2782                 return sprintf(buf, "%pU\n", ids->nguid);
2783         }
2784         return sprintf(buf, "%pU\n", &ids->uuid);
2785 }
2786 static DEVICE_ATTR_RO(uuid);
2787
2788 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2789                 char *buf)
2790 {
2791         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2792 }
2793 static DEVICE_ATTR_RO(eui);
2794
2795 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2796                 char *buf)
2797 {
2798         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2799 }
2800 static DEVICE_ATTR_RO(nsid);
2801
2802 static struct attribute *nvme_ns_id_attrs[] = {
2803         &dev_attr_wwid.attr,
2804         &dev_attr_uuid.attr,
2805         &dev_attr_nguid.attr,
2806         &dev_attr_eui.attr,
2807         &dev_attr_nsid.attr,
2808 #ifdef CONFIG_NVME_MULTIPATH
2809         &dev_attr_ana_grpid.attr,
2810         &dev_attr_ana_state.attr,
2811 #endif
2812         NULL,
2813 };
2814
2815 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2816                 struct attribute *a, int n)
2817 {
2818         struct device *dev = container_of(kobj, struct device, kobj);
2819         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2820
2821         if (a == &dev_attr_uuid.attr) {
2822                 if (uuid_is_null(&ids->uuid) &&
2823                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2824                         return 0;
2825         }
2826         if (a == &dev_attr_nguid.attr) {
2827                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2828                         return 0;
2829         }
2830         if (a == &dev_attr_eui.attr) {
2831                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2832                         return 0;
2833         }
2834 #ifdef CONFIG_NVME_MULTIPATH
2835         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2836                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2837                         return 0;
2838                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2839                         return 0;
2840         }
2841 #endif
2842         return a->mode;
2843 }
2844
2845 const struct attribute_group nvme_ns_id_attr_group = {
2846         .attrs          = nvme_ns_id_attrs,
2847         .is_visible     = nvme_ns_id_attrs_are_visible,
2848 };
2849
2850 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2851         &nvme_ns_id_attr_group,
2852 #ifdef CONFIG_NVM
2853         &nvme_nvm_attr_group,
2854 #endif
2855         NULL,
2856 };
2857
2858 #define nvme_show_str_function(field)                                           \
2859 static ssize_t  field##_show(struct device *dev,                                \
2860                             struct device_attribute *attr, char *buf)           \
2861 {                                                                               \
2862         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2863         return sprintf(buf, "%.*s\n",                                           \
2864                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2865 }                                                                               \
2866 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2867
2868 nvme_show_str_function(model);
2869 nvme_show_str_function(serial);
2870 nvme_show_str_function(firmware_rev);
2871
2872 #define nvme_show_int_function(field)                                           \
2873 static ssize_t  field##_show(struct device *dev,                                \
2874                             struct device_attribute *attr, char *buf)           \
2875 {                                                                               \
2876         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2877         return sprintf(buf, "%d\n", ctrl->field);       \
2878 }                                                                               \
2879 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2880
2881 nvme_show_int_function(cntlid);
2882
2883 static ssize_t nvme_sysfs_delete(struct device *dev,
2884                                 struct device_attribute *attr, const char *buf,
2885                                 size_t count)
2886 {
2887         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2888
2889         /* Can't delete non-created controllers */
2890         if (!ctrl->created)
2891                 return -EBUSY;
2892
2893         if (device_remove_file_self(dev, attr))
2894                 nvme_delete_ctrl_sync(ctrl);
2895         return count;
2896 }
2897 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2898
2899 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2900                                          struct device_attribute *attr,
2901                                          char *buf)
2902 {
2903         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2904
2905         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2906 }
2907 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2908
2909 static ssize_t nvme_sysfs_show_state(struct device *dev,
2910                                      struct device_attribute *attr,
2911                                      char *buf)
2912 {
2913         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2914         static const char *const state_name[] = {
2915                 [NVME_CTRL_NEW]         = "new",
2916                 [NVME_CTRL_LIVE]        = "live",
2917                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2918                 [NVME_CTRL_RESETTING]   = "resetting",
2919                 [NVME_CTRL_CONNECTING]  = "connecting",
2920                 [NVME_CTRL_DELETING]    = "deleting",
2921                 [NVME_CTRL_DEAD]        = "dead",
2922         };
2923
2924         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2925             state_name[ctrl->state])
2926                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2927
2928         return sprintf(buf, "unknown state\n");
2929 }
2930
2931 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2932
2933 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2934                                          struct device_attribute *attr,
2935                                          char *buf)
2936 {
2937         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2938
2939         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2940 }
2941 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2942
2943 static ssize_t nvme_sysfs_show_address(struct device *dev,
2944                                          struct device_attribute *attr,
2945                                          char *buf)
2946 {
2947         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2948
2949         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2950 }
2951 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2952
2953 static struct attribute *nvme_dev_attrs[] = {
2954         &dev_attr_reset_controller.attr,
2955         &dev_attr_rescan_controller.attr,
2956         &dev_attr_model.attr,
2957         &dev_attr_serial.attr,
2958         &dev_attr_firmware_rev.attr,
2959         &dev_attr_cntlid.attr,
2960         &dev_attr_delete_controller.attr,
2961         &dev_attr_transport.attr,
2962         &dev_attr_subsysnqn.attr,
2963         &dev_attr_address.attr,
2964         &dev_attr_state.attr,
2965         NULL
2966 };
2967
2968 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2969                 struct attribute *a, int n)
2970 {
2971         struct device *dev = container_of(kobj, struct device, kobj);
2972         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2973
2974         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2975                 return 0;
2976         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2977                 return 0;
2978
2979         return a->mode;
2980 }
2981
2982 static struct attribute_group nvme_dev_attrs_group = {
2983         .attrs          = nvme_dev_attrs,
2984         .is_visible     = nvme_dev_attrs_are_visible,
2985 };
2986
2987 static const struct attribute_group *nvme_dev_attr_groups[] = {
2988         &nvme_dev_attrs_group,
2989         NULL,
2990 };
2991
2992 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2993                 unsigned nsid)
2994 {
2995         struct nvme_ns_head *h;
2996
2997         lockdep_assert_held(&subsys->lock);
2998
2999         list_for_each_entry(h, &subsys->nsheads, entry) {
3000                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3001                         return h;
3002         }
3003
3004         return NULL;
3005 }
3006
3007 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3008                 struct nvme_ns_head *new)
3009 {
3010         struct nvme_ns_head *h;
3011
3012         lockdep_assert_held(&subsys->lock);
3013
3014         list_for_each_entry(h, &subsys->nsheads, entry) {
3015                 if (nvme_ns_ids_valid(&new->ids) &&
3016                     !list_empty(&h->list) &&
3017                     nvme_ns_ids_equal(&new->ids, &h->ids))
3018                         return -EINVAL;
3019         }
3020
3021         return 0;
3022 }
3023
3024 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3025                 unsigned nsid, struct nvme_id_ns *id)
3026 {
3027         struct nvme_ns_head *head;
3028         int ret = -ENOMEM;
3029
3030         head = kzalloc(sizeof(*head), GFP_KERNEL);
3031         if (!head)
3032                 goto out;
3033         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3034         if (ret < 0)
3035                 goto out_free_head;
3036         head->instance = ret;
3037         INIT_LIST_HEAD(&head->list);
3038         ret = init_srcu_struct(&head->srcu);
3039         if (ret)
3040                 goto out_ida_remove;
3041         head->subsys = ctrl->subsys;
3042         head->ns_id = nsid;
3043         kref_init(&head->ref);
3044
3045         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3046
3047         ret = __nvme_check_ids(ctrl->subsys, head);
3048         if (ret) {
3049                 dev_err(ctrl->device,
3050                         "duplicate IDs for nsid %d\n", nsid);
3051                 goto out_cleanup_srcu;
3052         }
3053
3054         ret = nvme_mpath_alloc_disk(ctrl, head);
3055         if (ret)
3056                 goto out_cleanup_srcu;
3057
3058         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3059
3060         kref_get(&ctrl->subsys->ref);
3061
3062         return head;
3063 out_cleanup_srcu:
3064         cleanup_srcu_struct(&head->srcu);
3065 out_ida_remove:
3066         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3067 out_free_head:
3068         kfree(head);
3069 out:
3070         return ERR_PTR(ret);
3071 }
3072
3073 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3074                 struct nvme_id_ns *id)
3075 {
3076         struct nvme_ctrl *ctrl = ns->ctrl;
3077         bool is_shared = id->nmic & (1 << 0);
3078         struct nvme_ns_head *head = NULL;
3079         int ret = 0;
3080
3081         mutex_lock(&ctrl->subsys->lock);
3082         if (is_shared)
3083                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3084         if (!head) {
3085                 head = nvme_alloc_ns_head(ctrl, nsid, id);
3086                 if (IS_ERR(head)) {
3087                         ret = PTR_ERR(head);
3088                         goto out_unlock;
3089                 }
3090         } else {
3091                 struct nvme_ns_ids ids;
3092
3093                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3094                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3095                         dev_err(ctrl->device,
3096                                 "IDs don't match for shared namespace %d\n",
3097                                         nsid);
3098                         ret = -EINVAL;
3099                         goto out_unlock;
3100                 }
3101         }
3102
3103         list_add_tail(&ns->siblings, &head->list);
3104         ns->head = head;
3105
3106 out_unlock:
3107         mutex_unlock(&ctrl->subsys->lock);
3108         return ret;
3109 }
3110
3111 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3112 {
3113         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3114         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3115
3116         return nsa->head->ns_id - nsb->head->ns_id;
3117 }
3118
3119 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3120 {
3121         struct nvme_ns *ns, *ret = NULL;
3122
3123         down_read(&ctrl->namespaces_rwsem);
3124         list_for_each_entry(ns, &ctrl->namespaces, list) {
3125                 if (ns->head->ns_id == nsid) {
3126                         if (!kref_get_unless_zero(&ns->kref))
3127                                 continue;
3128                         ret = ns;
3129                         break;
3130                 }
3131                 if (ns->head->ns_id > nsid)
3132                         break;
3133         }
3134         up_read(&ctrl->namespaces_rwsem);
3135         return ret;
3136 }
3137
3138 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3139 {
3140         struct streams_directive_params s;
3141         int ret;
3142
3143         if (!ctrl->nr_streams)
3144                 return 0;
3145
3146         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3147         if (ret)
3148                 return ret;
3149
3150         ns->sws = le32_to_cpu(s.sws);
3151         ns->sgs = le16_to_cpu(s.sgs);
3152
3153         if (ns->sws) {
3154                 unsigned int bs = 1 << ns->lba_shift;
3155
3156                 blk_queue_io_min(ns->queue, bs * ns->sws);
3157                 if (ns->sgs)
3158                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3159         }
3160
3161         return 0;
3162 }
3163
3164 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3165 {
3166         struct nvme_ns *ns;
3167         struct gendisk *disk;
3168         struct nvme_id_ns *id;
3169         char disk_name[DISK_NAME_LEN];
3170         int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
3171
3172         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3173         if (!ns)
3174                 return;
3175
3176         ns->queue = blk_mq_init_queue(ctrl->tagset);
3177         if (IS_ERR(ns->queue))
3178                 goto out_free_ns;
3179         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3180         ns->queue->queuedata = ns;
3181         ns->ctrl = ctrl;
3182
3183         kref_init(&ns->kref);
3184         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3185
3186         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3187         nvme_set_queue_limits(ctrl, ns->queue);
3188
3189         id = nvme_identify_ns(ctrl, nsid);
3190         if (!id)
3191                 goto out_free_queue;
3192
3193         if (id->ncap == 0)
3194                 goto out_free_id;
3195
3196         if (nvme_init_ns_head(ns, nsid, id))
3197                 goto out_free_id;
3198         nvme_setup_streams_ns(ctrl, ns);
3199         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3200
3201         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3202                 if (nvme_nvm_register(ns, disk_name, node)) {
3203                         dev_warn(ctrl->device, "LightNVM init failure\n");
3204                         goto out_unlink_ns;
3205                 }
3206         }
3207
3208         disk = alloc_disk_node(0, node);
3209         if (!disk)
3210                 goto out_unlink_ns;
3211
3212         disk->fops = &nvme_fops;
3213         disk->private_data = ns;
3214         disk->queue = ns->queue;
3215         disk->flags = flags;
3216         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3217         ns->disk = disk;
3218
3219         __nvme_revalidate_disk(disk, id);
3220
3221         down_write(&ctrl->namespaces_rwsem);
3222         list_add_tail(&ns->list, &ctrl->namespaces);
3223         up_write(&ctrl->namespaces_rwsem);
3224
3225         nvme_get_ctrl(ctrl);
3226
3227         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3228
3229         nvme_mpath_add_disk(ns, id);
3230         nvme_fault_inject_init(ns);
3231         kfree(id);
3232
3233         return;
3234  out_unlink_ns:
3235         mutex_lock(&ctrl->subsys->lock);
3236         list_del_rcu(&ns->siblings);
3237         mutex_unlock(&ctrl->subsys->lock);
3238  out_free_id:
3239         kfree(id);
3240  out_free_queue:
3241         blk_cleanup_queue(ns->queue);
3242  out_free_ns:
3243         kfree(ns);
3244 }
3245
3246 static void nvme_ns_remove(struct nvme_ns *ns)
3247 {
3248         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3249                 return;
3250
3251         nvme_fault_inject_fini(ns);
3252
3253         mutex_lock(&ns->ctrl->subsys->lock);
3254         list_del_rcu(&ns->siblings);
3255         mutex_unlock(&ns->ctrl->subsys->lock);
3256         synchronize_rcu(); /* guarantee not available in head->list */
3257         nvme_mpath_clear_current_path(ns);
3258         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3259
3260         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3261                 del_gendisk(ns->disk);
3262                 blk_cleanup_queue(ns->queue);
3263                 if (blk_get_integrity(ns->disk))
3264                         blk_integrity_unregister(ns->disk);
3265         }
3266
3267         down_write(&ns->ctrl->namespaces_rwsem);
3268         list_del_init(&ns->list);
3269         up_write(&ns->ctrl->namespaces_rwsem);
3270
3271         nvme_mpath_check_last_path(ns);
3272         nvme_put_ns(ns);
3273 }
3274
3275 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3276 {
3277         struct nvme_ns *ns;
3278
3279         ns = nvme_find_get_ns(ctrl, nsid);
3280         if (ns) {
3281                 if (ns->disk && revalidate_disk(ns->disk))
3282                         nvme_ns_remove(ns);
3283                 nvme_put_ns(ns);
3284         } else
3285                 nvme_alloc_ns(ctrl, nsid);
3286 }
3287
3288 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3289                                         unsigned nsid)
3290 {
3291         struct nvme_ns *ns, *next;
3292         LIST_HEAD(rm_list);
3293
3294         down_write(&ctrl->namespaces_rwsem);
3295         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3296                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3297                         list_move_tail(&ns->list, &rm_list);
3298         }
3299         up_write(&ctrl->namespaces_rwsem);
3300
3301         list_for_each_entry_safe(ns, next, &rm_list, list)
3302                 nvme_ns_remove(ns);
3303
3304 }
3305
3306 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3307 {
3308         struct nvme_ns *ns;
3309         __le32 *ns_list;
3310         unsigned i, j, nsid, prev = 0;
3311         unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3312         int ret = 0;
3313
3314         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3315         if (!ns_list)
3316                 return -ENOMEM;
3317
3318         for (i = 0; i < num_lists; i++) {
3319                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3320                 if (ret)
3321                         goto free;
3322
3323                 for (j = 0; j < min(nn, 1024U); j++) {
3324                         nsid = le32_to_cpu(ns_list[j]);
3325                         if (!nsid)
3326                                 goto out;
3327
3328                         nvme_validate_ns(ctrl, nsid);
3329
3330                         while (++prev < nsid) {
3331                                 ns = nvme_find_get_ns(ctrl, prev);
3332                                 if (ns) {
3333                                         nvme_ns_remove(ns);
3334                                         nvme_put_ns(ns);
3335                                 }
3336                         }
3337                 }
3338                 nn -= j;
3339         }
3340  out:
3341         nvme_remove_invalid_namespaces(ctrl, prev);
3342  free:
3343         kfree(ns_list);
3344         return ret;
3345 }
3346
3347 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3348 {
3349         unsigned i;
3350
3351         for (i = 1; i <= nn; i++)
3352                 nvme_validate_ns(ctrl, i);
3353
3354         nvme_remove_invalid_namespaces(ctrl, nn);
3355 }
3356
3357 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3358 {
3359         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3360         __le32 *log;
3361         int error;
3362
3363         log = kzalloc(log_size, GFP_KERNEL);
3364         if (!log)
3365                 return;
3366
3367         /*
3368          * We need to read the log to clear the AEN, but we don't want to rely
3369          * on it for the changed namespace information as userspace could have
3370          * raced with us in reading the log page, which could cause us to miss
3371          * updates.
3372          */
3373         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3374                         log_size, 0);
3375         if (error)
3376                 dev_warn(ctrl->device,
3377                         "reading changed ns log failed: %d\n", error);
3378
3379         kfree(log);
3380 }
3381
3382 static void nvme_scan_work(struct work_struct *work)
3383 {
3384         struct nvme_ctrl *ctrl =
3385                 container_of(work, struct nvme_ctrl, scan_work);
3386         struct nvme_id_ctrl *id;
3387         unsigned nn;
3388
3389         if (ctrl->state != NVME_CTRL_LIVE)
3390                 return;
3391
3392         WARN_ON_ONCE(!ctrl->tagset);
3393
3394         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3395                 dev_info(ctrl->device, "rescanning namespaces.\n");
3396                 nvme_clear_changed_ns_log(ctrl);
3397         }
3398
3399         if (nvme_identify_ctrl(ctrl, &id))
3400                 return;
3401
3402         mutex_lock(&ctrl->scan_lock);
3403         nn = le32_to_cpu(id->nn);
3404         if (!nvme_ctrl_limited_cns(ctrl)) {
3405                 if (!nvme_scan_ns_list(ctrl, nn))
3406                         goto out_free_id;
3407         }
3408         nvme_scan_ns_sequential(ctrl, nn);
3409 out_free_id:
3410         mutex_unlock(&ctrl->scan_lock);
3411         kfree(id);
3412         down_write(&ctrl->namespaces_rwsem);
3413         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3414         up_write(&ctrl->namespaces_rwsem);
3415 }
3416
3417 /*
3418  * This function iterates the namespace list unlocked to allow recovery from
3419  * controller failure. It is up to the caller to ensure the namespace list is
3420  * not modified by scan work while this function is executing.
3421  */
3422 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3423 {
3424         struct nvme_ns *ns, *next;
3425         LIST_HEAD(ns_list);
3426
3427         /* prevent racing with ns scanning */
3428         flush_work(&ctrl->scan_work);
3429
3430         /*
3431          * The dead states indicates the controller was not gracefully
3432          * disconnected. In that case, we won't be able to flush any data while
3433          * removing the namespaces' disks; fail all the queues now to avoid
3434          * potentially having to clean up the failed sync later.
3435          */
3436         if (ctrl->state == NVME_CTRL_DEAD)
3437                 nvme_kill_queues(ctrl);
3438
3439         down_write(&ctrl->namespaces_rwsem);
3440         list_splice_init(&ctrl->namespaces, &ns_list);
3441         up_write(&ctrl->namespaces_rwsem);
3442
3443         list_for_each_entry_safe(ns, next, &ns_list, list)
3444                 nvme_ns_remove(ns);
3445 }
3446 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3447
3448 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3449 {
3450         char *envp[2] = { NULL, NULL };
3451         u32 aen_result = ctrl->aen_result;
3452
3453         ctrl->aen_result = 0;
3454         if (!aen_result)
3455                 return;
3456
3457         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3458         if (!envp[0])
3459                 return;
3460         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3461         kfree(envp[0]);
3462 }
3463
3464 static void nvme_async_event_work(struct work_struct *work)
3465 {
3466         struct nvme_ctrl *ctrl =
3467                 container_of(work, struct nvme_ctrl, async_event_work);
3468
3469         nvme_aen_uevent(ctrl);
3470
3471         /*
3472          * The transport drivers must guarantee AER submission here is safe by
3473          * flushing ctrl async_event_work after changing the controller state
3474          * from LIVE and before freeing the admin queue.
3475         */
3476         if (ctrl->state == NVME_CTRL_LIVE)
3477                 ctrl->ops->submit_async_event(ctrl);
3478 }
3479
3480 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3481 {
3482
3483         u32 csts;
3484
3485         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3486                 return false;
3487
3488         if (csts == ~0)
3489                 return false;
3490
3491         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3492 }
3493
3494 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3495 {
3496         struct nvme_fw_slot_info_log *log;
3497
3498         log = kmalloc(sizeof(*log), GFP_KERNEL);
3499         if (!log)
3500                 return;
3501
3502         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, log,
3503                         sizeof(*log), 0))
3504                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3505         kfree(log);
3506 }
3507
3508 static void nvme_fw_act_work(struct work_struct *work)
3509 {
3510         struct nvme_ctrl *ctrl = container_of(work,
3511                                 struct nvme_ctrl, fw_act_work);
3512         unsigned long fw_act_timeout;
3513
3514         if (ctrl->mtfa)
3515                 fw_act_timeout = jiffies +
3516                                 msecs_to_jiffies(ctrl->mtfa * 100);
3517         else
3518                 fw_act_timeout = jiffies +
3519                                 msecs_to_jiffies(admin_timeout * 1000);
3520
3521         nvme_stop_queues(ctrl);
3522         while (nvme_ctrl_pp_status(ctrl)) {
3523                 if (time_after(jiffies, fw_act_timeout)) {
3524                         dev_warn(ctrl->device,
3525                                 "Fw activation timeout, reset controller\n");
3526                         nvme_reset_ctrl(ctrl);
3527                         break;
3528                 }
3529                 msleep(100);
3530         }
3531
3532         if (ctrl->state != NVME_CTRL_LIVE)
3533                 return;
3534
3535         nvme_start_queues(ctrl);
3536         /* read FW slot information to clear the AER */
3537         nvme_get_fw_slot_info(ctrl);
3538 }
3539
3540 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3541 {
3542         switch ((result & 0xff00) >> 8) {
3543         case NVME_AER_NOTICE_NS_CHANGED:
3544                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3545                 nvme_queue_scan(ctrl);
3546                 break;
3547         case NVME_AER_NOTICE_FW_ACT_STARTING:
3548                 queue_work(nvme_wq, &ctrl->fw_act_work);
3549                 break;
3550 #ifdef CONFIG_NVME_MULTIPATH
3551         case NVME_AER_NOTICE_ANA:
3552                 if (!ctrl->ana_log_buf)
3553                         break;
3554                 queue_work(nvme_wq, &ctrl->ana_work);
3555                 break;
3556 #endif
3557         default:
3558                 dev_warn(ctrl->device, "async event result %08x\n", result);
3559         }
3560 }
3561
3562 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3563                 volatile union nvme_result *res)
3564 {
3565         u32 result = le32_to_cpu(res->u32);
3566
3567         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3568                 return;
3569
3570         switch (result & 0x7) {
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                 ctrl->aen_result = result;
3579                 break;
3580         default:
3581                 break;
3582         }
3583         queue_work(nvme_wq, &ctrl->async_event_work);
3584 }
3585 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3586
3587 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3588 {
3589         nvme_mpath_stop(ctrl);
3590         nvme_stop_keep_alive(ctrl);
3591         flush_work(&ctrl->async_event_work);
3592         cancel_work_sync(&ctrl->fw_act_work);
3593         if (ctrl->ops->stop_ctrl)
3594                 ctrl->ops->stop_ctrl(ctrl);
3595 }
3596 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3597
3598 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3599 {
3600         if (ctrl->kato)
3601                 nvme_start_keep_alive(ctrl);
3602
3603         if (ctrl->queue_count > 1) {
3604                 nvme_queue_scan(ctrl);
3605                 nvme_enable_aen(ctrl);
3606                 queue_work(nvme_wq, &ctrl->async_event_work);
3607                 nvme_start_queues(ctrl);
3608         }
3609         ctrl->created = true;
3610 }
3611 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3612
3613 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3614 {
3615         dev_pm_qos_hide_latency_tolerance(ctrl->device);
3616         cdev_device_del(&ctrl->cdev, ctrl->device);
3617 }
3618 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3619
3620 static void nvme_free_ctrl(struct device *dev)
3621 {
3622         struct nvme_ctrl *ctrl =
3623                 container_of(dev, struct nvme_ctrl, ctrl_device);
3624         struct nvme_subsystem *subsys = ctrl->subsys;
3625
3626         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3627         kfree(ctrl->effects);
3628         nvme_mpath_uninit(ctrl);
3629         __free_page(ctrl->discard_page);
3630
3631         if (subsys) {
3632                 mutex_lock(&subsys->lock);
3633                 list_del(&ctrl->subsys_entry);
3634                 mutex_unlock(&subsys->lock);
3635                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3636         }
3637
3638         ctrl->ops->free_ctrl(ctrl);
3639
3640         if (subsys)
3641                 nvme_put_subsystem(subsys);
3642 }
3643
3644 /*
3645  * Initialize a NVMe controller structures.  This needs to be called during
3646  * earliest initialization so that we have the initialized structured around
3647  * during probing.
3648  */
3649 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3650                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3651 {
3652         int ret;
3653
3654         ctrl->state = NVME_CTRL_NEW;
3655         spin_lock_init(&ctrl->lock);
3656         mutex_init(&ctrl->scan_lock);
3657         INIT_LIST_HEAD(&ctrl->namespaces);
3658         init_rwsem(&ctrl->namespaces_rwsem);
3659         ctrl->dev = dev;
3660         ctrl->ops = ops;
3661         ctrl->quirks = quirks;
3662         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3663         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3664         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3665         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3666
3667         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3668         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3669         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3670
3671         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3672                         PAGE_SIZE);
3673         ctrl->discard_page = alloc_page(GFP_KERNEL);
3674         if (!ctrl->discard_page) {
3675                 ret = -ENOMEM;
3676                 goto out;
3677         }
3678
3679         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3680         if (ret < 0)
3681                 goto out;
3682         ctrl->instance = ret;
3683
3684         device_initialize(&ctrl->ctrl_device);
3685         ctrl->device = &ctrl->ctrl_device;
3686         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3687         ctrl->device->class = nvme_class;
3688         ctrl->device->parent = ctrl->dev;
3689         ctrl->device->groups = nvme_dev_attr_groups;
3690         ctrl->device->release = nvme_free_ctrl;
3691         dev_set_drvdata(ctrl->device, ctrl);
3692         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3693         if (ret)
3694                 goto out_release_instance;
3695
3696         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3697         ctrl->cdev.owner = ops->module;
3698         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3699         if (ret)
3700                 goto out_free_name;
3701
3702         /*
3703          * Initialize latency tolerance controls.  The sysfs files won't
3704          * be visible to userspace unless the device actually supports APST.
3705          */
3706         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3707         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3708                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3709
3710         return 0;
3711 out_free_name:
3712         kfree_const(ctrl->device->kobj.name);
3713 out_release_instance:
3714         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3715 out:
3716         if (ctrl->discard_page)
3717                 __free_page(ctrl->discard_page);
3718         return ret;
3719 }
3720 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3721
3722 /**
3723  * nvme_kill_queues(): Ends all namespace queues
3724  * @ctrl: the dead controller that needs to end
3725  *
3726  * Call this function when the driver determines it is unable to get the
3727  * controller in a state capable of servicing IO.
3728  */
3729 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3730 {
3731         struct nvme_ns *ns;
3732
3733         down_read(&ctrl->namespaces_rwsem);
3734
3735         /* Forcibly unquiesce queues to avoid blocking dispatch */
3736         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3737                 blk_mq_unquiesce_queue(ctrl->admin_q);
3738
3739         list_for_each_entry(ns, &ctrl->namespaces, list)
3740                 nvme_set_queue_dying(ns);
3741
3742         up_read(&ctrl->namespaces_rwsem);
3743 }
3744 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3745
3746 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3747 {
3748         struct nvme_ns *ns;
3749
3750         down_read(&ctrl->namespaces_rwsem);
3751         list_for_each_entry(ns, &ctrl->namespaces, list)
3752                 blk_mq_unfreeze_queue(ns->queue);
3753         up_read(&ctrl->namespaces_rwsem);
3754 }
3755 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3756
3757 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3758 {
3759         struct nvme_ns *ns;
3760
3761         down_read(&ctrl->namespaces_rwsem);
3762         list_for_each_entry(ns, &ctrl->namespaces, list) {
3763                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3764                 if (timeout <= 0)
3765                         break;
3766         }
3767         up_read(&ctrl->namespaces_rwsem);
3768 }
3769 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3770
3771 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3772 {
3773         struct nvme_ns *ns;
3774
3775         down_read(&ctrl->namespaces_rwsem);
3776         list_for_each_entry(ns, &ctrl->namespaces, list)
3777                 blk_mq_freeze_queue_wait(ns->queue);
3778         up_read(&ctrl->namespaces_rwsem);
3779 }
3780 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3781
3782 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3783 {
3784         struct nvme_ns *ns;
3785
3786         down_read(&ctrl->namespaces_rwsem);
3787         list_for_each_entry(ns, &ctrl->namespaces, list)
3788                 blk_freeze_queue_start(ns->queue);
3789         up_read(&ctrl->namespaces_rwsem);
3790 }
3791 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3792
3793 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3794 {
3795         struct nvme_ns *ns;
3796
3797         down_read(&ctrl->namespaces_rwsem);
3798         list_for_each_entry(ns, &ctrl->namespaces, list)
3799                 blk_mq_quiesce_queue(ns->queue);
3800         up_read(&ctrl->namespaces_rwsem);
3801 }
3802 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3803
3804 void nvme_start_queues(struct nvme_ctrl *ctrl)
3805 {
3806         struct nvme_ns *ns;
3807
3808         down_read(&ctrl->namespaces_rwsem);
3809         list_for_each_entry(ns, &ctrl->namespaces, list)
3810                 blk_mq_unquiesce_queue(ns->queue);
3811         up_read(&ctrl->namespaces_rwsem);
3812 }
3813 EXPORT_SYMBOL_GPL(nvme_start_queues);
3814
3815 int __init nvme_core_init(void)
3816 {
3817         int result = -ENOMEM;
3818
3819         nvme_wq = alloc_workqueue("nvme-wq",
3820                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3821         if (!nvme_wq)
3822                 goto out;
3823
3824         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3825                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3826         if (!nvme_reset_wq)
3827                 goto destroy_wq;
3828
3829         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3830                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3831         if (!nvme_delete_wq)
3832                 goto destroy_reset_wq;
3833
3834         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3835         if (result < 0)
3836                 goto destroy_delete_wq;
3837
3838         nvme_class = class_create(THIS_MODULE, "nvme");
3839         if (IS_ERR(nvme_class)) {
3840                 result = PTR_ERR(nvme_class);
3841                 goto unregister_chrdev;
3842         }
3843
3844         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3845         if (IS_ERR(nvme_subsys_class)) {
3846                 result = PTR_ERR(nvme_subsys_class);
3847                 goto destroy_class;
3848         }
3849         return 0;
3850
3851 destroy_class:
3852         class_destroy(nvme_class);
3853 unregister_chrdev:
3854         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3855 destroy_delete_wq:
3856         destroy_workqueue(nvme_delete_wq);
3857 destroy_reset_wq:
3858         destroy_workqueue(nvme_reset_wq);
3859 destroy_wq:
3860         destroy_workqueue(nvme_wq);
3861 out:
3862         return result;
3863 }
3864
3865 void nvme_core_exit(void)
3866 {
3867         ida_destroy(&nvme_subsystems_ida);
3868         class_destroy(nvme_subsys_class);
3869         class_destroy(nvme_class);
3870         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3871         destroy_workqueue(nvme_delete_wq);
3872         destroy_workqueue(nvme_reset_wq);
3873         destroy_workqueue(nvme_wq);
3874 }
3875
3876 MODULE_LICENSE("GPL");
3877 MODULE_VERSION("1.0");
3878 module_init(nvme_core_init);
3879 module_exit(nvme_core_exit);