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(&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         if (ctrl->effects)
1261                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1262         else
1263                 effects = nvme_known_admin_effects(opcode);
1264
1265         /*
1266          * For simplicity, IO to all namespaces is quiesced even if the command
1267          * effects say only one namespace is affected.
1268          */
1269         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1270                 mutex_lock(&ctrl->scan_lock);
1271                 nvme_start_freeze(ctrl);
1272                 nvme_wait_freeze(ctrl);
1273         }
1274         return effects;
1275 }
1276
1277 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1278 {
1279         struct nvme_ns *ns;
1280
1281         down_read(&ctrl->namespaces_rwsem);
1282         list_for_each_entry(ns, &ctrl->namespaces, list)
1283                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1284                         nvme_set_queue_dying(ns);
1285         up_read(&ctrl->namespaces_rwsem);
1286
1287         nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1288 }
1289
1290 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1291 {
1292         /*
1293          * Revalidate LBA changes prior to unfreezing. This is necessary to
1294          * prevent memory corruption if a logical block size was changed by
1295          * this command.
1296          */
1297         if (effects & NVME_CMD_EFFECTS_LBCC)
1298                 nvme_update_formats(ctrl);
1299         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1300                 nvme_unfreeze(ctrl);
1301                 mutex_unlock(&ctrl->scan_lock);
1302         }
1303         if (effects & NVME_CMD_EFFECTS_CCC)
1304                 nvme_init_identify(ctrl);
1305         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1306                 nvme_queue_scan(ctrl);
1307 }
1308
1309 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1310                         struct nvme_passthru_cmd __user *ucmd)
1311 {
1312         struct nvme_passthru_cmd cmd;
1313         struct nvme_command c;
1314         unsigned timeout = 0;
1315         u32 effects;
1316         int status;
1317
1318         if (!capable(CAP_SYS_ADMIN))
1319                 return -EACCES;
1320         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1321                 return -EFAULT;
1322         if (cmd.flags)
1323                 return -EINVAL;
1324
1325         memset(&c, 0, sizeof(c));
1326         c.common.opcode = cmd.opcode;
1327         c.common.flags = cmd.flags;
1328         c.common.nsid = cpu_to_le32(cmd.nsid);
1329         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1330         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1331         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1332         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1333         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1334         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1335         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1336         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1337
1338         if (cmd.timeout_ms)
1339                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1340
1341         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1342         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1343                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1344                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1345                         0, &cmd.result, timeout);
1346         nvme_passthru_end(ctrl, effects);
1347
1348         if (status >= 0) {
1349                 if (put_user(cmd.result, &ucmd->result))
1350                         return -EFAULT;
1351         }
1352
1353         return status;
1354 }
1355
1356 /*
1357  * Issue ioctl requests on the first available path.  Note that unlike normal
1358  * block layer requests we will not retry failed request on another controller.
1359  */
1360 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1361                 struct nvme_ns_head **head, int *srcu_idx)
1362 {
1363 #ifdef CONFIG_NVME_MULTIPATH
1364         if (disk->fops == &nvme_ns_head_ops) {
1365                 *head = disk->private_data;
1366                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1367                 return nvme_find_path(*head);
1368         }
1369 #endif
1370         *head = NULL;
1371         *srcu_idx = -1;
1372         return disk->private_data;
1373 }
1374
1375 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1376 {
1377         if (head)
1378                 srcu_read_unlock(&head->srcu, idx);
1379 }
1380
1381 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1382 {
1383         switch (cmd) {
1384         case NVME_IOCTL_ID:
1385                 force_successful_syscall_return();
1386                 return ns->head->ns_id;
1387         case NVME_IOCTL_ADMIN_CMD:
1388                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1389         case NVME_IOCTL_IO_CMD:
1390                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1391         case NVME_IOCTL_SUBMIT_IO:
1392                 return nvme_submit_io(ns, (void __user *)arg);
1393         default:
1394 #ifdef CONFIG_NVM
1395                 if (ns->ndev)
1396                         return nvme_nvm_ioctl(ns, cmd, arg);
1397 #endif
1398                 if (is_sed_ioctl(cmd))
1399                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
1400                                          (void __user *) arg);
1401                 return -ENOTTY;
1402         }
1403 }
1404
1405 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1406                 unsigned int cmd, unsigned long arg)
1407 {
1408         struct nvme_ns_head *head = NULL;
1409         struct nvme_ns *ns;
1410         int srcu_idx, ret;
1411
1412         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1413         if (unlikely(!ns))
1414                 ret = -EWOULDBLOCK;
1415         else
1416                 ret = nvme_ns_ioctl(ns, cmd, arg);
1417         nvme_put_ns_from_disk(head, srcu_idx);
1418         return ret;
1419 }
1420
1421 static int nvme_open(struct block_device *bdev, fmode_t mode)
1422 {
1423         struct nvme_ns *ns = bdev->bd_disk->private_data;
1424
1425 #ifdef CONFIG_NVME_MULTIPATH
1426         /* should never be called due to GENHD_FL_HIDDEN */
1427         if (WARN_ON_ONCE(ns->head->disk))
1428                 goto fail;
1429 #endif
1430         if (!kref_get_unless_zero(&ns->kref))
1431                 goto fail;
1432         if (!try_module_get(ns->ctrl->ops->module))
1433                 goto fail_put_ns;
1434
1435         return 0;
1436
1437 fail_put_ns:
1438         nvme_put_ns(ns);
1439 fail:
1440         return -ENXIO;
1441 }
1442
1443 static void nvme_release(struct gendisk *disk, fmode_t mode)
1444 {
1445         struct nvme_ns *ns = disk->private_data;
1446
1447         module_put(ns->ctrl->ops->module);
1448         nvme_put_ns(ns);
1449 }
1450
1451 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1452 {
1453         /* some standard values */
1454         geo->heads = 1 << 6;
1455         geo->sectors = 1 << 5;
1456         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1457         return 0;
1458 }
1459
1460 #ifdef CONFIG_BLK_DEV_INTEGRITY
1461 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1462 {
1463         struct blk_integrity integrity;
1464
1465         memset(&integrity, 0, sizeof(integrity));
1466         switch (pi_type) {
1467         case NVME_NS_DPS_PI_TYPE3:
1468                 integrity.profile = &t10_pi_type3_crc;
1469                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1470                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1471                 break;
1472         case NVME_NS_DPS_PI_TYPE1:
1473         case NVME_NS_DPS_PI_TYPE2:
1474                 integrity.profile = &t10_pi_type1_crc;
1475                 integrity.tag_size = sizeof(u16);
1476                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1477                 break;
1478         default:
1479                 integrity.profile = NULL;
1480                 break;
1481         }
1482         integrity.tuple_size = ms;
1483         blk_integrity_register(disk, &integrity);
1484         blk_queue_max_integrity_segments(disk->queue, 1);
1485 }
1486 #else
1487 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1488 {
1489 }
1490 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1491
1492 static void nvme_set_chunk_size(struct nvme_ns *ns)
1493 {
1494         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1495         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1496 }
1497
1498 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1499 {
1500         struct nvme_ctrl *ctrl = ns->ctrl;
1501         struct request_queue *queue = disk->queue;
1502         u32 size = queue_logical_block_size(queue);
1503
1504         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1505                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1506                 return;
1507         }
1508
1509         if (ctrl->nr_streams && ns->sws && ns->sgs)
1510                 size *= ns->sws * ns->sgs;
1511
1512         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1513                         NVME_DSM_MAX_RANGES);
1514
1515         queue->limits.discard_alignment = 0;
1516         queue->limits.discard_granularity = size;
1517
1518         /* If discard is already enabled, don't reset queue limits */
1519         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1520                 return;
1521
1522         blk_queue_max_discard_sectors(queue, UINT_MAX);
1523         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1524
1525         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1526                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1527 }
1528
1529 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1530 {
1531         u32 max_sectors;
1532         unsigned short bs = 1 << ns->lba_shift;
1533
1534         if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1535             (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1536                 return;
1537         /*
1538          * Even though NVMe spec explicitly states that MDTS is not
1539          * applicable to the write-zeroes:- "The restriction does not apply to
1540          * commands that do not transfer data between the host and the
1541          * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1542          * In order to be more cautious use controller's max_hw_sectors value
1543          * to configure the maximum sectors for the write-zeroes which is
1544          * configured based on the controller's MDTS field in the
1545          * nvme_init_identify() if available.
1546          */
1547         if (ns->ctrl->max_hw_sectors == UINT_MAX)
1548                 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1549         else
1550                 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1551
1552         blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1553 }
1554
1555 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1556                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1557 {
1558         memset(ids, 0, sizeof(*ids));
1559
1560         if (ctrl->vs >= NVME_VS(1, 1, 0))
1561                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1562         if (ctrl->vs >= NVME_VS(1, 2, 0))
1563                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1564         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1565                  /* Don't treat error as fatal we potentially
1566                   * already have a NGUID or EUI-64
1567                   */
1568                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1569                         dev_warn(ctrl->device,
1570                                  "%s: Identify Descriptors failed\n", __func__);
1571         }
1572 }
1573
1574 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1575 {
1576         return !uuid_is_null(&ids->uuid) ||
1577                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1578                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1579 }
1580
1581 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1582 {
1583         return uuid_equal(&a->uuid, &b->uuid) &&
1584                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1585                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1586 }
1587
1588 static void nvme_update_disk_info(struct gendisk *disk,
1589                 struct nvme_ns *ns, struct nvme_id_ns *id)
1590 {
1591         sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1592         unsigned short bs = 1 << ns->lba_shift;
1593
1594         if (ns->lba_shift > PAGE_SHIFT) {
1595                 /* unsupported block size, set capacity to 0 later */
1596                 bs = (1 << 9);
1597         }
1598         blk_mq_freeze_queue(disk->queue);
1599         blk_integrity_unregister(disk);
1600
1601         blk_queue_logical_block_size(disk->queue, bs);
1602         blk_queue_physical_block_size(disk->queue, bs);
1603         blk_queue_io_min(disk->queue, bs);
1604
1605         if (ns->ms && !ns->ext &&
1606             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1607                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1608         if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1609             ns->lba_shift > PAGE_SHIFT)
1610                 capacity = 0;
1611
1612         set_capacity(disk, capacity);
1613
1614         nvme_config_discard(disk, ns);
1615         nvme_config_write_zeroes(disk, ns);
1616
1617         if (id->nsattr & (1 << 0))
1618                 set_disk_ro(disk, true);
1619         else
1620                 set_disk_ro(disk, false);
1621
1622         blk_mq_unfreeze_queue(disk->queue);
1623 }
1624
1625 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1626 {
1627         struct nvme_ns *ns = disk->private_data;
1628
1629         /*
1630          * If identify namespace failed, use default 512 byte block size so
1631          * block layer can use before failing read/write for 0 capacity.
1632          */
1633         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1634         if (ns->lba_shift == 0)
1635                 ns->lba_shift = 9;
1636         ns->noiob = le16_to_cpu(id->noiob);
1637         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1638         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1639         /* the PI implementation requires metadata equal t10 pi tuple size */
1640         if (ns->ms == sizeof(struct t10_pi_tuple))
1641                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1642         else
1643                 ns->pi_type = 0;
1644
1645         if (ns->noiob)
1646                 nvme_set_chunk_size(ns);
1647         nvme_update_disk_info(disk, ns, id);
1648 #ifdef CONFIG_NVME_MULTIPATH
1649         if (ns->head->disk) {
1650                 nvme_update_disk_info(ns->head->disk, ns, id);
1651                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1652         }
1653 #endif
1654 }
1655
1656 static int nvme_revalidate_disk(struct gendisk *disk)
1657 {
1658         struct nvme_ns *ns = disk->private_data;
1659         struct nvme_ctrl *ctrl = ns->ctrl;
1660         struct nvme_id_ns *id;
1661         struct nvme_ns_ids ids;
1662         int ret = 0;
1663
1664         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1665                 set_capacity(disk, 0);
1666                 return -ENODEV;
1667         }
1668
1669         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1670         if (!id)
1671                 return -ENODEV;
1672
1673         if (id->ncap == 0) {
1674                 ret = -ENODEV;
1675                 goto out;
1676         }
1677
1678         __nvme_revalidate_disk(disk, id);
1679         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1680         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1681                 dev_err(ctrl->device,
1682                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1683                 ret = -ENODEV;
1684         }
1685
1686 out:
1687         kfree(id);
1688         return ret;
1689 }
1690
1691 static char nvme_pr_type(enum pr_type type)
1692 {
1693         switch (type) {
1694         case PR_WRITE_EXCLUSIVE:
1695                 return 1;
1696         case PR_EXCLUSIVE_ACCESS:
1697                 return 2;
1698         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1699                 return 3;
1700         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1701                 return 4;
1702         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1703                 return 5;
1704         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1705                 return 6;
1706         default:
1707                 return 0;
1708         }
1709 };
1710
1711 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1712                                 u64 key, u64 sa_key, u8 op)
1713 {
1714         struct nvme_ns_head *head = NULL;
1715         struct nvme_ns *ns;
1716         struct nvme_command c;
1717         int srcu_idx, ret;
1718         u8 data[16] = { 0, };
1719
1720         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1721         if (unlikely(!ns))
1722                 return -EWOULDBLOCK;
1723
1724         put_unaligned_le64(key, &data[0]);
1725         put_unaligned_le64(sa_key, &data[8]);
1726
1727         memset(&c, 0, sizeof(c));
1728         c.common.opcode = op;
1729         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1730         c.common.cdw10 = cpu_to_le32(cdw10);
1731
1732         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1733         nvme_put_ns_from_disk(head, srcu_idx);
1734         return ret;
1735 }
1736
1737 static int nvme_pr_register(struct block_device *bdev, u64 old,
1738                 u64 new, unsigned flags)
1739 {
1740         u32 cdw10;
1741
1742         if (flags & ~PR_FL_IGNORE_KEY)
1743                 return -EOPNOTSUPP;
1744
1745         cdw10 = old ? 2 : 0;
1746         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1747         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1748         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1749 }
1750
1751 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1752                 enum pr_type type, unsigned flags)
1753 {
1754         u32 cdw10;
1755
1756         if (flags & ~PR_FL_IGNORE_KEY)
1757                 return -EOPNOTSUPP;
1758
1759         cdw10 = nvme_pr_type(type) << 8;
1760         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1761         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1762 }
1763
1764 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1765                 enum pr_type type, bool abort)
1766 {
1767         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1768         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1769 }
1770
1771 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1772 {
1773         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1774         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1775 }
1776
1777 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1778 {
1779         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1780         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1781 }
1782
1783 static const struct pr_ops nvme_pr_ops = {
1784         .pr_register    = nvme_pr_register,
1785         .pr_reserve     = nvme_pr_reserve,
1786         .pr_release     = nvme_pr_release,
1787         .pr_preempt     = nvme_pr_preempt,
1788         .pr_clear       = nvme_pr_clear,
1789 };
1790
1791 #ifdef CONFIG_BLK_SED_OPAL
1792 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1793                 bool send)
1794 {
1795         struct nvme_ctrl *ctrl = data;
1796         struct nvme_command cmd;
1797
1798         memset(&cmd, 0, sizeof(cmd));
1799         if (send)
1800                 cmd.common.opcode = nvme_admin_security_send;
1801         else
1802                 cmd.common.opcode = nvme_admin_security_recv;
1803         cmd.common.nsid = 0;
1804         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1805         cmd.common.cdw11 = cpu_to_le32(len);
1806
1807         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1808                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1809 }
1810 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1811 #endif /* CONFIG_BLK_SED_OPAL */
1812
1813 static const struct block_device_operations nvme_fops = {
1814         .owner          = THIS_MODULE,
1815         .ioctl          = nvme_ioctl,
1816         .compat_ioctl   = nvme_ioctl,
1817         .open           = nvme_open,
1818         .release        = nvme_release,
1819         .getgeo         = nvme_getgeo,
1820         .revalidate_disk= nvme_revalidate_disk,
1821         .pr_ops         = &nvme_pr_ops,
1822 };
1823
1824 #ifdef CONFIG_NVME_MULTIPATH
1825 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1826 {
1827         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1828
1829         if (!kref_get_unless_zero(&head->ref))
1830                 return -ENXIO;
1831         return 0;
1832 }
1833
1834 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1835 {
1836         nvme_put_ns_head(disk->private_data);
1837 }
1838
1839 const struct block_device_operations nvme_ns_head_ops = {
1840         .owner          = THIS_MODULE,
1841         .open           = nvme_ns_head_open,
1842         .release        = nvme_ns_head_release,
1843         .ioctl          = nvme_ioctl,
1844         .compat_ioctl   = nvme_ioctl,
1845         .getgeo         = nvme_getgeo,
1846         .pr_ops         = &nvme_pr_ops,
1847 };
1848 #endif /* CONFIG_NVME_MULTIPATH */
1849
1850 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1851 {
1852         unsigned long timeout =
1853                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1854         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1855         int ret;
1856
1857         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1858                 if (csts == ~0)
1859                         return -ENODEV;
1860                 if ((csts & NVME_CSTS_RDY) == bit)
1861                         break;
1862
1863                 msleep(100);
1864                 if (fatal_signal_pending(current))
1865                         return -EINTR;
1866                 if (time_after(jiffies, timeout)) {
1867                         dev_err(ctrl->device,
1868                                 "Device not ready; aborting %s\n", enabled ?
1869                                                 "initialisation" : "reset");
1870                         return -ENODEV;
1871                 }
1872         }
1873
1874         return ret;
1875 }
1876
1877 /*
1878  * If the device has been passed off to us in an enabled state, just clear
1879  * the enabled bit.  The spec says we should set the 'shutdown notification
1880  * bits', but doing so may cause the device to complete commands to the
1881  * admin queue ... and we don't know what memory that might be pointing at!
1882  */
1883 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1884 {
1885         int ret;
1886
1887         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1888         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1889
1890         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1891         if (ret)
1892                 return ret;
1893
1894         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1895                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1896
1897         return nvme_wait_ready(ctrl, cap, false);
1898 }
1899 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1900
1901 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1902 {
1903         /*
1904          * Default to a 4K page size, with the intention to update this
1905          * path in the future to accomodate architectures with differing
1906          * kernel and IO page sizes.
1907          */
1908         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1909         int ret;
1910
1911         if (page_shift < dev_page_min) {
1912                 dev_err(ctrl->device,
1913                         "Minimum device page size %u too large for host (%u)\n",
1914                         1 << dev_page_min, 1 << page_shift);
1915                 return -ENODEV;
1916         }
1917
1918         ctrl->page_size = 1 << page_shift;
1919
1920         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1921         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1922         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1923         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1924         ctrl->ctrl_config |= NVME_CC_ENABLE;
1925
1926         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1927         if (ret)
1928                 return ret;
1929         return nvme_wait_ready(ctrl, cap, true);
1930 }
1931 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1932
1933 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1934 {
1935         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1936         u32 csts;
1937         int ret;
1938
1939         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1940         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1941
1942         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1943         if (ret)
1944                 return ret;
1945
1946         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1947                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1948                         break;
1949
1950                 msleep(100);
1951                 if (fatal_signal_pending(current))
1952                         return -EINTR;
1953                 if (time_after(jiffies, timeout)) {
1954                         dev_err(ctrl->device,
1955                                 "Device shutdown incomplete; abort shutdown\n");
1956                         return -ENODEV;
1957                 }
1958         }
1959
1960         return ret;
1961 }
1962 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1963
1964 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1965                 struct request_queue *q)
1966 {
1967         bool vwc = false;
1968
1969         if (ctrl->max_hw_sectors) {
1970                 u32 max_segments =
1971                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1972
1973                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1974                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1975                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1976         }
1977         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1978             is_power_of_2(ctrl->max_hw_sectors))
1979                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1980         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1981         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1982                 vwc = true;
1983         blk_queue_write_cache(q, vwc, vwc);
1984 }
1985
1986 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1987 {
1988         __le64 ts;
1989         int ret;
1990
1991         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1992                 return 0;
1993
1994         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1995         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1996                         NULL);
1997         if (ret)
1998                 dev_warn_once(ctrl->device,
1999                         "could not set timestamp (%d)\n", ret);
2000         return ret;
2001 }
2002
2003 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2004 {
2005         struct nvme_feat_host_behavior *host;
2006         int ret;
2007
2008         /* Don't bother enabling the feature if retry delay is not reported */
2009         if (!ctrl->crdt[0])
2010                 return 0;
2011
2012         host = kzalloc(sizeof(*host), GFP_KERNEL);
2013         if (!host)
2014                 return 0;
2015
2016         host->acre = NVME_ENABLE_ACRE;
2017         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2018                                 host, sizeof(*host), NULL);
2019         kfree(host);
2020         return ret;
2021 }
2022
2023 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2024 {
2025         /*
2026          * APST (Autonomous Power State Transition) lets us program a
2027          * table of power state transitions that the controller will
2028          * perform automatically.  We configure it with a simple
2029          * heuristic: we are willing to spend at most 2% of the time
2030          * transitioning between power states.  Therefore, when running
2031          * in any given state, we will enter the next lower-power
2032          * non-operational state after waiting 50 * (enlat + exlat)
2033          * microseconds, as long as that state's exit latency is under
2034          * the requested maximum latency.
2035          *
2036          * We will not autonomously enter any non-operational state for
2037          * which the total latency exceeds ps_max_latency_us.  Users
2038          * can set ps_max_latency_us to zero to turn off APST.
2039          */
2040
2041         unsigned apste;
2042         struct nvme_feat_auto_pst *table;
2043         u64 max_lat_us = 0;
2044         int max_ps = -1;
2045         int ret;
2046
2047         /*
2048          * If APST isn't supported or if we haven't been initialized yet,
2049          * then don't do anything.
2050          */
2051         if (!ctrl->apsta)
2052                 return 0;
2053
2054         if (ctrl->npss > 31) {
2055                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2056                 return 0;
2057         }
2058
2059         table = kzalloc(sizeof(*table), GFP_KERNEL);
2060         if (!table)
2061                 return 0;
2062
2063         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2064                 /* Turn off APST. */
2065                 apste = 0;
2066                 dev_dbg(ctrl->device, "APST disabled\n");
2067         } else {
2068                 __le64 target = cpu_to_le64(0);
2069                 int state;
2070
2071                 /*
2072                  * Walk through all states from lowest- to highest-power.
2073                  * According to the spec, lower-numbered states use more
2074                  * power.  NPSS, despite the name, is the index of the
2075                  * lowest-power state, not the number of states.
2076                  */
2077                 for (state = (int)ctrl->npss; state >= 0; state--) {
2078                         u64 total_latency_us, exit_latency_us, transition_ms;
2079
2080                         if (target)
2081                                 table->entries[state] = target;
2082
2083                         /*
2084                          * Don't allow transitions to the deepest state
2085                          * if it's quirked off.
2086                          */
2087                         if (state == ctrl->npss &&
2088                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2089                                 continue;
2090
2091                         /*
2092                          * Is this state a useful non-operational state for
2093                          * higher-power states to autonomously transition to?
2094                          */
2095                         if (!(ctrl->psd[state].flags &
2096                               NVME_PS_FLAGS_NON_OP_STATE))
2097                                 continue;
2098
2099                         exit_latency_us =
2100                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2101                         if (exit_latency_us > ctrl->ps_max_latency_us)
2102                                 continue;
2103
2104                         total_latency_us =
2105                                 exit_latency_us +
2106                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2107
2108                         /*
2109                          * This state is good.  Use it as the APST idle
2110                          * target for higher power states.
2111                          */
2112                         transition_ms = total_latency_us + 19;
2113                         do_div(transition_ms, 20);
2114                         if (transition_ms > (1 << 24) - 1)
2115                                 transition_ms = (1 << 24) - 1;
2116
2117                         target = cpu_to_le64((state << 3) |
2118                                              (transition_ms << 8));
2119
2120                         if (max_ps == -1)
2121                                 max_ps = state;
2122
2123                         if (total_latency_us > max_lat_us)
2124                                 max_lat_us = total_latency_us;
2125                 }
2126
2127                 apste = 1;
2128
2129                 if (max_ps == -1) {
2130                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2131                 } else {
2132                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2133                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2134                 }
2135         }
2136
2137         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2138                                 table, sizeof(*table), NULL);
2139         if (ret)
2140                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2141
2142         kfree(table);
2143         return ret;
2144 }
2145
2146 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2147 {
2148         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2149         u64 latency;
2150
2151         switch (val) {
2152         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2153         case PM_QOS_LATENCY_ANY:
2154                 latency = U64_MAX;
2155                 break;
2156
2157         default:
2158                 latency = val;
2159         }
2160
2161         if (ctrl->ps_max_latency_us != latency) {
2162                 ctrl->ps_max_latency_us = latency;
2163                 nvme_configure_apst(ctrl);
2164         }
2165 }
2166
2167 struct nvme_core_quirk_entry {
2168         /*
2169          * NVMe model and firmware strings are padded with spaces.  For
2170          * simplicity, strings in the quirk table are padded with NULLs
2171          * instead.
2172          */
2173         u16 vid;
2174         const char *mn;
2175         const char *fr;
2176         unsigned long quirks;
2177 };
2178
2179 static const struct nvme_core_quirk_entry core_quirks[] = {
2180         {
2181                 /*
2182                  * This Toshiba device seems to die using any APST states.  See:
2183                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2184                  */
2185                 .vid = 0x1179,
2186                 .mn = "THNSF5256GPUK TOSHIBA",
2187                 .quirks = NVME_QUIRK_NO_APST,
2188         }
2189 };
2190
2191 /* match is null-terminated but idstr is space-padded. */
2192 static bool string_matches(const char *idstr, const char *match, size_t len)
2193 {
2194         size_t matchlen;
2195
2196         if (!match)
2197                 return true;
2198
2199         matchlen = strlen(match);
2200         WARN_ON_ONCE(matchlen > len);
2201
2202         if (memcmp(idstr, match, matchlen))
2203                 return false;
2204
2205         for (; matchlen < len; matchlen++)
2206                 if (idstr[matchlen] != ' ')
2207                         return false;
2208
2209         return true;
2210 }
2211
2212 static bool quirk_matches(const struct nvme_id_ctrl *id,
2213                           const struct nvme_core_quirk_entry *q)
2214 {
2215         return q->vid == le16_to_cpu(id->vid) &&
2216                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2217                 string_matches(id->fr, q->fr, sizeof(id->fr));
2218 }
2219
2220 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2221                 struct nvme_id_ctrl *id)
2222 {
2223         size_t nqnlen;
2224         int off;
2225
2226         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2227                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2228                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2229                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2230                         return;
2231                 }
2232
2233                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2234                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2235         }
2236
2237         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2238         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2239                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2240                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2241         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2242         off += sizeof(id->sn);
2243         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2244         off += sizeof(id->mn);
2245         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2246 }
2247
2248 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2249 {
2250         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2251         kfree(subsys);
2252 }
2253
2254 static void nvme_release_subsystem(struct device *dev)
2255 {
2256         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2257 }
2258
2259 static void nvme_destroy_subsystem(struct kref *ref)
2260 {
2261         struct nvme_subsystem *subsys =
2262                         container_of(ref, struct nvme_subsystem, ref);
2263
2264         mutex_lock(&nvme_subsystems_lock);
2265         list_del(&subsys->entry);
2266         mutex_unlock(&nvme_subsystems_lock);
2267
2268         ida_destroy(&subsys->ns_ida);
2269         device_del(&subsys->dev);
2270         put_device(&subsys->dev);
2271 }
2272
2273 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2274 {
2275         kref_put(&subsys->ref, nvme_destroy_subsystem);
2276 }
2277
2278 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2279 {
2280         struct nvme_subsystem *subsys;
2281
2282         lockdep_assert_held(&nvme_subsystems_lock);
2283
2284         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2285                 if (strcmp(subsys->subnqn, subsysnqn))
2286                         continue;
2287                 if (!kref_get_unless_zero(&subsys->ref))
2288                         continue;
2289                 return subsys;
2290         }
2291
2292         return NULL;
2293 }
2294
2295 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2296         struct device_attribute subsys_attr_##_name = \
2297                 __ATTR(_name, _mode, _show, NULL)
2298
2299 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2300                                     struct device_attribute *attr,
2301                                     char *buf)
2302 {
2303         struct nvme_subsystem *subsys =
2304                 container_of(dev, struct nvme_subsystem, dev);
2305
2306         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2307 }
2308 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2309
2310 #define nvme_subsys_show_str_function(field)                            \
2311 static ssize_t subsys_##field##_show(struct device *dev,                \
2312                             struct device_attribute *attr, char *buf)   \
2313 {                                                                       \
2314         struct nvme_subsystem *subsys =                                 \
2315                 container_of(dev, struct nvme_subsystem, dev);          \
2316         return sprintf(buf, "%.*s\n",                                   \
2317                        (int)sizeof(subsys->field), subsys->field);      \
2318 }                                                                       \
2319 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2320
2321 nvme_subsys_show_str_function(model);
2322 nvme_subsys_show_str_function(serial);
2323 nvme_subsys_show_str_function(firmware_rev);
2324
2325 static struct attribute *nvme_subsys_attrs[] = {
2326         &subsys_attr_model.attr,
2327         &subsys_attr_serial.attr,
2328         &subsys_attr_firmware_rev.attr,
2329         &subsys_attr_subsysnqn.attr,
2330 #ifdef CONFIG_NVME_MULTIPATH
2331         &subsys_attr_iopolicy.attr,
2332 #endif
2333         NULL,
2334 };
2335
2336 static struct attribute_group nvme_subsys_attrs_group = {
2337         .attrs = nvme_subsys_attrs,
2338 };
2339
2340 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2341         &nvme_subsys_attrs_group,
2342         NULL,
2343 };
2344
2345 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2346 {
2347         int count = 0;
2348         struct nvme_ctrl *ctrl;
2349
2350         mutex_lock(&subsys->lock);
2351         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2352                 if (ctrl->state != NVME_CTRL_DELETING &&
2353                     ctrl->state != NVME_CTRL_DEAD)
2354                         count++;
2355         }
2356         mutex_unlock(&subsys->lock);
2357
2358         return count;
2359 }
2360
2361 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2362 {
2363         struct nvme_subsystem *subsys, *found;
2364         int ret;
2365
2366         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2367         if (!subsys)
2368                 return -ENOMEM;
2369         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2370         if (ret < 0) {
2371                 kfree(subsys);
2372                 return ret;
2373         }
2374         subsys->instance = ret;
2375         mutex_init(&subsys->lock);
2376         kref_init(&subsys->ref);
2377         INIT_LIST_HEAD(&subsys->ctrls);
2378         INIT_LIST_HEAD(&subsys->nsheads);
2379         nvme_init_subnqn(subsys, ctrl, id);
2380         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2381         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2382         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2383         subsys->vendor_id = le16_to_cpu(id->vid);
2384         subsys->cmic = id->cmic;
2385 #ifdef CONFIG_NVME_MULTIPATH
2386         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2387 #endif
2388
2389         subsys->dev.class = nvme_subsys_class;
2390         subsys->dev.release = nvme_release_subsystem;
2391         subsys->dev.groups = nvme_subsys_attrs_groups;
2392         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2393         device_initialize(&subsys->dev);
2394
2395         mutex_lock(&nvme_subsystems_lock);
2396         found = __nvme_find_get_subsystem(subsys->subnqn);
2397         if (found) {
2398                 /*
2399                  * Verify that the subsystem actually supports multiple
2400                  * controllers, else bail out.
2401                  */
2402                 if (!(ctrl->opts && ctrl->opts->discovery_nqn) &&
2403                     nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2404                         dev_err(ctrl->device,
2405                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2406                                 found->subnqn);
2407                         nvme_put_subsystem(found);
2408                         ret = -EINVAL;
2409                         goto out_unlock;
2410                 }
2411
2412                 __nvme_release_subsystem(subsys);
2413                 subsys = found;
2414         } else {
2415                 ret = device_add(&subsys->dev);
2416                 if (ret) {
2417                         dev_err(ctrl->device,
2418                                 "failed to register subsystem device.\n");
2419                         goto out_unlock;
2420                 }
2421                 ida_init(&subsys->ns_ida);
2422                 list_add_tail(&subsys->entry, &nvme_subsystems);
2423         }
2424
2425         ctrl->subsys = subsys;
2426         mutex_unlock(&nvme_subsystems_lock);
2427
2428         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2429                         dev_name(ctrl->device))) {
2430                 dev_err(ctrl->device,
2431                         "failed to create sysfs link from subsystem.\n");
2432                 /* the transport driver will eventually put the subsystem */
2433                 return -EINVAL;
2434         }
2435
2436         mutex_lock(&subsys->lock);
2437         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2438         mutex_unlock(&subsys->lock);
2439
2440         return 0;
2441
2442 out_unlock:
2443         mutex_unlock(&nvme_subsystems_lock);
2444         put_device(&subsys->dev);
2445         return ret;
2446 }
2447
2448 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2449                 void *log, size_t size, u64 offset)
2450 {
2451         struct nvme_command c = { };
2452         unsigned long dwlen = size / 4 - 1;
2453
2454         c.get_log_page.opcode = nvme_admin_get_log_page;
2455         c.get_log_page.nsid = cpu_to_le32(nsid);
2456         c.get_log_page.lid = log_page;
2457         c.get_log_page.lsp = lsp;
2458         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2459         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2460         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2461         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2462
2463         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2464 }
2465
2466 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2467 {
2468         int ret;
2469
2470         if (!ctrl->effects)
2471                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2472
2473         if (!ctrl->effects)
2474                 return 0;
2475
2476         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2477                         ctrl->effects, sizeof(*ctrl->effects), 0);
2478         if (ret) {
2479                 kfree(ctrl->effects);
2480                 ctrl->effects = NULL;
2481         }
2482         return ret;
2483 }
2484
2485 /*
2486  * Initialize the cached copies of the Identify data and various controller
2487  * register in our nvme_ctrl structure.  This should be called as soon as
2488  * the admin queue is fully up and running.
2489  */
2490 int nvme_init_identify(struct nvme_ctrl *ctrl)
2491 {
2492         struct nvme_id_ctrl *id;
2493         u64 cap;
2494         int ret, page_shift;
2495         u32 max_hw_sectors;
2496         bool prev_apst_enabled;
2497
2498         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2499         if (ret) {
2500                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2501                 return ret;
2502         }
2503
2504         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2505         if (ret) {
2506                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2507                 return ret;
2508         }
2509         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2510
2511         if (ctrl->vs >= NVME_VS(1, 1, 0))
2512                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2513
2514         ret = nvme_identify_ctrl(ctrl, &id);
2515         if (ret) {
2516                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2517                 return -EIO;
2518         }
2519
2520         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2521                 ret = nvme_get_effects_log(ctrl);
2522                 if (ret < 0)
2523                         goto out_free;
2524         }
2525
2526         if (!ctrl->identified) {
2527                 int i;
2528
2529                 ret = nvme_init_subsystem(ctrl, id);
2530                 if (ret)
2531                         goto out_free;
2532
2533                 /*
2534                  * Check for quirks.  Quirk can depend on firmware version,
2535                  * so, in principle, the set of quirks present can change
2536                  * across a reset.  As a possible future enhancement, we
2537                  * could re-scan for quirks every time we reinitialize
2538                  * the device, but we'd have to make sure that the driver
2539                  * behaves intelligently if the quirks change.
2540                  */
2541                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2542                         if (quirk_matches(id, &core_quirks[i]))
2543                                 ctrl->quirks |= core_quirks[i].quirks;
2544                 }
2545         }
2546
2547         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2548                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2549                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2550         }
2551
2552         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2553         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2554         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2555
2556         ctrl->oacs = le16_to_cpu(id->oacs);
2557         ctrl->oncs = le16_to_cpu(id->oncs);
2558         ctrl->oaes = le32_to_cpu(id->oaes);
2559         atomic_set(&ctrl->abort_limit, id->acl + 1);
2560         ctrl->vwc = id->vwc;
2561         if (id->mdts)
2562                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2563         else
2564                 max_hw_sectors = UINT_MAX;
2565         ctrl->max_hw_sectors =
2566                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2567
2568         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2569         ctrl->sgls = le32_to_cpu(id->sgls);
2570         ctrl->kas = le16_to_cpu(id->kas);
2571         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2572         ctrl->ctratt = le32_to_cpu(id->ctratt);
2573
2574         if (id->rtd3e) {
2575                 /* us -> s */
2576                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2577
2578                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2579                                                  shutdown_timeout, 60);
2580
2581                 if (ctrl->shutdown_timeout != shutdown_timeout)
2582                         dev_info(ctrl->device,
2583                                  "Shutdown timeout set to %u seconds\n",
2584                                  ctrl->shutdown_timeout);
2585         } else
2586                 ctrl->shutdown_timeout = shutdown_timeout;
2587
2588         ctrl->npss = id->npss;
2589         ctrl->apsta = id->apsta;
2590         prev_apst_enabled = ctrl->apst_enabled;
2591         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2592                 if (force_apst && id->apsta) {
2593                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2594                         ctrl->apst_enabled = true;
2595                 } else {
2596                         ctrl->apst_enabled = false;
2597                 }
2598         } else {
2599                 ctrl->apst_enabled = id->apsta;
2600         }
2601         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2602
2603         if (ctrl->ops->flags & NVME_F_FABRICS) {
2604                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2605                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2606                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2607                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2608
2609                 /*
2610                  * In fabrics we need to verify the cntlid matches the
2611                  * admin connect
2612                  */
2613                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2614                         ret = -EINVAL;
2615                         goto out_free;
2616                 }
2617
2618                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2619                         dev_err(ctrl->device,
2620                                 "keep-alive support is mandatory for fabrics\n");
2621                         ret = -EINVAL;
2622                         goto out_free;
2623                 }
2624         } else {
2625                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2626                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2627                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2628                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2629                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2630         }
2631
2632         ret = nvme_mpath_init(ctrl, id);
2633         kfree(id);
2634
2635         if (ret < 0)
2636                 return ret;
2637
2638         if (ctrl->apst_enabled && !prev_apst_enabled)
2639                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2640         else if (!ctrl->apst_enabled && prev_apst_enabled)
2641                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2642
2643         ret = nvme_configure_apst(ctrl);
2644         if (ret < 0)
2645                 return ret;
2646
2647         ret = nvme_configure_timestamp(ctrl);
2648         if (ret < 0)
2649                 return ret;
2650
2651         ret = nvme_configure_directives(ctrl);
2652         if (ret < 0)
2653                 return ret;
2654
2655         ret = nvme_configure_acre(ctrl);
2656         if (ret < 0)
2657                 return ret;
2658
2659         ctrl->identified = true;
2660
2661         return 0;
2662
2663 out_free:
2664         kfree(id);
2665         return ret;
2666 }
2667 EXPORT_SYMBOL_GPL(nvme_init_identify);
2668
2669 static int nvme_dev_open(struct inode *inode, struct file *file)
2670 {
2671         struct nvme_ctrl *ctrl =
2672                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2673
2674         switch (ctrl->state) {
2675         case NVME_CTRL_LIVE:
2676         case NVME_CTRL_ADMIN_ONLY:
2677                 break;
2678         default:
2679                 return -EWOULDBLOCK;
2680         }
2681
2682         file->private_data = ctrl;
2683         return 0;
2684 }
2685
2686 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2687 {
2688         struct nvme_ns *ns;
2689         int ret;
2690
2691         down_read(&ctrl->namespaces_rwsem);
2692         if (list_empty(&ctrl->namespaces)) {
2693                 ret = -ENOTTY;
2694                 goto out_unlock;
2695         }
2696
2697         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2698         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2699                 dev_warn(ctrl->device,
2700                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2701                 ret = -EINVAL;
2702                 goto out_unlock;
2703         }
2704
2705         dev_warn(ctrl->device,
2706                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2707         kref_get(&ns->kref);
2708         up_read(&ctrl->namespaces_rwsem);
2709
2710         ret = nvme_user_cmd(ctrl, ns, argp);
2711         nvme_put_ns(ns);
2712         return ret;
2713
2714 out_unlock:
2715         up_read(&ctrl->namespaces_rwsem);
2716         return ret;
2717 }
2718
2719 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2720                 unsigned long arg)
2721 {
2722         struct nvme_ctrl *ctrl = file->private_data;
2723         void __user *argp = (void __user *)arg;
2724
2725         switch (cmd) {
2726         case NVME_IOCTL_ADMIN_CMD:
2727                 return nvme_user_cmd(ctrl, NULL, argp);
2728         case NVME_IOCTL_IO_CMD:
2729                 return nvme_dev_user_cmd(ctrl, argp);
2730         case NVME_IOCTL_RESET:
2731                 dev_warn(ctrl->device, "resetting controller\n");
2732                 return nvme_reset_ctrl_sync(ctrl);
2733         case NVME_IOCTL_SUBSYS_RESET:
2734                 return nvme_reset_subsystem(ctrl);
2735         case NVME_IOCTL_RESCAN:
2736                 nvme_queue_scan(ctrl);
2737                 return 0;
2738         default:
2739                 return -ENOTTY;
2740         }
2741 }
2742
2743 static const struct file_operations nvme_dev_fops = {
2744         .owner          = THIS_MODULE,
2745         .open           = nvme_dev_open,
2746         .unlocked_ioctl = nvme_dev_ioctl,
2747         .compat_ioctl   = nvme_dev_ioctl,
2748 };
2749
2750 static ssize_t nvme_sysfs_reset(struct device *dev,
2751                                 struct device_attribute *attr, const char *buf,
2752                                 size_t count)
2753 {
2754         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2755         int ret;
2756
2757         ret = nvme_reset_ctrl_sync(ctrl);
2758         if (ret < 0)
2759                 return ret;
2760         return count;
2761 }
2762 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2763
2764 static ssize_t nvme_sysfs_rescan(struct device *dev,
2765                                 struct device_attribute *attr, const char *buf,
2766                                 size_t count)
2767 {
2768         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2769
2770         nvme_queue_scan(ctrl);
2771         return count;
2772 }
2773 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2774
2775 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2776 {
2777         struct gendisk *disk = dev_to_disk(dev);
2778
2779         if (disk->fops == &nvme_fops)
2780                 return nvme_get_ns_from_dev(dev)->head;
2781         else
2782                 return disk->private_data;
2783 }
2784
2785 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2786                 char *buf)
2787 {
2788         struct nvme_ns_head *head = dev_to_ns_head(dev);
2789         struct nvme_ns_ids *ids = &head->ids;
2790         struct nvme_subsystem *subsys = head->subsys;
2791         int serial_len = sizeof(subsys->serial);
2792         int model_len = sizeof(subsys->model);
2793
2794         if (!uuid_is_null(&ids->uuid))
2795                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2796
2797         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2798                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2799
2800         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2801                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2802
2803         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2804                                   subsys->serial[serial_len - 1] == '\0'))
2805                 serial_len--;
2806         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2807                                  subsys->model[model_len - 1] == '\0'))
2808                 model_len--;
2809
2810         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2811                 serial_len, subsys->serial, model_len, subsys->model,
2812                 head->ns_id);
2813 }
2814 static DEVICE_ATTR_RO(wwid);
2815
2816 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2817                 char *buf)
2818 {
2819         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2820 }
2821 static DEVICE_ATTR_RO(nguid);
2822
2823 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2824                 char *buf)
2825 {
2826         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2827
2828         /* For backward compatibility expose the NGUID to userspace if
2829          * we have no UUID set
2830          */
2831         if (uuid_is_null(&ids->uuid)) {
2832                 printk_ratelimited(KERN_WARNING
2833                                    "No UUID available providing old NGUID\n");
2834                 return sprintf(buf, "%pU\n", ids->nguid);
2835         }
2836         return sprintf(buf, "%pU\n", &ids->uuid);
2837 }
2838 static DEVICE_ATTR_RO(uuid);
2839
2840 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2841                 char *buf)
2842 {
2843         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2844 }
2845 static DEVICE_ATTR_RO(eui);
2846
2847 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2848                 char *buf)
2849 {
2850         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2851 }
2852 static DEVICE_ATTR_RO(nsid);
2853
2854 static struct attribute *nvme_ns_id_attrs[] = {
2855         &dev_attr_wwid.attr,
2856         &dev_attr_uuid.attr,
2857         &dev_attr_nguid.attr,
2858         &dev_attr_eui.attr,
2859         &dev_attr_nsid.attr,
2860 #ifdef CONFIG_NVME_MULTIPATH
2861         &dev_attr_ana_grpid.attr,
2862         &dev_attr_ana_state.attr,
2863 #endif
2864         NULL,
2865 };
2866
2867 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2868                 struct attribute *a, int n)
2869 {
2870         struct device *dev = container_of(kobj, struct device, kobj);
2871         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2872
2873         if (a == &dev_attr_uuid.attr) {
2874                 if (uuid_is_null(&ids->uuid) &&
2875                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2876                         return 0;
2877         }
2878         if (a == &dev_attr_nguid.attr) {
2879                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2880                         return 0;
2881         }
2882         if (a == &dev_attr_eui.attr) {
2883                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2884                         return 0;
2885         }
2886 #ifdef CONFIG_NVME_MULTIPATH
2887         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2888                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2889                         return 0;
2890                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2891                         return 0;
2892         }
2893 #endif
2894         return a->mode;
2895 }
2896
2897 static const struct attribute_group nvme_ns_id_attr_group = {
2898         .attrs          = nvme_ns_id_attrs,
2899         .is_visible     = nvme_ns_id_attrs_are_visible,
2900 };
2901
2902 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2903         &nvme_ns_id_attr_group,
2904 #ifdef CONFIG_NVM
2905         &nvme_nvm_attr_group,
2906 #endif
2907         NULL,
2908 };
2909
2910 #define nvme_show_str_function(field)                                           \
2911 static ssize_t  field##_show(struct device *dev,                                \
2912                             struct device_attribute *attr, char *buf)           \
2913 {                                                                               \
2914         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2915         return sprintf(buf, "%.*s\n",                                           \
2916                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2917 }                                                                               \
2918 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2919
2920 nvme_show_str_function(model);
2921 nvme_show_str_function(serial);
2922 nvme_show_str_function(firmware_rev);
2923
2924 #define nvme_show_int_function(field)                                           \
2925 static ssize_t  field##_show(struct device *dev,                                \
2926                             struct device_attribute *attr, char *buf)           \
2927 {                                                                               \
2928         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2929         return sprintf(buf, "%d\n", ctrl->field);       \
2930 }                                                                               \
2931 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2932
2933 nvme_show_int_function(cntlid);
2934 nvme_show_int_function(numa_node);
2935
2936 static ssize_t nvme_sysfs_delete(struct device *dev,
2937                                 struct device_attribute *attr, const char *buf,
2938                                 size_t count)
2939 {
2940         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2941
2942         if (device_remove_file_self(dev, attr))
2943                 nvme_delete_ctrl_sync(ctrl);
2944         return count;
2945 }
2946 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2947
2948 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2949                                          struct device_attribute *attr,
2950                                          char *buf)
2951 {
2952         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2953
2954         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2955 }
2956 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2957
2958 static ssize_t nvme_sysfs_show_state(struct device *dev,
2959                                      struct device_attribute *attr,
2960                                      char *buf)
2961 {
2962         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2963         static const char *const state_name[] = {
2964                 [NVME_CTRL_NEW]         = "new",
2965                 [NVME_CTRL_LIVE]        = "live",
2966                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2967                 [NVME_CTRL_RESETTING]   = "resetting",
2968                 [NVME_CTRL_CONNECTING]  = "connecting",
2969                 [NVME_CTRL_DELETING]    = "deleting",
2970                 [NVME_CTRL_DEAD]        = "dead",
2971         };
2972
2973         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2974             state_name[ctrl->state])
2975                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2976
2977         return sprintf(buf, "unknown state\n");
2978 }
2979
2980 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2981
2982 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2983                                          struct device_attribute *attr,
2984                                          char *buf)
2985 {
2986         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2987
2988         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2989 }
2990 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2991
2992 static ssize_t nvme_sysfs_show_address(struct device *dev,
2993                                          struct device_attribute *attr,
2994                                          char *buf)
2995 {
2996         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2997
2998         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2999 }
3000 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3001
3002 static struct attribute *nvme_dev_attrs[] = {
3003         &dev_attr_reset_controller.attr,
3004         &dev_attr_rescan_controller.attr,
3005         &dev_attr_model.attr,
3006         &dev_attr_serial.attr,
3007         &dev_attr_firmware_rev.attr,
3008         &dev_attr_cntlid.attr,
3009         &dev_attr_delete_controller.attr,
3010         &dev_attr_transport.attr,
3011         &dev_attr_subsysnqn.attr,
3012         &dev_attr_address.attr,
3013         &dev_attr_state.attr,
3014         &dev_attr_numa_node.attr,
3015         NULL
3016 };
3017
3018 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3019                 struct attribute *a, int n)
3020 {
3021         struct device *dev = container_of(kobj, struct device, kobj);
3022         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3023
3024         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3025                 return 0;
3026         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3027                 return 0;
3028
3029         return a->mode;
3030 }
3031
3032 static struct attribute_group nvme_dev_attrs_group = {
3033         .attrs          = nvme_dev_attrs,
3034         .is_visible     = nvme_dev_attrs_are_visible,
3035 };
3036
3037 static const struct attribute_group *nvme_dev_attr_groups[] = {
3038         &nvme_dev_attrs_group,
3039         NULL,
3040 };
3041
3042 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3043                 unsigned nsid)
3044 {
3045         struct nvme_ns_head *h;
3046
3047         lockdep_assert_held(&subsys->lock);
3048
3049         list_for_each_entry(h, &subsys->nsheads, entry) {
3050                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3051                         return h;
3052         }
3053
3054         return NULL;
3055 }
3056
3057 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3058                 struct nvme_ns_head *new)
3059 {
3060         struct nvme_ns_head *h;
3061
3062         lockdep_assert_held(&subsys->lock);
3063
3064         list_for_each_entry(h, &subsys->nsheads, entry) {
3065                 if (nvme_ns_ids_valid(&new->ids) &&
3066                     !list_empty(&h->list) &&
3067                     nvme_ns_ids_equal(&new->ids, &h->ids))
3068                         return -EINVAL;
3069         }
3070
3071         return 0;
3072 }
3073
3074 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3075                 unsigned nsid, struct nvme_id_ns *id)
3076 {
3077         struct nvme_ns_head *head;
3078         size_t size = sizeof(*head);
3079         int ret = -ENOMEM;
3080
3081 #ifdef CONFIG_NVME_MULTIPATH
3082         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3083 #endif
3084
3085         head = kzalloc(size, GFP_KERNEL);
3086         if (!head)
3087                 goto out;
3088         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3089         if (ret < 0)
3090                 goto out_free_head;
3091         head->instance = ret;
3092         INIT_LIST_HEAD(&head->list);
3093         ret = init_srcu_struct(&head->srcu);
3094         if (ret)
3095                 goto out_ida_remove;
3096         head->subsys = ctrl->subsys;
3097         head->ns_id = nsid;
3098         kref_init(&head->ref);
3099
3100         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3101
3102         ret = __nvme_check_ids(ctrl->subsys, head);
3103         if (ret) {
3104                 dev_err(ctrl->device,
3105                         "duplicate IDs for nsid %d\n", nsid);
3106                 goto out_cleanup_srcu;
3107         }
3108
3109         ret = nvme_mpath_alloc_disk(ctrl, head);
3110         if (ret)
3111                 goto out_cleanup_srcu;
3112
3113         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3114
3115         kref_get(&ctrl->subsys->ref);
3116
3117         return head;
3118 out_cleanup_srcu:
3119         cleanup_srcu_struct(&head->srcu);
3120 out_ida_remove:
3121         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3122 out_free_head:
3123         kfree(head);
3124 out:
3125         return ERR_PTR(ret);
3126 }
3127
3128 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3129                 struct nvme_id_ns *id)
3130 {
3131         struct nvme_ctrl *ctrl = ns->ctrl;
3132         bool is_shared = id->nmic & (1 << 0);
3133         struct nvme_ns_head *head = NULL;
3134         int ret = 0;
3135
3136         mutex_lock(&ctrl->subsys->lock);
3137         if (is_shared)
3138                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3139         if (!head) {
3140                 head = nvme_alloc_ns_head(ctrl, nsid, id);
3141                 if (IS_ERR(head)) {
3142                         ret = PTR_ERR(head);
3143                         goto out_unlock;
3144                 }
3145         } else {
3146                 struct nvme_ns_ids ids;
3147
3148                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3149                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3150                         dev_err(ctrl->device,
3151                                 "IDs don't match for shared namespace %d\n",
3152                                         nsid);
3153                         ret = -EINVAL;
3154                         goto out_unlock;
3155                 }
3156         }
3157
3158         list_add_tail(&ns->siblings, &head->list);
3159         ns->head = head;
3160
3161 out_unlock:
3162         mutex_unlock(&ctrl->subsys->lock);
3163         return ret;
3164 }
3165
3166 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3167 {
3168         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3169         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3170
3171         return nsa->head->ns_id - nsb->head->ns_id;
3172 }
3173
3174 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3175 {
3176         struct nvme_ns *ns, *ret = NULL;
3177
3178         down_read(&ctrl->namespaces_rwsem);
3179         list_for_each_entry(ns, &ctrl->namespaces, list) {
3180                 if (ns->head->ns_id == nsid) {
3181                         if (!kref_get_unless_zero(&ns->kref))
3182                                 continue;
3183                         ret = ns;
3184                         break;
3185                 }
3186                 if (ns->head->ns_id > nsid)
3187                         break;
3188         }
3189         up_read(&ctrl->namespaces_rwsem);
3190         return ret;
3191 }
3192
3193 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3194 {
3195         struct streams_directive_params s;
3196         int ret;
3197
3198         if (!ctrl->nr_streams)
3199                 return 0;
3200
3201         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3202         if (ret)
3203                 return ret;
3204
3205         ns->sws = le32_to_cpu(s.sws);
3206         ns->sgs = le16_to_cpu(s.sgs);
3207
3208         if (ns->sws) {
3209                 unsigned int bs = 1 << ns->lba_shift;
3210
3211                 blk_queue_io_min(ns->queue, bs * ns->sws);
3212                 if (ns->sgs)
3213                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3214         }
3215
3216         return 0;
3217 }
3218
3219 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3220 {
3221         struct nvme_ns *ns;
3222         struct gendisk *disk;
3223         struct nvme_id_ns *id;
3224         char disk_name[DISK_NAME_LEN];
3225         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3226
3227         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3228         if (!ns)
3229                 return -ENOMEM;
3230
3231         ns->queue = blk_mq_init_queue(ctrl->tagset);
3232         if (IS_ERR(ns->queue)) {
3233                 ret = PTR_ERR(ns->queue);
3234                 goto out_free_ns;
3235         }
3236
3237         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3238         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3239                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3240
3241         ns->queue->queuedata = ns;
3242         ns->ctrl = ctrl;
3243
3244         kref_init(&ns->kref);
3245         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3246
3247         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3248         nvme_set_queue_limits(ctrl, ns->queue);
3249
3250         id = nvme_identify_ns(ctrl, nsid);
3251         if (!id) {
3252                 ret = -EIO;
3253                 goto out_free_queue;
3254         }
3255
3256         if (id->ncap == 0) {
3257                 ret = -EINVAL;
3258                 goto out_free_id;
3259         }
3260
3261         ret = nvme_init_ns_head(ns, nsid, id);
3262         if (ret)
3263                 goto out_free_id;
3264         nvme_setup_streams_ns(ctrl, ns);
3265         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3266
3267         disk = alloc_disk_node(0, node);
3268         if (!disk) {
3269                 ret = -ENOMEM;
3270                 goto out_unlink_ns;
3271         }
3272
3273         disk->fops = &nvme_fops;
3274         disk->private_data = ns;
3275         disk->queue = ns->queue;
3276         disk->flags = flags;
3277         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3278         ns->disk = disk;
3279
3280         __nvme_revalidate_disk(disk, id);
3281
3282         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3283                 ret = nvme_nvm_register(ns, disk_name, node);
3284                 if (ret) {
3285                         dev_warn(ctrl->device, "LightNVM init failure\n");
3286                         goto out_put_disk;
3287                 }
3288         }
3289
3290         down_write(&ctrl->namespaces_rwsem);
3291         list_add_tail(&ns->list, &ctrl->namespaces);
3292         up_write(&ctrl->namespaces_rwsem);
3293
3294         nvme_get_ctrl(ctrl);
3295
3296         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3297
3298         nvme_mpath_add_disk(ns, id);
3299         nvme_fault_inject_init(ns);
3300         kfree(id);
3301
3302         return 0;
3303  out_put_disk:
3304         put_disk(ns->disk);
3305  out_unlink_ns:
3306         mutex_lock(&ctrl->subsys->lock);
3307         list_del_rcu(&ns->siblings);
3308         mutex_unlock(&ctrl->subsys->lock);
3309         nvme_put_ns_head(ns->head);
3310  out_free_id:
3311         kfree(id);
3312  out_free_queue:
3313         blk_cleanup_queue(ns->queue);
3314  out_free_ns:
3315         kfree(ns);
3316         return ret;
3317 }
3318
3319 static void nvme_ns_remove(struct nvme_ns *ns)
3320 {
3321         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3322                 return;
3323
3324         nvme_fault_inject_fini(ns);
3325         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3326                 del_gendisk(ns->disk);
3327                 blk_cleanup_queue(ns->queue);
3328                 if (blk_get_integrity(ns->disk))
3329                         blk_integrity_unregister(ns->disk);
3330         }
3331
3332         mutex_lock(&ns->ctrl->subsys->lock);
3333         list_del_rcu(&ns->siblings);
3334         nvme_mpath_clear_current_path(ns);
3335         mutex_unlock(&ns->ctrl->subsys->lock);
3336
3337         down_write(&ns->ctrl->namespaces_rwsem);
3338         list_del_init(&ns->list);
3339         up_write(&ns->ctrl->namespaces_rwsem);
3340
3341         synchronize_srcu(&ns->head->srcu);
3342         nvme_mpath_check_last_path(ns);
3343         nvme_put_ns(ns);
3344 }
3345
3346 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3347 {
3348         struct nvme_ns *ns;
3349
3350         ns = nvme_find_get_ns(ctrl, nsid);
3351         if (ns) {
3352                 if (ns->disk && revalidate_disk(ns->disk))
3353                         nvme_ns_remove(ns);
3354                 nvme_put_ns(ns);
3355         } else
3356                 nvme_alloc_ns(ctrl, nsid);
3357 }
3358
3359 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3360                                         unsigned nsid)
3361 {
3362         struct nvme_ns *ns, *next;
3363         LIST_HEAD(rm_list);
3364
3365         down_write(&ctrl->namespaces_rwsem);
3366         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3367                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3368                         list_move_tail(&ns->list, &rm_list);
3369         }
3370         up_write(&ctrl->namespaces_rwsem);
3371
3372         list_for_each_entry_safe(ns, next, &rm_list, list)
3373                 nvme_ns_remove(ns);
3374
3375 }
3376
3377 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3378 {
3379         struct nvme_ns *ns;
3380         __le32 *ns_list;
3381         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3382         int ret = 0;
3383
3384         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3385         if (!ns_list)
3386                 return -ENOMEM;
3387
3388         for (i = 0; i < num_lists; i++) {
3389                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3390                 if (ret)
3391                         goto free;
3392
3393                 for (j = 0; j < min(nn, 1024U); j++) {
3394                         nsid = le32_to_cpu(ns_list[j]);
3395                         if (!nsid)
3396                                 goto out;
3397
3398                         nvme_validate_ns(ctrl, nsid);
3399
3400                         while (++prev < nsid) {
3401                                 ns = nvme_find_get_ns(ctrl, prev);
3402                                 if (ns) {
3403                                         nvme_ns_remove(ns);
3404                                         nvme_put_ns(ns);
3405                                 }
3406                         }
3407                 }
3408                 nn -= j;
3409         }
3410  out:
3411         nvme_remove_invalid_namespaces(ctrl, prev);
3412  free:
3413         kfree(ns_list);
3414         return ret;
3415 }
3416
3417 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3418 {
3419         unsigned i;
3420
3421         for (i = 1; i <= nn; i++)
3422                 nvme_validate_ns(ctrl, i);
3423
3424         nvme_remove_invalid_namespaces(ctrl, nn);
3425 }
3426
3427 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3428 {
3429         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3430         __le32 *log;
3431         int error;
3432
3433         log = kzalloc(log_size, GFP_KERNEL);
3434         if (!log)
3435                 return;
3436
3437         /*
3438          * We need to read the log to clear the AEN, but we don't want to rely
3439          * on it for the changed namespace information as userspace could have
3440          * raced with us in reading the log page, which could cause us to miss
3441          * updates.
3442          */
3443         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3444                         log_size, 0);
3445         if (error)
3446                 dev_warn(ctrl->device,
3447                         "reading changed ns log failed: %d\n", error);
3448
3449         kfree(log);
3450 }
3451
3452 static void nvme_scan_work(struct work_struct *work)
3453 {
3454         struct nvme_ctrl *ctrl =
3455                 container_of(work, struct nvme_ctrl, scan_work);
3456         struct nvme_id_ctrl *id;
3457         unsigned nn;
3458
3459         if (ctrl->state != NVME_CTRL_LIVE)
3460                 return;
3461
3462         WARN_ON_ONCE(!ctrl->tagset);
3463
3464         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3465                 dev_info(ctrl->device, "rescanning namespaces.\n");
3466                 nvme_clear_changed_ns_log(ctrl);
3467         }
3468
3469         if (nvme_identify_ctrl(ctrl, &id))
3470                 return;
3471
3472         mutex_lock(&ctrl->scan_lock);
3473         nn = le32_to_cpu(id->nn);
3474         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3475             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3476                 if (!nvme_scan_ns_list(ctrl, nn))
3477                         goto out_free_id;
3478         }
3479         nvme_scan_ns_sequential(ctrl, nn);
3480 out_free_id:
3481         mutex_unlock(&ctrl->scan_lock);
3482         kfree(id);
3483         down_write(&ctrl->namespaces_rwsem);
3484         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3485         up_write(&ctrl->namespaces_rwsem);
3486 }
3487
3488 /*
3489  * This function iterates the namespace list unlocked to allow recovery from
3490  * controller failure. It is up to the caller to ensure the namespace list is
3491  * not modified by scan work while this function is executing.
3492  */
3493 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3494 {
3495         struct nvme_ns *ns, *next;
3496         LIST_HEAD(ns_list);
3497
3498         /* prevent racing with ns scanning */
3499         flush_work(&ctrl->scan_work);
3500
3501         /*
3502          * The dead states indicates the controller was not gracefully
3503          * disconnected. In that case, we won't be able to flush any data while
3504          * removing the namespaces' disks; fail all the queues now to avoid
3505          * potentially having to clean up the failed sync later.
3506          */
3507         if (ctrl->state == NVME_CTRL_DEAD)
3508                 nvme_kill_queues(ctrl);
3509
3510         down_write(&ctrl->namespaces_rwsem);
3511         list_splice_init(&ctrl->namespaces, &ns_list);
3512         up_write(&ctrl->namespaces_rwsem);
3513
3514         list_for_each_entry_safe(ns, next, &ns_list, list)
3515                 nvme_ns_remove(ns);
3516 }
3517 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3518
3519 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3520 {
3521         char *envp[2] = { NULL, NULL };
3522         u32 aen_result = ctrl->aen_result;
3523
3524         ctrl->aen_result = 0;
3525         if (!aen_result)
3526                 return;
3527
3528         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3529         if (!envp[0])
3530                 return;
3531         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3532         kfree(envp[0]);
3533 }
3534
3535 static void nvme_async_event_work(struct work_struct *work)
3536 {
3537         struct nvme_ctrl *ctrl =
3538                 container_of(work, struct nvme_ctrl, async_event_work);
3539
3540         nvme_aen_uevent(ctrl);
3541         ctrl->ops->submit_async_event(ctrl);
3542 }
3543
3544 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3545 {
3546
3547         u32 csts;
3548
3549         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3550                 return false;
3551
3552         if (csts == ~0)
3553                 return false;
3554
3555         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3556 }
3557
3558 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3559 {
3560         struct nvme_fw_slot_info_log *log;
3561
3562         log = kmalloc(sizeof(*log), GFP_KERNEL);
3563         if (!log)
3564                 return;
3565
3566         if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3567                         sizeof(*log), 0))
3568                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3569         kfree(log);
3570 }
3571
3572 static void nvme_fw_act_work(struct work_struct *work)
3573 {
3574         struct nvme_ctrl *ctrl = container_of(work,
3575                                 struct nvme_ctrl, fw_act_work);
3576         unsigned long fw_act_timeout;
3577
3578         if (ctrl->mtfa)
3579                 fw_act_timeout = jiffies +
3580                                 msecs_to_jiffies(ctrl->mtfa * 100);
3581         else
3582                 fw_act_timeout = jiffies +
3583                                 msecs_to_jiffies(admin_timeout * 1000);
3584
3585         nvme_stop_queues(ctrl);
3586         while (nvme_ctrl_pp_status(ctrl)) {
3587                 if (time_after(jiffies, fw_act_timeout)) {
3588                         dev_warn(ctrl->device,
3589                                 "Fw activation timeout, reset controller\n");
3590                         nvme_reset_ctrl(ctrl);
3591                         break;
3592                 }
3593                 msleep(100);
3594         }
3595
3596         if (ctrl->state != NVME_CTRL_LIVE)
3597                 return;
3598
3599         nvme_start_queues(ctrl);
3600         /* read FW slot information to clear the AER */
3601         nvme_get_fw_slot_info(ctrl);
3602 }
3603
3604 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3605 {
3606         u32 aer_notice_type = (result & 0xff00) >> 8;
3607
3608         switch (aer_notice_type) {
3609         case NVME_AER_NOTICE_NS_CHANGED:
3610                 trace_nvme_async_event(ctrl, aer_notice_type);
3611                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3612                 nvme_queue_scan(ctrl);
3613                 break;
3614         case NVME_AER_NOTICE_FW_ACT_STARTING:
3615                 trace_nvme_async_event(ctrl, aer_notice_type);
3616                 queue_work(nvme_wq, &ctrl->fw_act_work);
3617                 break;
3618 #ifdef CONFIG_NVME_MULTIPATH
3619         case NVME_AER_NOTICE_ANA:
3620                 trace_nvme_async_event(ctrl, aer_notice_type);
3621                 if (!ctrl->ana_log_buf)
3622                         break;
3623                 queue_work(nvme_wq, &ctrl->ana_work);
3624                 break;
3625 #endif
3626         default:
3627                 dev_warn(ctrl->device, "async event result %08x\n", result);
3628         }
3629 }
3630
3631 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3632                 volatile union nvme_result *res)
3633 {
3634         u32 result = le32_to_cpu(res->u32);
3635         u32 aer_type = result & 0x07;
3636
3637         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3638                 return;
3639
3640         switch (aer_type) {
3641         case NVME_AER_NOTICE:
3642                 nvme_handle_aen_notice(ctrl, result);
3643                 break;
3644         case NVME_AER_ERROR:
3645         case NVME_AER_SMART:
3646         case NVME_AER_CSS:
3647         case NVME_AER_VS:
3648                 trace_nvme_async_event(ctrl, aer_type);
3649                 ctrl->aen_result = result;
3650                 break;
3651         default:
3652                 break;
3653         }
3654         queue_work(nvme_wq, &ctrl->async_event_work);
3655 }
3656 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3657
3658 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3659 {
3660         nvme_mpath_stop(ctrl);
3661         nvme_stop_keep_alive(ctrl);
3662         flush_work(&ctrl->async_event_work);
3663         cancel_work_sync(&ctrl->fw_act_work);
3664 }
3665 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3666
3667 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3668 {
3669         if (ctrl->kato)
3670                 nvme_start_keep_alive(ctrl);
3671
3672         if (ctrl->queue_count > 1) {
3673                 nvme_queue_scan(ctrl);
3674                 nvme_enable_aen(ctrl);
3675                 queue_work(nvme_wq, &ctrl->async_event_work);
3676                 nvme_start_queues(ctrl);
3677         }
3678 }
3679 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3680
3681 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3682 {
3683         cdev_device_del(&ctrl->cdev, ctrl->device);
3684 }
3685 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3686
3687 static void nvme_free_ctrl(struct device *dev)
3688 {
3689         struct nvme_ctrl *ctrl =
3690                 container_of(dev, struct nvme_ctrl, ctrl_device);
3691         struct nvme_subsystem *subsys = ctrl->subsys;
3692
3693         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3694         kfree(ctrl->effects);
3695         nvme_mpath_uninit(ctrl);
3696         __free_page(ctrl->discard_page);
3697
3698         if (subsys) {
3699                 mutex_lock(&subsys->lock);
3700                 list_del(&ctrl->subsys_entry);
3701                 mutex_unlock(&subsys->lock);
3702                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3703         }
3704
3705         ctrl->ops->free_ctrl(ctrl);
3706
3707         if (subsys)
3708                 nvme_put_subsystem(subsys);
3709 }
3710
3711 /*
3712  * Initialize a NVMe controller structures.  This needs to be called during
3713  * earliest initialization so that we have the initialized structured around
3714  * during probing.
3715  */
3716 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3717                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3718 {
3719         int ret;
3720
3721         ctrl->state = NVME_CTRL_NEW;
3722         spin_lock_init(&ctrl->lock);
3723         mutex_init(&ctrl->scan_lock);
3724         INIT_LIST_HEAD(&ctrl->namespaces);
3725         init_rwsem(&ctrl->namespaces_rwsem);
3726         ctrl->dev = dev;
3727         ctrl->ops = ops;
3728         ctrl->quirks = quirks;
3729         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3730         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3731         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3732         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3733
3734         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3735         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3736         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3737
3738         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3739                         PAGE_SIZE);
3740         ctrl->discard_page = alloc_page(GFP_KERNEL);
3741         if (!ctrl->discard_page) {
3742                 ret = -ENOMEM;
3743                 goto out;
3744         }
3745
3746         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3747         if (ret < 0)
3748                 goto out;
3749         ctrl->instance = ret;
3750
3751         device_initialize(&ctrl->ctrl_device);
3752         ctrl->device = &ctrl->ctrl_device;
3753         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3754         ctrl->device->class = nvme_class;
3755         ctrl->device->parent = ctrl->dev;
3756         ctrl->device->groups = nvme_dev_attr_groups;
3757         ctrl->device->release = nvme_free_ctrl;
3758         dev_set_drvdata(ctrl->device, ctrl);
3759         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3760         if (ret)
3761                 goto out_release_instance;
3762
3763         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3764         ctrl->cdev.owner = ops->module;
3765         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3766         if (ret)
3767                 goto out_free_name;
3768
3769         /*
3770          * Initialize latency tolerance controls.  The sysfs files won't
3771          * be visible to userspace unless the device actually supports APST.
3772          */
3773         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3774         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3775                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3776
3777         return 0;
3778 out_free_name:
3779         kfree_const(ctrl->device->kobj.name);
3780 out_release_instance:
3781         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3782 out:
3783         if (ctrl->discard_page)
3784                 __free_page(ctrl->discard_page);
3785         return ret;
3786 }
3787 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3788
3789 /**
3790  * nvme_kill_queues(): Ends all namespace queues
3791  * @ctrl: the dead controller that needs to end
3792  *
3793  * Call this function when the driver determines it is unable to get the
3794  * controller in a state capable of servicing IO.
3795  */
3796 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3797 {
3798         struct nvme_ns *ns;
3799
3800         down_read(&ctrl->namespaces_rwsem);
3801
3802         /* Forcibly unquiesce queues to avoid blocking dispatch */
3803         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3804                 blk_mq_unquiesce_queue(ctrl->admin_q);
3805
3806         list_for_each_entry(ns, &ctrl->namespaces, list)
3807                 nvme_set_queue_dying(ns);
3808
3809         up_read(&ctrl->namespaces_rwsem);
3810 }
3811 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3812
3813 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3814 {
3815         struct nvme_ns *ns;
3816
3817         down_read(&ctrl->namespaces_rwsem);
3818         list_for_each_entry(ns, &ctrl->namespaces, list)
3819                 blk_mq_unfreeze_queue(ns->queue);
3820         up_read(&ctrl->namespaces_rwsem);
3821 }
3822 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3823
3824 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3825 {
3826         struct nvme_ns *ns;
3827
3828         down_read(&ctrl->namespaces_rwsem);
3829         list_for_each_entry(ns, &ctrl->namespaces, list) {
3830                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3831                 if (timeout <= 0)
3832                         break;
3833         }
3834         up_read(&ctrl->namespaces_rwsem);
3835 }
3836 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3837
3838 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3839 {
3840         struct nvme_ns *ns;
3841
3842         down_read(&ctrl->namespaces_rwsem);
3843         list_for_each_entry(ns, &ctrl->namespaces, list)
3844                 blk_mq_freeze_queue_wait(ns->queue);
3845         up_read(&ctrl->namespaces_rwsem);
3846 }
3847 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3848
3849 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3850 {
3851         struct nvme_ns *ns;
3852
3853         down_read(&ctrl->namespaces_rwsem);
3854         list_for_each_entry(ns, &ctrl->namespaces, list)
3855                 blk_freeze_queue_start(ns->queue);
3856         up_read(&ctrl->namespaces_rwsem);
3857 }
3858 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3859
3860 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3861 {
3862         struct nvme_ns *ns;
3863
3864         down_read(&ctrl->namespaces_rwsem);
3865         list_for_each_entry(ns, &ctrl->namespaces, list)
3866                 blk_mq_quiesce_queue(ns->queue);
3867         up_read(&ctrl->namespaces_rwsem);
3868 }
3869 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3870
3871 void nvme_start_queues(struct nvme_ctrl *ctrl)
3872 {
3873         struct nvme_ns *ns;
3874
3875         down_read(&ctrl->namespaces_rwsem);
3876         list_for_each_entry(ns, &ctrl->namespaces, list)
3877                 blk_mq_unquiesce_queue(ns->queue);
3878         up_read(&ctrl->namespaces_rwsem);
3879 }
3880 EXPORT_SYMBOL_GPL(nvme_start_queues);
3881
3882 /*
3883  * Check we didn't inadvertently grow the command structure sizes:
3884  */
3885 static inline void _nvme_check_size(void)
3886 {
3887         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
3888         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
3889         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
3890         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
3891         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
3892         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
3893         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
3894         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
3895         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
3896         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
3897         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
3898         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
3899         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
3900         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
3901         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
3902         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
3903         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
3904 }
3905
3906
3907 static int __init nvme_core_init(void)
3908 {
3909         int result = -ENOMEM;
3910
3911         _nvme_check_size();
3912
3913         nvme_wq = alloc_workqueue("nvme-wq",
3914                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3915         if (!nvme_wq)
3916                 goto out;
3917
3918         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3919                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3920         if (!nvme_reset_wq)
3921                 goto destroy_wq;
3922
3923         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3924                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3925         if (!nvme_delete_wq)
3926                 goto destroy_reset_wq;
3927
3928         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3929         if (result < 0)
3930                 goto destroy_delete_wq;
3931
3932         nvme_class = class_create(THIS_MODULE, "nvme");
3933         if (IS_ERR(nvme_class)) {
3934                 result = PTR_ERR(nvme_class);
3935                 goto unregister_chrdev;
3936         }
3937
3938         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3939         if (IS_ERR(nvme_subsys_class)) {
3940                 result = PTR_ERR(nvme_subsys_class);
3941                 goto destroy_class;
3942         }
3943         return 0;
3944
3945 destroy_class:
3946         class_destroy(nvme_class);
3947 unregister_chrdev:
3948         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3949 destroy_delete_wq:
3950         destroy_workqueue(nvme_delete_wq);
3951 destroy_reset_wq:
3952         destroy_workqueue(nvme_reset_wq);
3953 destroy_wq:
3954         destroy_workqueue(nvme_wq);
3955 out:
3956         return result;
3957 }
3958
3959 static void __exit nvme_core_exit(void)
3960 {
3961         ida_destroy(&nvme_subsystems_ida);
3962         class_destroy(nvme_subsys_class);
3963         class_destroy(nvme_class);
3964         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3965         destroy_workqueue(nvme_delete_wq);
3966         destroy_workqueue(nvme_reset_wq);
3967         destroy_workqueue(nvme_wq);
3968 }
3969
3970 MODULE_LICENSE("GPL");
3971 MODULE_VERSION("1.0");
3972 module_init(nvme_core_init);
3973 module_exit(nvme_core_exit);