drivers/nvme/host/core.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * NVM Express device driver
   4  * Copyright (c) 2011-2014, Intel Corporation.
   5  */
   6
   7 #include <linux/blkdev.h>
   8 #include <linux/blk-mq.h>
   9 #include <linux/delay.h>
  10 #include <linux/errno.h>
  11 #include <linux/hdreg.h>
  12 #include <linux/kernel.h>
  13 #include <linux/module.h>
  14 #include <linux/list_sort.h>
  15 #include <linux/slab.h>
  16 #include <linux/types.h>
  17 #include <linux/pr.h>
  18 #include <linux/ptrace.h>
  19 #include <linux/nvme_ioctl.h>
  20 #include <linux/t10-pi.h>
  21 #include <linux/pm_qos.h>
  22 #include <asm/unaligned.h>
  23
  24 #define CREATE_TRACE_POINTS
  25 #include "trace.h"
  26
  27 #include "nvme.h"
  28 #include "fabrics.h"
  29
  30 #define NVME_MINORS             (1U << MINORBITS)
  31
  32 unsigned int admin_timeout = 60;
  33 module_param(admin_timeout, uint, 0644);
  34 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  35 EXPORT_SYMBOL_GPL(admin_timeout);
  36
  37 unsigned int nvme_io_timeout = 30;
  38 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  39 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  40 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  41
  42 static unsigned char shutdown_timeout = 5;
  43 module_param(shutdown_timeout, byte, 0644);
  44 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  45
  46 static u8 nvme_max_retries = 5;
  47 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  48 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  49
  50 static unsigned long default_ps_max_latency_us = 100000;
  51 module_param(default_ps_max_latency_us, ulong, 0644);
  52 MODULE_PARM_DESC(default_ps_max_latency_us,
  53                  "max power saving latency for new devices; use PM QOS to change per device");
  54
  55 static bool force_apst;
  56 module_param(force_apst, bool, 0644);
  57 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  58
  59 static bool streams;
  60 module_param(streams, bool, 0644);
  61 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  62
  63 /*
  64  * nvme_wq - hosts nvme related works that are not reset or delete
  65  * nvme_reset_wq - hosts nvme reset works
  66  * nvme_delete_wq - hosts nvme delete works
  67  *
  68  * nvme_wq will host works such are scan, aen handling, fw activation,
  69  * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
  70  * runs reset works which also flush works hosted on nvme_wq for
  71  * serialization purposes. nvme_delete_wq host controller deletion
  72  * works which flush reset works for serialization.
  73  */
  74 struct workqueue_struct *nvme_wq;
  75 EXPORT_SYMBOL_GPL(nvme_wq);
  76
  77 struct workqueue_struct *nvme_reset_wq;
  78 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  79
  80 struct workqueue_struct *nvme_delete_wq;
  81 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  82
  83 static DEFINE_IDA(nvme_subsystems_ida);
  84 static LIST_HEAD(nvme_subsystems);
  85 static DEFINE_MUTEX(nvme_subsystems_lock);
  86
  87 static DEFINE_IDA(nvme_instance_ida);
  88 static dev_t nvme_chr_devt;
  89 static struct class *nvme_class;
  90 static struct class *nvme_subsys_class;
  91
  92 static int nvme_revalidate_disk(struct gendisk *disk);
  93 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
  94 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
  95                                            unsigned nsid);
  96
  97 static void nvme_set_queue_dying(struct nvme_ns *ns)
  98 {
  99         /*
 100          * Revalidating a dead namespace sets capacity to 0. This will end
 101          * buffered writers dirtying pages that can't be synced.
 102          */
 103         if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 104                 return;
 105         revalidate_disk(ns->disk);
 106         blk_set_queue_dying(ns->queue);
 107         /* Forcibly unquiesce queues to avoid blocking dispatch */
 108         blk_mq_unquiesce_queue(ns->queue);
 109 }
 110
 111 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 112 {
 113         /*
 114          * Only new queue scan work when admin and IO queues are both alive
 115          */
 116         if (ctrl->state == NVME_CTRL_LIVE)
 117                 queue_work(nvme_wq, &ctrl->scan_work);
 118 }
 119
 120 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 121 {
 122         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 123                 return -EBUSY;
 124         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 125                 return -EBUSY;
 126         return 0;
 127 }
 128 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 129
 130 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 131 {
 132         int ret;
 133
 134         ret = nvme_reset_ctrl(ctrl);
 135         if (!ret) {
 136                 flush_work(&ctrl->reset_work);
 137                 if (ctrl->state != NVME_CTRL_LIVE &&
 138                     ctrl->state != NVME_CTRL_ADMIN_ONLY)
 139                         ret = -ENETRESET;
 140         }
 141
 142         return ret;
 143 }
 144 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 145
 146 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 147 {
 148         dev_info(ctrl->device,
 149                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 150
 151         flush_work(&ctrl->reset_work);
 152         nvme_stop_ctrl(ctrl);
 153         nvme_remove_namespaces(ctrl);
 154         ctrl->ops->delete_ctrl(ctrl);
 155         nvme_uninit_ctrl(ctrl);
 156         nvme_put_ctrl(ctrl);
 157 }
 158
 159 static void nvme_delete_ctrl_work(struct work_struct *work)
 160 {
 161         struct nvme_ctrl *ctrl =
 162                 container_of(work, struct nvme_ctrl, delete_work);
 163
 164         nvme_do_delete_ctrl(ctrl);
 165 }
 166
 167 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 168 {
 169         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 170                 return -EBUSY;
 171         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 172                 return -EBUSY;
 173         return 0;
 174 }
 175 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 176
 177 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 178 {
 179         int ret = 0;
 180
 181         /*
 182          * Keep a reference until nvme_do_delete_ctrl() complete,
 183          * since ->delete_ctrl can free the controller.
 184          */
 185         nvme_get_ctrl(ctrl);
 186         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 187                 ret = -EBUSY;
 188         if (!ret)
 189                 nvme_do_delete_ctrl(ctrl);
 190         nvme_put_ctrl(ctrl);
 191         return ret;
 192 }
 193
 194 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 195 {
 196         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 197 }
 198
 199 static blk_status_t nvme_error_status(struct request *req)
 200 {
 201         switch (nvme_req(req)->status & 0x7ff) {
 202         case NVME_SC_SUCCESS:
 203                 return BLK_STS_OK;
 204         case NVME_SC_CAP_EXCEEDED:
 205                 return BLK_STS_NOSPC;
 206         case NVME_SC_LBA_RANGE:
 207                 return BLK_STS_TARGET;
 208         case NVME_SC_BAD_ATTRIBUTES:
 209         case NVME_SC_ONCS_NOT_SUPPORTED:
 210         case NVME_SC_INVALID_OPCODE:
 211         case NVME_SC_INVALID_FIELD:
 212         case NVME_SC_INVALID_NS:
 213                 return BLK_STS_NOTSUPP;
 214         case NVME_SC_WRITE_FAULT:
 215         case NVME_SC_READ_ERROR:
 216         case NVME_SC_UNWRITTEN_BLOCK:
 217         case NVME_SC_ACCESS_DENIED:
 218         case NVME_SC_READ_ONLY:
 219         case NVME_SC_COMPARE_FAILED:
 220                 return BLK_STS_MEDIUM;
 221         case NVME_SC_GUARD_CHECK:
 222         case NVME_SC_APPTAG_CHECK:
 223         case NVME_SC_REFTAG_CHECK:
 224         case NVME_SC_INVALID_PI:
 225                 return BLK_STS_PROTECTION;
 226         case NVME_SC_RESERVATION_CONFLICT:
 227                 return BLK_STS_NEXUS;
 228         default:
 229                 return BLK_STS_IOERR;
 230         }
 231 }
 232
 233 static inline bool nvme_req_needs_retry(struct request *req)
 234 {
 235         if (blk_noretry_request(req))
 236                 return false;
 237         if (nvme_req(req)->status & NVME_SC_DNR)
 238                 return false;
 239         if (nvme_req(req)->retries >= nvme_max_retries)
 240                 return false;
 241         return true;
 242 }
 243
 244 static void nvme_retry_req(struct request *req)
 245 {
 246         struct nvme_ns *ns = req->q->queuedata;
 247         unsigned long delay = 0;
 248         u16 crd;
 249
 250         /* The mask and shift result must be <= 3 */
 251         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 252         if (ns && crd)
 253                 delay = ns->ctrl->crdt[crd - 1] * 100;
 254
 255         nvme_req(req)->retries++;
 256         blk_mq_requeue_request(req, false);
 257         blk_mq_delay_kick_requeue_list(req->q, delay);
 258 }
 259
 260 void nvme_complete_rq(struct request *req)
 261 {
 262         blk_status_t status = nvme_error_status(req);
 263
 264         trace_nvme_complete_rq(req);
 265
 266         if (nvme_req(req)->ctrl->kas)
 267                 nvme_req(req)->ctrl->comp_seen = true;
 268
 269         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 270                 if ((req->cmd_flags & REQ_NVME_MPATH) &&
 271                     blk_path_error(status)) {
 272                         nvme_failover_req(req);
 273                         return;
 274                 }
 275
 276                 if (!blk_queue_dying(req->q)) {
 277                         nvme_retry_req(req);
 278                         return;
 279                 }
 280         }
 281         blk_mq_end_request(req, status);
 282 }
 283 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 284
 285 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 286 {
 287         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 288                                 "Cancelling I/O %d", req->tag);
 289
 290         nvme_req(req)->status = NVME_SC_ABORT_REQ;
 291         blk_mq_complete_request_sync(req);
 292         return true;
 293 }
 294 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 295
 296 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 297                 enum nvme_ctrl_state new_state)
 298 {
 299         enum nvme_ctrl_state old_state;
 300         unsigned long flags;
 301         bool changed = false;
 302
 303         spin_lock_irqsave(&ctrl->lock, flags);
 304
 305         old_state = ctrl->state;
 306         switch (new_state) {
 307         case NVME_CTRL_ADMIN_ONLY:
 308                 switch (old_state) {
 309                 case NVME_CTRL_CONNECTING:
 310                         changed = true;
 311                         /* FALLTHRU */
 312                 default:
 313                         break;
 314                 }
 315                 break;
 316         case NVME_CTRL_LIVE:
 317                 switch (old_state) {
 318                 case NVME_CTRL_NEW:
 319                 case NVME_CTRL_RESETTING:
 320                 case NVME_CTRL_CONNECTING:
 321                         changed = true;
 322                         /* FALLTHRU */
 323                 default:
 324                         break;
 325                 }
 326                 break;
 327         case NVME_CTRL_RESETTING:
 328                 switch (old_state) {
 329                 case NVME_CTRL_NEW:
 330                 case NVME_CTRL_LIVE:
 331                 case NVME_CTRL_ADMIN_ONLY:
 332                         changed = true;
 333                         /* FALLTHRU */
 334                 default:
 335                         break;
 336                 }
 337                 break;
 338         case NVME_CTRL_CONNECTING:
 339                 switch (old_state) {
 340                 case NVME_CTRL_NEW:
 341                 case NVME_CTRL_RESETTING:
 342                         changed = true;
 343                         /* FALLTHRU */
 344                 default:
 345                         break;
 346                 }
 347                 break;
 348         case NVME_CTRL_DELETING:
 349                 switch (old_state) {
 350                 case NVME_CTRL_LIVE:
 351                 case NVME_CTRL_ADMIN_ONLY:
 352                 case NVME_CTRL_RESETTING:
 353                 case NVME_CTRL_CONNECTING:
 354                         changed = true;
 355                         /* FALLTHRU */
 356                 default:
 357                         break;
 358                 }
 359                 break;
 360         case NVME_CTRL_DEAD:
 361                 switch (old_state) {
 362                 case NVME_CTRL_DELETING:
 363                         changed = true;
 364                         /* FALLTHRU */
 365                 default:
 366                         break;
 367                 }
 368                 break;
 369         default:
 370                 break;
 371         }
 372
 373         if (changed)
 374                 ctrl->state = new_state;
 375
 376         spin_unlock_irqrestore(&ctrl->lock, flags);
 377         if (changed && ctrl->state == NVME_CTRL_LIVE)
 378                 nvme_kick_requeue_lists(ctrl);
 379         return changed;
 380 }
 381 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 382
 383 static void nvme_free_ns_head(struct kref *ref)
 384 {
 385         struct nvme_ns_head *head =
 386                 container_of(ref, struct nvme_ns_head, ref);
 387
 388         nvme_mpath_remove_disk(head);
 389         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 390         list_del_init(&head->entry);
 391         cleanup_srcu_struct_quiesced(&head->srcu);
 392         nvme_put_subsystem(head->subsys);
 393         kfree(head);
 394 }
 395
 396 static void nvme_put_ns_head(struct nvme_ns_head *head)
 397 {
 398         kref_put(&head->ref, nvme_free_ns_head);
 399 }
 400
 401 static void nvme_free_ns(struct kref *kref)
 402 {
 403         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 404
 405         if (ns->ndev)
 406                 nvme_nvm_unregister(ns);
 407
 408         put_disk(ns->disk);
 409         nvme_put_ns_head(ns->head);
 410         nvme_put_ctrl(ns->ctrl);
 411         kfree(ns);
 412 }
 413
 414 static void nvme_put_ns(struct nvme_ns *ns)
 415 {
 416         kref_put(&ns->kref, nvme_free_ns);
 417 }
 418
 419 static inline void nvme_clear_nvme_request(struct request *req)
 420 {
 421         if (!(req->rq_flags & RQF_DONTPREP)) {
 422                 nvme_req(req)->retries = 0;
 423                 nvme_req(req)->flags = 0;
 424                 req->rq_flags |= RQF_DONTPREP;
 425         }
 426 }
 427
 428 struct request *nvme_alloc_request(struct request_queue *q,
 429                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 430 {
 431         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 432         struct request *req;
 433
 434         if (qid == NVME_QID_ANY) {
 435                 req = blk_mq_alloc_request(q, op, flags);
 436         } else {
 437                 req = blk_mq_alloc_request_hctx(q, op, flags,
 438                                 qid ? qid - 1 : 0);
 439         }
 440         if (IS_ERR(req))
 441                 return req;
 442
 443         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 444         nvme_clear_nvme_request(req);
 445         nvme_req(req)->cmd = cmd;
 446
 447         return req;
 448 }
 449 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 450
 451 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 452 {
 453         struct nvme_command c;
 454
 455         memset(&c, 0, sizeof(c));
 456
 457         c.directive.opcode = nvme_admin_directive_send;
 458         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 459         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 460         c.directive.dtype = NVME_DIR_IDENTIFY;
 461         c.directive.tdtype = NVME_DIR_STREAMS;
 462         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 463
 464         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 465 }
 466
 467 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 468 {
 469         return nvme_toggle_streams(ctrl, false);
 470 }
 471
 472 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 473 {
 474         return nvme_toggle_streams(ctrl, true);
 475 }
 476
 477 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 478                                   struct streams_directive_params *s, u32 nsid)
 479 {
 480         struct nvme_command c;
 481
 482         memset(&c, 0, sizeof(c));
 483         memset(s, 0, sizeof(*s));
 484
 485         c.directive.opcode = nvme_admin_directive_recv;
 486         c.directive.nsid = cpu_to_le32(nsid);
 487         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
 488         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 489         c.directive.dtype = NVME_DIR_STREAMS;
 490
 491         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 492 }
 493
 494 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 495 {
 496         struct streams_directive_params s;
 497         int ret;
 498
 499         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 500                 return 0;
 501         if (!streams)
 502                 return 0;
 503
 504         ret = nvme_enable_streams(ctrl);
 505         if (ret)
 506                 return ret;
 507
 508         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 509         if (ret)
 510                 return ret;
 511
 512         ctrl->nssa = le16_to_cpu(s.nssa);
 513         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 514                 dev_info(ctrl->device, "too few streams (%u) available\n",
 515                                         ctrl->nssa);
 516                 nvme_disable_streams(ctrl);
 517                 return 0;
 518         }
 519
 520         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 521         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 522         return 0;
 523 }
 524
 525 /*
 526  * Check if 'req' has a write hint associated with it. If it does, assign
 527  * a valid namespace stream to the write.
 528  */
 529 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 530                                      struct request *req, u16 *control,
 531                                      u32 *dsmgmt)
 532 {
 533         enum rw_hint streamid = req->write_hint;
 534
 535         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 536                 streamid = 0;
 537         else {
 538                 streamid--;
 539                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 540                         return;
 541
 542                 *control |= NVME_RW_DTYPE_STREAMS;
 543                 *dsmgmt |= streamid << 16;
 544         }
 545
 546         if (streamid < ARRAY_SIZE(req->q->write_hints))
 547                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 548 }
 549
 550 static inline void nvme_setup_flush(struct nvme_ns *ns,
 551                 struct nvme_command *cmnd)
 552 {
 553         cmnd->common.opcode = nvme_cmd_flush;
 554         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 555 }
 556
 557 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 558                 struct nvme_command *cmnd)
 559 {
 560         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 561         struct nvme_dsm_range *range;
 562         struct bio *bio;
 563
 564         range = kmalloc_array(segments, sizeof(*range),
 565                                 GFP_ATOMIC | __GFP_NOWARN);
 566         if (!range) {
 567                 /*
 568                  * If we fail allocation our range, fallback to the controller
 569                  * discard page. If that's also busy, it's safe to return
 570                  * busy, as we know we can make progress once that's freed.
 571                  */
 572                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 573                         return BLK_STS_RESOURCE;
 574
 575                 range = page_address(ns->ctrl->discard_page);
 576         }
 577
 578         __rq_for_each_bio(bio, req) {
 579                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 580                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 581
 582                 if (n < segments) {
 583                         range[n].cattr = cpu_to_le32(0);
 584                         range[n].nlb = cpu_to_le32(nlb);
 585                         range[n].slba = cpu_to_le64(slba);
 586                 }
 587                 n++;
 588         }
 589
 590         if (WARN_ON_ONCE(n != segments)) {
 591                 if (virt_to_page(range) == ns->ctrl->discard_page)
 592                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 593                 else
 594                         kfree(range);
 595                 return BLK_STS_IOERR;
 596         }
 597
 598         cmnd->dsm.opcode = nvme_cmd_dsm;
 599         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 600         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 601         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 602
 603         req->special_vec.bv_page = virt_to_page(range);
 604         req->special_vec.bv_offset = offset_in_page(range);
 605         req->special_vec.bv_len = sizeof(*range) * segments;
 606         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 607
 608         return BLK_STS_OK;
 609 }
 610
 611 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 612                 struct request *req, struct nvme_command *cmnd)
 613 {
 614         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 615                 return nvme_setup_discard(ns, req, cmnd);
 616
 617         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 618         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 619         cmnd->write_zeroes.slba =
 620                 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 621         cmnd->write_zeroes.length =
 622                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 623         cmnd->write_zeroes.control = 0;
 624         return BLK_STS_OK;
 625 }
 626
 627 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 628                 struct request *req, struct nvme_command *cmnd)
 629 {
 630         struct nvme_ctrl *ctrl = ns->ctrl;
 631         u16 control = 0;
 632         u32 dsmgmt = 0;
 633
 634         if (req->cmd_flags & REQ_FUA)
 635                 control |= NVME_RW_FUA;
 636         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 637                 control |= NVME_RW_LR;
 638
 639         if (req->cmd_flags & REQ_RAHEAD)
 640                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 641
 642         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 643         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 644         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 645         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 646
 647         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 648                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 649
 650         if (ns->ms) {
 651                 /*
 652                  * If formated with metadata, the block layer always provides a
 653                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 654                  * we enable the PRACT bit for protection information or set the
 655                  * namespace capacity to zero to prevent any I/O.
 656                  */
 657                 if (!blk_integrity_rq(req)) {
 658                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 659                                 return BLK_STS_NOTSUPP;
 660                         control |= NVME_RW_PRINFO_PRACT;
 661                 } else if (req_op(req) == REQ_OP_WRITE) {
 662                         t10_pi_prepare(req, ns->pi_type);
 663                 }
 664
 665                 switch (ns->pi_type) {
 666                 case NVME_NS_DPS_PI_TYPE3:
 667                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 668                         break;
 669                 case NVME_NS_DPS_PI_TYPE1:
 670                 case NVME_NS_DPS_PI_TYPE2:
 671                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 672                                         NVME_RW_PRINFO_PRCHK_REF;
 673                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 674                         break;
 675                 }
 676         }
 677
 678         cmnd->rw.control = cpu_to_le16(control);
 679         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 680         return 0;
 681 }
 682
 683 void nvme_cleanup_cmd(struct request *req)
 684 {
 685         if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
 686             nvme_req(req)->status == 0) {
 687                 struct nvme_ns *ns = req->rq_disk->private_data;
 688
 689                 t10_pi_complete(req, ns->pi_type,
 690                                 blk_rq_bytes(req) >> ns->lba_shift);
 691         }
 692         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 693                 struct nvme_ns *ns = req->rq_disk->private_data;
 694                 struct page *page = req->special_vec.bv_page;
 695
 696                 if (page == ns->ctrl->discard_page)
 697                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 698                 else
 699                         kfree(page_address(page) + req->special_vec.bv_offset);
 700         }
 701 }
 702 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 703
 704 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 705                 struct nvme_command *cmd)
 706 {
 707         blk_status_t ret = BLK_STS_OK;
 708
 709         nvme_clear_nvme_request(req);
 710
 711         memset(cmd, 0, sizeof(*cmd));
 712         switch (req_op(req)) {
 713         case REQ_OP_DRV_IN:
 714         case REQ_OP_DRV_OUT:
 715                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 716                 break;
 717         case REQ_OP_FLUSH:
 718                 nvme_setup_flush(ns, cmd);
 719                 break;
 720         case REQ_OP_WRITE_ZEROES:
 721                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 722                 break;
 723         case REQ_OP_DISCARD:
 724                 ret = nvme_setup_discard(ns, req, cmd);
 725                 break;
 726         case REQ_OP_READ:
 727         case REQ_OP_WRITE:
 728                 ret = nvme_setup_rw(ns, req, cmd);
 729                 break;
 730         default:
 731                 WARN_ON_ONCE(1);
 732                 return BLK_STS_IOERR;
 733         }
 734
 735         cmd->common.command_id = req->tag;
 736         trace_nvme_setup_cmd(req, cmd);
 737         return ret;
 738 }
 739 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 740
 741 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 742 {
 743         struct completion *waiting = rq->end_io_data;
 744
 745         rq->end_io_data = NULL;
 746         complete(waiting);
 747 }
 748
 749 static void nvme_execute_rq_polled(struct request_queue *q,
 750                 struct gendisk *bd_disk, struct request *rq, int at_head)
 751 {
 752         DECLARE_COMPLETION_ONSTACK(wait);
 753
 754         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
 755
 756         rq->cmd_flags |= REQ_HIPRI;
 757         rq->end_io_data = &wait;
 758         blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
 759
 760         while (!completion_done(&wait)) {
 761                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
 762                 cond_resched();
 763         }
 764 }
 765
 766 /*
 767  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 768  * if the result is positive, it's an NVM Express status code
 769  */
 770 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 771                 union nvme_result *result, void *buffer, unsigned bufflen,
 772                 unsigned timeout, int qid, int at_head,
 773                 blk_mq_req_flags_t flags, bool poll)
 774 {
 775         struct request *req;
 776         int ret;
 777
 778         req = nvme_alloc_request(q, cmd, flags, qid);
 779         if (IS_ERR(req))
 780                 return PTR_ERR(req);
 781
 782         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 783
 784         if (buffer && bufflen) {
 785                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 786                 if (ret)
 787                         goto out;
 788         }
 789
 790         if (poll)
 791                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
 792         else
 793                 blk_execute_rq(req->q, NULL, req, at_head);
 794         if (result)
 795                 *result = nvme_req(req)->result;
 796         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 797                 ret = -EINTR;
 798         else
 799                 ret = nvme_req(req)->status;
 800  out:
 801         blk_mq_free_request(req);
 802         return ret;
 803 }
 804 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 805
 806 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 807                 void *buffer, unsigned bufflen)
 808 {
 809         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 810                         NVME_QID_ANY, 0, 0, false);
 811 }
 812 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 813
 814 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 815                 unsigned len, u32 seed, bool write)
 816 {
 817         struct bio_integrity_payload *bip;
 818         int ret = -ENOMEM;
 819         void *buf;
 820
 821         buf = kmalloc(len, GFP_KERNEL);
 822         if (!buf)
 823                 goto out;
 824
 825         ret = -EFAULT;
 826         if (write && copy_from_user(buf, ubuf, len))
 827                 goto out_free_meta;
 828
 829         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 830         if (IS_ERR(bip)) {
 831                 ret = PTR_ERR(bip);
 832                 goto out_free_meta;
 833         }
 834
 835         bip->bip_iter.bi_size = len;
 836         bip->bip_iter.bi_sector = seed;
 837         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 838                         offset_in_page(buf));
 839         if (ret == len)
 840                 return buf;
 841         ret = -ENOMEM;
 842 out_free_meta:
 843         kfree(buf);
 844 out:
 845         return ERR_PTR(ret);
 846 }
 847
 848 static int nvme_submit_user_cmd(struct request_queue *q,
 849                 struct nvme_command *cmd, void __user *ubuffer,
 850                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 851                 u32 meta_seed, u32 *result, unsigned timeout)
 852 {
 853         bool write = nvme_is_write(cmd);
 854         struct nvme_ns *ns = q->queuedata;
 855         struct gendisk *disk = ns ? ns->disk : NULL;
 856         struct request *req;
 857         struct bio *bio = NULL;
 858         void *meta = NULL;
 859         int ret;
 860
 861         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 862         if (IS_ERR(req))
 863                 return PTR_ERR(req);
 864
 865         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 866         nvme_req(req)->flags |= NVME_REQ_USERCMD;
 867
 868         if (ubuffer && bufflen) {
 869                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 870                                 GFP_KERNEL);
 871                 if (ret)
 872                         goto out;
 873                 bio = req->bio;
 874                 bio->bi_disk = disk;
 875                 if (disk && meta_buffer && meta_len) {
 876                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 877                                         meta_seed, write);
 878                         if (IS_ERR(meta)) {
 879                                 ret = PTR_ERR(meta);
 880                                 goto out_unmap;
 881                         }
 882                         req->cmd_flags |= REQ_INTEGRITY;
 883                 }
 884         }
 885
 886         blk_execute_rq(req->q, disk, req, 0);
 887         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 888                 ret = -EINTR;
 889         else
 890                 ret = nvme_req(req)->status;
 891         if (result)
 892                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 893         if (meta && !ret && !write) {
 894                 if (copy_to_user(meta_buffer, meta, meta_len))
 895                         ret = -EFAULT;
 896         }
 897         kfree(meta);
 898  out_unmap:
 899         if (bio)
 900                 blk_rq_unmap_user(bio);
 901  out:
 902         blk_mq_free_request(req);
 903         return ret;
 904 }
 905
 906 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 907 {
 908         struct nvme_ctrl *ctrl = rq->end_io_data;
 909         unsigned long flags;
 910         bool startka = false;
 911
 912         blk_mq_free_request(rq);
 913
 914         if (status) {
 915                 dev_err(ctrl->device,
 916                         "failed nvme_keep_alive_end_io error=%d\n",
 917                                 status);
 918                 return;
 919         }
 920
 921         ctrl->comp_seen = false;
 922         spin_lock_irqsave(&ctrl->lock, flags);
 923         if (ctrl->state == NVME_CTRL_LIVE ||
 924             ctrl->state == NVME_CTRL_CONNECTING)
 925                 startka = true;
 926         spin_unlock_irqrestore(&ctrl->lock, flags);
 927         if (startka)
 928                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 929 }
 930
 931 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 932 {
 933         struct request *rq;
 934
 935         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
 936                         NVME_QID_ANY);
 937         if (IS_ERR(rq))
 938                 return PTR_ERR(rq);
 939
 940         rq->timeout = ctrl->kato * HZ;
 941         rq->end_io_data = ctrl;
 942
 943         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 944
 945         return 0;
 946 }
 947
 948 static void nvme_keep_alive_work(struct work_struct *work)
 949 {
 950         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 951                         struct nvme_ctrl, ka_work);
 952         bool comp_seen = ctrl->comp_seen;
 953
 954         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
 955                 dev_dbg(ctrl->device,
 956                         "reschedule traffic based keep-alive timer\n");
 957                 ctrl->comp_seen = false;
 958                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 959                 return;
 960         }
 961
 962         if (nvme_keep_alive(ctrl)) {
 963                 /* allocation failure, reset the controller */
 964                 dev_err(ctrl->device, "keep-alive failed\n");
 965                 nvme_reset_ctrl(ctrl);
 966                 return;
 967         }
 968 }
 969
 970 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 971 {
 972         if (unlikely(ctrl->kato == 0))
 973                 return;
 974
 975         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 976 }
 977
 978 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 979 {
 980         if (unlikely(ctrl->kato == 0))
 981                 return;
 982
 983         cancel_delayed_work_sync(&ctrl->ka_work);
 984 }
 985 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 986
 987 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 988 {
 989         struct nvme_command c = { };
 990         int error;
 991
 992         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 993         c.identify.opcode = nvme_admin_identify;
 994         c.identify.cns = NVME_ID_CNS_CTRL;
 995
 996         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 997         if (!*id)
 998                 return -ENOMEM;
 999
1000         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1001                         sizeof(struct nvme_id_ctrl));
1002         if (error)
1003                 kfree(*id);
1004         return error;
1005 }
1006
1007 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1008                 struct nvme_ns_ids *ids)
1009 {
1010         struct nvme_command c = { };
1011         int status;
1012         void *data;
1013         int pos;
1014         int len;
1015
1016         c.identify.opcode = nvme_admin_identify;
1017         c.identify.nsid = cpu_to_le32(nsid);
1018         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1019
1020         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1021         if (!data)
1022                 return -ENOMEM;
1023
1024         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1025                                       NVME_IDENTIFY_DATA_SIZE);
1026         if (status)
1027                 goto free_data;
1028
1029         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1030                 struct nvme_ns_id_desc *cur = data + pos;
1031
1032                 if (cur->nidl == 0)
1033                         break;
1034
1035                 switch (cur->nidt) {
1036                 case NVME_NIDT_EUI64:
1037                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1038                                 dev_warn(ctrl->device,
1039                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1040                                          cur->nidl);
1041                                 goto free_data;
1042                         }
1043                         len = NVME_NIDT_EUI64_LEN;
1044                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1045                         break;
1046                 case NVME_NIDT_NGUID:
1047                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1048                                 dev_warn(ctrl->device,
1049                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1050                                          cur->nidl);
1051                                 goto free_data;
1052                         }
1053                         len = NVME_NIDT_NGUID_LEN;
1054                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1055                         break;
1056                 case NVME_NIDT_UUID:
1057                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
1058                                 dev_warn(ctrl->device,
1059                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1060                                          cur->nidl);
1061                                 goto free_data;
1062                         }
1063                         len = NVME_NIDT_UUID_LEN;
1064                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1065                         break;
1066                 default:
1067                         /* Skip unknown types */
1068                         len = cur->nidl;
1069                         break;
1070                 }
1071
1072                 len += sizeof(*cur);
1073         }
1074 free_data:
1075         kfree(data);
1076         return status;
1077 }
1078
1079 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1080 {
1081         struct nvme_command c = { };
1082
1083         c.identify.opcode = nvme_admin_identify;
1084         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1085         c.identify.nsid = cpu_to_le32(nsid);
1086         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1087                                     NVME_IDENTIFY_DATA_SIZE);
1088 }
1089
1090 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
1091                 unsigned nsid)
1092 {
1093         struct nvme_id_ns *id;
1094         struct nvme_command c = { };
1095         int error;
1096
1097         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1098         c.identify.opcode = nvme_admin_identify;
1099         c.identify.nsid = cpu_to_le32(nsid);
1100         c.identify.cns = NVME_ID_CNS_NS;
1101
1102         id = kmalloc(sizeof(*id), GFP_KERNEL);
1103         if (!id)
1104                 return NULL;
1105
1106         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1107         if (error) {
1108                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1109                 kfree(id);
1110                 return NULL;
1111         }
1112
1113         return id;
1114 }
1115
1116 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
1117                       void *buffer, size_t buflen, u32 *result)
1118 {
1119         struct nvme_command c;
1120         union nvme_result res;
1121         int ret;
1122
1123         memset(&c, 0, sizeof(c));
1124         c.features.opcode = nvme_admin_set_features;
1125         c.features.fid = cpu_to_le32(fid);
1126         c.features.dword11 = cpu_to_le32(dword11);
1127
1128         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1129                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1130         if (ret >= 0 && result)
1131                 *result = le32_to_cpu(res.u32);
1132         return ret;
1133 }
1134
1135 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1136 {
1137         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1138         u32 result;
1139         int status, nr_io_queues;
1140
1141         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1142                         &result);
1143         if (status < 0)
1144                 return status;
1145
1146         /*
1147          * Degraded controllers might return an error when setting the queue
1148          * count.  We still want to be able to bring them online and offer
1149          * access to the admin queue, as that might be only way to fix them up.
1150          */
1151         if (status > 0) {
1152                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1153                 *count = 0;
1154         } else {
1155                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1156                 *count = min(*count, nr_io_queues);
1157         }
1158
1159         return 0;
1160 }
1161 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1162
1163 #define NVME_AEN_SUPPORTED \
1164         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1165
1166 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1167 {
1168         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1169         int status;
1170
1171         if (!supported_aens)
1172                 return;
1173
1174         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1175                         NULL, 0, &result);
1176         if (status)
1177                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1178                          supported_aens);
1179 }
1180
1181 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1182 {
1183         struct nvme_user_io io;
1184         struct nvme_command c;
1185         unsigned length, meta_len;
1186         void __user *metadata;
1187
1188         if (copy_from_user(&io, uio, sizeof(io)))
1189                 return -EFAULT;
1190         if (io.flags)
1191                 return -EINVAL;
1192
1193         switch (io.opcode) {
1194         case nvme_cmd_write:
1195         case nvme_cmd_read:
1196         case nvme_cmd_compare:
1197                 break;
1198         default:
1199                 return -EINVAL;
1200         }
1201
1202         length = (io.nblocks + 1) << ns->lba_shift;
1203         meta_len = (io.nblocks + 1) * ns->ms;
1204         metadata = (void __user *)(uintptr_t)io.metadata;
1205
1206         if (ns->ext) {
1207                 length += meta_len;
1208                 meta_len = 0;
1209         } else if (meta_len) {
1210                 if ((io.metadata & 3) || !io.metadata)
1211                         return -EINVAL;
1212         }
1213
1214         memset(&c, 0, sizeof(c));
1215         c.rw.opcode = io.opcode;
1216         c.rw.flags = io.flags;
1217         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1218         c.rw.slba = cpu_to_le64(io.slba);
1219         c.rw.length = cpu_to_le16(io.nblocks);
1220         c.rw.control = cpu_to_le16(io.control);
1221         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1222         c.rw.reftag = cpu_to_le32(io.reftag);
1223         c.rw.apptag = cpu_to_le16(io.apptag);
1224         c.rw.appmask = cpu_to_le16(io.appmask);
1225
1226         return nvme_submit_user_cmd(ns->queue, &c,
1227                         (void __user *)(uintptr_t)io.addr, length,
1228                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1229 }
1230
1231 static u32 nvme_known_admin_effects(u8 opcode)
1232 {
1233         switch (opcode) {
1234         case nvme_admin_format_nvm:
1235                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1236                                         NVME_CMD_EFFECTS_CSE_MASK;
1237         case nvme_admin_sanitize_nvm:
1238                 return NVME_CMD_EFFECTS_CSE_MASK;
1239         default:
1240                 break;
1241         }
1242         return 0;
1243 }
1244
1245 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1246                                                                 u8 opcode)
1247 {
1248         u32 effects = 0;
1249
1250         if (ns) {
1251                 if (ctrl->effects)
1252                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1253                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1254                         dev_warn(ctrl->device,
1255                                  "IO command:%02x has unhandled effects:%08x\n",
1256                                  opcode, effects);
1257                 return 0;
1258         }
1259
1260         effects |= nvme_known_admin_effects(opcode);
1261         if (ctrl->effects)
1262                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1263
1264         /*
1265          * For simplicity, IO to all namespaces is quiesced even if the command
1266          * effects say only one namespace is affected.
1267          */
1268         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1269                 mutex_lock(&ctrl->scan_lock);
1270                 nvme_start_freeze(ctrl);
1271                 nvme_wait_freeze(ctrl);
1272         }
1273         return effects;
1274 }
1275
1276 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1277 {
1278         struct nvme_ns *ns;
1279
1280         down_read(&ctrl->namespaces_rwsem);
1281         list_for_each_entry(ns, &ctrl->namespaces, list)
1282                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1283                         nvme_set_queue_dying(ns);
1284         up_read(&ctrl->namespaces_rwsem);
1285
1286         nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1287 }
1288
1289 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1290 {
1291         /*
1292          * Revalidate LBA changes prior to unfreezing. This is necessary to
1293          * prevent memory corruption if a logical block size was changed by
1294          * this command.
1295          */
1296         if (effects & NVME_CMD_EFFECTS_LBCC)
1297                 nvme_update_formats(ctrl);
1298         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1299                 nvme_unfreeze(ctrl);
1300                 mutex_unlock(&ctrl->scan_lock);
1301         }
1302         if (effects & NVME_CMD_EFFECTS_CCC)
1303                 nvme_init_identify(ctrl);
1304         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1305                 nvme_queue_scan(ctrl);
1306 }
1307
1308 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1309                         struct nvme_passthru_cmd __user *ucmd)
1310 {
1311         struct nvme_passthru_cmd cmd;
1312         struct nvme_command c;
1313         unsigned timeout = 0;
1314         u32 effects;
1315         int status;
1316
1317         if (!capable(CAP_SYS_ADMIN))
1318                 return -EACCES;
1319         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1320                 return -EFAULT;
1321         if (cmd.flags)
1322                 return -EINVAL;
1323
1324         memset(&c, 0, sizeof(c));
1325         c.common.opcode = cmd.opcode;
1326         c.common.flags = cmd.flags;
1327         c.common.nsid = cpu_to_le32(cmd.nsid);
1328         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1329         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1330         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1331         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1332         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1333         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1334         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1335         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1336
1337         if (cmd.timeout_ms)
1338                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1339
1340         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1341         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1342                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1343                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1344                         0, &cmd.result, timeout);
1345         nvme_passthru_end(ctrl, effects);
1346
1347         if (status >= 0) {
1348                 if (put_user(cmd.result, &ucmd->result))
1349                         return -EFAULT;
1350         }
1351
1352         return status;
1353 }
1354
1355 /*
1356  * Issue ioctl requests on the first available path.  Note that unlike normal
1357  * block layer requests we will not retry failed request on another controller.
1358  */
1359 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1360                 struct nvme_ns_head **head, int *srcu_idx)
1361 {
1362 #ifdef CONFIG_NVME_MULTIPATH
1363         if (disk->fops == &nvme_ns_head_ops) {
1364                 *head = disk->private_data;
1365                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1366                 return nvme_find_path(*head);
1367         }
1368 #endif
1369         *head = NULL;
1370         *srcu_idx = -1;
1371         return disk->private_data;
1372 }
1373
1374 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1375 {
1376         if (head)
1377                 srcu_read_unlock(&head->srcu, idx);
1378 }
1379
1380 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1381 {
1382         switch (cmd) {
1383         case NVME_IOCTL_ID:
1384                 force_successful_syscall_return();
1385                 return ns->head->ns_id;
1386         case NVME_IOCTL_ADMIN_CMD:
1387                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1388         case NVME_IOCTL_IO_CMD:
1389                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1390         case NVME_IOCTL_SUBMIT_IO:
1391                 return nvme_submit_io(ns, (void __user *)arg);
1392         default:
1393 #ifdef CONFIG_NVM
1394                 if (ns->ndev)
1395                         return nvme_nvm_ioctl(ns, cmd, arg);
1396 #endif
1397                 if (is_sed_ioctl(cmd))
1398                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
1399                                          (void __user *) arg);
1400                 return -ENOTTY;
1401         }
1402 }
1403
1404 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1405                 unsigned int cmd, unsigned long arg)
1406 {
1407         struct nvme_ns_head *head = NULL;
1408         struct nvme_ns *ns;
1409         int srcu_idx, ret;
1410
1411         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1412         if (unlikely(!ns))
1413                 ret = -EWOULDBLOCK;
1414         else
1415                 ret = nvme_ns_ioctl(ns, cmd, arg);
1416         nvme_put_ns_from_disk(head, srcu_idx);
1417         return ret;
1418 }
1419
1420 static int nvme_open(struct block_device *bdev, fmode_t mode)
1421 {
1422         struct nvme_ns *ns = bdev->bd_disk->private_data;
1423
1424 #ifdef CONFIG_NVME_MULTIPATH
1425         /* should never be called due to GENHD_FL_HIDDEN */
1426         if (WARN_ON_ONCE(ns->head->disk))
1427                 goto fail;
1428 #endif
1429         if (!kref_get_unless_zero(&ns->kref))
1430                 goto fail;
1431         if (!try_module_get(ns->ctrl->ops->module))
1432                 goto fail_put_ns;
1433
1434         return 0;
1435
1436 fail_put_ns:
1437         nvme_put_ns(ns);
1438 fail:
1439         return -ENXIO;
1440 }
1441
1442 static void nvme_release(struct gendisk *disk, fmode_t mode)
1443 {
1444         struct nvme_ns *ns = disk->private_data;
1445
1446         module_put(ns->ctrl->ops->module);
1447         nvme_put_ns(ns);
1448 }
1449
1450 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1451 {
1452         /* some standard values */
1453         geo->heads = 1 << 6;
1454         geo->sectors = 1 << 5;
1455         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1456         return 0;
1457 }
1458
1459 #ifdef CONFIG_BLK_DEV_INTEGRITY
1460 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1461 {
1462         struct blk_integrity integrity;
1463
1464         memset(&integrity, 0, sizeof(integrity));
1465         switch (pi_type) {
1466         case NVME_NS_DPS_PI_TYPE3:
1467                 integrity.profile = &t10_pi_type3_crc;
1468                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1469                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1470                 break;
1471         case NVME_NS_DPS_PI_TYPE1:
1472         case NVME_NS_DPS_PI_TYPE2:
1473                 integrity.profile = &t10_pi_type1_crc;
1474                 integrity.tag_size = sizeof(u16);
1475                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1476                 break;
1477         default:
1478                 integrity.profile = NULL;
1479                 break;
1480         }
1481         integrity.tuple_size = ms;
1482         blk_integrity_register(disk, &integrity);
1483         blk_queue_max_integrity_segments(disk->queue, 1);
1484 }
1485 #else
1486 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1487 {
1488 }
1489 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1490
1491 static void nvme_set_chunk_size(struct nvme_ns *ns)
1492 {
1493         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1494         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1495 }
1496
1497 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1498 {
1499         struct nvme_ctrl *ctrl = ns->ctrl;
1500         struct request_queue *queue = disk->queue;
1501         u32 size = queue_logical_block_size(queue);
1502
1503         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1504                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1505                 return;
1506         }
1507
1508         if (ctrl->nr_streams && ns->sws && ns->sgs)
1509                 size *= ns->sws * ns->sgs;
1510
1511         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1512                         NVME_DSM_MAX_RANGES);
1513
1514         queue->limits.discard_alignment = 0;
1515         queue->limits.discard_granularity = size;
1516
1517         /* If discard is already enabled, don't reset queue limits */
1518         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1519                 return;
1520
1521         blk_queue_max_discard_sectors(queue, UINT_MAX);
1522         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1523
1524         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1525                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1526 }
1527
1528 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1529 {
1530         u32 max_sectors;
1531         unsigned short bs = 1 << ns->lba_shift;
1532
1533         if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1534             (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1535                 return;
1536         /*
1537          * Even though NVMe spec explicitly states that MDTS is not
1538          * applicable to the write-zeroes:- "The restriction does not apply to
1539          * commands that do not transfer data between the host and the
1540          * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1541          * In order to be more cautious use controller's max_hw_sectors value
1542          * to configure the maximum sectors for the write-zeroes which is
1543          * configured based on the controller's MDTS field in the
1544          * nvme_init_identify() if available.
1545          */
1546         if (ns->ctrl->max_hw_sectors == UINT_MAX)
1547                 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1548         else
1549                 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1550
1551         blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1552 }
1553
1554 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1555                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1556 {
1557         memset(ids, 0, sizeof(*ids));
1558
1559         if (ctrl->vs >= NVME_VS(1, 1, 0))
1560                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1561         if (ctrl->vs >= NVME_VS(1, 2, 0))
1562                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1563         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1564                  /* Don't treat error as fatal we potentially
1565                   * already have a NGUID or EUI-64
1566                   */
1567                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1568                         dev_warn(ctrl->device,
1569                                  "%s: Identify Descriptors failed\n", __func__);
1570         }
1571 }
1572
1573 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1574 {
1575         return !uuid_is_null(&ids->uuid) ||
1576                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1577                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1578 }
1579
1580 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1581 {
1582         return uuid_equal(&a->uuid, &b->uuid) &&
1583                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1584                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1585 }
1586
1587 static void nvme_update_disk_info(struct gendisk *disk,
1588                 struct nvme_ns *ns, struct nvme_id_ns *id)
1589 {
1590         sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1591         unsigned short bs = 1 << ns->lba_shift;
1592
1593         if (ns->lba_shift > PAGE_SHIFT) {
1594                 /* unsupported block size, set capacity to 0 later */
1595                 bs = (1 << 9);
1596         }
1597         blk_mq_freeze_queue(disk->queue);
1598         blk_integrity_unregister(disk);
1599
1600         blk_queue_logical_block_size(disk->queue, bs);
1601         blk_queue_physical_block_size(disk->queue, bs);
1602         blk_queue_io_min(disk->queue, bs);
1603
1604         if (ns->ms && !ns->ext &&
1605             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1606                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1607         if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1608             ns->lba_shift > PAGE_SHIFT)
1609                 capacity = 0;
1610
1611         set_capacity(disk, capacity);
1612
1613         nvme_config_discard(disk, ns);
1614         nvme_config_write_zeroes(disk, ns);
1615
1616         if (id->nsattr & (1 << 0))
1617                 set_disk_ro(disk, true);
1618         else
1619                 set_disk_ro(disk, false);
1620
1621         blk_mq_unfreeze_queue(disk->queue);
1622 }
1623
1624 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1625 {
1626         struct nvme_ns *ns = disk->private_data;
1627
1628         /*
1629          * If identify namespace failed, use default 512 byte block size so
1630          * block layer can use before failing read/write for 0 capacity.
1631          */
1632         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1633         if (ns->lba_shift == 0)
1634                 ns->lba_shift = 9;
1635         ns->noiob = le16_to_cpu(id->noiob);
1636         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1637         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1638         /* the PI implementation requires metadata equal t10 pi tuple size */
1639         if (ns->ms == sizeof(struct t10_pi_tuple))
1640                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1641         else
1642                 ns->pi_type = 0;
1643
1644         if (ns->noiob)
1645                 nvme_set_chunk_size(ns);
1646         nvme_update_disk_info(disk, ns, id);
1647 #ifdef CONFIG_NVME_MULTIPATH
1648         if (ns->head->disk) {
1649                 nvme_update_disk_info(ns->head->disk, ns, id);
1650                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1651         }
1652 #endif
1653 }
1654
1655 static int nvme_revalidate_disk(struct gendisk *disk)
1656 {
1657         struct nvme_ns *ns = disk->private_data;
1658         struct nvme_ctrl *ctrl = ns->ctrl;
1659         struct nvme_id_ns *id;
1660         struct nvme_ns_ids ids;
1661         int ret = 0;
1662
1663         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1664                 set_capacity(disk, 0);
1665                 return -ENODEV;
1666         }
1667
1668         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1669         if (!id)
1670                 return -ENODEV;
1671
1672         if (id->ncap == 0) {
1673                 ret = -ENODEV;
1674                 goto out;
1675         }
1676
1677         __nvme_revalidate_disk(disk, id);
1678         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1679         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1680                 dev_err(ctrl->device,
1681                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1682                 ret = -ENODEV;
1683         }
1684
1685 out:
1686         kfree(id);
1687         return ret;
1688 }
1689
1690 static char nvme_pr_type(enum pr_type type)
1691 {
1692         switch (type) {
1693         case PR_WRITE_EXCLUSIVE:
1694                 return 1;
1695         case PR_EXCLUSIVE_ACCESS:
1696                 return 2;
1697         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1698                 return 3;
1699         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1700                 return 4;
1701         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1702                 return 5;
1703         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1704                 return 6;
1705         default:
1706                 return 0;
1707         }
1708 };
1709
1710 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1711                                 u64 key, u64 sa_key, u8 op)
1712 {
1713         struct nvme_ns_head *head = NULL;
1714         struct nvme_ns *ns;
1715         struct nvme_command c;
1716         int srcu_idx, ret;
1717         u8 data[16] = { 0, };
1718
1719         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1720         if (unlikely(!ns))
1721                 return -EWOULDBLOCK;
1722
1723         put_unaligned_le64(key, &data[0]);
1724         put_unaligned_le64(sa_key, &data[8]);
1725
1726         memset(&c, 0, sizeof(c));
1727         c.common.opcode = op;
1728         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1729         c.common.cdw10 = cpu_to_le32(cdw10);
1730
1731         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1732         nvme_put_ns_from_disk(head, srcu_idx);
1733         return ret;
1734 }
1735
1736 static int nvme_pr_register(struct block_device *bdev, u64 old,
1737                 u64 new, unsigned flags)
1738 {
1739         u32 cdw10;
1740
1741         if (flags & ~PR_FL_IGNORE_KEY)
1742                 return -EOPNOTSUPP;
1743
1744         cdw10 = old ? 2 : 0;
1745         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1746         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1747         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1748 }
1749
1750 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1751                 enum pr_type type, unsigned flags)
1752 {
1753         u32 cdw10;
1754
1755         if (flags & ~PR_FL_IGNORE_KEY)
1756                 return -EOPNOTSUPP;
1757
1758         cdw10 = nvme_pr_type(type) << 8;
1759         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1760         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1761 }
1762
1763 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1764                 enum pr_type type, bool abort)
1765 {
1766         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1767         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1768 }
1769
1770 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1771 {
1772         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1773         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1774 }
1775
1776 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1777 {
1778         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1779         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1780 }
1781
1782 static const struct pr_ops nvme_pr_ops = {
1783         .pr_register    = nvme_pr_register,
1784         .pr_reserve     = nvme_pr_reserve,
1785         .pr_release     = nvme_pr_release,
1786         .pr_preempt     = nvme_pr_preempt,
1787         .pr_clear       = nvme_pr_clear,
1788 };
1789
1790 #ifdef CONFIG_BLK_SED_OPAL
1791 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1792                 bool send)
1793 {
1794         struct nvme_ctrl *ctrl = data;
1795         struct nvme_command cmd;
1796
1797         memset(&cmd, 0, sizeof(cmd));
1798         if (send)
1799                 cmd.common.opcode = nvme_admin_security_send;
1800         else
1801                 cmd.common.opcode = nvme_admin_security_recv;
1802         cmd.common.nsid = 0;
1803         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1804         cmd.common.cdw11 = cpu_to_le32(len);
1805
1806         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1807                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1808 }
1809 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1810 #endif /* CONFIG_BLK_SED_OPAL */
1811
1812 static const struct block_device_operations nvme_fops = {
1813         .owner          = THIS_MODULE,
1814         .ioctl          = nvme_ioctl,
1815         .compat_ioctl   = nvme_ioctl,
1816         .open           = nvme_open,
1817         .release        = nvme_release,
1818         .getgeo         = nvme_getgeo,
1819         .revalidate_disk= nvme_revalidate_disk,
1820         .pr_ops         = &nvme_pr_ops,
1821 };
1822
1823 #ifdef CONFIG_NVME_MULTIPATH
1824 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1825 {
1826         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1827
1828         if (!kref_get_unless_zero(&head->ref))
1829                 return -ENXIO;
1830         return 0;
1831 }
1832
1833 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1834 {
1835         nvme_put_ns_head(disk->private_data);
1836 }
1837
1838 const struct block_device_operations nvme_ns_head_ops = {
1839         .owner          = THIS_MODULE,
1840         .open           = nvme_ns_head_open,
1841         .release        = nvme_ns_head_release,
1842         .ioctl          = nvme_ioctl,
1843         .compat_ioctl   = nvme_ioctl,
1844         .getgeo         = nvme_getgeo,
1845         .pr_ops         = &nvme_pr_ops,
1846 };
1847 #endif /* CONFIG_NVME_MULTIPATH */
1848
1849 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1850 {
1851         unsigned long timeout =
1852                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1853         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1854         int ret;
1855
1856         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1857                 if (csts == ~0)
1858                         return -ENODEV;
1859                 if ((csts & NVME_CSTS_RDY) == bit)
1860                         break;
1861
1862                 msleep(100);
1863                 if (fatal_signal_pending(current))
1864                         return -EINTR;
1865                 if (time_after(jiffies, timeout)) {
1866                         dev_err(ctrl->device,
1867                                 "Device not ready; aborting %s\n", enabled ?
1868                                                 "initialisation" : "reset");
1869                         return -ENODEV;
1870                 }
1871         }
1872
1873         return ret;
1874 }
1875
1876 /*
1877  * If the device has been passed off to us in an enabled state, just clear
1878  * the enabled bit.  The spec says we should set the 'shutdown notification
1879  * bits', but doing so may cause the device to complete commands to the
1880  * admin queue ... and we don't know what memory that might be pointing at!
1881  */
1882 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1883 {
1884         int ret;
1885
1886         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1887         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1888
1889         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1890         if (ret)
1891                 return ret;
1892
1893         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1894                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1895
1896         return nvme_wait_ready(ctrl, cap, false);
1897 }
1898 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1899
1900 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1901 {
1902         /*
1903          * Default to a 4K page size, with the intention to update this
1904          * path in the future to accomodate architectures with differing
1905          * kernel and IO page sizes.
1906          */
1907         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1908         int ret;
1909
1910         if (page_shift < dev_page_min) {
1911                 dev_err(ctrl->device,
1912                         "Minimum device page size %u too large for host (%u)\n",
1913                         1 << dev_page_min, 1 << page_shift);
1914                 return -ENODEV;
1915         }
1916
1917         ctrl->page_size = 1 << page_shift;
1918
1919         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1920         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1921         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1922         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1923         ctrl->ctrl_config |= NVME_CC_ENABLE;
1924
1925         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1926         if (ret)
1927                 return ret;
1928         return nvme_wait_ready(ctrl, cap, true);
1929 }
1930 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1931
1932 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1933 {
1934         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1935         u32 csts;
1936         int ret;
1937
1938         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1939         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1940
1941         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1942         if (ret)
1943                 return ret;
1944
1945         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1946                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1947                         break;
1948
1949                 msleep(100);
1950                 if (fatal_signal_pending(current))
1951                         return -EINTR;
1952                 if (time_after(jiffies, timeout)) {
1953                         dev_err(ctrl->device,
1954                                 "Device shutdown incomplete; abort shutdown\n");
1955                         return -ENODEV;
1956                 }
1957         }
1958
1959         return ret;
1960 }
1961 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1962
1963 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1964                 struct request_queue *q)
1965 {
1966         bool vwc = false;
1967
1968         if (ctrl->max_hw_sectors) {
1969                 u32 max_segments =
1970                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1971
1972                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1973                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1974                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1975         }
1976         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1977             is_power_of_2(ctrl->max_hw_sectors))
1978                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1979         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1980         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1981                 vwc = true;
1982         blk_queue_write_cache(q, vwc, vwc);
1983 }
1984
1985 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1986 {
1987         __le64 ts;
1988         int ret;
1989
1990         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1991                 return 0;
1992
1993         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1994         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1995                         NULL);
1996         if (ret)
1997                 dev_warn_once(ctrl->device,
1998                         "could not set timestamp (%d)\n", ret);
1999         return ret;
2000 }
2001
2002 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2003 {
2004         struct nvme_feat_host_behavior *host;
2005         int ret;
2006
2007         /* Don't bother enabling the feature if retry delay is not reported */
2008         if (!ctrl->crdt[0])
2009                 return 0;
2010
2011         host = kzalloc(sizeof(*host), GFP_KERNEL);
2012         if (!host)
2013                 return 0;
2014
2015         host->acre = NVME_ENABLE_ACRE;
2016         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2017                                 host, sizeof(*host), NULL);
2018         kfree(host);
2019         return ret;
2020 }
2021
2022 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2023 {
2024         /*
2025          * APST (Autonomous Power State Transition) lets us program a
2026          * table of power state transitions that the controller will
2027          * perform automatically.  We configure it with a simple
2028          * heuristic: we are willing to spend at most 2% of the time
2029          * transitioning between power states.  Therefore, when running
2030          * in any given state, we will enter the next lower-power
2031          * non-operational state after waiting 50 * (enlat + exlat)
2032          * microseconds, as long as that state's exit latency is under
2033          * the requested maximum latency.
2034          *
2035          * We will not autonomously enter any non-operational state for
2036          * which the total latency exceeds ps_max_latency_us.  Users
2037          * can set ps_max_latency_us to zero to turn off APST.
2038          */
2039
2040         unsigned apste;
2041         struct nvme_feat_auto_pst *table;
2042         u64 max_lat_us = 0;
2043         int max_ps = -1;
2044         int ret;
2045
2046         /*
2047          * If APST isn't supported or if we haven't been initialized yet,
2048          * then don't do anything.
2049          */
2050         if (!ctrl->apsta)
2051                 return 0;
2052
2053         if (ctrl->npss > 31) {
2054                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2055                 return 0;
2056         }
2057
2058         table = kzalloc(sizeof(*table), GFP_KERNEL);
2059         if (!table)
2060                 return 0;
2061
2062         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2063                 /* Turn off APST. */
2064                 apste = 0;
2065                 dev_dbg(ctrl->device, "APST disabled\n");
2066         } else {
2067                 __le64 target = cpu_to_le64(0);
2068                 int state;
2069
2070                 /*
2071                  * Walk through all states from lowest- to highest-power.
2072                  * According to the spec, lower-numbered states use more
2073                  * power.  NPSS, despite the name, is the index of the
2074                  * lowest-power state, not the number of states.
2075                  */
2076                 for (state = (int)ctrl->npss; state >= 0; state--) {
2077                         u64 total_latency_us, exit_latency_us, transition_ms;
2078
2079                         if (target)
2080                                 table->entries[state] = target;
2081
2082                         /*
2083                          * Don't allow transitions to the deepest state
2084                          * if it's quirked off.
2085                          */
2086                         if (state == ctrl->npss &&
2087                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2088                                 continue;
2089
2090                         /*
2091                          * Is this state a useful non-operational state for
2092                          * higher-power states to autonomously transition to?
2093                          */
2094                         if (!(ctrl->psd[state].flags &
2095                               NVME_PS_FLAGS_NON_OP_STATE))
2096                                 continue;
2097
2098                         exit_latency_us =
2099                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2100                         if (exit_latency_us > ctrl->ps_max_latency_us)
2101                                 continue;
2102
2103                         total_latency_us =
2104                                 exit_latency_us +
2105                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2106
2107                         /*
2108                          * This state is good.  Use it as the APST idle
2109                          * target for higher power states.
2110                          */
2111                         transition_ms = total_latency_us + 19;
2112                         do_div(transition_ms, 20);
2113                         if (transition_ms > (1 << 24) - 1)
2114                                 transition_ms = (1 << 24) - 1;
2115
2116                         target = cpu_to_le64((state << 3) |
2117                                              (transition_ms << 8));
2118
2119                         if (max_ps == -1)
2120                                 max_ps = state;
2121
2122                         if (total_latency_us > max_lat_us)
2123                                 max_lat_us = total_latency_us;
2124                 }
2125
2126                 apste = 1;
2127
2128                 if (max_ps == -1) {
2129                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2130                 } else {
2131                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2132                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2133                 }
2134         }
2135
2136         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2137                                 table, sizeof(*table), NULL);
2138         if (ret)
2139                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2140
2141         kfree(table);
2142         return ret;
2143 }
2144
2145 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2146 {
2147         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2148         u64 latency;
2149
2150         switch (val) {
2151         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2152         case PM_QOS_LATENCY_ANY:
2153                 latency = U64_MAX;
2154                 break;
2155
2156         default:
2157                 latency = val;
2158         }
2159
2160         if (ctrl->ps_max_latency_us != latency) {
2161                 ctrl->ps_max_latency_us = latency;
2162                 nvme_configure_apst(ctrl);
2163         }
2164 }
2165
2166 struct nvme_core_quirk_entry {
2167         /*
2168          * NVMe model and firmware strings are padded with spaces.  For
2169          * simplicity, strings in the quirk table are padded with NULLs
2170          * instead.
2171          */
2172         u16 vid;
2173         const char *mn;
2174         const char *fr;
2175         unsigned long quirks;
2176 };
2177
2178 static const struct nvme_core_quirk_entry core_quirks[] = {
2179         {
2180                 /*
2181                  * This Toshiba device seems to die using any APST states.  See:
2182                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2183                  */
2184                 .vid = 0x1179,
2185                 .mn = "THNSF5256GPUK TOSHIBA",
2186                 .quirks = NVME_QUIRK_NO_APST,
2187         }
2188 };
2189
2190 /* match is null-terminated but idstr is space-padded. */
2191 static bool string_matches(const char *idstr, const char *match, size_t len)
2192 {
2193         size_t matchlen;
2194
2195         if (!match)
2196                 return true;
2197
2198         matchlen = strlen(match);
2199         WARN_ON_ONCE(matchlen > len);
2200
2201         if (memcmp(idstr, match, matchlen))
2202                 return false;
2203
2204         for (; matchlen < len; matchlen++)
2205                 if (idstr[matchlen] != ' ')
2206                         return false;
2207
2208         return true;
2209 }
2210
2211 static bool quirk_matches(const struct nvme_id_ctrl *id,
2212                           const struct nvme_core_quirk_entry *q)
2213 {
2214         return q->vid == le16_to_cpu(id->vid) &&
2215                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2216                 string_matches(id->fr, q->fr, sizeof(id->fr));
2217 }
2218
2219 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2220                 struct nvme_id_ctrl *id)
2221 {
2222         size_t nqnlen;
2223         int off;
2224
2225         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2226                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2227                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2228                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2229                         return;
2230                 }
2231
2232                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2233                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2234         }
2235
2236         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2237         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2238                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2239                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2240         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2241         off += sizeof(id->sn);
2242         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2243         off += sizeof(id->mn);
2244         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2245 }
2246
2247 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2248 {
2249         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2250         kfree(subsys);
2251 }
2252
2253 static void nvme_release_subsystem(struct device *dev)
2254 {
2255         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2256 }
2257
2258 static void nvme_destroy_subsystem(struct kref *ref)
2259 {
2260         struct nvme_subsystem *subsys =
2261                         container_of(ref, struct nvme_subsystem, ref);
2262
2263         mutex_lock(&nvme_subsystems_lock);
2264         list_del(&subsys->entry);
2265         mutex_unlock(&nvme_subsystems_lock);
2266
2267         ida_destroy(&subsys->ns_ida);
2268         device_del(&subsys->dev);
2269         put_device(&subsys->dev);
2270 }
2271
2272 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2273 {
2274         kref_put(&subsys->ref, nvme_destroy_subsystem);
2275 }
2276
2277 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2278 {
2279         struct nvme_subsystem *subsys;
2280
2281         lockdep_assert_held(&nvme_subsystems_lock);
2282
2283         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2284                 if (strcmp(subsys->subnqn, subsysnqn))
2285                         continue;
2286                 if (!kref_get_unless_zero(&subsys->ref))
2287                         continue;
2288                 return subsys;
2289         }
2290
2291         return NULL;
2292 }
2293
2294 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2295         struct device_attribute subsys_attr_##_name = \
2296                 __ATTR(_name, _mode, _show, NULL)
2297
2298 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2299                                     struct device_attribute *attr,
2300                                     char *buf)
2301 {
2302         struct nvme_subsystem *subsys =
2303                 container_of(dev, struct nvme_subsystem, dev);
2304
2305         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2306 }
2307 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2308
2309 #define nvme_subsys_show_str_function(field)                            \
2310 static ssize_t subsys_##field##_show(struct device *dev,                \
2311                             struct device_attribute *attr, char *buf)   \
2312 {                                                                       \
2313         struct nvme_subsystem *subsys =                                 \
2314                 container_of(dev, struct nvme_subsystem, dev);          \
2315         return sprintf(buf, "%.*s\n",                                   \
2316                        (int)sizeof(subsys->field), subsys->field);      \
2317 }                                                                       \
2318 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2319
2320 nvme_subsys_show_str_function(model);
2321 nvme_subsys_show_str_function(serial);
2322 nvme_subsys_show_str_function(firmware_rev);
2323
2324 static struct attribute *nvme_subsys_attrs[] = {
2325         &subsys_attr_model.attr,
2326         &subsys_attr_serial.attr,
2327         &subsys_attr_firmware_rev.attr,
2328         &subsys_attr_subsysnqn.attr,
2329 #ifdef CONFIG_NVME_MULTIPATH
2330         &subsys_attr_iopolicy.attr,
2331 #endif
2332         NULL,
2333 };
2334
2335 static struct attribute_group nvme_subsys_attrs_group = {
2336         .attrs = nvme_subsys_attrs,
2337 };
2338
2339 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2340         &nvme_subsys_attrs_group,
2341         NULL,
2342 };
2343
2344 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2345 {
2346         int count = 0;
2347         struct nvme_ctrl *ctrl;
2348
2349         mutex_lock(&subsys->lock);
2350         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2351                 if (ctrl->state != NVME_CTRL_DELETING &&
2352                     ctrl->state != NVME_CTRL_DEAD)
2353                         count++;
2354         }
2355         mutex_unlock(&subsys->lock);
2356
2357         return count;
2358 }
2359
2360 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2361 {
2362         struct nvme_subsystem *subsys, *found;
2363         int ret;
2364
2365         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2366         if (!subsys)
2367                 return -ENOMEM;
2368         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2369         if (ret < 0) {
2370                 kfree(subsys);
2371                 return ret;
2372         }
2373         subsys->instance = ret;
2374         mutex_init(&subsys->lock);
2375         kref_init(&subsys->ref);
2376         INIT_LIST_HEAD(&subsys->ctrls);
2377         INIT_LIST_HEAD(&subsys->nsheads);
2378         nvme_init_subnqn(subsys, ctrl, id);
2379         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2380         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2381         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2382         subsys->vendor_id = le16_to_cpu(id->vid);
2383         subsys->cmic = id->cmic;
2384 #ifdef CONFIG_NVME_MULTIPATH
2385         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2386 #endif
2387
2388         subsys->dev.class = nvme_subsys_class;
2389         subsys->dev.release = nvme_release_subsystem;
2390         subsys->dev.groups = nvme_subsys_attrs_groups;
2391         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2392         device_initialize(&subsys->dev);
2393
2394         mutex_lock(&nvme_subsystems_lock);
2395         found = __nvme_find_get_subsystem(subsys->subnqn);
2396         if (found) {
2397                 /*
2398                  * Verify that the subsystem actually supports multiple
2399                  * controllers, else bail out.
2400                  */
2401                 if (!(ctrl->opts && ctrl->opts->discovery_nqn) &&
2402                     nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2403                         dev_err(ctrl->device,
2404                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2405                                 found->subnqn);
2406                         nvme_put_subsystem(found);
2407                         ret = -EINVAL;
2408                         goto out_unlock;
2409                 }
2410
2411                 __nvme_release_subsystem(subsys);
2412                 subsys = found;
2413         } else {
2414                 ret = device_add(&subsys->dev);
2415                 if (ret) {
2416                         dev_err(ctrl->device,
2417                                 "failed to register subsystem device.\n");
2418                         goto out_unlock;
2419                 }
2420                 ida_init(&subsys->ns_ida);
2421                 list_add_tail(&subsys->entry, &nvme_subsystems);
2422         }
2423
2424         ctrl->subsys = subsys;
2425         mutex_unlock(&nvme_subsystems_lock);
2426
2427         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2428                         dev_name(ctrl->device))) {
2429                 dev_err(ctrl->device,
2430                         "failed to create sysfs link from subsystem.\n");
2431                 /* the transport driver will eventually put the subsystem */
2432                 return -EINVAL;
2433         }
2434
2435         mutex_lock(&subsys->lock);
2436         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2437         mutex_unlock(&subsys->lock);
2438
2439         return 0;
2440
2441 out_unlock:
2442         mutex_unlock(&nvme_subsystems_lock);
2443         put_device(&subsys->dev);
2444         return ret;
2445 }
2446
2447 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2448                 void *log, size_t size, u64 offset)
2449 {
2450         struct nvme_command c = { };
2451         unsigned long dwlen = size / 4 - 1;
2452
2453         c.get_log_page.opcode = nvme_admin_get_log_page;
2454         c.get_log_page.nsid = cpu_to_le32(nsid);
2455         c.get_log_page.lid = log_page;
2456         c.get_log_page.lsp = lsp;
2457         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2458         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2459         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2460         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2461
2462         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2463 }
2464
2465 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2466 {
2467         int ret;
2468
2469         if (!ctrl->effects)
2470                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2471
2472         if (!ctrl->effects)
2473                 return 0;
2474
2475         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2476                         ctrl->effects, sizeof(*ctrl->effects), 0);
2477         if (ret) {
2478                 kfree(ctrl->effects);
2479                 ctrl->effects = NULL;
2480         }
2481         return ret;
2482 }
2483
2484 /*
2485  * Initialize the cached copies of the Identify data and various controller
2486  * register in our nvme_ctrl structure.  This should be called as soon as
2487  * the admin queue is fully up and running.
2488  */
2489 int nvme_init_identify(struct nvme_ctrl *ctrl)
2490 {
2491         struct nvme_id_ctrl *id;
2492         u64 cap;
2493         int ret, page_shift;
2494         u32 max_hw_sectors;
2495         bool prev_apst_enabled;
2496
2497         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2498         if (ret) {
2499                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2500                 return ret;
2501         }
2502
2503         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2504         if (ret) {
2505                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2506                 return ret;
2507         }
2508         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2509
2510         if (ctrl->vs >= NVME_VS(1, 1, 0))
2511                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2512
2513         ret = nvme_identify_ctrl(ctrl, &id);
2514         if (ret) {
2515                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2516                 return -EIO;
2517         }
2518
2519         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2520                 ret = nvme_get_effects_log(ctrl);
2521                 if (ret < 0)
2522                         goto out_free;
2523         }
2524
2525         if (!ctrl->identified) {
2526                 int i;
2527
2528                 ret = nvme_init_subsystem(ctrl, id);
2529                 if (ret)
2530                         goto out_free;
2531
2532                 /*
2533                  * Check for quirks.  Quirk can depend on firmware version,
2534                  * so, in principle, the set of quirks present can change
2535                  * across a reset.  As a possible future enhancement, we
2536                  * could re-scan for quirks every time we reinitialize
2537                  * the device, but we'd have to make sure that the driver
2538                  * behaves intelligently if the quirks change.
2539                  */
2540                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2541                         if (quirk_matches(id, &core_quirks[i]))
2542                                 ctrl->quirks |= core_quirks[i].quirks;
2543                 }
2544         }
2545
2546         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2547                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2548                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2549         }
2550
2551         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2552         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2553         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2554
2555         ctrl->oacs = le16_to_cpu(id->oacs);
2556         ctrl->oncs = le16_to_cpu(id->oncs);
2557         ctrl->oaes = le32_to_cpu(id->oaes);
2558         atomic_set(&ctrl->abort_limit, id->acl + 1);
2559         ctrl->vwc = id->vwc;
2560         if (id->mdts)
2561                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2562         else
2563                 max_hw_sectors = UINT_MAX;
2564         ctrl->max_hw_sectors =
2565                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2566
2567         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2568         ctrl->sgls = le32_to_cpu(id->sgls);
2569         ctrl->kas = le16_to_cpu(id->kas);
2570         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2571         ctrl->ctratt = le32_to_cpu(id->ctratt);
2572
2573         if (id->rtd3e) {
2574                 /* us -> s */
2575                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2576
2577                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2578                                                  shutdown_timeout, 60);
2579
2580                 if (ctrl->shutdown_timeout != shutdown_timeout)
2581                         dev_info(ctrl->device,
2582                                  "Shutdown timeout set to %u seconds\n",
2583                                  ctrl->shutdown_timeout);
2584         } else
2585                 ctrl->shutdown_timeout = shutdown_timeout;
2586
2587         ctrl->npss = id->npss;
2588         ctrl->apsta = id->apsta;
2589         prev_apst_enabled = ctrl->apst_enabled;
2590         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2591                 if (force_apst && id->apsta) {
2592                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2593                         ctrl->apst_enabled = true;
2594                 } else {
2595                         ctrl->apst_enabled = false;
2596                 }
2597         } else {
2598                 ctrl->apst_enabled = id->apsta;
2599         }
2600         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2601
2602         if (ctrl->ops->flags & NVME_F_FABRICS) {
2603                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2604                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2605                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2606                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2607
2608                 /*
2609                  * In fabrics we need to verify the cntlid matches the
2610                  * admin connect
2611                  */
2612                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2613                         ret = -EINVAL;
2614                         goto out_free;
2615                 }
2616
2617                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2618                         dev_err(ctrl->device,
2619                                 "keep-alive support is mandatory for fabrics\n");
2620                         ret = -EINVAL;
2621                         goto out_free;
2622                 }
2623         } else {
2624                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2625                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2626                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2627                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2628                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2629         }
2630
2631         ret = nvme_mpath_init(ctrl, id);
2632         kfree(id);
2633
2634         if (ret < 0)
2635                 return ret;
2636
2637         if (ctrl->apst_enabled && !prev_apst_enabled)
2638                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2639         else if (!ctrl->apst_enabled && prev_apst_enabled)
2640                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2641
2642         ret = nvme_configure_apst(ctrl);
2643         if (ret < 0)
2644                 return ret;
2645
2646         ret = nvme_configure_timestamp(ctrl);
2647         if (ret < 0)
2648                 return ret;
2649
2650         ret = nvme_configure_directives(ctrl);
2651         if (ret < 0)
2652                 return ret;
2653
2654         ret = nvme_configure_acre(ctrl);
2655         if (ret < 0)
2656                 return ret;
2657
2658         ctrl->identified = true;
2659
2660         return 0;
2661
2662 out_free:
2663         kfree(id);
2664         return ret;
2665 }
2666 EXPORT_SYMBOL_GPL(nvme_init_identify);
2667
2668 static int nvme_dev_open(struct inode *inode, struct file *file)
2669 {
2670         struct nvme_ctrl *ctrl =
2671                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2672
2673         switch (ctrl->state) {
2674         case NVME_CTRL_LIVE:
2675         case NVME_CTRL_ADMIN_ONLY:
2676                 break;
2677         default:
2678                 return -EWOULDBLOCK;
2679         }
2680
2681         file->private_data = ctrl;
2682         return 0;
2683 }
2684
2685 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2686 {
2687         struct nvme_ns *ns;
2688         int ret;
2689
2690         down_read(&ctrl->namespaces_rwsem);
2691         if (list_empty(&ctrl->namespaces)) {
2692                 ret = -ENOTTY;
2693                 goto out_unlock;
2694         }
2695
2696         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2697         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2698                 dev_warn(ctrl->device,
2699                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2700                 ret = -EINVAL;
2701                 goto out_unlock;
2702         }
2703
2704         dev_warn(ctrl->device,
2705                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2706         kref_get(&ns->kref);
2707         up_read(&ctrl->namespaces_rwsem);
2708
2709         ret = nvme_user_cmd(ctrl, ns, argp);
2710         nvme_put_ns(ns);
2711         return ret;
2712
2713 out_unlock:
2714         up_read(&ctrl->namespaces_rwsem);
2715         return ret;
2716 }
2717
2718 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2719                 unsigned long arg)
2720 {
2721         struct nvme_ctrl *ctrl = file->private_data;
2722         void __user *argp = (void __user *)arg;
2723
2724         switch (cmd) {
2725         case NVME_IOCTL_ADMIN_CMD:
2726                 return nvme_user_cmd(ctrl, NULL, argp);
2727         case NVME_IOCTL_IO_CMD:
2728                 return nvme_dev_user_cmd(ctrl, argp);
2729         case NVME_IOCTL_RESET:
2730                 dev_warn(ctrl->device, "resetting controller\n");
2731                 return nvme_reset_ctrl_sync(ctrl);
2732         case NVME_IOCTL_SUBSYS_RESET:
2733                 return nvme_reset_subsystem(ctrl);
2734         case NVME_IOCTL_RESCAN:
2735                 nvme_queue_scan(ctrl);
2736                 return 0;
2737         default:
2738                 return -ENOTTY;
2739         }
2740 }
2741
2742 static const struct file_operations nvme_dev_fops = {
2743         .owner          = THIS_MODULE,
2744         .open           = nvme_dev_open,
2745         .unlocked_ioctl = nvme_dev_ioctl,
2746         .compat_ioctl   = nvme_dev_ioctl,
2747 };
2748
2749 static ssize_t nvme_sysfs_reset(struct device *dev,
2750                                 struct device_attribute *attr, const char *buf,
2751                                 size_t count)
2752 {
2753         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2754         int ret;
2755
2756         ret = nvme_reset_ctrl_sync(ctrl);
2757         if (ret < 0)
2758                 return ret;
2759         return count;
2760 }
2761 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2762
2763 static ssize_t nvme_sysfs_rescan(struct device *dev,
2764                                 struct device_attribute *attr, const char *buf,
2765                                 size_t count)
2766 {
2767         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2768
2769         nvme_queue_scan(ctrl);
2770         return count;
2771 }
2772 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2773
2774 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2775 {
2776         struct gendisk *disk = dev_to_disk(dev);
2777
2778         if (disk->fops == &nvme_fops)
2779                 return nvme_get_ns_from_dev(dev)->head;
2780         else
2781                 return disk->private_data;
2782 }
2783
2784 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2785                 char *buf)
2786 {
2787         struct nvme_ns_head *head = dev_to_ns_head(dev);
2788         struct nvme_ns_ids *ids = &head->ids;
2789         struct nvme_subsystem *subsys = head->subsys;
2790         int serial_len = sizeof(subsys->serial);
2791         int model_len = sizeof(subsys->model);
2792
2793         if (!uuid_is_null(&ids->uuid))
2794                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2795
2796         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2797                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2798
2799         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2800                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2801
2802         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2803                                   subsys->serial[serial_len - 1] == '\0'))
2804                 serial_len--;
2805         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2806                                  subsys->model[model_len - 1] == '\0'))
2807                 model_len--;
2808
2809         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2810                 serial_len, subsys->serial, model_len, subsys->model,
2811                 head->ns_id);
2812 }
2813 static DEVICE_ATTR_RO(wwid);
2814
2815 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2816                 char *buf)
2817 {
2818         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2819 }
2820 static DEVICE_ATTR_RO(nguid);
2821
2822 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2823                 char *buf)
2824 {
2825         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2826
2827         /* For backward compatibility expose the NGUID to userspace if
2828          * we have no UUID set
2829          */
2830         if (uuid_is_null(&ids->uuid)) {
2831                 printk_ratelimited(KERN_WARNING
2832                                    "No UUID available providing old NGUID\n");
2833                 return sprintf(buf, "%pU\n", ids->nguid);
2834         }
2835         return sprintf(buf, "%pU\n", &ids->uuid);
2836 }
2837 static DEVICE_ATTR_RO(uuid);
2838
2839 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2840                 char *buf)
2841 {
2842         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2843 }
2844 static DEVICE_ATTR_RO(eui);
2845
2846 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2847                 char *buf)
2848 {
2849         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2850 }
2851 static DEVICE_ATTR_RO(nsid);
2852
2853 static struct attribute *nvme_ns_id_attrs[] = {
2854         &dev_attr_wwid.attr,
2855         &dev_attr_uuid.attr,
2856         &dev_attr_nguid.attr,
2857         &dev_attr_eui.attr,
2858         &dev_attr_nsid.attr,
2859 #ifdef CONFIG_NVME_MULTIPATH
2860         &dev_attr_ana_grpid.attr,
2861         &dev_attr_ana_state.attr,
2862 #endif
2863         NULL,
2864 };
2865
2866 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2867                 struct attribute *a, int n)
2868 {
2869         struct device *dev = container_of(kobj, struct device, kobj);
2870         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2871
2872         if (a == &dev_attr_uuid.attr) {
2873                 if (uuid_is_null(&ids->uuid) &&
2874                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2875                         return 0;
2876         }
2877         if (a == &dev_attr_nguid.attr) {
2878                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2879                         return 0;
2880         }
2881         if (a == &dev_attr_eui.attr) {
2882                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2883                         return 0;
2884         }
2885 #ifdef CONFIG_NVME_MULTIPATH
2886         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2887                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2888                         return 0;
2889                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2890                         return 0;
2891         }
2892 #endif
2893         return a->mode;
2894 }
2895
2896 static const struct attribute_group nvme_ns_id_attr_group = {
2897         .attrs          = nvme_ns_id_attrs,
2898         .is_visible     = nvme_ns_id_attrs_are_visible,
2899 };
2900
2901 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2902         &nvme_ns_id_attr_group,
2903 #ifdef CONFIG_NVM
2904         &nvme_nvm_attr_group,
2905 #endif
2906         NULL,
2907 };
2908
2909 #define nvme_show_str_function(field)                                           \
2910 static ssize_t  field##_show(struct device *dev,                                \
2911                             struct device_attribute *attr, char *buf)           \
2912 {                                                                               \
2913         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2914         return sprintf(buf, "%.*s\n",                                           \
2915                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2916 }                                                                               \
2917 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2918
2919 nvme_show_str_function(model);
2920 nvme_show_str_function(serial);
2921 nvme_show_str_function(firmware_rev);
2922
2923 #define nvme_show_int_function(field)                                           \
2924 static ssize_t  field##_show(struct device *dev,                                \
2925                             struct device_attribute *attr, char *buf)           \
2926 {                                                                               \
2927         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2928         return sprintf(buf, "%d\n", ctrl->field);       \
2929 }                                                                               \
2930 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2931
2932 nvme_show_int_function(cntlid);
2933 nvme_show_int_function(numa_node);
2934
2935 static ssize_t nvme_sysfs_delete(struct device *dev,
2936                                 struct device_attribute *attr, const char *buf,
2937                                 size_t count)
2938 {
2939         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2940
2941         if (device_remove_file_self(dev, attr))
2942                 nvme_delete_ctrl_sync(ctrl);
2943         return count;
2944 }
2945 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2946
2947 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2948                                          struct device_attribute *attr,
2949                                          char *buf)
2950 {
2951         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2952
2953         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2954 }
2955 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2956
2957 static ssize_t nvme_sysfs_show_state(struct device *dev,
2958                                      struct device_attribute *attr,
2959                                      char *buf)
2960 {
2961         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2962         static const char *const state_name[] = {
2963                 [NVME_CTRL_NEW]         = "new",
2964                 [NVME_CTRL_LIVE]        = "live",
2965                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2966                 [NVME_CTRL_RESETTING]   = "resetting",
2967                 [NVME_CTRL_CONNECTING]  = "connecting",
2968                 [NVME_CTRL_DELETING]    = "deleting",
2969                 [NVME_CTRL_DEAD]        = "dead",
2970         };
2971
2972         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2973             state_name[ctrl->state])
2974                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2975
2976         return sprintf(buf, "unknown state\n");
2977 }
2978
2979 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2980
2981 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2982                                          struct device_attribute *attr,
2983                                          char *buf)
2984 {
2985         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2986
2987         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2988 }
2989 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2990
2991 static ssize_t nvme_sysfs_show_address(struct device *dev,
2992                                          struct device_attribute *attr,
2993                                          char *buf)
2994 {
2995         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2996
2997         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2998 }
2999 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3000
3001 static struct attribute *nvme_dev_attrs[] = {
3002         &dev_attr_reset_controller.attr,
3003         &dev_attr_rescan_controller.attr,
3004         &dev_attr_model.attr,
3005         &dev_attr_serial.attr,
3006         &dev_attr_firmware_rev.attr,
3007         &dev_attr_cntlid.attr,
3008         &dev_attr_delete_controller.attr,
3009         &dev_attr_transport.attr,
3010         &dev_attr_subsysnqn.attr,
3011         &dev_attr_address.attr,
3012         &dev_attr_state.attr,
3013         &dev_attr_numa_node.attr,
3014         NULL
3015 };
3016
3017 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3018                 struct attribute *a, int n)
3019 {
3020         struct device *dev = container_of(kobj, struct device, kobj);
3021         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3022
3023         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3024                 return 0;
3025         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3026                 return 0;
3027
3028         return a->mode;
3029 }
3030
3031 static struct attribute_group nvme_dev_attrs_group = {
3032         .attrs          = nvme_dev_attrs,
3033         .is_visible     = nvme_dev_attrs_are_visible,
3034 };
3035
3036 static const struct attribute_group *nvme_dev_attr_groups[] = {
3037         &nvme_dev_attrs_group,
3038         NULL,
3039 };
3040
3041 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3042                 unsigned nsid)
3043 {
3044         struct nvme_ns_head *h;
3045
3046         lockdep_assert_held(&subsys->lock);
3047
3048         list_for_each_entry(h, &subsys->nsheads, entry) {
3049                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3050                         return h;
3051         }
3052
3053         return NULL;
3054 }
3055
3056 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3057                 struct nvme_ns_head *new)
3058 {
3059         struct nvme_ns_head *h;
3060
3061         lockdep_assert_held(&subsys->lock);
3062
3063         list_for_each_entry(h, &subsys->nsheads, entry) {
3064                 if (nvme_ns_ids_valid(&new->ids) &&
3065                     !list_empty(&h->list) &&
3066                     nvme_ns_ids_equal(&new->ids, &h->ids))
3067                         return -EINVAL;
3068         }
3069
3070         return 0;
3071 }
3072
3073 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3074                 unsigned nsid, struct nvme_id_ns *id)
3075 {
3076         struct nvme_ns_head *head;
3077         size_t size = sizeof(*head);
3078         int ret = -ENOMEM;
3079
3080 #ifdef CONFIG_NVME_MULTIPATH
3081         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3082 #endif
3083
3084         head = kzalloc(size, GFP_KERNEL);
3085         if (!head)
3086                 goto out;
3087         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3088         if (ret < 0)
3089                 goto out_free_head;
3090         head->instance = ret;
3091         INIT_LIST_HEAD(&head->list);
3092         ret = init_srcu_struct(&head->srcu);
3093         if (ret)
3094                 goto out_ida_remove;
3095         head->subsys = ctrl->subsys;
3096         head->ns_id = nsid;
3097         kref_init(&head->ref);
3098
3099         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3100
3101         ret = __nvme_check_ids(ctrl->subsys, head);
3102         if (ret) {
3103                 dev_err(ctrl->device,
3104                         "duplicate IDs for nsid %d\n", nsid);
3105                 goto out_cleanup_srcu;
3106         }
3107
3108         ret = nvme_mpath_alloc_disk(ctrl, head);
3109         if (ret)
3110                 goto out_cleanup_srcu;
3111
3112         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3113
3114         kref_get(&ctrl->subsys->ref);
3115
3116         return head;
3117 out_cleanup_srcu:
3118         cleanup_srcu_struct(&head->srcu);
3119 out_ida_remove:
3120         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3121 out_free_head:
3122         kfree(head);
3123 out:
3124         return ERR_PTR(ret);
3125 }
3126
3127 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3128                 struct nvme_id_ns *id)
3129 {
3130         struct nvme_ctrl *ctrl = ns->ctrl;
3131         bool is_shared = id->nmic & (1 << 0);
3132         struct nvme_ns_head *head = NULL;
3133         int ret = 0;
3134
3135         mutex_lock(&ctrl->subsys->lock);
3136         if (is_shared)
3137                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3138         if (!head) {
3139                 head = nvme_alloc_ns_head(ctrl, nsid, id);
3140                 if (IS_ERR(head)) {
3141                         ret = PTR_ERR(head);
3142                         goto out_unlock;
3143                 }
3144         } else {
3145                 struct nvme_ns_ids ids;
3146
3147                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3148                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3149                         dev_err(ctrl->device,
3150                                 "IDs don't match for shared namespace %d\n",
3151                                         nsid);
3152                         ret = -EINVAL;
3153                         goto out_unlock;
3154                 }
3155         }
3156
3157         list_add_tail(&ns->siblings, &head->list);
3158         ns->head = head;
3159
3160 out_unlock:
3161         mutex_unlock(&ctrl->subsys->lock);
3162         return ret;
3163 }
3164
3165 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3166 {
3167         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3168         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3169
3170         return nsa->head->ns_id - nsb->head->ns_id;
3171 }
3172
3173 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3174 {
3175         struct nvme_ns *ns, *ret = NULL;
3176
3177         down_read(&ctrl->namespaces_rwsem);
3178         list_for_each_entry(ns, &ctrl->namespaces, list) {
3179                 if (ns->head->ns_id == nsid) {
3180                         if (!kref_get_unless_zero(&ns->kref))
3181                                 continue;
3182                         ret = ns;
3183                         break;
3184                 }
3185                 if (ns->head->ns_id > nsid)
3186                         break;
3187         }
3188         up_read(&ctrl->namespaces_rwsem);
3189         return ret;
3190 }
3191
3192 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3193 {
3194         struct streams_directive_params s;
3195         int ret;
3196
3197         if (!ctrl->nr_streams)
3198                 return 0;
3199
3200         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3201         if (ret)
3202                 return ret;
3203
3204         ns->sws = le32_to_cpu(s.sws);
3205         ns->sgs = le16_to_cpu(s.sgs);
3206
3207         if (ns->sws) {
3208                 unsigned int bs = 1 << ns->lba_shift;
3209
3210                 blk_queue_io_min(ns->queue, bs * ns->sws);
3211                 if (ns->sgs)
3212                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3213         }
3214
3215         return 0;
3216 }
3217
3218 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3219 {
3220         struct nvme_ns *ns;
3221         struct gendisk *disk;
3222         struct nvme_id_ns *id;
3223         char disk_name[DISK_NAME_LEN];
3224         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3225
3226         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3227         if (!ns)
3228                 return -ENOMEM;
3229
3230         ns->queue = blk_mq_init_queue(ctrl->tagset);
3231         if (IS_ERR(ns->queue)) {
3232                 ret = PTR_ERR(ns->queue);
3233                 goto out_free_ns;
3234         }
3235
3236         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3237         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3238                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3239
3240         ns->queue->queuedata = ns;
3241         ns->ctrl = ctrl;
3242
3243         kref_init(&ns->kref);
3244         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3245
3246         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3247         nvme_set_queue_limits(ctrl, ns->queue);
3248
3249         id = nvme_identify_ns(ctrl, nsid);
3250         if (!id) {
3251                 ret = -EIO;
3252                 goto out_free_queue;
3253         }
3254
3255         if (id->ncap == 0) {
3256                 ret = -EINVAL;
3257                 goto out_free_id;
3258         }
3259
3260         ret = nvme_init_ns_head(ns, nsid, id);
3261         if (ret)
3262                 goto out_free_id;
3263         nvme_setup_streams_ns(ctrl, ns);
3264         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3265
3266         disk = alloc_disk_node(0, node);
3267         if (!disk) {
3268                 ret = -ENOMEM;
3269                 goto out_unlink_ns;
3270         }
3271
3272         disk->fops = &nvme_fops;
3273         disk->private_data = ns;
3274         disk->queue = ns->queue;
3275         disk->flags = flags;
3276         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3277         ns->disk = disk;
3278
3279         __nvme_revalidate_disk(disk, id);
3280
3281         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3282                 ret = nvme_nvm_register(ns, disk_name, node);
3283                 if (ret) {
3284                         dev_warn(ctrl->device, "LightNVM init failure\n");
3285                         goto out_put_disk;
3286                 }
3287         }
3288
3289         down_write(&ctrl->namespaces_rwsem);
3290         list_add_tail(&ns->list, &ctrl->namespaces);
3291         up_write(&ctrl->namespaces_rwsem);
3292
3293         nvme_get_ctrl(ctrl);
3294
3295         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3296
3297         nvme_mpath_add_disk(ns, id);
3298         nvme_fault_inject_init(ns);
3299         kfree(id);
3300
3301         return 0;
3302  out_put_disk:
3303         put_disk(ns->disk);
3304  out_unlink_ns:
3305         mutex_lock(&ctrl->subsys->lock);
3306         list_del_rcu(&ns->siblings);
3307         mutex_unlock(&ctrl->subsys->lock);
3308         nvme_put_ns_head(ns->head);
3309  out_free_id:
3310         kfree(id);
3311  out_free_queue:
3312         blk_cleanup_queue(ns->queue);
3313  out_free_ns:
3314         kfree(ns);
3315         return ret;
3316 }
3317
3318 static void nvme_ns_remove(struct nvme_ns *ns)
3319 {
3320         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3321                 return;
3322
3323         nvme_fault_inject_fini(ns);
3324         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3325                 del_gendisk(ns->disk);
3326                 blk_cleanup_queue(ns->queue);
3327                 if (blk_get_integrity(ns->disk))
3328                         blk_integrity_unregister(ns->disk);
3329         }
3330
3331         mutex_lock(&ns->ctrl->subsys->lock);
3332         list_del_rcu(&ns->siblings);
3333         nvme_mpath_clear_current_path(ns);
3334         mutex_unlock(&ns->ctrl->subsys->lock);
3335
3336         down_write(&ns->ctrl->namespaces_rwsem);
3337         list_del_init(&ns->list);
3338         up_write(&ns->ctrl->namespaces_rwsem);
3339
3340         synchronize_srcu(&ns->head->srcu);
3341         nvme_mpath_check_last_path(ns);
3342         nvme_put_ns(ns);
3343 }
3344
3345 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3346 {
3347         struct nvme_ns *ns;
3348
3349         ns = nvme_find_get_ns(ctrl, nsid);
3350         if (ns) {
3351                 if (ns->disk && revalidate_disk(ns->disk))
3352                         nvme_ns_remove(ns);
3353                 nvme_put_ns(ns);
3354         } else
3355                 nvme_alloc_ns(ctrl, nsid);
3356 }
3357
3358 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3359                                         unsigned nsid)
3360 {
3361         struct nvme_ns *ns, *next;
3362         LIST_HEAD(rm_list);
3363
3364         down_write(&ctrl->namespaces_rwsem);
3365         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3366                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3367                         list_move_tail(&ns->list, &rm_list);
3368         }
3369         up_write(&ctrl->namespaces_rwsem);
3370
3371         list_for_each_entry_safe(ns, next, &rm_list, list)
3372                 nvme_ns_remove(ns);
3373
3374 }
3375
3376 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3377 {
3378         struct nvme_ns *ns;
3379         __le32 *ns_list;
3380         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3381         int ret = 0;
3382
3383         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3384         if (!ns_list)
3385                 return -ENOMEM;
3386
3387         for (i = 0; i < num_lists; i++) {
3388                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3389                 if (ret)
3390                         goto free;
3391
3392                 for (j = 0; j < min(nn, 1024U); j++) {
3393                         nsid = le32_to_cpu(ns_list[j]);
3394                         if (!nsid)
3395                                 goto out;
3396
3397                         nvme_validate_ns(ctrl, nsid);
3398
3399                         while (++prev < nsid) {
3400                                 ns = nvme_find_get_ns(ctrl, prev);
3401                                 if (ns) {
3402                                         nvme_ns_remove(ns);
3403                                         nvme_put_ns(ns);
3404                                 }
3405                         }
3406                 }
3407                 nn -= j;
3408         }
3409  out:
3410         nvme_remove_invalid_namespaces(ctrl, prev);
3411  free:
3412         kfree(ns_list);
3413         return ret;
3414 }
3415
3416 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3417 {
3418         unsigned i;
3419
3420         for (i = 1; i <= nn; i++)
3421                 nvme_validate_ns(ctrl, i);
3422
3423         nvme_remove_invalid_namespaces(ctrl, nn);
3424 }
3425
3426 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3427 {
3428         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3429         __le32 *log;
3430         int error;
3431
3432         log = kzalloc(log_size, GFP_KERNEL);
3433         if (!log)
3434                 return;
3435
3436         /*
3437          * We need to read the log to clear the AEN, but we don't want to rely
3438          * on it for the changed namespace information as userspace could have
3439          * raced with us in reading the log page, which could cause us to miss
3440          * updates.
3441          */
3442         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3443                         log_size, 0);
3444         if (error)
3445                 dev_warn(ctrl->device,
3446                         "reading changed ns log failed: %d\n", error);
3447
3448         kfree(log);
3449 }
3450
3451 static void nvme_scan_work(struct work_struct *work)
3452 {
3453         struct nvme_ctrl *ctrl =
3454                 container_of(work, struct nvme_ctrl, scan_work);
3455         struct nvme_id_ctrl *id;
3456         unsigned nn;
3457
3458         if (ctrl->state != NVME_CTRL_LIVE)
3459                 return;
3460
3461         WARN_ON_ONCE(!ctrl->tagset);
3462
3463         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3464                 dev_info(ctrl->device, "rescanning namespaces.\n");
3465                 nvme_clear_changed_ns_log(ctrl);
3466         }
3467
3468         if (nvme_identify_ctrl(ctrl, &id))
3469                 return;
3470
3471         mutex_lock(&ctrl->scan_lock);
3472         nn = le32_to_cpu(id->nn);
3473         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3474             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3475                 if (!nvme_scan_ns_list(ctrl, nn))
3476                         goto out_free_id;
3477         }
3478         nvme_scan_ns_sequential(ctrl, nn);
3479 out_free_id:
3480         mutex_unlock(&ctrl->scan_lock);
3481         kfree(id);
3482         down_write(&ctrl->namespaces_rwsem);
3483         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3484         up_write(&ctrl->namespaces_rwsem);
3485 }
3486
3487 /*
3488  * This function iterates the namespace list unlocked to allow recovery from
3489  * controller failure. It is up to the caller to ensure the namespace list is
3490  * not modified by scan work while this function is executing.
3491  */
3492 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3493 {
3494         struct nvme_ns *ns, *next;
3495         LIST_HEAD(ns_list);
3496
3497         /* prevent racing with ns scanning */
3498         flush_work(&ctrl->scan_work);
3499
3500         /*
3501          * The dead states indicates the controller was not gracefully
3502          * disconnected. In that case, we won't be able to flush any data while
3503          * removing the namespaces' disks; fail all the queues now to avoid
3504          * potentially having to clean up the failed sync later.
3505          */
3506         if (ctrl->state == NVME_CTRL_DEAD)
3507                 nvme_kill_queues(ctrl);
3508
3509         down_write(&ctrl->namespaces_rwsem);
3510         list_splice_init(&ctrl->namespaces, &ns_list);
3511         up_write(&ctrl->namespaces_rwsem);
3512
3513         list_for_each_entry_safe(ns, next, &ns_list, list)
3514                 nvme_ns_remove(ns);
3515 }
3516 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3517
3518 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3519 {
3520         char *envp[2] = { NULL, NULL };
3521         u32 aen_result = ctrl->aen_result;
3522
3523         ctrl->aen_result = 0;
3524         if (!aen_result)
3525                 return;
3526
3527         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3528         if (!envp[0])
3529                 return;
3530         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3531         kfree(envp[0]);
3532 }
3533
3534 static void nvme_async_event_work(struct work_struct *work)
3535 {
3536         struct nvme_ctrl *ctrl =
3537                 container_of(work, struct nvme_ctrl, async_event_work);
3538
3539         nvme_aen_uevent(ctrl);
3540         ctrl->ops->submit_async_event(ctrl);
3541 }
3542
3543 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3544 {
3545
3546         u32 csts;
3547
3548         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3549                 return false;
3550
3551         if (csts == ~0)
3552                 return false;
3553
3554         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3555 }
3556
3557 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3558 {
3559         struct nvme_fw_slot_info_log *log;
3560
3561         log = kmalloc(sizeof(*log), GFP_KERNEL);
3562         if (!log)
3563                 return;
3564
3565         if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3566                         sizeof(*log), 0))
3567                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3568         kfree(log);
3569 }
3570
3571 static void nvme_fw_act_work(struct work_struct *work)
3572 {
3573         struct nvme_ctrl *ctrl = container_of(work,
3574                                 struct nvme_ctrl, fw_act_work);
3575         unsigned long fw_act_timeout;
3576
3577         if (ctrl->mtfa)
3578                 fw_act_timeout = jiffies +
3579                                 msecs_to_jiffies(ctrl->mtfa * 100);
3580         else
3581                 fw_act_timeout = jiffies +
3582                                 msecs_to_jiffies(admin_timeout * 1000);
3583
3584         nvme_stop_queues(ctrl);
3585         while (nvme_ctrl_pp_status(ctrl)) {
3586                 if (time_after(jiffies, fw_act_timeout)) {
3587                         dev_warn(ctrl->device,
3588                                 "Fw activation timeout, reset controller\n");
3589                         nvme_reset_ctrl(ctrl);
3590                         break;
3591                 }
3592                 msleep(100);
3593         }
3594
3595         if (ctrl->state != NVME_CTRL_LIVE)
3596                 return;
3597
3598         nvme_start_queues(ctrl);
3599         /* read FW slot information to clear the AER */
3600         nvme_get_fw_slot_info(ctrl);
3601 }
3602
3603 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3604 {
3605         u32 aer_notice_type = (result & 0xff00) >> 8;
3606
3607         switch (aer_notice_type) {
3608         case NVME_AER_NOTICE_NS_CHANGED:
3609                 trace_nvme_async_event(ctrl, aer_notice_type);
3610                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3611                 nvme_queue_scan(ctrl);
3612                 break;
3613         case NVME_AER_NOTICE_FW_ACT_STARTING:
3614                 trace_nvme_async_event(ctrl, aer_notice_type);
3615                 queue_work(nvme_wq, &ctrl->fw_act_work);
3616                 break;
3617 #ifdef CONFIG_NVME_MULTIPATH
3618         case NVME_AER_NOTICE_ANA:
3619                 trace_nvme_async_event(ctrl, aer_notice_type);
3620                 if (!ctrl->ana_log_buf)
3621                         break;
3622                 queue_work(nvme_wq, &ctrl->ana_work);
3623                 break;
3624 #endif
3625         default:
3626                 dev_warn(ctrl->device, "async event result %08x\n", result);
3627         }
3628 }
3629
3630 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3631                 volatile union nvme_result *res)
3632 {
3633         u32 result = le32_to_cpu(res->u32);
3634         u32 aer_type = result & 0x07;
3635
3636         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3637                 return;
3638
3639         switch (aer_type) {
3640         case NVME_AER_NOTICE:
3641                 nvme_handle_aen_notice(ctrl, result);
3642                 break;
3643         case NVME_AER_ERROR:
3644         case NVME_AER_SMART:
3645         case NVME_AER_CSS:
3646         case NVME_AER_VS:
3647                 trace_nvme_async_event(ctrl, aer_type);
3648                 ctrl->aen_result = result;
3649                 break;
3650         default:
3651                 break;
3652         }
3653         queue_work(nvme_wq, &ctrl->async_event_work);
3654 }
3655 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3656
3657 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3658 {
3659         nvme_mpath_stop(ctrl);
3660         nvme_stop_keep_alive(ctrl);
3661         flush_work(&ctrl->async_event_work);
3662         cancel_work_sync(&ctrl->fw_act_work);
3663 }
3664 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3665
3666 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3667 {
3668         if (ctrl->kato)
3669                 nvme_start_keep_alive(ctrl);
3670
3671         if (ctrl->queue_count > 1) {
3672                 nvme_queue_scan(ctrl);
3673                 nvme_enable_aen(ctrl);
3674                 queue_work(nvme_wq, &ctrl->async_event_work);
3675                 nvme_start_queues(ctrl);
3676         }
3677 }
3678 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3679
3680 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3681 {
3682         cdev_device_del(&ctrl->cdev, ctrl->device);
3683 }
3684 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3685
3686 static void nvme_free_ctrl(struct device *dev)
3687 {
3688         struct nvme_ctrl *ctrl =
3689                 container_of(dev, struct nvme_ctrl, ctrl_device);
3690         struct nvme_subsystem *subsys = ctrl->subsys;
3691
3692         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3693         kfree(ctrl->effects);
3694         nvme_mpath_uninit(ctrl);
3695         __free_page(ctrl->discard_page);
3696
3697         if (subsys) {
3698                 mutex_lock(&subsys->lock);
3699                 list_del(&ctrl->subsys_entry);
3700                 mutex_unlock(&subsys->lock);
3701                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3702         }
3703
3704         ctrl->ops->free_ctrl(ctrl);
3705
3706         if (subsys)
3707                 nvme_put_subsystem(subsys);
3708 }
3709
3710 /*
3711  * Initialize a NVMe controller structures.  This needs to be called during
3712  * earliest initialization so that we have the initialized structured around
3713  * during probing.
3714  */
3715 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3716                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3717 {
3718         int ret;
3719
3720         ctrl->state = NVME_CTRL_NEW;
3721         spin_lock_init(&ctrl->lock);
3722         mutex_init(&ctrl->scan_lock);
3723         INIT_LIST_HEAD(&ctrl->namespaces);
3724         init_rwsem(&ctrl->namespaces_rwsem);
3725         ctrl->dev = dev;
3726         ctrl->ops = ops;
3727         ctrl->quirks = quirks;
3728         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3729         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3730         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3731         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3732
3733         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3734         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3735         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3736
3737         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3738                         PAGE_SIZE);
3739         ctrl->discard_page = alloc_page(GFP_KERNEL);
3740         if (!ctrl->discard_page) {
3741                 ret = -ENOMEM;
3742                 goto out;
3743         }
3744
3745         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3746         if (ret < 0)
3747                 goto out;
3748         ctrl->instance = ret;
3749
3750         device_initialize(&ctrl->ctrl_device);
3751         ctrl->device = &ctrl->ctrl_device;
3752         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3753         ctrl->device->class = nvme_class;
3754         ctrl->device->parent = ctrl->dev;
3755         ctrl->device->groups = nvme_dev_attr_groups;
3756         ctrl->device->release = nvme_free_ctrl;
3757         dev_set_drvdata(ctrl->device, ctrl);
3758         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3759         if (ret)
3760                 goto out_release_instance;
3761
3762         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3763         ctrl->cdev.owner = ops->module;
3764         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3765         if (ret)
3766                 goto out_free_name;
3767
3768         /*
3769          * Initialize latency tolerance controls.  The sysfs files won't
3770          * be visible to userspace unless the device actually supports APST.
3771          */
3772         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3773         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3774                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3775
3776         return 0;
3777 out_free_name:
3778         kfree_const(ctrl->device->kobj.name);
3779 out_release_instance:
3780         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3781 out:
3782         if (ctrl->discard_page)
3783                 __free_page(ctrl->discard_page);
3784         return ret;
3785 }
3786 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3787
3788 /**
3789  * nvme_kill_queues(): Ends all namespace queues
3790  * @ctrl: the dead controller that needs to end
3791  *
3792  * Call this function when the driver determines it is unable to get the
3793  * controller in a state capable of servicing IO.
3794  */
3795 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3796 {
3797         struct nvme_ns *ns;
3798
3799         down_read(&ctrl->namespaces_rwsem);
3800
3801         /* Forcibly unquiesce queues to avoid blocking dispatch */
3802         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3803                 blk_mq_unquiesce_queue(ctrl->admin_q);
3804
3805         list_for_each_entry(ns, &ctrl->namespaces, list)
3806                 nvme_set_queue_dying(ns);
3807
3808         up_read(&ctrl->namespaces_rwsem);
3809 }
3810 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3811
3812 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3813 {
3814         struct nvme_ns *ns;
3815
3816         down_read(&ctrl->namespaces_rwsem);
3817         list_for_each_entry(ns, &ctrl->namespaces, list)
3818                 blk_mq_unfreeze_queue(ns->queue);
3819         up_read(&ctrl->namespaces_rwsem);
3820 }
3821 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3822
3823 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3824 {
3825         struct nvme_ns *ns;
3826
3827         down_read(&ctrl->namespaces_rwsem);
3828         list_for_each_entry(ns, &ctrl->namespaces, list) {
3829                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3830                 if (timeout <= 0)
3831                         break;
3832         }
3833         up_read(&ctrl->namespaces_rwsem);
3834 }
3835 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3836
3837 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3838 {
3839         struct nvme_ns *ns;
3840
3841         down_read(&ctrl->namespaces_rwsem);
3842         list_for_each_entry(ns, &ctrl->namespaces, list)
3843                 blk_mq_freeze_queue_wait(ns->queue);
3844         up_read(&ctrl->namespaces_rwsem);
3845 }
3846 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3847
3848 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3849 {
3850         struct nvme_ns *ns;
3851
3852         down_read(&ctrl->namespaces_rwsem);
3853         list_for_each_entry(ns, &ctrl->namespaces, list)
3854                 blk_freeze_queue_start(ns->queue);
3855         up_read(&ctrl->namespaces_rwsem);
3856 }
3857 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3858
3859 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3860 {
3861         struct nvme_ns *ns;
3862
3863         down_read(&ctrl->namespaces_rwsem);
3864         list_for_each_entry(ns, &ctrl->namespaces, list)
3865                 blk_mq_quiesce_queue(ns->queue);
3866         up_read(&ctrl->namespaces_rwsem);
3867 }
3868 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3869
3870 void nvme_start_queues(struct nvme_ctrl *ctrl)
3871 {
3872         struct nvme_ns *ns;
3873
3874         down_read(&ctrl->namespaces_rwsem);
3875         list_for_each_entry(ns, &ctrl->namespaces, list)
3876                 blk_mq_unquiesce_queue(ns->queue);
3877         up_read(&ctrl->namespaces_rwsem);
3878 }
3879 EXPORT_SYMBOL_GPL(nvme_start_queues);
3880
3881 /*
3882  * Check we didn't inadvertently grow the command structure sizes:
3883  */
3884 static inline void _nvme_check_size(void)
3885 {
3886         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
3887         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
3888         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
3889         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
3890         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
3891         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
3892         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
3893         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
3894         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
3895         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
3896         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
3897         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
3898         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
3899         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
3900         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
3901         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
3902         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
3903 }
3904
3905
3906 static int __init nvme_core_init(void)
3907 {
3908         int result = -ENOMEM;
3909
3910         _nvme_check_size();
3911
3912         nvme_wq = alloc_workqueue("nvme-wq",
3913                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3914         if (!nvme_wq)
3915                 goto out;
3916
3917         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3918                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3919         if (!nvme_reset_wq)
3920                 goto destroy_wq;
3921
3922         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3923                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3924         if (!nvme_delete_wq)
3925                 goto destroy_reset_wq;
3926
3927         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3928         if (result < 0)
3929                 goto destroy_delete_wq;
3930
3931         nvme_class = class_create(THIS_MODULE, "nvme");
3932         if (IS_ERR(nvme_class)) {
3933                 result = PTR_ERR(nvme_class);
3934                 goto unregister_chrdev;
3935         }
3936
3937         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3938         if (IS_ERR(nvme_subsys_class)) {
3939                 result = PTR_ERR(nvme_subsys_class);
3940                 goto destroy_class;
3941         }
3942         return 0;
3943
3944 destroy_class:
3945         class_destroy(nvme_class);
3946 unregister_chrdev:
3947         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3948 destroy_delete_wq:
3949         destroy_workqueue(nvme_delete_wq);
3950 destroy_reset_wq:
3951         destroy_workqueue(nvme_reset_wq);
3952 destroy_wq:
3953         destroy_workqueue(nvme_wq);
3954 out:
3955         return result;
3956 }
3957
3958 static void __exit nvme_core_exit(void)
3959 {
3960         ida_destroy(&nvme_subsystems_ida);
3961         class_destroy(nvme_subsys_class);
3962         class_destroy(nvme_class);
3963         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3964         destroy_workqueue(nvme_delete_wq);
3965         destroy_workqueue(nvme_reset_wq);
3966         destroy_workqueue(nvme_wq);
3967 }
3968
3969 MODULE_LICENSE("GPL");
3970 MODULE_VERSION("1.0");
3971 module_init(nvme_core_init);
3972 module_exit(nvme_core_exit);