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         revalidate_disk(ns->disk);
 115         blk_set_queue_dying(ns->queue);
 116         /* Forcibly unquiesce queues to avoid blocking dispatch */
 117         blk_mq_unquiesce_queue(ns->queue);
 118 }
 119
 120 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 121 {
 122         /*
 123          * Only new queue scan work when admin and IO queues are both alive
 124          */
 125         if (ctrl->state == NVME_CTRL_LIVE)
 126                 queue_work(nvme_wq, &ctrl->scan_work);
 127 }
 128
 129 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 130 {
 131         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 132                 return -EBUSY;
 133         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 134                 return -EBUSY;
 135         return 0;
 136 }
 137 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 138
 139 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 140 {
 141         int ret;
 142
 143         ret = nvme_reset_ctrl(ctrl);
 144         if (!ret) {
 145                 flush_work(&ctrl->reset_work);
 146                 if (ctrl->state != NVME_CTRL_LIVE &&
 147                     ctrl->state != NVME_CTRL_ADMIN_ONLY)
 148                         ret = -ENETRESET;
 149         }
 150
 151         return ret;
 152 }
 153 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 154
 155 static void nvme_delete_ctrl_work(struct work_struct *work)
 156 {
 157         struct nvme_ctrl *ctrl =
 158                 container_of(work, struct nvme_ctrl, delete_work);
 159
 160         dev_info(ctrl->device,
 161                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 162
 163         flush_work(&ctrl->reset_work);
 164         nvme_stop_ctrl(ctrl);
 165         nvme_remove_namespaces(ctrl);
 166         ctrl->ops->delete_ctrl(ctrl);
 167         nvme_uninit_ctrl(ctrl);
 168         nvme_put_ctrl(ctrl);
 169 }
 170
 171 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 172 {
 173         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 174                 return -EBUSY;
 175         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 176                 return -EBUSY;
 177         return 0;
 178 }
 179 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 180
 181 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 182 {
 183         int ret = 0;
 184
 185         /*
 186          * Keep a reference until the work is flushed since ->delete_ctrl
 187          * can free the controller.
 188          */
 189         nvme_get_ctrl(ctrl);
 190         ret = nvme_delete_ctrl(ctrl);
 191         if (!ret)
 192                 flush_work(&ctrl->delete_work);
 193         nvme_put_ctrl(ctrl);
 194         return ret;
 195 }
 196 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
 197
 198 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 199 {
 200         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 201 }
 202
 203 static blk_status_t nvme_error_status(struct request *req)
 204 {
 205         switch (nvme_req(req)->status & 0x7ff) {
 206         case NVME_SC_SUCCESS:
 207                 return BLK_STS_OK;
 208         case NVME_SC_CAP_EXCEEDED:
 209                 return BLK_STS_NOSPC;
 210         case NVME_SC_LBA_RANGE:
 211                 return BLK_STS_TARGET;
 212         case NVME_SC_BAD_ATTRIBUTES:
 213         case NVME_SC_ONCS_NOT_SUPPORTED:
 214         case NVME_SC_INVALID_OPCODE:
 215         case NVME_SC_INVALID_FIELD:
 216         case NVME_SC_INVALID_NS:
 217                 return BLK_STS_NOTSUPP;
 218         case NVME_SC_WRITE_FAULT:
 219         case NVME_SC_READ_ERROR:
 220         case NVME_SC_UNWRITTEN_BLOCK:
 221         case NVME_SC_ACCESS_DENIED:
 222         case NVME_SC_READ_ONLY:
 223         case NVME_SC_COMPARE_FAILED:
 224                 return BLK_STS_MEDIUM;
 225         case NVME_SC_GUARD_CHECK:
 226         case NVME_SC_APPTAG_CHECK:
 227         case NVME_SC_REFTAG_CHECK:
 228         case NVME_SC_INVALID_PI:
 229                 return BLK_STS_PROTECTION;
 230         case NVME_SC_RESERVATION_CONFLICT:
 231                 return BLK_STS_NEXUS;
 232         default:
 233                 return BLK_STS_IOERR;
 234         }
 235 }
 236
 237 static inline bool nvme_req_needs_retry(struct request *req)
 238 {
 239         if (blk_noretry_request(req))
 240                 return false;
 241         if (nvme_req(req)->status & NVME_SC_DNR)
 242                 return false;
 243         if (nvme_req(req)->retries >= nvme_max_retries)
 244                 return false;
 245         return true;
 246 }
 247
 248 void nvme_complete_rq(struct request *req)
 249 {
 250         blk_status_t status = nvme_error_status(req);
 251
 252         trace_nvme_complete_rq(req);
 253
 254         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 255                 if ((req->cmd_flags & REQ_NVME_MPATH) &&
 256                     blk_path_error(status)) {
 257                         nvme_failover_req(req);
 258                         return;
 259                 }
 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         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 552         if (!range)
 553                 return BLK_STS_RESOURCE;
 554
 555         __rq_for_each_bio(bio, req) {
 556                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 557                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 558
 559                 if (n < segments) {
 560                         range[n].cattr = cpu_to_le32(0);
 561                         range[n].nlb = cpu_to_le32(nlb);
 562                         range[n].slba = cpu_to_le64(slba);
 563                 }
 564                 n++;
 565         }
 566
 567         if (WARN_ON_ONCE(n != segments)) {
 568                 kfree(range);
 569                 return BLK_STS_IOERR;
 570         }
 571
 572         memset(cmnd, 0, sizeof(*cmnd));
 573         cmnd->dsm.opcode = nvme_cmd_dsm;
 574         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 575         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 576         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 577
 578         req->special_vec.bv_page = virt_to_page(range);
 579         req->special_vec.bv_offset = offset_in_page(range);
 580         req->special_vec.bv_len = sizeof(*range) * segments;
 581         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 582
 583         return BLK_STS_OK;
 584 }
 585
 586 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 587                 struct request *req, struct nvme_command *cmnd)
 588 {
 589         struct nvme_ctrl *ctrl = ns->ctrl;
 590         u16 control = 0;
 591         u32 dsmgmt = 0;
 592
 593         if (req->cmd_flags & REQ_FUA)
 594                 control |= NVME_RW_FUA;
 595         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 596                 control |= NVME_RW_LR;
 597
 598         if (req->cmd_flags & REQ_RAHEAD)
 599                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 600
 601         memset(cmnd, 0, sizeof(*cmnd));
 602         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 603         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 604         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 605         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 606
 607         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 608                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 609
 610         if (ns->ms) {
 611                 /*
 612                  * If formated with metadata, the block layer always provides a
 613                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 614                  * we enable the PRACT bit for protection information or set the
 615                  * namespace capacity to zero to prevent any I/O.
 616                  */
 617                 if (!blk_integrity_rq(req)) {
 618                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 619                                 return BLK_STS_NOTSUPP;
 620                         control |= NVME_RW_PRINFO_PRACT;
 621                 } else if (req_op(req) == REQ_OP_WRITE) {
 622                         t10_pi_prepare(req, ns->pi_type);
 623                 }
 624
 625                 switch (ns->pi_type) {
 626                 case NVME_NS_DPS_PI_TYPE3:
 627                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 628                         break;
 629                 case NVME_NS_DPS_PI_TYPE1:
 630                 case NVME_NS_DPS_PI_TYPE2:
 631                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 632                                         NVME_RW_PRINFO_PRCHK_REF;
 633                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 634                         break;
 635                 }
 636         }
 637
 638         cmnd->rw.control = cpu_to_le16(control);
 639         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 640         return 0;
 641 }
 642
 643 void nvme_cleanup_cmd(struct request *req)
 644 {
 645         if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
 646             nvme_req(req)->status == 0) {
 647                 struct nvme_ns *ns = req->rq_disk->private_data;
 648
 649                 t10_pi_complete(req, ns->pi_type,
 650                                 blk_rq_bytes(req) >> ns->lba_shift);
 651         }
 652         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 653                 kfree(page_address(req->special_vec.bv_page) +
 654                       req->special_vec.bv_offset);
 655         }
 656 }
 657 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 658
 659 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 660                 struct nvme_command *cmd)
 661 {
 662         blk_status_t ret = BLK_STS_OK;
 663
 664         nvme_clear_nvme_request(req);
 665
 666         switch (req_op(req)) {
 667         case REQ_OP_DRV_IN:
 668         case REQ_OP_DRV_OUT:
 669                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 670                 break;
 671         case REQ_OP_FLUSH:
 672                 nvme_setup_flush(ns, cmd);
 673                 break;
 674         case REQ_OP_WRITE_ZEROES:
 675                 /* currently only aliased to deallocate for a few ctrls: */
 676         case REQ_OP_DISCARD:
 677                 ret = nvme_setup_discard(ns, req, cmd);
 678                 break;
 679         case REQ_OP_READ:
 680         case REQ_OP_WRITE:
 681                 ret = nvme_setup_rw(ns, req, cmd);
 682                 break;
 683         default:
 684                 WARN_ON_ONCE(1);
 685                 return BLK_STS_IOERR;
 686         }
 687
 688         cmd->common.command_id = req->tag;
 689         trace_nvme_setup_cmd(req, cmd);
 690         return ret;
 691 }
 692 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 693
 694 /*
 695  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 696  * if the result is positive, it's an NVM Express status code
 697  */
 698 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 699                 union nvme_result *result, void *buffer, unsigned bufflen,
 700                 unsigned timeout, int qid, int at_head,
 701                 blk_mq_req_flags_t flags)
 702 {
 703         struct request *req;
 704         int ret;
 705
 706         req = nvme_alloc_request(q, cmd, flags, qid);
 707         if (IS_ERR(req))
 708                 return PTR_ERR(req);
 709
 710         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 711
 712         if (buffer && bufflen) {
 713                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 714                 if (ret)
 715                         goto out;
 716         }
 717
 718         blk_execute_rq(req->q, NULL, req, at_head);
 719         if (result)
 720                 *result = nvme_req(req)->result;
 721         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 722                 ret = -EINTR;
 723         else
 724                 ret = nvme_req(req)->status;
 725  out:
 726         blk_mq_free_request(req);
 727         return ret;
 728 }
 729 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 730
 731 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 732                 void *buffer, unsigned bufflen)
 733 {
 734         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 735                         NVME_QID_ANY, 0, 0);
 736 }
 737 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 738
 739 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 740                 unsigned len, u32 seed, bool write)
 741 {
 742         struct bio_integrity_payload *bip;
 743         int ret = -ENOMEM;
 744         void *buf;
 745
 746         buf = kmalloc(len, GFP_KERNEL);
 747         if (!buf)
 748                 goto out;
 749
 750         ret = -EFAULT;
 751         if (write && copy_from_user(buf, ubuf, len))
 752                 goto out_free_meta;
 753
 754         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 755         if (IS_ERR(bip)) {
 756                 ret = PTR_ERR(bip);
 757                 goto out_free_meta;
 758         }
 759
 760         bip->bip_iter.bi_size = len;
 761         bip->bip_iter.bi_sector = seed;
 762         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 763                         offset_in_page(buf));
 764         if (ret == len)
 765                 return buf;
 766         ret = -ENOMEM;
 767 out_free_meta:
 768         kfree(buf);
 769 out:
 770         return ERR_PTR(ret);
 771 }
 772
 773 static int nvme_submit_user_cmd(struct request_queue *q,
 774                 struct nvme_command *cmd, void __user *ubuffer,
 775                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 776                 u32 meta_seed, u32 *result, unsigned timeout)
 777 {
 778         bool write = nvme_is_write(cmd);
 779         struct nvme_ns *ns = q->queuedata;
 780         struct gendisk *disk = ns ? ns->disk : NULL;
 781         struct request *req;
 782         struct bio *bio = NULL;
 783         void *meta = NULL;
 784         int ret;
 785
 786         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 787         if (IS_ERR(req))
 788                 return PTR_ERR(req);
 789
 790         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 791         nvme_req(req)->flags |= NVME_REQ_USERCMD;
 792
 793         if (ubuffer && bufflen) {
 794                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 795                                 GFP_KERNEL);
 796                 if (ret)
 797                         goto out;
 798                 bio = req->bio;
 799                 bio->bi_disk = disk;
 800                 if (disk && meta_buffer && meta_len) {
 801                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 802                                         meta_seed, write);
 803                         if (IS_ERR(meta)) {
 804                                 ret = PTR_ERR(meta);
 805                                 goto out_unmap;
 806                         }
 807                         req->cmd_flags |= REQ_INTEGRITY;
 808                 }
 809         }
 810
 811         blk_execute_rq(req->q, disk, req, 0);
 812         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 813                 ret = -EINTR;
 814         else
 815                 ret = nvme_req(req)->status;
 816         if (result)
 817                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 818         if (meta && !ret && !write) {
 819                 if (copy_to_user(meta_buffer, meta, meta_len))
 820                         ret = -EFAULT;
 821         }
 822         kfree(meta);
 823  out_unmap:
 824         if (bio)
 825                 blk_rq_unmap_user(bio);
 826  out:
 827         blk_mq_free_request(req);
 828         return ret;
 829 }
 830
 831 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 832 {
 833         struct nvme_ctrl *ctrl = rq->end_io_data;
 834         unsigned long flags;
 835         bool startka = false;
 836
 837         blk_mq_free_request(rq);
 838
 839         if (status) {
 840                 dev_err(ctrl->device,
 841                         "failed nvme_keep_alive_end_io error=%d\n",
 842                                 status);
 843                 return;
 844         }
 845
 846         spin_lock_irqsave(&ctrl->lock, flags);
 847         if (ctrl->state == NVME_CTRL_LIVE ||
 848             ctrl->state == NVME_CTRL_CONNECTING)
 849                 startka = true;
 850         spin_unlock_irqrestore(&ctrl->lock, flags);
 851         if (startka)
 852                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 853 }
 854
 855 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 856 {
 857         struct request *rq;
 858
 859         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
 860                         NVME_QID_ANY);
 861         if (IS_ERR(rq))
 862                 return PTR_ERR(rq);
 863
 864         rq->timeout = ctrl->kato * HZ;
 865         rq->end_io_data = ctrl;
 866
 867         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 868
 869         return 0;
 870 }
 871
 872 static void nvme_keep_alive_work(struct work_struct *work)
 873 {
 874         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 875                         struct nvme_ctrl, ka_work);
 876
 877         if (nvme_keep_alive(ctrl)) {
 878                 /* allocation failure, reset the controller */
 879                 dev_err(ctrl->device, "keep-alive failed\n");
 880                 nvme_reset_ctrl(ctrl);
 881                 return;
 882         }
 883 }
 884
 885 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 886 {
 887         if (unlikely(ctrl->kato == 0))
 888                 return;
 889
 890         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 891 }
 892
 893 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 894 {
 895         if (unlikely(ctrl->kato == 0))
 896                 return;
 897
 898         cancel_delayed_work_sync(&ctrl->ka_work);
 899 }
 900 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 901
 902 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 903 {
 904         struct nvme_command c = { };
 905         int error;
 906
 907         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 908         c.identify.opcode = nvme_admin_identify;
 909         c.identify.cns = NVME_ID_CNS_CTRL;
 910
 911         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 912         if (!*id)
 913                 return -ENOMEM;
 914
 915         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 916                         sizeof(struct nvme_id_ctrl));
 917         if (error)
 918                 kfree(*id);
 919         return error;
 920 }
 921
 922 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
 923                 struct nvme_ns_ids *ids)
 924 {
 925         struct nvme_command c = { };
 926         int status;
 927         void *data;
 928         int pos;
 929         int len;
 930
 931         c.identify.opcode = nvme_admin_identify;
 932         c.identify.nsid = cpu_to_le32(nsid);
 933         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
 934
 935         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
 936         if (!data)
 937                 return -ENOMEM;
 938
 939         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
 940                                       NVME_IDENTIFY_DATA_SIZE);
 941         if (status)
 942                 goto free_data;
 943
 944         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
 945                 struct nvme_ns_id_desc *cur = data + pos;
 946
 947                 if (cur->nidl == 0)
 948                         break;
 949
 950                 switch (cur->nidt) {
 951                 case NVME_NIDT_EUI64:
 952                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
 953                                 dev_warn(ctrl->device,
 954                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
 955                                          cur->nidl);
 956                                 goto free_data;
 957                         }
 958                         len = NVME_NIDT_EUI64_LEN;
 959                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
 960                         break;
 961                 case NVME_NIDT_NGUID:
 962                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
 963                                 dev_warn(ctrl->device,
 964                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
 965                                          cur->nidl);
 966                                 goto free_data;
 967                         }
 968                         len = NVME_NIDT_NGUID_LEN;
 969                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
 970                         break;
 971                 case NVME_NIDT_UUID:
 972                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
 973                                 dev_warn(ctrl->device,
 974                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
 975                                          cur->nidl);
 976                                 goto free_data;
 977                         }
 978                         len = NVME_NIDT_UUID_LEN;
 979                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
 980                         break;
 981                 default:
 982                         /* Skip unnkown types */
 983                         len = cur->nidl;
 984                         break;
 985                 }
 986
 987                 len += sizeof(*cur);
 988         }
 989 free_data:
 990         kfree(data);
 991         return status;
 992 }
 993
 994 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 995 {
 996         struct nvme_command c = { };
 997
 998         c.identify.opcode = nvme_admin_identify;
 999         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1000         c.identify.nsid = cpu_to_le32(nsid);
1001         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1002                                     NVME_IDENTIFY_DATA_SIZE);
1003 }
1004
1005 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
1006                 unsigned nsid)
1007 {
1008         struct nvme_id_ns *id;
1009         struct nvme_command c = { };
1010         int error;
1011
1012         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1013         c.identify.opcode = nvme_admin_identify;
1014         c.identify.nsid = cpu_to_le32(nsid);
1015         c.identify.cns = NVME_ID_CNS_NS;
1016
1017         id = kmalloc(sizeof(*id), GFP_KERNEL);
1018         if (!id)
1019                 return NULL;
1020
1021         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1022         if (error) {
1023                 dev_warn(ctrl->device, "Identify namespace failed\n");
1024                 kfree(id);
1025                 return NULL;
1026         }
1027
1028         return id;
1029 }
1030
1031 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
1032                       void *buffer, size_t buflen, u32 *result)
1033 {
1034         struct nvme_command c;
1035         union nvme_result res;
1036         int ret;
1037
1038         memset(&c, 0, sizeof(c));
1039         c.features.opcode = nvme_admin_set_features;
1040         c.features.fid = cpu_to_le32(fid);
1041         c.features.dword11 = cpu_to_le32(dword11);
1042
1043         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1044                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1045         if (ret >= 0 && result)
1046                 *result = le32_to_cpu(res.u32);
1047         return ret;
1048 }
1049
1050 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1051 {
1052         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1053         u32 result;
1054         int status, nr_io_queues;
1055
1056         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1057                         &result);
1058         if (status < 0)
1059                 return status;
1060
1061         /*
1062          * Degraded controllers might return an error when setting the queue
1063          * count.  We still want to be able to bring them online and offer
1064          * access to the admin queue, as that might be only way to fix them up.
1065          */
1066         if (status > 0) {
1067                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1068                 *count = 0;
1069         } else {
1070                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1071                 *count = min(*count, nr_io_queues);
1072         }
1073
1074         return 0;
1075 }
1076 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1077
1078 #define NVME_AEN_SUPPORTED \
1079         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1080
1081 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1082 {
1083         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1084         int status;
1085
1086         if (!supported_aens)
1087                 return;
1088
1089         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1090                         NULL, 0, &result);
1091         if (status)
1092                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1093                          supported_aens);
1094 }
1095
1096 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1097 {
1098         struct nvme_user_io io;
1099         struct nvme_command c;
1100         unsigned length, meta_len;
1101         void __user *metadata;
1102
1103         if (copy_from_user(&io, uio, sizeof(io)))
1104                 return -EFAULT;
1105         if (io.flags)
1106                 return -EINVAL;
1107
1108         switch (io.opcode) {
1109         case nvme_cmd_write:
1110         case nvme_cmd_read:
1111         case nvme_cmd_compare:
1112                 break;
1113         default:
1114                 return -EINVAL;
1115         }
1116
1117         length = (io.nblocks + 1) << ns->lba_shift;
1118         meta_len = (io.nblocks + 1) * ns->ms;
1119         metadata = (void __user *)(uintptr_t)io.metadata;
1120
1121         if (ns->ext) {
1122                 length += meta_len;
1123                 meta_len = 0;
1124         } else if (meta_len) {
1125                 if ((io.metadata & 3) || !io.metadata)
1126                         return -EINVAL;
1127         }
1128
1129         memset(&c, 0, sizeof(c));
1130         c.rw.opcode = io.opcode;
1131         c.rw.flags = io.flags;
1132         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1133         c.rw.slba = cpu_to_le64(io.slba);
1134         c.rw.length = cpu_to_le16(io.nblocks);
1135         c.rw.control = cpu_to_le16(io.control);
1136         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1137         c.rw.reftag = cpu_to_le32(io.reftag);
1138         c.rw.apptag = cpu_to_le16(io.apptag);
1139         c.rw.appmask = cpu_to_le16(io.appmask);
1140
1141         return nvme_submit_user_cmd(ns->queue, &c,
1142                         (void __user *)(uintptr_t)io.addr, length,
1143                         metadata, meta_len, io.slba, NULL, 0);
1144 }
1145
1146 static u32 nvme_known_admin_effects(u8 opcode)
1147 {
1148         switch (opcode) {
1149         case nvme_admin_format_nvm:
1150                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1151                                         NVME_CMD_EFFECTS_CSE_MASK;
1152         case nvme_admin_sanitize_nvm:
1153                 return NVME_CMD_EFFECTS_CSE_MASK;
1154         default:
1155                 break;
1156         }
1157         return 0;
1158 }
1159
1160 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1161                                                                 u8 opcode)
1162 {
1163         u32 effects = 0;
1164
1165         if (ns) {
1166                 if (ctrl->effects)
1167                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1168                 if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1169                         dev_warn(ctrl->device,
1170                                  "IO command:%02x has unhandled effects:%08x\n",
1171                                  opcode, effects);
1172                 return 0;
1173         }
1174
1175         if (ctrl->effects)
1176                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1177         else
1178                 effects = nvme_known_admin_effects(opcode);
1179
1180         /*
1181          * For simplicity, IO to all namespaces is quiesced even if the command
1182          * effects say only one namespace is affected.
1183          */
1184         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1185                 mutex_lock(&ctrl->scan_lock);
1186                 nvme_start_freeze(ctrl);
1187                 nvme_wait_freeze(ctrl);
1188         }
1189         return effects;
1190 }
1191
1192 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1193 {
1194         struct nvme_ns *ns;
1195
1196         down_read(&ctrl->namespaces_rwsem);
1197         list_for_each_entry(ns, &ctrl->namespaces, list)
1198                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1199                         nvme_set_queue_dying(ns);
1200         up_read(&ctrl->namespaces_rwsem);
1201
1202         nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1203 }
1204
1205 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1206 {
1207         /*
1208          * Revalidate LBA changes prior to unfreezing. This is necessary to
1209          * prevent memory corruption if a logical block size was changed by
1210          * this command.
1211          */
1212         if (effects & NVME_CMD_EFFECTS_LBCC)
1213                 nvme_update_formats(ctrl);
1214         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1215                 nvme_unfreeze(ctrl);
1216                 mutex_unlock(&ctrl->scan_lock);
1217         }
1218         if (effects & NVME_CMD_EFFECTS_CCC)
1219                 nvme_init_identify(ctrl);
1220         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1221                 nvme_queue_scan(ctrl);
1222 }
1223
1224 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1225                         struct nvme_passthru_cmd __user *ucmd)
1226 {
1227         struct nvme_passthru_cmd cmd;
1228         struct nvme_command c;
1229         unsigned timeout = 0;
1230         u32 effects;
1231         int status;
1232
1233         if (!capable(CAP_SYS_ADMIN))
1234                 return -EACCES;
1235         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1236                 return -EFAULT;
1237         if (cmd.flags)
1238                 return -EINVAL;
1239
1240         memset(&c, 0, sizeof(c));
1241         c.common.opcode = cmd.opcode;
1242         c.common.flags = cmd.flags;
1243         c.common.nsid = cpu_to_le32(cmd.nsid);
1244         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1245         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1246         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1247         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1248         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1249         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1250         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1251         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1252
1253         if (cmd.timeout_ms)
1254                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1255
1256         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1257         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1258                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1259                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1260                         0, &cmd.result, timeout);
1261         nvme_passthru_end(ctrl, effects);
1262
1263         if (status >= 0) {
1264                 if (put_user(cmd.result, &ucmd->result))
1265                         return -EFAULT;
1266         }
1267
1268         return status;
1269 }
1270
1271 /*
1272  * Issue ioctl requests on the first available path.  Note that unlike normal
1273  * block layer requests we will not retry failed request on another controller.
1274  */
1275 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1276                 struct nvme_ns_head **head, int *srcu_idx)
1277 {
1278 #ifdef CONFIG_NVME_MULTIPATH
1279         if (disk->fops == &nvme_ns_head_ops) {
1280                 struct nvme_ns *ns;
1281
1282                 *head = disk->private_data;
1283                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1284                 ns = nvme_find_path(*head);
1285                 if (!ns)
1286                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1287                 return ns;
1288         }
1289 #endif
1290         *head = NULL;
1291         *srcu_idx = -1;
1292         return disk->private_data;
1293 }
1294
1295 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1296 {
1297         if (head)
1298                 srcu_read_unlock(&head->srcu, idx);
1299 }
1300
1301 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1302                 unsigned int cmd, unsigned long arg)
1303 {
1304         struct nvme_ns_head *head = NULL;
1305         void __user *argp = (void __user *)arg;
1306         struct nvme_ns *ns;
1307         int srcu_idx, ret;
1308
1309         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1310         if (unlikely(!ns))
1311                 return -EWOULDBLOCK;
1312
1313         /*
1314          * Handle ioctls that apply to the controller instead of the namespace
1315          * seperately and drop the ns SRCU reference early.  This avoids a
1316          * deadlock when deleting namespaces using the passthrough interface.
1317          */
1318         if (cmd == NVME_IOCTL_ADMIN_CMD || is_sed_ioctl(cmd)) {
1319                 struct nvme_ctrl *ctrl = ns->ctrl;
1320
1321                 nvme_get_ctrl(ns->ctrl);
1322                 nvme_put_ns_from_disk(head, srcu_idx);
1323
1324                 if (cmd == NVME_IOCTL_ADMIN_CMD)
1325                         ret = nvme_user_cmd(ctrl, NULL, argp);
1326                 else
1327                         ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1328
1329                 nvme_put_ctrl(ctrl);
1330                 return ret;
1331         }
1332
1333         switch (cmd) {
1334         case NVME_IOCTL_ID:
1335                 force_successful_syscall_return();
1336                 ret = ns->head->ns_id;
1337                 break;
1338         case NVME_IOCTL_IO_CMD:
1339                 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1340                 break;
1341         case NVME_IOCTL_SUBMIT_IO:
1342                 ret = nvme_submit_io(ns, argp);
1343                 break;
1344         default:
1345                 if (ns->ndev)
1346                         ret = nvme_nvm_ioctl(ns, cmd, arg);
1347                 else
1348                         ret = -ENOTTY;
1349         }
1350
1351         nvme_put_ns_from_disk(head, srcu_idx);
1352         return ret;
1353 }
1354
1355 static int nvme_open(struct block_device *bdev, fmode_t mode)
1356 {
1357         struct nvme_ns *ns = bdev->bd_disk->private_data;
1358
1359 #ifdef CONFIG_NVME_MULTIPATH
1360         /* should never be called due to GENHD_FL_HIDDEN */
1361         if (WARN_ON_ONCE(ns->head->disk))
1362                 goto fail;
1363 #endif
1364         if (!kref_get_unless_zero(&ns->kref))
1365                 goto fail;
1366         if (!try_module_get(ns->ctrl->ops->module))
1367                 goto fail_put_ns;
1368
1369         return 0;
1370
1371 fail_put_ns:
1372         nvme_put_ns(ns);
1373 fail:
1374         return -ENXIO;
1375 }
1376
1377 static void nvme_release(struct gendisk *disk, fmode_t mode)
1378 {
1379         struct nvme_ns *ns = disk->private_data;
1380
1381         module_put(ns->ctrl->ops->module);
1382         nvme_put_ns(ns);
1383 }
1384
1385 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1386 {
1387         /* some standard values */
1388         geo->heads = 1 << 6;
1389         geo->sectors = 1 << 5;
1390         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1391         return 0;
1392 }
1393
1394 #ifdef CONFIG_BLK_DEV_INTEGRITY
1395 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1396 {
1397         struct blk_integrity integrity;
1398
1399         memset(&integrity, 0, sizeof(integrity));
1400         switch (pi_type) {
1401         case NVME_NS_DPS_PI_TYPE3:
1402                 integrity.profile = &t10_pi_type3_crc;
1403                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1404                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1405                 break;
1406         case NVME_NS_DPS_PI_TYPE1:
1407         case NVME_NS_DPS_PI_TYPE2:
1408                 integrity.profile = &t10_pi_type1_crc;
1409                 integrity.tag_size = sizeof(u16);
1410                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1411                 break;
1412         default:
1413                 integrity.profile = NULL;
1414                 break;
1415         }
1416         integrity.tuple_size = ms;
1417         blk_integrity_register(disk, &integrity);
1418         blk_queue_max_integrity_segments(disk->queue, 1);
1419 }
1420 #else
1421 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1422 {
1423 }
1424 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1425
1426 static void nvme_set_chunk_size(struct nvme_ns *ns)
1427 {
1428         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1429         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1430 }
1431
1432 static void nvme_config_discard(struct nvme_ns *ns)
1433 {
1434         struct nvme_ctrl *ctrl = ns->ctrl;
1435         struct request_queue *queue = ns->queue;
1436         u32 size = queue_logical_block_size(queue);
1437
1438         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1439                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1440                 return;
1441         }
1442
1443         if (ctrl->nr_streams && ns->sws && ns->sgs)
1444                 size *= ns->sws * ns->sgs;
1445
1446         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1447                         NVME_DSM_MAX_RANGES);
1448
1449         queue->limits.discard_alignment = 0;
1450         queue->limits.discard_granularity = size;
1451
1452         /* If discard is already enabled, don't reset queue limits */
1453         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1454                 return;
1455
1456         blk_queue_max_discard_sectors(queue, UINT_MAX);
1457         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1458
1459         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1460                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1461 }
1462
1463 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1464                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1465 {
1466         memset(ids, 0, sizeof(*ids));
1467
1468         if (ctrl->vs >= NVME_VS(1, 1, 0))
1469                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1470         if (ctrl->vs >= NVME_VS(1, 2, 0))
1471                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1472         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1473                  /* Don't treat error as fatal we potentially
1474                   * already have a NGUID or EUI-64
1475                   */
1476                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1477                         dev_warn(ctrl->device,
1478                                  "%s: Identify Descriptors failed\n", __func__);
1479         }
1480 }
1481
1482 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1483 {
1484         return !uuid_is_null(&ids->uuid) ||
1485                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1486                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1487 }
1488
1489 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1490 {
1491         return uuid_equal(&a->uuid, &b->uuid) &&
1492                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1493                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1494 }
1495
1496 static void nvme_update_disk_info(struct gendisk *disk,
1497                 struct nvme_ns *ns, struct nvme_id_ns *id)
1498 {
1499         sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1500         unsigned short bs = 1 << ns->lba_shift;
1501
1502         if (ns->lba_shift > PAGE_SHIFT) {
1503                 /* unsupported block size, set capacity to 0 later */
1504                 bs = (1 << 9);
1505         }
1506         blk_mq_freeze_queue(disk->queue);
1507         blk_integrity_unregister(disk);
1508
1509         blk_queue_logical_block_size(disk->queue, bs);
1510         blk_queue_physical_block_size(disk->queue, bs);
1511         blk_queue_io_min(disk->queue, bs);
1512
1513         if (ns->ms && !ns->ext &&
1514             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1515                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1516         if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1517             ns->lba_shift > PAGE_SHIFT)
1518                 capacity = 0;
1519
1520         set_capacity(disk, capacity);
1521         nvme_config_discard(ns);
1522
1523         if (id->nsattr & (1 << 0))
1524                 set_disk_ro(disk, true);
1525         else
1526                 set_disk_ro(disk, false);
1527
1528         blk_mq_unfreeze_queue(disk->queue);
1529 }
1530
1531 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1532 {
1533         struct nvme_ns *ns = disk->private_data;
1534
1535         /*
1536          * If identify namespace failed, use default 512 byte block size so
1537          * block layer can use before failing read/write for 0 capacity.
1538          */
1539         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1540         if (ns->lba_shift == 0)
1541                 ns->lba_shift = 9;
1542         ns->noiob = le16_to_cpu(id->noiob);
1543         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1544         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1545         /* the PI implementation requires metadata equal t10 pi tuple size */
1546         if (ns->ms == sizeof(struct t10_pi_tuple))
1547                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1548         else
1549                 ns->pi_type = 0;
1550
1551         if (ns->noiob)
1552                 nvme_set_chunk_size(ns);
1553         nvme_update_disk_info(disk, ns, id);
1554         if (ns->ndev)
1555                 nvme_nvm_update_nvm_info(ns);
1556 #ifdef CONFIG_NVME_MULTIPATH
1557         if (ns->head->disk) {
1558                 nvme_update_disk_info(ns->head->disk, ns, id);
1559                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1560         }
1561 #endif
1562 }
1563
1564 static int nvme_revalidate_disk(struct gendisk *disk)
1565 {
1566         struct nvme_ns *ns = disk->private_data;
1567         struct nvme_ctrl *ctrl = ns->ctrl;
1568         struct nvme_id_ns *id;
1569         struct nvme_ns_ids ids;
1570         int ret = 0;
1571
1572         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1573                 set_capacity(disk, 0);
1574                 return -ENODEV;
1575         }
1576
1577         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1578         if (!id)
1579                 return -ENODEV;
1580
1581         if (id->ncap == 0) {
1582                 ret = -ENODEV;
1583                 goto out;
1584         }
1585
1586         __nvme_revalidate_disk(disk, id);
1587         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1588         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1589                 dev_err(ctrl->device,
1590                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1591                 ret = -ENODEV;
1592         }
1593
1594 out:
1595         kfree(id);
1596         return ret;
1597 }
1598
1599 static char nvme_pr_type(enum pr_type type)
1600 {
1601         switch (type) {
1602         case PR_WRITE_EXCLUSIVE:
1603                 return 1;
1604         case PR_EXCLUSIVE_ACCESS:
1605                 return 2;
1606         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1607                 return 3;
1608         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1609                 return 4;
1610         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1611                 return 5;
1612         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1613                 return 6;
1614         default:
1615                 return 0;
1616         }
1617 };
1618
1619 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1620                                 u64 key, u64 sa_key, u8 op)
1621 {
1622         struct nvme_ns_head *head = NULL;
1623         struct nvme_ns *ns;
1624         struct nvme_command c;
1625         int srcu_idx, ret;
1626         u8 data[16] = { 0, };
1627
1628         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1629         if (unlikely(!ns))
1630                 return -EWOULDBLOCK;
1631
1632         put_unaligned_le64(key, &data[0]);
1633         put_unaligned_le64(sa_key, &data[8]);
1634
1635         memset(&c, 0, sizeof(c));
1636         c.common.opcode = op;
1637         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1638         c.common.cdw10[0] = cpu_to_le32(cdw10);
1639
1640         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1641         nvme_put_ns_from_disk(head, srcu_idx);
1642         return ret;
1643 }
1644
1645 static int nvme_pr_register(struct block_device *bdev, u64 old,
1646                 u64 new, unsigned flags)
1647 {
1648         u32 cdw10;
1649
1650         if (flags & ~PR_FL_IGNORE_KEY)
1651                 return -EOPNOTSUPP;
1652
1653         cdw10 = old ? 2 : 0;
1654         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1655         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1656         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1657 }
1658
1659 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1660                 enum pr_type type, unsigned flags)
1661 {
1662         u32 cdw10;
1663
1664         if (flags & ~PR_FL_IGNORE_KEY)
1665                 return -EOPNOTSUPP;
1666
1667         cdw10 = nvme_pr_type(type) << 8;
1668         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1669         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1670 }
1671
1672 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1673                 enum pr_type type, bool abort)
1674 {
1675         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1676         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1677 }
1678
1679 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1680 {
1681         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1682         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1683 }
1684
1685 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1686 {
1687         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1688         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1689 }
1690
1691 static const struct pr_ops nvme_pr_ops = {
1692         .pr_register    = nvme_pr_register,
1693         .pr_reserve     = nvme_pr_reserve,
1694         .pr_release     = nvme_pr_release,
1695         .pr_preempt     = nvme_pr_preempt,
1696         .pr_clear       = nvme_pr_clear,
1697 };
1698
1699 #ifdef CONFIG_BLK_SED_OPAL
1700 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1701                 bool send)
1702 {
1703         struct nvme_ctrl *ctrl = data;
1704         struct nvme_command cmd;
1705
1706         memset(&cmd, 0, sizeof(cmd));
1707         if (send)
1708                 cmd.common.opcode = nvme_admin_security_send;
1709         else
1710                 cmd.common.opcode = nvme_admin_security_recv;
1711         cmd.common.nsid = 0;
1712         cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1713         cmd.common.cdw10[1] = cpu_to_le32(len);
1714
1715         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1716                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1717 }
1718 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1719 #endif /* CONFIG_BLK_SED_OPAL */
1720
1721 static const struct block_device_operations nvme_fops = {
1722         .owner          = THIS_MODULE,
1723         .ioctl          = nvme_ioctl,
1724         .compat_ioctl   = nvme_ioctl,
1725         .open           = nvme_open,
1726         .release        = nvme_release,
1727         .getgeo         = nvme_getgeo,
1728         .revalidate_disk= nvme_revalidate_disk,
1729         .pr_ops         = &nvme_pr_ops,
1730 };
1731
1732 #ifdef CONFIG_NVME_MULTIPATH
1733 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1734 {
1735         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1736
1737         if (!kref_get_unless_zero(&head->ref))
1738                 return -ENXIO;
1739         return 0;
1740 }
1741
1742 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1743 {
1744         nvme_put_ns_head(disk->private_data);
1745 }
1746
1747 const struct block_device_operations nvme_ns_head_ops = {
1748         .owner          = THIS_MODULE,
1749         .open           = nvme_ns_head_open,
1750         .release        = nvme_ns_head_release,
1751         .ioctl          = nvme_ioctl,
1752         .compat_ioctl   = nvme_ioctl,
1753         .getgeo         = nvme_getgeo,
1754         .pr_ops         = &nvme_pr_ops,
1755 };
1756 #endif /* CONFIG_NVME_MULTIPATH */
1757
1758 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1759 {
1760         unsigned long timeout =
1761                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1762         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1763         int ret;
1764
1765         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1766                 if (csts == ~0)
1767                         return -ENODEV;
1768                 if ((csts & NVME_CSTS_RDY) == bit)
1769                         break;
1770
1771                 msleep(100);
1772                 if (fatal_signal_pending(current))
1773                         return -EINTR;
1774                 if (time_after(jiffies, timeout)) {
1775                         dev_err(ctrl->device,
1776                                 "Device not ready; aborting %s\n", enabled ?
1777                                                 "initialisation" : "reset");
1778                         return -ENODEV;
1779                 }
1780         }
1781
1782         return ret;
1783 }
1784
1785 /*
1786  * If the device has been passed off to us in an enabled state, just clear
1787  * the enabled bit.  The spec says we should set the 'shutdown notification
1788  * bits', but doing so may cause the device to complete commands to the
1789  * admin queue ... and we don't know what memory that might be pointing at!
1790  */
1791 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1792 {
1793         int ret;
1794
1795         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1796         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1797
1798         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1799         if (ret)
1800                 return ret;
1801
1802         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1803                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1804
1805         return nvme_wait_ready(ctrl, cap, false);
1806 }
1807 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1808
1809 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1810 {
1811         /*
1812          * Default to a 4K page size, with the intention to update this
1813          * path in the future to accomodate architectures with differing
1814          * kernel and IO page sizes.
1815          */
1816         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1817         int ret;
1818
1819         if (page_shift < dev_page_min) {
1820                 dev_err(ctrl->device,
1821                         "Minimum device page size %u too large for host (%u)\n",
1822                         1 << dev_page_min, 1 << page_shift);
1823                 return -ENODEV;
1824         }
1825
1826         ctrl->page_size = 1 << page_shift;
1827
1828         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1829         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1830         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1831         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1832         ctrl->ctrl_config |= NVME_CC_ENABLE;
1833
1834         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1835         if (ret)
1836                 return ret;
1837         return nvme_wait_ready(ctrl, cap, true);
1838 }
1839 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1840
1841 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1842 {
1843         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1844         u32 csts;
1845         int ret;
1846
1847         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1848         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1849
1850         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1851         if (ret)
1852                 return ret;
1853
1854         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1855                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1856                         break;
1857
1858                 msleep(100);
1859                 if (fatal_signal_pending(current))
1860                         return -EINTR;
1861                 if (time_after(jiffies, timeout)) {
1862                         dev_err(ctrl->device,
1863                                 "Device shutdown incomplete; abort shutdown\n");
1864                         return -ENODEV;
1865                 }
1866         }
1867
1868         return ret;
1869 }
1870 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1871
1872 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1873                 struct request_queue *q)
1874 {
1875         bool vwc = false;
1876
1877         if (ctrl->max_hw_sectors) {
1878                 u32 max_segments =
1879                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1880
1881                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1882                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1883                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1884         }
1885         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1886             is_power_of_2(ctrl->max_hw_sectors))
1887                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1888         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1889         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1890                 vwc = true;
1891         blk_queue_write_cache(q, vwc, vwc);
1892 }
1893
1894 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1895 {
1896         __le64 ts;
1897         int ret;
1898
1899         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1900                 return 0;
1901
1902         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1903         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1904                         NULL);
1905         if (ret)
1906                 dev_warn_once(ctrl->device,
1907                         "could not set timestamp (%d)\n", ret);
1908         return ret;
1909 }
1910
1911 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1912 {
1913         /*
1914          * APST (Autonomous Power State Transition) lets us program a
1915          * table of power state transitions that the controller will
1916          * perform automatically.  We configure it with a simple
1917          * heuristic: we are willing to spend at most 2% of the time
1918          * transitioning between power states.  Therefore, when running
1919          * in any given state, we will enter the next lower-power
1920          * non-operational state after waiting 50 * (enlat + exlat)
1921          * microseconds, as long as that state's exit latency is under
1922          * the requested maximum latency.
1923          *
1924          * We will not autonomously enter any non-operational state for
1925          * which the total latency exceeds ps_max_latency_us.  Users
1926          * can set ps_max_latency_us to zero to turn off APST.
1927          */
1928
1929         unsigned apste;
1930         struct nvme_feat_auto_pst *table;
1931         u64 max_lat_us = 0;
1932         int max_ps = -1;
1933         int ret;
1934
1935         /*
1936          * If APST isn't supported or if we haven't been initialized yet,
1937          * then don't do anything.
1938          */
1939         if (!ctrl->apsta)
1940                 return 0;
1941
1942         if (ctrl->npss > 31) {
1943                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1944                 return 0;
1945         }
1946
1947         table = kzalloc(sizeof(*table), GFP_KERNEL);
1948         if (!table)
1949                 return 0;
1950
1951         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1952                 /* Turn off APST. */
1953                 apste = 0;
1954                 dev_dbg(ctrl->device, "APST disabled\n");
1955         } else {
1956                 __le64 target = cpu_to_le64(0);
1957                 int state;
1958
1959                 /*
1960                  * Walk through all states from lowest- to highest-power.
1961                  * According to the spec, lower-numbered states use more
1962                  * power.  NPSS, despite the name, is the index of the
1963                  * lowest-power state, not the number of states.
1964                  */
1965                 for (state = (int)ctrl->npss; state >= 0; state--) {
1966                         u64 total_latency_us, exit_latency_us, transition_ms;
1967
1968                         if (target)
1969                                 table->entries[state] = target;
1970
1971                         /*
1972                          * Don't allow transitions to the deepest state
1973                          * if it's quirked off.
1974                          */
1975                         if (state == ctrl->npss &&
1976                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1977                                 continue;
1978
1979                         /*
1980                          * Is this state a useful non-operational state for
1981                          * higher-power states to autonomously transition to?
1982                          */
1983                         if (!(ctrl->psd[state].flags &
1984                               NVME_PS_FLAGS_NON_OP_STATE))
1985                                 continue;
1986
1987                         exit_latency_us =
1988                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1989                         if (exit_latency_us > ctrl->ps_max_latency_us)
1990                                 continue;
1991
1992                         total_latency_us =
1993                                 exit_latency_us +
1994                                 le32_to_cpu(ctrl->psd[state].entry_lat);
1995
1996                         /*
1997                          * This state is good.  Use it as the APST idle
1998                          * target for higher power states.
1999                          */
2000                         transition_ms = total_latency_us + 19;
2001                         do_div(transition_ms, 20);
2002                         if (transition_ms > (1 << 24) - 1)
2003                                 transition_ms = (1 << 24) - 1;
2004
2005                         target = cpu_to_le64((state << 3) |
2006                                              (transition_ms << 8));
2007
2008                         if (max_ps == -1)
2009                                 max_ps = state;
2010
2011                         if (total_latency_us > max_lat_us)
2012                                 max_lat_us = total_latency_us;
2013                 }
2014
2015                 apste = 1;
2016
2017                 if (max_ps == -1) {
2018                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2019                 } else {
2020                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2021                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2022                 }
2023         }
2024
2025         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2026                                 table, sizeof(*table), NULL);
2027         if (ret)
2028                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2029
2030         kfree(table);
2031         return ret;
2032 }
2033
2034 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2035 {
2036         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2037         u64 latency;
2038
2039         switch (val) {
2040         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2041         case PM_QOS_LATENCY_ANY:
2042                 latency = U64_MAX;
2043                 break;
2044
2045         default:
2046                 latency = val;
2047         }
2048
2049         if (ctrl->ps_max_latency_us != latency) {
2050                 ctrl->ps_max_latency_us = latency;
2051                 nvme_configure_apst(ctrl);
2052         }
2053 }
2054
2055 struct nvme_core_quirk_entry {
2056         /*
2057          * NVMe model and firmware strings are padded with spaces.  For
2058          * simplicity, strings in the quirk table are padded with NULLs
2059          * instead.
2060          */
2061         u16 vid;
2062         const char *mn;
2063         const char *fr;
2064         unsigned long quirks;
2065 };
2066
2067 static const struct nvme_core_quirk_entry core_quirks[] = {
2068         {
2069                 /*
2070                  * This Toshiba device seems to die using any APST states.  See:
2071                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2072                  */
2073                 .vid = 0x1179,
2074                 .mn = "THNSF5256GPUK TOSHIBA",
2075                 .quirks = NVME_QUIRK_NO_APST,
2076         }
2077 };
2078
2079 /* match is null-terminated but idstr is space-padded. */
2080 static bool string_matches(const char *idstr, const char *match, size_t len)
2081 {
2082         size_t matchlen;
2083
2084         if (!match)
2085                 return true;
2086
2087         matchlen = strlen(match);
2088         WARN_ON_ONCE(matchlen > len);
2089
2090         if (memcmp(idstr, match, matchlen))
2091                 return false;
2092
2093         for (; matchlen < len; matchlen++)
2094                 if (idstr[matchlen] != ' ')
2095                         return false;
2096
2097         return true;
2098 }
2099
2100 static bool quirk_matches(const struct nvme_id_ctrl *id,
2101                           const struct nvme_core_quirk_entry *q)
2102 {
2103         return q->vid == le16_to_cpu(id->vid) &&
2104                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2105                 string_matches(id->fr, q->fr, sizeof(id->fr));
2106 }
2107
2108 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2109                 struct nvme_id_ctrl *id)
2110 {
2111         size_t nqnlen;
2112         int off;
2113
2114         nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2115         if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2116                 strncpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2117                 return;
2118         }
2119
2120         if (ctrl->vs >= NVME_VS(1, 2, 1))
2121                 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2122
2123         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2124         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2125                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2126                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2127         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2128         off += sizeof(id->sn);
2129         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2130         off += sizeof(id->mn);
2131         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2132 }
2133
2134 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2135 {
2136         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2137         kfree(subsys);
2138 }
2139
2140 static void nvme_release_subsystem(struct device *dev)
2141 {
2142         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2143 }
2144
2145 static void nvme_destroy_subsystem(struct kref *ref)
2146 {
2147         struct nvme_subsystem *subsys =
2148                         container_of(ref, struct nvme_subsystem, ref);
2149
2150         mutex_lock(&nvme_subsystems_lock);
2151         list_del(&subsys->entry);
2152         mutex_unlock(&nvme_subsystems_lock);
2153
2154         ida_destroy(&subsys->ns_ida);
2155         device_del(&subsys->dev);
2156         put_device(&subsys->dev);
2157 }
2158
2159 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2160 {
2161         kref_put(&subsys->ref, nvme_destroy_subsystem);
2162 }
2163
2164 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2165 {
2166         struct nvme_subsystem *subsys;
2167
2168         lockdep_assert_held(&nvme_subsystems_lock);
2169
2170         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2171                 if (strcmp(subsys->subnqn, subsysnqn))
2172                         continue;
2173                 if (!kref_get_unless_zero(&subsys->ref))
2174                         continue;
2175                 return subsys;
2176         }
2177
2178         return NULL;
2179 }
2180
2181 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2182         struct device_attribute subsys_attr_##_name = \
2183                 __ATTR(_name, _mode, _show, NULL)
2184
2185 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2186                                     struct device_attribute *attr,
2187                                     char *buf)
2188 {
2189         struct nvme_subsystem *subsys =
2190                 container_of(dev, struct nvme_subsystem, dev);
2191
2192         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2193 }
2194 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2195
2196 #define nvme_subsys_show_str_function(field)                            \
2197 static ssize_t subsys_##field##_show(struct device *dev,                \
2198                             struct device_attribute *attr, char *buf)   \
2199 {                                                                       \
2200         struct nvme_subsystem *subsys =                                 \
2201                 container_of(dev, struct nvme_subsystem, dev);          \
2202         return sprintf(buf, "%.*s\n",                                   \
2203                        (int)sizeof(subsys->field), subsys->field);      \
2204 }                                                                       \
2205 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2206
2207 nvme_subsys_show_str_function(model);
2208 nvme_subsys_show_str_function(serial);
2209 nvme_subsys_show_str_function(firmware_rev);
2210
2211 static struct attribute *nvme_subsys_attrs[] = {
2212         &subsys_attr_model.attr,
2213         &subsys_attr_serial.attr,
2214         &subsys_attr_firmware_rev.attr,
2215         &subsys_attr_subsysnqn.attr,
2216         NULL,
2217 };
2218
2219 static struct attribute_group nvme_subsys_attrs_group = {
2220         .attrs = nvme_subsys_attrs,
2221 };
2222
2223 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2224         &nvme_subsys_attrs_group,
2225         NULL,
2226 };
2227
2228 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2229 {
2230         int count = 0;
2231         struct nvme_ctrl *ctrl;
2232
2233         mutex_lock(&subsys->lock);
2234         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2235                 if (ctrl->state != NVME_CTRL_DELETING &&
2236                     ctrl->state != NVME_CTRL_DEAD)
2237                         count++;
2238         }
2239         mutex_unlock(&subsys->lock);
2240
2241         return count;
2242 }
2243
2244 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2245 {
2246         struct nvme_subsystem *subsys, *found;
2247         int ret;
2248
2249         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2250         if (!subsys)
2251                 return -ENOMEM;
2252         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2253         if (ret < 0) {
2254                 kfree(subsys);
2255                 return ret;
2256         }
2257         subsys->instance = ret;
2258         mutex_init(&subsys->lock);
2259         kref_init(&subsys->ref);
2260         INIT_LIST_HEAD(&subsys->ctrls);
2261         INIT_LIST_HEAD(&subsys->nsheads);
2262         nvme_init_subnqn(subsys, ctrl, id);
2263         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2264         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2265         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2266         subsys->vendor_id = le16_to_cpu(id->vid);
2267         subsys->cmic = id->cmic;
2268
2269         subsys->dev.class = nvme_subsys_class;
2270         subsys->dev.release = nvme_release_subsystem;
2271         subsys->dev.groups = nvme_subsys_attrs_groups;
2272         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2273         device_initialize(&subsys->dev);
2274
2275         mutex_lock(&nvme_subsystems_lock);
2276         found = __nvme_find_get_subsystem(subsys->subnqn);
2277         if (found) {
2278                 /*
2279                  * Verify that the subsystem actually supports multiple
2280                  * controllers, else bail out.
2281                  */
2282                 if (!(ctrl->opts && ctrl->opts->discovery_nqn) &&
2283                     nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2284                         dev_err(ctrl->device,
2285                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2286                                 found->subnqn);
2287                         nvme_put_subsystem(found);
2288                         ret = -EINVAL;
2289                         goto out_unlock;
2290                 }
2291
2292                 __nvme_release_subsystem(subsys);
2293                 subsys = found;
2294         } else {
2295                 ret = device_add(&subsys->dev);
2296                 if (ret) {
2297                         dev_err(ctrl->device,
2298                                 "failed to register subsystem device.\n");
2299                         goto out_unlock;
2300                 }
2301                 ida_init(&subsys->ns_ida);
2302                 list_add_tail(&subsys->entry, &nvme_subsystems);
2303         }
2304
2305         ctrl->subsys = subsys;
2306         mutex_unlock(&nvme_subsystems_lock);
2307
2308         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2309                         dev_name(ctrl->device))) {
2310                 dev_err(ctrl->device,
2311                         "failed to create sysfs link from subsystem.\n");
2312                 /* the transport driver will eventually put the subsystem */
2313                 return -EINVAL;
2314         }
2315
2316         mutex_lock(&subsys->lock);
2317         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2318         mutex_unlock(&subsys->lock);
2319
2320         return 0;
2321
2322 out_unlock:
2323         mutex_unlock(&nvme_subsystems_lock);
2324         put_device(&subsys->dev);
2325         return ret;
2326 }
2327
2328 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2329                 void *log, size_t size, u64 offset)
2330 {
2331         struct nvme_command c = { };
2332         unsigned long dwlen = size / 4 - 1;
2333
2334         c.get_log_page.opcode = nvme_admin_get_log_page;
2335         c.get_log_page.nsid = cpu_to_le32(nsid);
2336         c.get_log_page.lid = log_page;
2337         c.get_log_page.lsp = lsp;
2338         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2339         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2340         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2341         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2342
2343         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2344 }
2345
2346 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2347 {
2348         int ret;
2349
2350         if (!ctrl->effects)
2351                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2352
2353         if (!ctrl->effects)
2354                 return 0;
2355
2356         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2357                         ctrl->effects, sizeof(*ctrl->effects), 0);
2358         if (ret) {
2359                 kfree(ctrl->effects);
2360                 ctrl->effects = NULL;
2361         }
2362         return ret;
2363 }
2364
2365 /*
2366  * Initialize the cached copies of the Identify data and various controller
2367  * register in our nvme_ctrl structure.  This should be called as soon as
2368  * the admin queue is fully up and running.
2369  */
2370 int nvme_init_identify(struct nvme_ctrl *ctrl)
2371 {
2372         struct nvme_id_ctrl *id;
2373         u64 cap;
2374         int ret, page_shift;
2375         u32 max_hw_sectors;
2376         bool prev_apst_enabled;
2377
2378         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2379         if (ret) {
2380                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2381                 return ret;
2382         }
2383
2384         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2385         if (ret) {
2386                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2387                 return ret;
2388         }
2389         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2390
2391         if (ctrl->vs >= NVME_VS(1, 1, 0))
2392                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2393
2394         ret = nvme_identify_ctrl(ctrl, &id);
2395         if (ret) {
2396                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2397                 return -EIO;
2398         }
2399
2400         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2401                 ret = nvme_get_effects_log(ctrl);
2402                 if (ret < 0)
2403                         goto out_free;
2404         }
2405
2406         if (!ctrl->identified) {
2407                 int i;
2408
2409                 ret = nvme_init_subsystem(ctrl, id);
2410                 if (ret)
2411                         goto out_free;
2412
2413                 /*
2414                  * Check for quirks.  Quirk can depend on firmware version,
2415                  * so, in principle, the set of quirks present can change
2416                  * across a reset.  As a possible future enhancement, we
2417                  * could re-scan for quirks every time we reinitialize
2418                  * the device, but we'd have to make sure that the driver
2419                  * behaves intelligently if the quirks change.
2420                  */
2421                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2422                         if (quirk_matches(id, &core_quirks[i]))
2423                                 ctrl->quirks |= core_quirks[i].quirks;
2424                 }
2425         }
2426
2427         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2428                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2429                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2430         }
2431
2432         ctrl->oacs = le16_to_cpu(id->oacs);
2433         ctrl->oncs = le16_to_cpup(&id->oncs);
2434         ctrl->oaes = le32_to_cpu(id->oaes);
2435         atomic_set(&ctrl->abort_limit, id->acl + 1);
2436         ctrl->vwc = id->vwc;
2437         ctrl->cntlid = le16_to_cpup(&id->cntlid);
2438         if (id->mdts)
2439                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2440         else
2441                 max_hw_sectors = UINT_MAX;
2442         ctrl->max_hw_sectors =
2443                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2444
2445         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2446         ctrl->sgls = le32_to_cpu(id->sgls);
2447         ctrl->kas = le16_to_cpu(id->kas);
2448         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2449
2450         if (id->rtd3e) {
2451                 /* us -> s */
2452                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2453
2454                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2455                                                  shutdown_timeout, 60);
2456
2457                 if (ctrl->shutdown_timeout != shutdown_timeout)
2458                         dev_info(ctrl->device,
2459                                  "Shutdown timeout set to %u seconds\n",
2460                                  ctrl->shutdown_timeout);
2461         } else
2462                 ctrl->shutdown_timeout = shutdown_timeout;
2463
2464         ctrl->npss = id->npss;
2465         ctrl->apsta = id->apsta;
2466         prev_apst_enabled = ctrl->apst_enabled;
2467         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2468                 if (force_apst && id->apsta) {
2469                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2470                         ctrl->apst_enabled = true;
2471                 } else {
2472                         ctrl->apst_enabled = false;
2473                 }
2474         } else {
2475                 ctrl->apst_enabled = id->apsta;
2476         }
2477         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2478
2479         if (ctrl->ops->flags & NVME_F_FABRICS) {
2480                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2481                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2482                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2483                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2484
2485                 /*
2486                  * In fabrics we need to verify the cntlid matches the
2487                  * admin connect
2488                  */
2489                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2490                         ret = -EINVAL;
2491                         goto out_free;
2492                 }
2493
2494                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2495                         dev_err(ctrl->device,
2496                                 "keep-alive support is mandatory for fabrics\n");
2497                         ret = -EINVAL;
2498                         goto out_free;
2499                 }
2500         } else {
2501                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2502                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2503                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2504                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2505                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2506         }
2507
2508         ret = nvme_mpath_init(ctrl, id);
2509         kfree(id);
2510
2511         if (ret < 0)
2512                 return ret;
2513
2514         if (ctrl->apst_enabled && !prev_apst_enabled)
2515                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2516         else if (!ctrl->apst_enabled && prev_apst_enabled)
2517                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2518
2519         ret = nvme_configure_apst(ctrl);
2520         if (ret < 0)
2521                 return ret;
2522
2523         ret = nvme_configure_timestamp(ctrl);
2524         if (ret < 0)
2525                 return ret;
2526
2527         ret = nvme_configure_directives(ctrl);
2528         if (ret < 0)
2529                 return ret;
2530
2531         ctrl->identified = true;
2532
2533         return 0;
2534
2535 out_free:
2536         kfree(id);
2537         return ret;
2538 }
2539 EXPORT_SYMBOL_GPL(nvme_init_identify);
2540
2541 static int nvme_dev_open(struct inode *inode, struct file *file)
2542 {
2543         struct nvme_ctrl *ctrl =
2544                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2545
2546         switch (ctrl->state) {
2547         case NVME_CTRL_LIVE:
2548         case NVME_CTRL_ADMIN_ONLY:
2549                 break;
2550         default:
2551                 return -EWOULDBLOCK;
2552         }
2553
2554         file->private_data = ctrl;
2555         return 0;
2556 }
2557
2558 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2559 {
2560         struct nvme_ns *ns;
2561         int ret;
2562
2563         down_read(&ctrl->namespaces_rwsem);
2564         if (list_empty(&ctrl->namespaces)) {
2565                 ret = -ENOTTY;
2566                 goto out_unlock;
2567         }
2568
2569         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2570         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2571                 dev_warn(ctrl->device,
2572                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2573                 ret = -EINVAL;
2574                 goto out_unlock;
2575         }
2576
2577         dev_warn(ctrl->device,
2578                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2579         kref_get(&ns->kref);
2580         up_read(&ctrl->namespaces_rwsem);
2581
2582         ret = nvme_user_cmd(ctrl, ns, argp);
2583         nvme_put_ns(ns);
2584         return ret;
2585
2586 out_unlock:
2587         up_read(&ctrl->namespaces_rwsem);
2588         return ret;
2589 }
2590
2591 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2592                 unsigned long arg)
2593 {
2594         struct nvme_ctrl *ctrl = file->private_data;
2595         void __user *argp = (void __user *)arg;
2596
2597         switch (cmd) {
2598         case NVME_IOCTL_ADMIN_CMD:
2599                 return nvme_user_cmd(ctrl, NULL, argp);
2600         case NVME_IOCTL_IO_CMD:
2601                 return nvme_dev_user_cmd(ctrl, argp);
2602         case NVME_IOCTL_RESET:
2603                 dev_warn(ctrl->device, "resetting controller\n");
2604                 return nvme_reset_ctrl_sync(ctrl);
2605         case NVME_IOCTL_SUBSYS_RESET:
2606                 return nvme_reset_subsystem(ctrl);
2607         case NVME_IOCTL_RESCAN:
2608                 nvme_queue_scan(ctrl);
2609                 return 0;
2610         default:
2611                 return -ENOTTY;
2612         }
2613 }
2614
2615 static const struct file_operations nvme_dev_fops = {
2616         .owner          = THIS_MODULE,
2617         .open           = nvme_dev_open,
2618         .unlocked_ioctl = nvme_dev_ioctl,
2619         .compat_ioctl   = nvme_dev_ioctl,
2620 };
2621
2622 static ssize_t nvme_sysfs_reset(struct device *dev,
2623                                 struct device_attribute *attr, const char *buf,
2624                                 size_t count)
2625 {
2626         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2627         int ret;
2628
2629         ret = nvme_reset_ctrl_sync(ctrl);
2630         if (ret < 0)
2631                 return ret;
2632         return count;
2633 }
2634 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2635
2636 static ssize_t nvme_sysfs_rescan(struct device *dev,
2637                                 struct device_attribute *attr, const char *buf,
2638                                 size_t count)
2639 {
2640         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2641
2642         nvme_queue_scan(ctrl);
2643         return count;
2644 }
2645 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2646
2647 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2648 {
2649         struct gendisk *disk = dev_to_disk(dev);
2650
2651         if (disk->fops == &nvme_fops)
2652                 return nvme_get_ns_from_dev(dev)->head;
2653         else
2654                 return disk->private_data;
2655 }
2656
2657 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2658                 char *buf)
2659 {
2660         struct nvme_ns_head *head = dev_to_ns_head(dev);
2661         struct nvme_ns_ids *ids = &head->ids;
2662         struct nvme_subsystem *subsys = head->subsys;
2663         int serial_len = sizeof(subsys->serial);
2664         int model_len = sizeof(subsys->model);
2665
2666         if (!uuid_is_null(&ids->uuid))
2667                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2668
2669         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2670                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2671
2672         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2673                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2674
2675         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2676                                   subsys->serial[serial_len - 1] == '\0'))
2677                 serial_len--;
2678         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2679                                  subsys->model[model_len - 1] == '\0'))
2680                 model_len--;
2681
2682         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2683                 serial_len, subsys->serial, model_len, subsys->model,
2684                 head->ns_id);
2685 }
2686 static DEVICE_ATTR_RO(wwid);
2687
2688 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2689                 char *buf)
2690 {
2691         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2692 }
2693 static DEVICE_ATTR_RO(nguid);
2694
2695 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2696                 char *buf)
2697 {
2698         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2699
2700         /* For backward compatibility expose the NGUID to userspace if
2701          * we have no UUID set
2702          */
2703         if (uuid_is_null(&ids->uuid)) {
2704                 printk_ratelimited(KERN_WARNING
2705                                    "No UUID available providing old NGUID\n");
2706                 return sprintf(buf, "%pU\n", ids->nguid);
2707         }
2708         return sprintf(buf, "%pU\n", &ids->uuid);
2709 }
2710 static DEVICE_ATTR_RO(uuid);
2711
2712 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2713                 char *buf)
2714 {
2715         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2716 }
2717 static DEVICE_ATTR_RO(eui);
2718
2719 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2720                 char *buf)
2721 {
2722         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2723 }
2724 static DEVICE_ATTR_RO(nsid);
2725
2726 static struct attribute *nvme_ns_id_attrs[] = {
2727         &dev_attr_wwid.attr,
2728         &dev_attr_uuid.attr,
2729         &dev_attr_nguid.attr,
2730         &dev_attr_eui.attr,
2731         &dev_attr_nsid.attr,
2732 #ifdef CONFIG_NVME_MULTIPATH
2733         &dev_attr_ana_grpid.attr,
2734         &dev_attr_ana_state.attr,
2735 #endif
2736         NULL,
2737 };
2738
2739 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2740                 struct attribute *a, int n)
2741 {
2742         struct device *dev = container_of(kobj, struct device, kobj);
2743         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2744
2745         if (a == &dev_attr_uuid.attr) {
2746                 if (uuid_is_null(&ids->uuid) &&
2747                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2748                         return 0;
2749         }
2750         if (a == &dev_attr_nguid.attr) {
2751                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2752                         return 0;
2753         }
2754         if (a == &dev_attr_eui.attr) {
2755                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2756                         return 0;
2757         }
2758 #ifdef CONFIG_NVME_MULTIPATH
2759         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2760                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2761                         return 0;
2762                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2763                         return 0;
2764         }
2765 #endif
2766         return a->mode;
2767 }
2768
2769 const struct attribute_group nvme_ns_id_attr_group = {
2770         .attrs          = nvme_ns_id_attrs,
2771         .is_visible     = nvme_ns_id_attrs_are_visible,
2772 };
2773
2774 #define nvme_show_str_function(field)                                           \
2775 static ssize_t  field##_show(struct device *dev,                                \
2776                             struct device_attribute *attr, char *buf)           \
2777 {                                                                               \
2778         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2779         return sprintf(buf, "%.*s\n",                                           \
2780                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2781 }                                                                               \
2782 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2783
2784 nvme_show_str_function(model);
2785 nvme_show_str_function(serial);
2786 nvme_show_str_function(firmware_rev);
2787
2788 #define nvme_show_int_function(field)                                           \
2789 static ssize_t  field##_show(struct device *dev,                                \
2790                             struct device_attribute *attr, char *buf)           \
2791 {                                                                               \
2792         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2793         return sprintf(buf, "%d\n", ctrl->field);       \
2794 }                                                                               \
2795 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2796
2797 nvme_show_int_function(cntlid);
2798
2799 static ssize_t nvme_sysfs_delete(struct device *dev,
2800                                 struct device_attribute *attr, const char *buf,
2801                                 size_t count)
2802 {
2803         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2804
2805         if (device_remove_file_self(dev, attr))
2806                 nvme_delete_ctrl_sync(ctrl);
2807         return count;
2808 }
2809 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2810
2811 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2812                                          struct device_attribute *attr,
2813                                          char *buf)
2814 {
2815         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2816
2817         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2818 }
2819 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2820
2821 static ssize_t nvme_sysfs_show_state(struct device *dev,
2822                                      struct device_attribute *attr,
2823                                      char *buf)
2824 {
2825         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2826         static const char *const state_name[] = {
2827                 [NVME_CTRL_NEW]         = "new",
2828                 [NVME_CTRL_LIVE]        = "live",
2829                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2830                 [NVME_CTRL_RESETTING]   = "resetting",
2831                 [NVME_CTRL_CONNECTING]  = "connecting",
2832                 [NVME_CTRL_DELETING]    = "deleting",
2833                 [NVME_CTRL_DEAD]        = "dead",
2834         };
2835
2836         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2837             state_name[ctrl->state])
2838                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2839
2840         return sprintf(buf, "unknown state\n");
2841 }
2842
2843 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2844
2845 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2846                                          struct device_attribute *attr,
2847                                          char *buf)
2848 {
2849         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2850
2851         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2852 }
2853 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2854
2855 static ssize_t nvme_sysfs_show_address(struct device *dev,
2856                                          struct device_attribute *attr,
2857                                          char *buf)
2858 {
2859         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2860
2861         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2862 }
2863 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2864
2865 static struct attribute *nvme_dev_attrs[] = {
2866         &dev_attr_reset_controller.attr,
2867         &dev_attr_rescan_controller.attr,
2868         &dev_attr_model.attr,
2869         &dev_attr_serial.attr,
2870         &dev_attr_firmware_rev.attr,
2871         &dev_attr_cntlid.attr,
2872         &dev_attr_delete_controller.attr,
2873         &dev_attr_transport.attr,
2874         &dev_attr_subsysnqn.attr,
2875         &dev_attr_address.attr,
2876         &dev_attr_state.attr,
2877         NULL
2878 };
2879
2880 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2881                 struct attribute *a, int n)
2882 {
2883         struct device *dev = container_of(kobj, struct device, kobj);
2884         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2885
2886         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2887                 return 0;
2888         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2889                 return 0;
2890
2891         return a->mode;
2892 }
2893
2894 static struct attribute_group nvme_dev_attrs_group = {
2895         .attrs          = nvme_dev_attrs,
2896         .is_visible     = nvme_dev_attrs_are_visible,
2897 };
2898
2899 static const struct attribute_group *nvme_dev_attr_groups[] = {
2900         &nvme_dev_attrs_group,
2901         NULL,
2902 };
2903
2904 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2905                 unsigned nsid)
2906 {
2907         struct nvme_ns_head *h;
2908
2909         lockdep_assert_held(&subsys->lock);
2910
2911         list_for_each_entry(h, &subsys->nsheads, entry) {
2912                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2913                         return h;
2914         }
2915
2916         return NULL;
2917 }
2918
2919 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2920                 struct nvme_ns_head *new)
2921 {
2922         struct nvme_ns_head *h;
2923
2924         lockdep_assert_held(&subsys->lock);
2925
2926         list_for_each_entry(h, &subsys->nsheads, entry) {
2927                 if (nvme_ns_ids_valid(&new->ids) &&
2928                     !list_empty(&h->list) &&
2929                     nvme_ns_ids_equal(&new->ids, &h->ids))
2930                         return -EINVAL;
2931         }
2932
2933         return 0;
2934 }
2935
2936 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2937                 unsigned nsid, struct nvme_id_ns *id)
2938 {
2939         struct nvme_ns_head *head;
2940         int ret = -ENOMEM;
2941
2942         head = kzalloc(sizeof(*head), GFP_KERNEL);
2943         if (!head)
2944                 goto out;
2945         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
2946         if (ret < 0)
2947                 goto out_free_head;
2948         head->instance = ret;
2949         INIT_LIST_HEAD(&head->list);
2950         ret = init_srcu_struct(&head->srcu);
2951         if (ret)
2952                 goto out_ida_remove;
2953         head->subsys = ctrl->subsys;
2954         head->ns_id = nsid;
2955         kref_init(&head->ref);
2956
2957         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
2958
2959         ret = __nvme_check_ids(ctrl->subsys, head);
2960         if (ret) {
2961                 dev_err(ctrl->device,
2962                         "duplicate IDs for nsid %d\n", nsid);
2963                 goto out_cleanup_srcu;
2964         }
2965
2966         ret = nvme_mpath_alloc_disk(ctrl, head);
2967         if (ret)
2968                 goto out_cleanup_srcu;
2969
2970         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
2971
2972         kref_get(&ctrl->subsys->ref);
2973
2974         return head;
2975 out_cleanup_srcu:
2976         cleanup_srcu_struct(&head->srcu);
2977 out_ida_remove:
2978         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
2979 out_free_head:
2980         kfree(head);
2981 out:
2982         return ERR_PTR(ret);
2983 }
2984
2985 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
2986                 struct nvme_id_ns *id)
2987 {
2988         struct nvme_ctrl *ctrl = ns->ctrl;
2989         bool is_shared = id->nmic & (1 << 0);
2990         struct nvme_ns_head *head = NULL;
2991         int ret = 0;
2992
2993         mutex_lock(&ctrl->subsys->lock);
2994         if (is_shared)
2995                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
2996         if (!head) {
2997                 head = nvme_alloc_ns_head(ctrl, nsid, id);
2998                 if (IS_ERR(head)) {
2999                         ret = PTR_ERR(head);
3000                         goto out_unlock;
3001                 }
3002         } else {
3003                 struct nvme_ns_ids ids;
3004
3005                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3006                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3007                         dev_err(ctrl->device,
3008                                 "IDs don't match for shared namespace %d\n",
3009                                         nsid);
3010                         ret = -EINVAL;
3011                         goto out_unlock;
3012                 }
3013         }
3014
3015         list_add_tail(&ns->siblings, &head->list);
3016         ns->head = head;
3017
3018 out_unlock:
3019         mutex_unlock(&ctrl->subsys->lock);
3020         return ret;
3021 }
3022
3023 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3024 {
3025         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3026         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3027
3028         return nsa->head->ns_id - nsb->head->ns_id;
3029 }
3030
3031 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3032 {
3033         struct nvme_ns *ns, *ret = NULL;
3034
3035         down_read(&ctrl->namespaces_rwsem);
3036         list_for_each_entry(ns, &ctrl->namespaces, list) {
3037                 if (ns->head->ns_id == nsid) {
3038                         if (!kref_get_unless_zero(&ns->kref))
3039                                 continue;
3040                         ret = ns;
3041                         break;
3042                 }
3043                 if (ns->head->ns_id > nsid)
3044                         break;
3045         }
3046         up_read(&ctrl->namespaces_rwsem);
3047         return ret;
3048 }
3049
3050 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3051 {
3052         struct streams_directive_params s;
3053         int ret;
3054
3055         if (!ctrl->nr_streams)
3056                 return 0;
3057
3058         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3059         if (ret)
3060                 return ret;
3061
3062         ns->sws = le32_to_cpu(s.sws);
3063         ns->sgs = le16_to_cpu(s.sgs);
3064
3065         if (ns->sws) {
3066                 unsigned int bs = 1 << ns->lba_shift;
3067
3068                 blk_queue_io_min(ns->queue, bs * ns->sws);
3069                 if (ns->sgs)
3070                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3071         }
3072
3073         return 0;
3074 }
3075
3076 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3077 {
3078         struct nvme_ns *ns;
3079         struct gendisk *disk;
3080         struct nvme_id_ns *id;
3081         char disk_name[DISK_NAME_LEN];
3082         int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
3083
3084         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3085         if (!ns)
3086                 return;
3087
3088         ns->queue = blk_mq_init_queue(ctrl->tagset);
3089         if (IS_ERR(ns->queue))
3090                 goto out_free_ns;
3091         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3092         ns->queue->queuedata = ns;
3093         ns->ctrl = ctrl;
3094
3095         kref_init(&ns->kref);
3096         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3097
3098         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3099         nvme_set_queue_limits(ctrl, ns->queue);
3100
3101         id = nvme_identify_ns(ctrl, nsid);
3102         if (!id)
3103                 goto out_free_queue;
3104
3105         if (id->ncap == 0)
3106                 goto out_free_id;
3107
3108         if (nvme_init_ns_head(ns, nsid, id))
3109                 goto out_free_id;
3110         nvme_setup_streams_ns(ctrl, ns);
3111         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3112
3113         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3114                 if (nvme_nvm_register(ns, disk_name, node)) {
3115                         dev_warn(ctrl->device, "LightNVM init failure\n");
3116                         goto out_unlink_ns;
3117                 }
3118         }
3119
3120         disk = alloc_disk_node(0, node);
3121         if (!disk)
3122                 goto out_unlink_ns;
3123
3124         disk->fops = &nvme_fops;
3125         disk->private_data = ns;
3126         disk->queue = ns->queue;
3127         disk->flags = flags;
3128         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3129         ns->disk = disk;
3130
3131         __nvme_revalidate_disk(disk, id);
3132
3133         down_write(&ctrl->namespaces_rwsem);
3134         list_add_tail(&ns->list, &ctrl->namespaces);
3135         up_write(&ctrl->namespaces_rwsem);
3136
3137         nvme_get_ctrl(ctrl);
3138
3139         device_add_disk(ctrl->device, ns->disk);
3140         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
3141                                         &nvme_ns_id_attr_group))
3142                 pr_warn("%s: failed to create sysfs group for identification\n",
3143                         ns->disk->disk_name);
3144         if (ns->ndev && nvme_nvm_register_sysfs(ns))
3145                 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
3146                         ns->disk->disk_name);
3147
3148         nvme_mpath_add_disk(ns, id);
3149         nvme_fault_inject_init(ns);
3150         kfree(id);
3151
3152         return;
3153  out_unlink_ns:
3154         mutex_lock(&ctrl->subsys->lock);
3155         list_del_rcu(&ns->siblings);
3156         mutex_unlock(&ctrl->subsys->lock);
3157  out_free_id:
3158         kfree(id);
3159  out_free_queue:
3160         blk_cleanup_queue(ns->queue);
3161  out_free_ns:
3162         kfree(ns);
3163 }
3164
3165 static void nvme_ns_remove(struct nvme_ns *ns)
3166 {
3167         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3168                 return;
3169
3170         nvme_fault_inject_fini(ns);
3171
3172         mutex_lock(&ns->ctrl->subsys->lock);
3173         list_del_rcu(&ns->siblings);
3174         mutex_unlock(&ns->ctrl->subsys->lock);
3175         synchronize_rcu(); /* guarantee not available in head->list */
3176         nvme_mpath_clear_current_path(ns);
3177         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3178
3179         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3180                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
3181                                         &nvme_ns_id_attr_group);
3182                 if (ns->ndev)
3183                         nvme_nvm_unregister_sysfs(ns);
3184                 del_gendisk(ns->disk);
3185                 blk_cleanup_queue(ns->queue);
3186                 if (blk_get_integrity(ns->disk))
3187                         blk_integrity_unregister(ns->disk);
3188         }
3189
3190         down_write(&ns->ctrl->namespaces_rwsem);
3191         list_del_init(&ns->list);
3192         up_write(&ns->ctrl->namespaces_rwsem);
3193
3194         nvme_mpath_check_last_path(ns);
3195         nvme_put_ns(ns);
3196 }
3197
3198 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3199 {
3200         struct nvme_ns *ns;
3201
3202         ns = nvme_find_get_ns(ctrl, nsid);
3203         if (ns) {
3204                 if (ns->disk && revalidate_disk(ns->disk))
3205                         nvme_ns_remove(ns);
3206                 nvme_put_ns(ns);
3207         } else
3208                 nvme_alloc_ns(ctrl, nsid);
3209 }
3210
3211 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3212                                         unsigned nsid)
3213 {
3214         struct nvme_ns *ns, *next;
3215         LIST_HEAD(rm_list);
3216
3217         down_write(&ctrl->namespaces_rwsem);
3218         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3219                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3220                         list_move_tail(&ns->list, &rm_list);
3221         }
3222         up_write(&ctrl->namespaces_rwsem);
3223
3224         list_for_each_entry_safe(ns, next, &rm_list, list)
3225                 nvme_ns_remove(ns);
3226
3227 }
3228
3229 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3230 {
3231         struct nvme_ns *ns;
3232         __le32 *ns_list;
3233         unsigned i, j, nsid, prev = 0;
3234         unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3235         int ret = 0;
3236
3237         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3238         if (!ns_list)
3239                 return -ENOMEM;
3240
3241         for (i = 0; i < num_lists; i++) {
3242                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3243                 if (ret)
3244                         goto free;
3245
3246                 for (j = 0; j < min(nn, 1024U); j++) {
3247                         nsid = le32_to_cpu(ns_list[j]);
3248                         if (!nsid)
3249                                 goto out;
3250
3251                         nvme_validate_ns(ctrl, nsid);
3252
3253                         while (++prev < nsid) {
3254                                 ns = nvme_find_get_ns(ctrl, prev);
3255                                 if (ns) {
3256                                         nvme_ns_remove(ns);
3257                                         nvme_put_ns(ns);
3258                                 }
3259                         }
3260                 }
3261                 nn -= j;
3262         }
3263  out:
3264         nvme_remove_invalid_namespaces(ctrl, prev);
3265  free:
3266         kfree(ns_list);
3267         return ret;
3268 }
3269
3270 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3271 {
3272         unsigned i;
3273
3274         for (i = 1; i <= nn; i++)
3275                 nvme_validate_ns(ctrl, i);
3276
3277         nvme_remove_invalid_namespaces(ctrl, nn);
3278 }
3279
3280 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3281 {
3282         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3283         __le32 *log;
3284         int error;
3285
3286         log = kzalloc(log_size, GFP_KERNEL);
3287         if (!log)
3288                 return;
3289
3290         /*
3291          * We need to read the log to clear the AEN, but we don't want to rely
3292          * on it for the changed namespace information as userspace could have
3293          * raced with us in reading the log page, which could cause us to miss
3294          * updates.
3295          */
3296         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3297                         log_size, 0);
3298         if (error)
3299                 dev_warn(ctrl->device,
3300                         "reading changed ns log failed: %d\n", error);
3301
3302         kfree(log);
3303 }
3304
3305 static void nvme_scan_work(struct work_struct *work)
3306 {
3307         struct nvme_ctrl *ctrl =
3308                 container_of(work, struct nvme_ctrl, scan_work);
3309         struct nvme_id_ctrl *id;
3310         unsigned nn;
3311
3312         if (ctrl->state != NVME_CTRL_LIVE)
3313                 return;
3314
3315         WARN_ON_ONCE(!ctrl->tagset);
3316
3317         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3318                 dev_info(ctrl->device, "rescanning namespaces.\n");
3319                 nvme_clear_changed_ns_log(ctrl);
3320         }
3321
3322         if (nvme_identify_ctrl(ctrl, &id))
3323                 return;
3324
3325         mutex_lock(&ctrl->scan_lock);
3326         nn = le32_to_cpu(id->nn);
3327         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3328             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3329                 if (!nvme_scan_ns_list(ctrl, nn))
3330                         goto out_free_id;
3331         }
3332         nvme_scan_ns_sequential(ctrl, nn);
3333 out_free_id:
3334         mutex_unlock(&ctrl->scan_lock);
3335         kfree(id);
3336         down_write(&ctrl->namespaces_rwsem);
3337         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3338         up_write(&ctrl->namespaces_rwsem);
3339 }
3340
3341 /*
3342  * This function iterates the namespace list unlocked to allow recovery from
3343  * controller failure. It is up to the caller to ensure the namespace list is
3344  * not modified by scan work while this function is executing.
3345  */
3346 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3347 {
3348         struct nvme_ns *ns, *next;
3349         LIST_HEAD(ns_list);
3350
3351         /* prevent racing with ns scanning */
3352         flush_work(&ctrl->scan_work);
3353
3354         /*
3355          * The dead states indicates the controller was not gracefully
3356          * disconnected. In that case, we won't be able to flush any data while
3357          * removing the namespaces' disks; fail all the queues now to avoid
3358          * potentially having to clean up the failed sync later.
3359          */
3360         if (ctrl->state == NVME_CTRL_DEAD)
3361                 nvme_kill_queues(ctrl);
3362
3363         down_write(&ctrl->namespaces_rwsem);
3364         list_splice_init(&ctrl->namespaces, &ns_list);
3365         up_write(&ctrl->namespaces_rwsem);
3366
3367         list_for_each_entry_safe(ns, next, &ns_list, list)
3368                 nvme_ns_remove(ns);
3369 }
3370 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3371
3372 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3373 {
3374         char *envp[2] = { NULL, NULL };
3375         u32 aen_result = ctrl->aen_result;
3376
3377         ctrl->aen_result = 0;
3378         if (!aen_result)
3379                 return;
3380
3381         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3382         if (!envp[0])
3383                 return;
3384         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3385         kfree(envp[0]);
3386 }
3387
3388 static void nvme_async_event_work(struct work_struct *work)
3389 {
3390         struct nvme_ctrl *ctrl =
3391                 container_of(work, struct nvme_ctrl, async_event_work);
3392
3393         nvme_aen_uevent(ctrl);
3394         ctrl->ops->submit_async_event(ctrl);
3395 }
3396
3397 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3398 {
3399
3400         u32 csts;
3401
3402         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3403                 return false;
3404
3405         if (csts == ~0)
3406                 return false;
3407
3408         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3409 }
3410
3411 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3412 {
3413         struct nvme_fw_slot_info_log *log;
3414
3415         log = kmalloc(sizeof(*log), GFP_KERNEL);
3416         if (!log)
3417                 return;
3418
3419         if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3420                         sizeof(*log), 0))
3421                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3422         kfree(log);
3423 }
3424
3425 static void nvme_fw_act_work(struct work_struct *work)
3426 {
3427         struct nvme_ctrl *ctrl = container_of(work,
3428                                 struct nvme_ctrl, fw_act_work);
3429         unsigned long fw_act_timeout;
3430
3431         if (ctrl->mtfa)
3432                 fw_act_timeout = jiffies +
3433                                 msecs_to_jiffies(ctrl->mtfa * 100);
3434         else
3435                 fw_act_timeout = jiffies +
3436                                 msecs_to_jiffies(admin_timeout * 1000);
3437
3438         nvme_stop_queues(ctrl);
3439         while (nvme_ctrl_pp_status(ctrl)) {
3440                 if (time_after(jiffies, fw_act_timeout)) {
3441                         dev_warn(ctrl->device,
3442                                 "Fw activation timeout, reset controller\n");
3443                         nvme_reset_ctrl(ctrl);
3444                         break;
3445                 }
3446                 msleep(100);
3447         }
3448
3449         if (ctrl->state != NVME_CTRL_LIVE)
3450                 return;
3451
3452         nvme_start_queues(ctrl);
3453         /* read FW slot information to clear the AER */
3454         nvme_get_fw_slot_info(ctrl);
3455 }
3456
3457 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3458 {
3459         switch ((result & 0xff00) >> 8) {
3460         case NVME_AER_NOTICE_NS_CHANGED:
3461                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3462                 nvme_queue_scan(ctrl);
3463                 break;
3464         case NVME_AER_NOTICE_FW_ACT_STARTING:
3465                 queue_work(nvme_wq, &ctrl->fw_act_work);
3466                 break;
3467 #ifdef CONFIG_NVME_MULTIPATH
3468         case NVME_AER_NOTICE_ANA:
3469                 if (!ctrl->ana_log_buf)
3470                         break;
3471                 queue_work(nvme_wq, &ctrl->ana_work);
3472                 break;
3473 #endif
3474         default:
3475                 dev_warn(ctrl->device, "async event result %08x\n", result);
3476         }
3477 }
3478
3479 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3480                 volatile union nvme_result *res)
3481 {
3482         u32 result = le32_to_cpu(res->u32);
3483
3484         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3485                 return;
3486
3487         switch (result & 0x7) {
3488         case NVME_AER_NOTICE:
3489                 nvme_handle_aen_notice(ctrl, result);
3490                 break;
3491         case NVME_AER_ERROR:
3492         case NVME_AER_SMART:
3493         case NVME_AER_CSS:
3494         case NVME_AER_VS:
3495                 ctrl->aen_result = result;
3496                 break;
3497         default:
3498                 break;
3499         }
3500         queue_work(nvme_wq, &ctrl->async_event_work);
3501 }
3502 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3503
3504 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3505 {
3506         nvme_mpath_stop(ctrl);
3507         nvme_stop_keep_alive(ctrl);
3508         flush_work(&ctrl->async_event_work);
3509         cancel_work_sync(&ctrl->fw_act_work);
3510         if (ctrl->ops->stop_ctrl)
3511                 ctrl->ops->stop_ctrl(ctrl);
3512 }
3513 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3514
3515 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3516 {
3517         if (ctrl->kato)
3518                 nvme_start_keep_alive(ctrl);
3519
3520         if (ctrl->queue_count > 1) {
3521                 nvme_queue_scan(ctrl);
3522                 nvme_enable_aen(ctrl);
3523                 queue_work(nvme_wq, &ctrl->async_event_work);
3524                 nvme_start_queues(ctrl);
3525         }
3526 }
3527 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3528
3529 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3530 {
3531         dev_pm_qos_hide_latency_tolerance(ctrl->device);
3532         cdev_device_del(&ctrl->cdev, ctrl->device);
3533 }
3534 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3535
3536 static void nvme_free_ctrl(struct device *dev)
3537 {
3538         struct nvme_ctrl *ctrl =
3539                 container_of(dev, struct nvme_ctrl, ctrl_device);
3540         struct nvme_subsystem *subsys = ctrl->subsys;
3541
3542         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3543         kfree(ctrl->effects);
3544         nvme_mpath_uninit(ctrl);
3545
3546         if (subsys) {
3547                 mutex_lock(&subsys->lock);
3548                 list_del(&ctrl->subsys_entry);
3549                 mutex_unlock(&subsys->lock);
3550                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3551         }
3552
3553         ctrl->ops->free_ctrl(ctrl);
3554
3555         if (subsys)
3556                 nvme_put_subsystem(subsys);
3557 }
3558
3559 /*
3560  * Initialize a NVMe controller structures.  This needs to be called during
3561  * earliest initialization so that we have the initialized structured around
3562  * during probing.
3563  */
3564 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3565                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3566 {
3567         int ret;
3568
3569         ctrl->state = NVME_CTRL_NEW;
3570         spin_lock_init(&ctrl->lock);
3571         mutex_init(&ctrl->scan_lock);
3572         INIT_LIST_HEAD(&ctrl->namespaces);
3573         init_rwsem(&ctrl->namespaces_rwsem);
3574         ctrl->dev = dev;
3575         ctrl->ops = ops;
3576         ctrl->quirks = quirks;
3577         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3578         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3579         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3580         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3581
3582         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3583         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3584         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3585
3586         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3587         if (ret < 0)
3588                 goto out;
3589         ctrl->instance = ret;
3590
3591         device_initialize(&ctrl->ctrl_device);
3592         ctrl->device = &ctrl->ctrl_device;
3593         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3594         ctrl->device->class = nvme_class;
3595         ctrl->device->parent = ctrl->dev;
3596         ctrl->device->groups = nvme_dev_attr_groups;
3597         ctrl->device->release = nvme_free_ctrl;
3598         dev_set_drvdata(ctrl->device, ctrl);
3599         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3600         if (ret)
3601                 goto out_release_instance;
3602
3603         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3604         ctrl->cdev.owner = ops->module;
3605         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3606         if (ret)
3607                 goto out_free_name;
3608
3609         /*
3610          * Initialize latency tolerance controls.  The sysfs files won't
3611          * be visible to userspace unless the device actually supports APST.
3612          */
3613         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3614         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3615                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3616
3617         return 0;
3618 out_free_name:
3619         kfree_const(dev->kobj.name);
3620 out_release_instance:
3621         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3622 out:
3623         return ret;
3624 }
3625 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3626
3627 /**
3628  * nvme_kill_queues(): Ends all namespace queues
3629  * @ctrl: the dead controller that needs to end
3630  *
3631  * Call this function when the driver determines it is unable to get the
3632  * controller in a state capable of servicing IO.
3633  */
3634 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3635 {
3636         struct nvme_ns *ns;
3637
3638         down_read(&ctrl->namespaces_rwsem);
3639
3640         /* Forcibly unquiesce queues to avoid blocking dispatch */
3641         if (ctrl->admin_q)
3642                 blk_mq_unquiesce_queue(ctrl->admin_q);
3643
3644         list_for_each_entry(ns, &ctrl->namespaces, list)
3645                 nvme_set_queue_dying(ns);
3646
3647         up_read(&ctrl->namespaces_rwsem);
3648 }
3649 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3650
3651 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3652 {
3653         struct nvme_ns *ns;
3654
3655         down_read(&ctrl->namespaces_rwsem);
3656         list_for_each_entry(ns, &ctrl->namespaces, list)
3657                 blk_mq_unfreeze_queue(ns->queue);
3658         up_read(&ctrl->namespaces_rwsem);
3659 }
3660 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3661
3662 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3663 {
3664         struct nvme_ns *ns;
3665
3666         down_read(&ctrl->namespaces_rwsem);
3667         list_for_each_entry(ns, &ctrl->namespaces, list) {
3668                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3669                 if (timeout <= 0)
3670                         break;
3671         }
3672         up_read(&ctrl->namespaces_rwsem);
3673 }
3674 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3675
3676 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3677 {
3678         struct nvme_ns *ns;
3679
3680         down_read(&ctrl->namespaces_rwsem);
3681         list_for_each_entry(ns, &ctrl->namespaces, list)
3682                 blk_mq_freeze_queue_wait(ns->queue);
3683         up_read(&ctrl->namespaces_rwsem);
3684 }
3685 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3686
3687 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3688 {
3689         struct nvme_ns *ns;
3690
3691         down_read(&ctrl->namespaces_rwsem);
3692         list_for_each_entry(ns, &ctrl->namespaces, list)
3693                 blk_freeze_queue_start(ns->queue);
3694         up_read(&ctrl->namespaces_rwsem);
3695 }
3696 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3697
3698 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3699 {
3700         struct nvme_ns *ns;
3701
3702         down_read(&ctrl->namespaces_rwsem);
3703         list_for_each_entry(ns, &ctrl->namespaces, list)
3704                 blk_mq_quiesce_queue(ns->queue);
3705         up_read(&ctrl->namespaces_rwsem);
3706 }
3707 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3708
3709 void nvme_start_queues(struct nvme_ctrl *ctrl)
3710 {
3711         struct nvme_ns *ns;
3712
3713         down_read(&ctrl->namespaces_rwsem);
3714         list_for_each_entry(ns, &ctrl->namespaces, list)
3715                 blk_mq_unquiesce_queue(ns->queue);
3716         up_read(&ctrl->namespaces_rwsem);
3717 }
3718 EXPORT_SYMBOL_GPL(nvme_start_queues);
3719
3720 int __init nvme_core_init(void)
3721 {
3722         int result = -ENOMEM;
3723
3724         nvme_wq = alloc_workqueue("nvme-wq",
3725                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3726         if (!nvme_wq)
3727                 goto out;
3728
3729         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3730                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3731         if (!nvme_reset_wq)
3732                 goto destroy_wq;
3733
3734         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3735                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3736         if (!nvme_delete_wq)
3737                 goto destroy_reset_wq;
3738
3739         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3740         if (result < 0)
3741                 goto destroy_delete_wq;
3742
3743         nvme_class = class_create(THIS_MODULE, "nvme");
3744         if (IS_ERR(nvme_class)) {
3745                 result = PTR_ERR(nvme_class);
3746                 goto unregister_chrdev;
3747         }
3748
3749         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3750         if (IS_ERR(nvme_subsys_class)) {
3751                 result = PTR_ERR(nvme_subsys_class);
3752                 goto destroy_class;
3753         }
3754         return 0;
3755
3756 destroy_class:
3757         class_destroy(nvme_class);
3758 unregister_chrdev:
3759         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3760 destroy_delete_wq:
3761         destroy_workqueue(nvme_delete_wq);
3762 destroy_reset_wq:
3763         destroy_workqueue(nvme_reset_wq);
3764 destroy_wq:
3765         destroy_workqueue(nvme_wq);
3766 out:
3767         return result;
3768 }
3769
3770 void nvme_core_exit(void)
3771 {
3772         ida_destroy(&nvme_subsystems_ida);
3773         class_destroy(nvme_subsys_class);
3774         class_destroy(nvme_class);
3775         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3776         destroy_workqueue(nvme_delete_wq);
3777         destroy_workqueue(nvme_reset_wq);
3778         destroy_workqueue(nvme_wq);
3779 }
3780
3781 MODULE_LICENSE("GPL");
3782 MODULE_VERSION("1.0");
3783 module_init(nvme_core_init);
3784 module_exit(nvme_core_exit);