drivers/nvme/host/core.c

   1 /*
   2  * NVM Express device driver
   3  * Copyright (c) 2011-2014, Intel Corporation.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  */
  14
  15 #include <linux/blkdev.h>
  16 #include <linux/blk-mq.h>
  17 #include <linux/delay.h>
  18 #include <linux/errno.h>
  19 #include <linux/hdreg.h>
  20 #include <linux/kernel.h>
  21 #include <linux/module.h>
  22 #include <linux/list_sort.h>
  23 #include <linux/slab.h>
  24 #include <linux/types.h>
  25 #include <linux/pr.h>
  26 #include <linux/ptrace.h>
  27 #include <linux/nvme_ioctl.h>
  28 #include <linux/t10-pi.h>
  29 #include <linux/pm_qos.h>
  30 #include <scsi/sg.h>
  31 #include <asm/unaligned.h>
  32
  33 #include "nvme.h"
  34 #include "fabrics.h"
  35
  36 #define NVME_MINORS             (1U << MINORBITS)
  37
  38 unsigned char admin_timeout = 60;
  39 module_param(admin_timeout, byte, 0644);
  40 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  41 EXPORT_SYMBOL_GPL(admin_timeout);
  42
  43 unsigned char nvme_io_timeout = 30;
  44 module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
  45 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  46 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  47
  48 unsigned char shutdown_timeout = 5;
  49 module_param(shutdown_timeout, byte, 0644);
  50 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  51
  52 static u8 nvme_max_retries = 5;
  53 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  54 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  55
  56 static int nvme_char_major;
  57 module_param(nvme_char_major, int, 0);
  58
  59 static unsigned long default_ps_max_latency_us = 25000;
  60 module_param(default_ps_max_latency_us, ulong, 0644);
  61 MODULE_PARM_DESC(default_ps_max_latency_us,
  62                  "max power saving latency for new devices; use PM QOS to change per device");
  63
  64 static bool force_apst;
  65 module_param(force_apst, bool, 0644);
  66 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  67
  68 static LIST_HEAD(nvme_ctrl_list);
  69 static DEFINE_SPINLOCK(dev_list_lock);
  70
  71 static struct class *nvme_class;
  72
  73 static blk_status_t nvme_error_status(struct request *req)
  74 {
  75         switch (nvme_req(req)->status & 0x7ff) {
  76         case NVME_SC_SUCCESS:
  77                 return BLK_STS_OK;
  78         case NVME_SC_CAP_EXCEEDED:
  79                 return BLK_STS_NOSPC;
  80         case NVME_SC_ONCS_NOT_SUPPORTED:
  81                 return BLK_STS_NOTSUPP;
  82         case NVME_SC_WRITE_FAULT:
  83         case NVME_SC_READ_ERROR:
  84         case NVME_SC_UNWRITTEN_BLOCK:
  85                 return BLK_STS_MEDIUM;
  86         default:
  87                 return BLK_STS_IOERR;
  88         }
  89 }
  90
  91 static inline bool nvme_req_needs_retry(struct request *req)
  92 {
  93         if (blk_noretry_request(req))
  94                 return false;
  95         if (nvme_req(req)->status & NVME_SC_DNR)
  96                 return false;
  97         if (jiffies - req->start_time >= req->timeout)
  98                 return false;
  99         if (nvme_req(req)->retries >= nvme_max_retries)
 100                 return false;
 101         return true;
 102 }
 103
 104 void nvme_complete_rq(struct request *req)
 105 {
 106         if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) {
 107                 nvme_req(req)->retries++;
 108                 blk_mq_requeue_request(req, !blk_mq_queue_stopped(req->q));
 109                 return;
 110         }
 111
 112         blk_mq_end_request(req, nvme_error_status(req));
 113 }
 114 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 115
 116 void nvme_cancel_request(struct request *req, void *data, bool reserved)
 117 {
 118         int status;
 119
 120         if (!blk_mq_request_started(req))
 121                 return;
 122
 123         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 124                                 "Cancelling I/O %d", req->tag);
 125
 126         status = NVME_SC_ABORT_REQ;
 127         if (blk_queue_dying(req->q))
 128                 status |= NVME_SC_DNR;
 129         nvme_req(req)->status = status;
 130         blk_mq_complete_request(req);
 131
 132 }
 133 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 134
 135 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 136                 enum nvme_ctrl_state new_state)
 137 {
 138         enum nvme_ctrl_state old_state;
 139         bool changed = false;
 140
 141         spin_lock_irq(&ctrl->lock);
 142
 143         old_state = ctrl->state;
 144         switch (new_state) {
 145         case NVME_CTRL_LIVE:
 146                 switch (old_state) {
 147                 case NVME_CTRL_NEW:
 148                 case NVME_CTRL_RESETTING:
 149                 case NVME_CTRL_RECONNECTING:
 150                         changed = true;
 151                         /* FALLTHRU */
 152                 default:
 153                         break;
 154                 }
 155                 break;
 156         case NVME_CTRL_RESETTING:
 157                 switch (old_state) {
 158                 case NVME_CTRL_NEW:
 159                 case NVME_CTRL_LIVE:
 160                 case NVME_CTRL_RECONNECTING:
 161                         changed = true;
 162                         /* FALLTHRU */
 163                 default:
 164                         break;
 165                 }
 166                 break;
 167         case NVME_CTRL_RECONNECTING:
 168                 switch (old_state) {
 169                 case NVME_CTRL_LIVE:
 170                         changed = true;
 171                         /* FALLTHRU */
 172                 default:
 173                         break;
 174                 }
 175                 break;
 176         case NVME_CTRL_DELETING:
 177                 switch (old_state) {
 178                 case NVME_CTRL_LIVE:
 179                 case NVME_CTRL_RESETTING:
 180                 case NVME_CTRL_RECONNECTING:
 181                         changed = true;
 182                         /* FALLTHRU */
 183                 default:
 184                         break;
 185                 }
 186                 break;
 187         case NVME_CTRL_DEAD:
 188                 switch (old_state) {
 189                 case NVME_CTRL_DELETING:
 190                         changed = true;
 191                         /* FALLTHRU */
 192                 default:
 193                         break;
 194                 }
 195                 break;
 196         default:
 197                 break;
 198         }
 199
 200         if (changed)
 201                 ctrl->state = new_state;
 202
 203         spin_unlock_irq(&ctrl->lock);
 204
 205         return changed;
 206 }
 207 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 208
 209 static void nvme_free_ns(struct kref *kref)
 210 {
 211         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 212
 213         if (ns->ndev)
 214                 nvme_nvm_unregister(ns);
 215
 216         if (ns->disk) {
 217                 spin_lock(&dev_list_lock);
 218                 ns->disk->private_data = NULL;
 219                 spin_unlock(&dev_list_lock);
 220         }
 221
 222         put_disk(ns->disk);
 223         ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
 224         nvme_put_ctrl(ns->ctrl);
 225         kfree(ns);
 226 }
 227
 228 static void nvme_put_ns(struct nvme_ns *ns)
 229 {
 230         kref_put(&ns->kref, nvme_free_ns);
 231 }
 232
 233 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
 234 {
 235         struct nvme_ns *ns;
 236
 237         spin_lock(&dev_list_lock);
 238         ns = disk->private_data;
 239         if (ns) {
 240                 if (!kref_get_unless_zero(&ns->kref))
 241                         goto fail;
 242                 if (!try_module_get(ns->ctrl->ops->module))
 243                         goto fail_put_ns;
 244         }
 245         spin_unlock(&dev_list_lock);
 246
 247         return ns;
 248
 249 fail_put_ns:
 250         kref_put(&ns->kref, nvme_free_ns);
 251 fail:
 252         spin_unlock(&dev_list_lock);
 253         return NULL;
 254 }
 255
 256 struct request *nvme_alloc_request(struct request_queue *q,
 257                 struct nvme_command *cmd, unsigned int flags, int qid)
 258 {
 259         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 260         struct request *req;
 261
 262         if (qid == NVME_QID_ANY) {
 263                 req = blk_mq_alloc_request(q, op, flags);
 264         } else {
 265                 req = blk_mq_alloc_request_hctx(q, op, flags,
 266                                 qid ? qid - 1 : 0);
 267         }
 268         if (IS_ERR(req))
 269                 return req;
 270
 271         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 272         nvme_req(req)->cmd = cmd;
 273
 274         return req;
 275 }
 276 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 277
 278 static inline void nvme_setup_flush(struct nvme_ns *ns,
 279                 struct nvme_command *cmnd)
 280 {
 281         memset(cmnd, 0, sizeof(*cmnd));
 282         cmnd->common.opcode = nvme_cmd_flush;
 283         cmnd->common.nsid = cpu_to_le32(ns->ns_id);
 284 }
 285
 286 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 287                 struct nvme_command *cmnd)
 288 {
 289         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 290         struct nvme_dsm_range *range;
 291         struct bio *bio;
 292
 293         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 294         if (!range)
 295                 return BLK_MQ_RQ_QUEUE_BUSY;
 296
 297         __rq_for_each_bio(bio, req) {
 298                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 299                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 300
 301                 range[n].cattr = cpu_to_le32(0);
 302                 range[n].nlb = cpu_to_le32(nlb);
 303                 range[n].slba = cpu_to_le64(slba);
 304                 n++;
 305         }
 306
 307         if (WARN_ON_ONCE(n != segments)) {
 308                 kfree(range);
 309                 return BLK_MQ_RQ_QUEUE_ERROR;
 310         }
 311
 312         memset(cmnd, 0, sizeof(*cmnd));
 313         cmnd->dsm.opcode = nvme_cmd_dsm;
 314         cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
 315         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 316         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 317
 318         req->special_vec.bv_page = virt_to_page(range);
 319         req->special_vec.bv_offset = offset_in_page(range);
 320         req->special_vec.bv_len = sizeof(*range) * segments;
 321         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 322
 323         return BLK_MQ_RQ_QUEUE_OK;
 324 }
 325
 326 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
 327                 struct nvme_command *cmnd)
 328 {
 329         u16 control = 0;
 330         u32 dsmgmt = 0;
 331
 332         if (req->cmd_flags & REQ_FUA)
 333                 control |= NVME_RW_FUA;
 334         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 335                 control |= NVME_RW_LR;
 336
 337         if (req->cmd_flags & REQ_RAHEAD)
 338                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 339
 340         memset(cmnd, 0, sizeof(*cmnd));
 341         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 342         cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
 343         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 344         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 345
 346         if (ns->ms) {
 347                 switch (ns->pi_type) {
 348                 case NVME_NS_DPS_PI_TYPE3:
 349                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 350                         break;
 351                 case NVME_NS_DPS_PI_TYPE1:
 352                 case NVME_NS_DPS_PI_TYPE2:
 353                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 354                                         NVME_RW_PRINFO_PRCHK_REF;
 355                         cmnd->rw.reftag = cpu_to_le32(
 356                                         nvme_block_nr(ns, blk_rq_pos(req)));
 357                         break;
 358                 }
 359                 if (!blk_integrity_rq(req))
 360                         control |= NVME_RW_PRINFO_PRACT;
 361         }
 362
 363         cmnd->rw.control = cpu_to_le16(control);
 364         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 365 }
 366
 367 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 368                 struct nvme_command *cmd)
 369 {
 370         int ret = BLK_MQ_RQ_QUEUE_OK;
 371
 372         if (!(req->rq_flags & RQF_DONTPREP)) {
 373                 nvme_req(req)->retries = 0;
 374                 nvme_req(req)->flags = 0;
 375                 req->rq_flags |= RQF_DONTPREP;
 376         }
 377
 378         switch (req_op(req)) {
 379         case REQ_OP_DRV_IN:
 380         case REQ_OP_DRV_OUT:
 381                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 382                 break;
 383         case REQ_OP_FLUSH:
 384                 nvme_setup_flush(ns, cmd);
 385                 break;
 386         case REQ_OP_WRITE_ZEROES:
 387                 /* currently only aliased to deallocate for a few ctrls: */
 388         case REQ_OP_DISCARD:
 389                 ret = nvme_setup_discard(ns, req, cmd);
 390                 break;
 391         case REQ_OP_READ:
 392         case REQ_OP_WRITE:
 393                 nvme_setup_rw(ns, req, cmd);
 394                 break;
 395         default:
 396                 WARN_ON_ONCE(1);
 397                 return BLK_MQ_RQ_QUEUE_ERROR;
 398         }
 399
 400         cmd->common.command_id = req->tag;
 401         return ret;
 402 }
 403 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 404
 405 /*
 406  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 407  * if the result is positive, it's an NVM Express status code
 408  */
 409 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 410                 union nvme_result *result, void *buffer, unsigned bufflen,
 411                 unsigned timeout, int qid, int at_head, int flags)
 412 {
 413         struct request *req;
 414         int ret;
 415
 416         req = nvme_alloc_request(q, cmd, flags, qid);
 417         if (IS_ERR(req))
 418                 return PTR_ERR(req);
 419
 420         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 421
 422         if (buffer && bufflen) {
 423                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 424                 if (ret)
 425                         goto out;
 426         }
 427
 428         blk_execute_rq(req->q, NULL, req, at_head);
 429         if (result)
 430                 *result = nvme_req(req)->result;
 431         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 432                 ret = -EINTR;
 433         else
 434                 ret = nvme_req(req)->status;
 435  out:
 436         blk_mq_free_request(req);
 437         return ret;
 438 }
 439 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 440
 441 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 442                 void *buffer, unsigned bufflen)
 443 {
 444         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 445                         NVME_QID_ANY, 0, 0);
 446 }
 447 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 448
 449 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 450                 void __user *ubuffer, unsigned bufflen,
 451                 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
 452                 u32 *result, unsigned timeout)
 453 {
 454         bool write = nvme_is_write(cmd);
 455         struct nvme_ns *ns = q->queuedata;
 456         struct gendisk *disk = ns ? ns->disk : NULL;
 457         struct request *req;
 458         struct bio *bio = NULL;
 459         void *meta = NULL;
 460         int ret;
 461
 462         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 463         if (IS_ERR(req))
 464                 return PTR_ERR(req);
 465
 466         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 467
 468         if (ubuffer && bufflen) {
 469                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 470                                 GFP_KERNEL);
 471                 if (ret)
 472                         goto out;
 473                 bio = req->bio;
 474
 475                 if (!disk)
 476                         goto submit;
 477                 bio->bi_bdev = bdget_disk(disk, 0);
 478                 if (!bio->bi_bdev) {
 479                         ret = -ENODEV;
 480                         goto out_unmap;
 481                 }
 482
 483                 if (meta_buffer && meta_len) {
 484                         struct bio_integrity_payload *bip;
 485
 486                         meta = kmalloc(meta_len, GFP_KERNEL);
 487                         if (!meta) {
 488                                 ret = -ENOMEM;
 489                                 goto out_unmap;
 490                         }
 491
 492                         if (write) {
 493                                 if (copy_from_user(meta, meta_buffer,
 494                                                 meta_len)) {
 495                                         ret = -EFAULT;
 496                                         goto out_free_meta;
 497                                 }
 498                         }
 499
 500                         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 501                         if (IS_ERR(bip)) {
 502                                 ret = PTR_ERR(bip);
 503                                 goto out_free_meta;
 504                         }
 505
 506                         bip->bip_iter.bi_size = meta_len;
 507                         bip->bip_iter.bi_sector = meta_seed;
 508
 509                         ret = bio_integrity_add_page(bio, virt_to_page(meta),
 510                                         meta_len, offset_in_page(meta));
 511                         if (ret != meta_len) {
 512                                 ret = -ENOMEM;
 513                                 goto out_free_meta;
 514                         }
 515                 }
 516         }
 517  submit:
 518         blk_execute_rq(req->q, disk, req, 0);
 519         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 520                 ret = -EINTR;
 521         else
 522                 ret = nvme_req(req)->status;
 523         if (result)
 524                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 525         if (meta && !ret && !write) {
 526                 if (copy_to_user(meta_buffer, meta, meta_len))
 527                         ret = -EFAULT;
 528         }
 529  out_free_meta:
 530         kfree(meta);
 531  out_unmap:
 532         if (bio) {
 533                 if (disk && bio->bi_bdev)
 534                         bdput(bio->bi_bdev);
 535                 blk_rq_unmap_user(bio);
 536         }
 537  out:
 538         blk_mq_free_request(req);
 539         return ret;
 540 }
 541
 542 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 543                 void __user *ubuffer, unsigned bufflen, u32 *result,
 544                 unsigned timeout)
 545 {
 546         return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
 547                         result, timeout);
 548 }
 549
 550 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 551 {
 552         struct nvme_ctrl *ctrl = rq->end_io_data;
 553
 554         blk_mq_free_request(rq);
 555
 556         if (status) {
 557                 dev_err(ctrl->device,
 558                         "failed nvme_keep_alive_end_io error=%d\n",
 559                                 status);
 560                 return;
 561         }
 562
 563         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 564 }
 565
 566 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 567 {
 568         struct nvme_command c;
 569         struct request *rq;
 570
 571         memset(&c, 0, sizeof(c));
 572         c.common.opcode = nvme_admin_keep_alive;
 573
 574         rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
 575                         NVME_QID_ANY);
 576         if (IS_ERR(rq))
 577                 return PTR_ERR(rq);
 578
 579         rq->timeout = ctrl->kato * HZ;
 580         rq->end_io_data = ctrl;
 581
 582         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 583
 584         return 0;
 585 }
 586
 587 static void nvme_keep_alive_work(struct work_struct *work)
 588 {
 589         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 590                         struct nvme_ctrl, ka_work);
 591
 592         if (nvme_keep_alive(ctrl)) {
 593                 /* allocation failure, reset the controller */
 594                 dev_err(ctrl->device, "keep-alive failed\n");
 595                 ctrl->ops->reset_ctrl(ctrl);
 596                 return;
 597         }
 598 }
 599
 600 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 601 {
 602         if (unlikely(ctrl->kato == 0))
 603                 return;
 604
 605         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
 606         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 607 }
 608 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
 609
 610 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 611 {
 612         if (unlikely(ctrl->kato == 0))
 613                 return;
 614
 615         cancel_delayed_work_sync(&ctrl->ka_work);
 616 }
 617 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 618
 619 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 620 {
 621         struct nvme_command c = { };
 622         int error;
 623
 624         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 625         c.identify.opcode = nvme_admin_identify;
 626         c.identify.cns = NVME_ID_CNS_CTRL;
 627
 628         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 629         if (!*id)
 630                 return -ENOMEM;
 631
 632         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 633                         sizeof(struct nvme_id_ctrl));
 634         if (error)
 635                 kfree(*id);
 636         return error;
 637 }
 638
 639 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 640 {
 641         struct nvme_command c = { };
 642
 643         c.identify.opcode = nvme_admin_identify;
 644         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
 645         c.identify.nsid = cpu_to_le32(nsid);
 646         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 647 }
 648
 649 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
 650                 struct nvme_id_ns **id)
 651 {
 652         struct nvme_command c = { };
 653         int error;
 654
 655         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 656         c.identify.opcode = nvme_admin_identify;
 657         c.identify.nsid = cpu_to_le32(nsid);
 658         c.identify.cns = NVME_ID_CNS_NS;
 659
 660         *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
 661         if (!*id)
 662                 return -ENOMEM;
 663
 664         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 665                         sizeof(struct nvme_id_ns));
 666         if (error)
 667                 kfree(*id);
 668         return error;
 669 }
 670
 671 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 672                       void *buffer, size_t buflen, u32 *result)
 673 {
 674         struct nvme_command c;
 675         union nvme_result res;
 676         int ret;
 677
 678         memset(&c, 0, sizeof(c));
 679         c.features.opcode = nvme_admin_get_features;
 680         c.features.nsid = cpu_to_le32(nsid);
 681         c.features.fid = cpu_to_le32(fid);
 682
 683         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, buffer, buflen, 0,
 684                         NVME_QID_ANY, 0, 0);
 685         if (ret >= 0 && result)
 686                 *result = le32_to_cpu(res.u32);
 687         return ret;
 688 }
 689
 690 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 691                       void *buffer, size_t buflen, u32 *result)
 692 {
 693         struct nvme_command c;
 694         union nvme_result res;
 695         int ret;
 696
 697         memset(&c, 0, sizeof(c));
 698         c.features.opcode = nvme_admin_set_features;
 699         c.features.fid = cpu_to_le32(fid);
 700         c.features.dword11 = cpu_to_le32(dword11);
 701
 702         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
 703                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
 704         if (ret >= 0 && result)
 705                 *result = le32_to_cpu(res.u32);
 706         return ret;
 707 }
 708
 709 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
 710 {
 711         struct nvme_command c = { };
 712         int error;
 713
 714         c.common.opcode = nvme_admin_get_log_page,
 715         c.common.nsid = cpu_to_le32(0xFFFFFFFF),
 716         c.common.cdw10[0] = cpu_to_le32(
 717                         (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
 718                          NVME_LOG_SMART),
 719
 720         *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
 721         if (!*log)
 722                 return -ENOMEM;
 723
 724         error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
 725                         sizeof(struct nvme_smart_log));
 726         if (error)
 727                 kfree(*log);
 728         return error;
 729 }
 730
 731 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
 732 {
 733         u32 q_count = (*count - 1) | ((*count - 1) << 16);
 734         u32 result;
 735         int status, nr_io_queues;
 736
 737         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
 738                         &result);
 739         if (status < 0)
 740                 return status;
 741
 742         /*
 743          * Degraded controllers might return an error when setting the queue
 744          * count.  We still want to be able to bring them online and offer
 745          * access to the admin queue, as that might be only way to fix them up.
 746          */
 747         if (status > 0) {
 748                 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
 749                 *count = 0;
 750         } else {
 751                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
 752                 *count = min(*count, nr_io_queues);
 753         }
 754
 755         return 0;
 756 }
 757 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
 758
 759 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 760 {
 761         struct nvme_user_io io;
 762         struct nvme_command c;
 763         unsigned length, meta_len;
 764         void __user *metadata;
 765
 766         if (copy_from_user(&io, uio, sizeof(io)))
 767                 return -EFAULT;
 768         if (io.flags)
 769                 return -EINVAL;
 770
 771         switch (io.opcode) {
 772         case nvme_cmd_write:
 773         case nvme_cmd_read:
 774         case nvme_cmd_compare:
 775                 break;
 776         default:
 777                 return -EINVAL;
 778         }
 779
 780         length = (io.nblocks + 1) << ns->lba_shift;
 781         meta_len = (io.nblocks + 1) * ns->ms;
 782         metadata = (void __user *)(uintptr_t)io.metadata;
 783
 784         if (ns->ext) {
 785                 length += meta_len;
 786                 meta_len = 0;
 787         } else if (meta_len) {
 788                 if ((io.metadata & 3) || !io.metadata)
 789                         return -EINVAL;
 790         }
 791
 792         memset(&c, 0, sizeof(c));
 793         c.rw.opcode = io.opcode;
 794         c.rw.flags = io.flags;
 795         c.rw.nsid = cpu_to_le32(ns->ns_id);
 796         c.rw.slba = cpu_to_le64(io.slba);
 797         c.rw.length = cpu_to_le16(io.nblocks);
 798         c.rw.control = cpu_to_le16(io.control);
 799         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
 800         c.rw.reftag = cpu_to_le32(io.reftag);
 801         c.rw.apptag = cpu_to_le16(io.apptag);
 802         c.rw.appmask = cpu_to_le16(io.appmask);
 803
 804         return __nvme_submit_user_cmd(ns->queue, &c,
 805                         (void __user *)(uintptr_t)io.addr, length,
 806                         metadata, meta_len, io.slba, NULL, 0);
 807 }
 808
 809 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 810                         struct nvme_passthru_cmd __user *ucmd)
 811 {
 812         struct nvme_passthru_cmd cmd;
 813         struct nvme_command c;
 814         unsigned timeout = 0;
 815         int status;
 816
 817         if (!capable(CAP_SYS_ADMIN))
 818                 return -EACCES;
 819         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
 820                 return -EFAULT;
 821         if (cmd.flags)
 822                 return -EINVAL;
 823
 824         memset(&c, 0, sizeof(c));
 825         c.common.opcode = cmd.opcode;
 826         c.common.flags = cmd.flags;
 827         c.common.nsid = cpu_to_le32(cmd.nsid);
 828         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
 829         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
 830         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
 831         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
 832         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
 833         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
 834         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
 835         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
 836
 837         if (cmd.timeout_ms)
 838                 timeout = msecs_to_jiffies(cmd.timeout_ms);
 839
 840         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
 841                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
 842                         &cmd.result, timeout);
 843         if (status >= 0) {
 844                 if (put_user(cmd.result, &ucmd->result))
 845                         return -EFAULT;
 846         }
 847
 848         return status;
 849 }
 850
 851 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 852                 unsigned int cmd, unsigned long arg)
 853 {
 854         struct nvme_ns *ns = bdev->bd_disk->private_data;
 855
 856         switch (cmd) {
 857         case NVME_IOCTL_ID:
 858                 force_successful_syscall_return();
 859                 return ns->ns_id;
 860         case NVME_IOCTL_ADMIN_CMD:
 861                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
 862         case NVME_IOCTL_IO_CMD:
 863                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
 864         case NVME_IOCTL_SUBMIT_IO:
 865                 return nvme_submit_io(ns, (void __user *)arg);
 866 #ifdef CONFIG_BLK_DEV_NVME_SCSI
 867         case SG_GET_VERSION_NUM:
 868                 return nvme_sg_get_version_num((void __user *)arg);
 869         case SG_IO:
 870                 return nvme_sg_io(ns, (void __user *)arg);
 871 #endif
 872         default:
 873 #ifdef CONFIG_NVM
 874                 if (ns->ndev)
 875                         return nvme_nvm_ioctl(ns, cmd, arg);
 876 #endif
 877                 if (is_sed_ioctl(cmd))
 878                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
 879                                          (void __user *) arg);
 880                 return -ENOTTY;
 881         }
 882 }
 883
 884 #ifdef CONFIG_COMPAT
 885 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 886                         unsigned int cmd, unsigned long arg)
 887 {
 888         switch (cmd) {
 889         case SG_IO:
 890                 return -ENOIOCTLCMD;
 891         }
 892         return nvme_ioctl(bdev, mode, cmd, arg);
 893 }
 894 #else
 895 #define nvme_compat_ioctl       NULL
 896 #endif
 897
 898 static int nvme_open(struct block_device *bdev, fmode_t mode)
 899 {
 900         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
 901 }
 902
 903 static void nvme_release(struct gendisk *disk, fmode_t mode)
 904 {
 905         struct nvme_ns *ns = disk->private_data;
 906
 907         module_put(ns->ctrl->ops->module);
 908         nvme_put_ns(ns);
 909 }
 910
 911 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 912 {
 913         /* some standard values */
 914         geo->heads = 1 << 6;
 915         geo->sectors = 1 << 5;
 916         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
 917         return 0;
 918 }
 919
 920 #ifdef CONFIG_BLK_DEV_INTEGRITY
 921 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
 922                 u16 bs)
 923 {
 924         struct nvme_ns *ns = disk->private_data;
 925         u16 old_ms = ns->ms;
 926         u8 pi_type = 0;
 927
 928         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
 929         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
 930
 931         /* PI implementation requires metadata equal t10 pi tuple size */
 932         if (ns->ms == sizeof(struct t10_pi_tuple))
 933                 pi_type = id->dps & NVME_NS_DPS_PI_MASK;
 934
 935         if (blk_get_integrity(disk) &&
 936             (ns->pi_type != pi_type || ns->ms != old_ms ||
 937              bs != queue_logical_block_size(disk->queue) ||
 938              (ns->ms && ns->ext)))
 939                 blk_integrity_unregister(disk);
 940
 941         ns->pi_type = pi_type;
 942 }
 943
 944 static void nvme_init_integrity(struct nvme_ns *ns)
 945 {
 946         struct blk_integrity integrity;
 947
 948         memset(&integrity, 0, sizeof(integrity));
 949         switch (ns->pi_type) {
 950         case NVME_NS_DPS_PI_TYPE3:
 951                 integrity.profile = &t10_pi_type3_crc;
 952                 integrity.tag_size = sizeof(u16) + sizeof(u32);
 953                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
 954                 break;
 955         case NVME_NS_DPS_PI_TYPE1:
 956         case NVME_NS_DPS_PI_TYPE2:
 957                 integrity.profile = &t10_pi_type1_crc;
 958                 integrity.tag_size = sizeof(u16);
 959                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
 960                 break;
 961         default:
 962                 integrity.profile = NULL;
 963                 break;
 964         }
 965         integrity.tuple_size = ns->ms;
 966         blk_integrity_register(ns->disk, &integrity);
 967         blk_queue_max_integrity_segments(ns->queue, 1);
 968 }
 969 #else
 970 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
 971                 u16 bs)
 972 {
 973 }
 974 static void nvme_init_integrity(struct nvme_ns *ns)
 975 {
 976 }
 977 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 978
 979 static void nvme_config_discard(struct nvme_ns *ns)
 980 {
 981         struct nvme_ctrl *ctrl = ns->ctrl;
 982         u32 logical_block_size = queue_logical_block_size(ns->queue);
 983
 984         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
 985                         NVME_DSM_MAX_RANGES);
 986
 987         ns->queue->limits.discard_alignment = logical_block_size;
 988         ns->queue->limits.discard_granularity = logical_block_size;
 989         blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
 990         blk_queue_max_discard_segments(ns->queue, NVME_DSM_MAX_RANGES);
 991         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
 992
 993         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 994                 blk_queue_max_write_zeroes_sectors(ns->queue, UINT_MAX);
 995 }
 996
 997 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
 998 {
 999         if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
1000                 dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
1001                 return -ENODEV;
1002         }
1003
1004         if ((*id)->ncap == 0) {
1005                 kfree(*id);
1006                 return -ENODEV;
1007         }
1008
1009         if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
1010                 memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
1011         if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
1012                 memcpy(ns->uuid, (*id)->nguid, sizeof(ns->uuid));
1013
1014         return 0;
1015 }
1016
1017 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1018 {
1019         struct nvme_ns *ns = disk->private_data;
1020         u16 bs;
1021
1022         /*
1023          * If identify namespace failed, use default 512 byte block size so
1024          * block layer can use before failing read/write for 0 capacity.
1025          */
1026         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1027         if (ns->lba_shift == 0)
1028                 ns->lba_shift = 9;
1029         bs = 1 << ns->lba_shift;
1030
1031         blk_mq_freeze_queue(disk->queue);
1032
1033         if (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED)
1034                 nvme_prep_integrity(disk, id, bs);
1035         blk_queue_logical_block_size(ns->queue, bs);
1036         if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
1037                 nvme_init_integrity(ns);
1038         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
1039                 set_capacity(disk, 0);
1040         else
1041                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
1042
1043         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1044                 nvme_config_discard(ns);
1045         blk_mq_unfreeze_queue(disk->queue);
1046 }
1047
1048 static int nvme_revalidate_disk(struct gendisk *disk)
1049 {
1050         struct nvme_ns *ns = disk->private_data;
1051         struct nvme_id_ns *id = NULL;
1052         int ret;
1053
1054         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1055                 set_capacity(disk, 0);
1056                 return -ENODEV;
1057         }
1058
1059         ret = nvme_revalidate_ns(ns, &id);
1060         if (ret)
1061                 return ret;
1062
1063         __nvme_revalidate_disk(disk, id);
1064         kfree(id);
1065
1066         return 0;
1067 }
1068
1069 static char nvme_pr_type(enum pr_type type)
1070 {
1071         switch (type) {
1072         case PR_WRITE_EXCLUSIVE:
1073                 return 1;
1074         case PR_EXCLUSIVE_ACCESS:
1075                 return 2;
1076         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1077                 return 3;
1078         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1079                 return 4;
1080         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1081                 return 5;
1082         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1083                 return 6;
1084         default:
1085                 return 0;
1086         }
1087 };
1088
1089 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1090                                 u64 key, u64 sa_key, u8 op)
1091 {
1092         struct nvme_ns *ns = bdev->bd_disk->private_data;
1093         struct nvme_command c;
1094         u8 data[16] = { 0, };
1095
1096         put_unaligned_le64(key, &data[0]);
1097         put_unaligned_le64(sa_key, &data[8]);
1098
1099         memset(&c, 0, sizeof(c));
1100         c.common.opcode = op;
1101         c.common.nsid = cpu_to_le32(ns->ns_id);
1102         c.common.cdw10[0] = cpu_to_le32(cdw10);
1103
1104         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1105 }
1106
1107 static int nvme_pr_register(struct block_device *bdev, u64 old,
1108                 u64 new, unsigned flags)
1109 {
1110         u32 cdw10;
1111
1112         if (flags & ~PR_FL_IGNORE_KEY)
1113                 return -EOPNOTSUPP;
1114
1115         cdw10 = old ? 2 : 0;
1116         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1117         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1118         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1119 }
1120
1121 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1122                 enum pr_type type, unsigned flags)
1123 {
1124         u32 cdw10;
1125
1126         if (flags & ~PR_FL_IGNORE_KEY)
1127                 return -EOPNOTSUPP;
1128
1129         cdw10 = nvme_pr_type(type) << 8;
1130         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1131         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1132 }
1133
1134 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1135                 enum pr_type type, bool abort)
1136 {
1137         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1138         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1139 }
1140
1141 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1142 {
1143         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1144         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1145 }
1146
1147 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1148 {
1149         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1150         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1151 }
1152
1153 static const struct pr_ops nvme_pr_ops = {
1154         .pr_register    = nvme_pr_register,
1155         .pr_reserve     = nvme_pr_reserve,
1156         .pr_release     = nvme_pr_release,
1157         .pr_preempt     = nvme_pr_preempt,
1158         .pr_clear       = nvme_pr_clear,
1159 };
1160
1161 #ifdef CONFIG_BLK_SED_OPAL
1162 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1163                 bool send)
1164 {
1165         struct nvme_ctrl *ctrl = data;
1166         struct nvme_command cmd;
1167
1168         memset(&cmd, 0, sizeof(cmd));
1169         if (send)
1170                 cmd.common.opcode = nvme_admin_security_send;
1171         else
1172                 cmd.common.opcode = nvme_admin_security_recv;
1173         cmd.common.nsid = 0;
1174         cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1175         cmd.common.cdw10[1] = cpu_to_le32(len);
1176
1177         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1178                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1179 }
1180 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1181 #endif /* CONFIG_BLK_SED_OPAL */
1182
1183 static const struct block_device_operations nvme_fops = {
1184         .owner          = THIS_MODULE,
1185         .ioctl          = nvme_ioctl,
1186         .compat_ioctl   = nvme_compat_ioctl,
1187         .open           = nvme_open,
1188         .release        = nvme_release,
1189         .getgeo         = nvme_getgeo,
1190         .revalidate_disk= nvme_revalidate_disk,
1191         .pr_ops         = &nvme_pr_ops,
1192 };
1193
1194 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1195 {
1196         unsigned long timeout =
1197                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1198         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1199         int ret;
1200
1201         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1202                 if (csts == ~0)
1203                         return -ENODEV;
1204                 if ((csts & NVME_CSTS_RDY) == bit)
1205                         break;
1206
1207                 msleep(100);
1208                 if (fatal_signal_pending(current))
1209                         return -EINTR;
1210                 if (time_after(jiffies, timeout)) {
1211                         dev_err(ctrl->device,
1212                                 "Device not ready; aborting %s\n", enabled ?
1213                                                 "initialisation" : "reset");
1214                         return -ENODEV;
1215                 }
1216         }
1217
1218         return ret;
1219 }
1220
1221 /*
1222  * If the device has been passed off to us in an enabled state, just clear
1223  * the enabled bit.  The spec says we should set the 'shutdown notification
1224  * bits', but doing so may cause the device to complete commands to the
1225  * admin queue ... and we don't know what memory that might be pointing at!
1226  */
1227 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1228 {
1229         int ret;
1230
1231         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1232         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1233
1234         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1235         if (ret)
1236                 return ret;
1237
1238         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1239                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1240
1241         return nvme_wait_ready(ctrl, cap, false);
1242 }
1243 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1244
1245 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1246 {
1247         /*
1248          * Default to a 4K page size, with the intention to update this
1249          * path in the future to accomodate architectures with differing
1250          * kernel and IO page sizes.
1251          */
1252         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1253         int ret;
1254
1255         if (page_shift < dev_page_min) {
1256                 dev_err(ctrl->device,
1257                         "Minimum device page size %u too large for host (%u)\n",
1258                         1 << dev_page_min, 1 << page_shift);
1259                 return -ENODEV;
1260         }
1261
1262         ctrl->page_size = 1 << page_shift;
1263
1264         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1265         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1266         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1267         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1268         ctrl->ctrl_config |= NVME_CC_ENABLE;
1269
1270         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1271         if (ret)
1272                 return ret;
1273         return nvme_wait_ready(ctrl, cap, true);
1274 }
1275 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1276
1277 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1278 {
1279         unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1280         u32 csts;
1281         int ret;
1282
1283         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1284         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1285
1286         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1287         if (ret)
1288                 return ret;
1289
1290         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1291                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1292                         break;
1293
1294                 msleep(100);
1295                 if (fatal_signal_pending(current))
1296                         return -EINTR;
1297                 if (time_after(jiffies, timeout)) {
1298                         dev_err(ctrl->device,
1299                                 "Device shutdown incomplete; abort shutdown\n");
1300                         return -ENODEV;
1301                 }
1302         }
1303
1304         return ret;
1305 }
1306 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1307
1308 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1309                 struct request_queue *q)
1310 {
1311         bool vwc = false;
1312
1313         if (ctrl->max_hw_sectors) {
1314                 u32 max_segments =
1315                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1316
1317                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1318                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1319         }
1320         if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1321                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1322         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1323         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1324                 vwc = true;
1325         blk_queue_write_cache(q, vwc, vwc);
1326 }
1327
1328 static void nvme_configure_apst(struct nvme_ctrl *ctrl)
1329 {
1330         /*
1331          * APST (Autonomous Power State Transition) lets us program a
1332          * table of power state transitions that the controller will
1333          * perform automatically.  We configure it with a simple
1334          * heuristic: we are willing to spend at most 2% of the time
1335          * transitioning between power states.  Therefore, when running
1336          * in any given state, we will enter the next lower-power
1337          * non-operational state after waiting 50 * (enlat + exlat)
1338          * microseconds, as long as that state's total latency is under
1339          * the requested maximum latency.
1340          *
1341          * We will not autonomously enter any non-operational state for
1342          * which the total latency exceeds ps_max_latency_us.  Users
1343          * can set ps_max_latency_us to zero to turn off APST.
1344          */
1345
1346         unsigned apste;
1347         struct nvme_feat_auto_pst *table;
1348         u64 max_lat_us = 0;
1349         int max_ps = -1;
1350         int ret;
1351
1352         /*
1353          * If APST isn't supported or if we haven't been initialized yet,
1354          * then don't do anything.
1355          */
1356         if (!ctrl->apsta)
1357                 return;
1358
1359         if (ctrl->npss > 31) {
1360                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1361                 return;
1362         }
1363
1364         table = kzalloc(sizeof(*table), GFP_KERNEL);
1365         if (!table)
1366                 return;
1367
1368         if (ctrl->ps_max_latency_us == 0) {
1369                 /* Turn off APST. */
1370                 apste = 0;
1371                 dev_dbg(ctrl->device, "APST disabled\n");
1372         } else {
1373                 __le64 target = cpu_to_le64(0);
1374                 int state;
1375
1376                 /*
1377                  * Walk through all states from lowest- to highest-power.
1378                  * According to the spec, lower-numbered states use more
1379                  * power.  NPSS, despite the name, is the index of the
1380                  * lowest-power state, not the number of states.
1381                  */
1382                 for (state = (int)ctrl->npss; state >= 0; state--) {
1383                         u64 total_latency_us, transition_ms;
1384
1385                         if (target)
1386                                 table->entries[state] = target;
1387
1388                         /*
1389                          * Don't allow transitions to the deepest state
1390                          * if it's quirked off.
1391                          */
1392                         if (state == ctrl->npss &&
1393                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1394                                 continue;
1395
1396                         /*
1397                          * Is this state a useful non-operational state for
1398                          * higher-power states to autonomously transition to?
1399                          */
1400                         if (!(ctrl->psd[state].flags &
1401                               NVME_PS_FLAGS_NON_OP_STATE))
1402                                 continue;
1403
1404                         total_latency_us =
1405                                 (u64)le32_to_cpu(ctrl->psd[state].entry_lat) +
1406                                 + le32_to_cpu(ctrl->psd[state].exit_lat);
1407                         if (total_latency_us > ctrl->ps_max_latency_us)
1408                                 continue;
1409
1410                         /*
1411                          * This state is good.  Use it as the APST idle
1412                          * target for higher power states.
1413                          */
1414                         transition_ms = total_latency_us + 19;
1415                         do_div(transition_ms, 20);
1416                         if (transition_ms > (1 << 24) - 1)
1417                                 transition_ms = (1 << 24) - 1;
1418
1419                         target = cpu_to_le64((state << 3) |
1420                                              (transition_ms << 8));
1421
1422                         if (max_ps == -1)
1423                                 max_ps = state;
1424
1425                         if (total_latency_us > max_lat_us)
1426                                 max_lat_us = total_latency_us;
1427                 }
1428
1429                 apste = 1;
1430
1431                 if (max_ps == -1) {
1432                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1433                 } else {
1434                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1435                                 max_ps, max_lat_us, (int)sizeof(*table), table);
1436                 }
1437         }
1438
1439         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1440                                 table, sizeof(*table), NULL);
1441         if (ret)
1442                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1443
1444         kfree(table);
1445 }
1446
1447 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1448 {
1449         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1450         u64 latency;
1451
1452         switch (val) {
1453         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1454         case PM_QOS_LATENCY_ANY:
1455                 latency = U64_MAX;
1456                 break;
1457
1458         default:
1459                 latency = val;
1460         }
1461
1462         if (ctrl->ps_max_latency_us != latency) {
1463                 ctrl->ps_max_latency_us = latency;
1464                 nvme_configure_apst(ctrl);
1465         }
1466 }
1467
1468 struct nvme_core_quirk_entry {
1469         /*
1470          * NVMe model and firmware strings are padded with spaces.  For
1471          * simplicity, strings in the quirk table are padded with NULLs
1472          * instead.
1473          */
1474         u16 vid;
1475         const char *mn;
1476         const char *fr;
1477         unsigned long quirks;
1478 };
1479
1480 static const struct nvme_core_quirk_entry core_quirks[] = {
1481         {
1482                 /*
1483                  * This Toshiba device seems to die using any APST states.  See:
1484                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1485                  */
1486                 .vid = 0x1179,
1487                 .mn = "THNSF5256GPUK TOSHIBA",
1488                 .quirks = NVME_QUIRK_NO_APST,
1489         }
1490 };
1491
1492 /* match is null-terminated but idstr is space-padded. */
1493 static bool string_matches(const char *idstr, const char *match, size_t len)
1494 {
1495         size_t matchlen;
1496
1497         if (!match)
1498                 return true;
1499
1500         matchlen = strlen(match);
1501         WARN_ON_ONCE(matchlen > len);
1502
1503         if (memcmp(idstr, match, matchlen))
1504                 return false;
1505
1506         for (; matchlen < len; matchlen++)
1507                 if (idstr[matchlen] != ' ')
1508                         return false;
1509
1510         return true;
1511 }
1512
1513 static bool quirk_matches(const struct nvme_id_ctrl *id,
1514                           const struct nvme_core_quirk_entry *q)
1515 {
1516         return q->vid == le16_to_cpu(id->vid) &&
1517                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1518                 string_matches(id->fr, q->fr, sizeof(id->fr));
1519 }
1520
1521 /*
1522  * Initialize the cached copies of the Identify data and various controller
1523  * register in our nvme_ctrl structure.  This should be called as soon as
1524  * the admin queue is fully up and running.
1525  */
1526 int nvme_init_identify(struct nvme_ctrl *ctrl)
1527 {
1528         struct nvme_id_ctrl *id;
1529         u64 cap;
1530         int ret, page_shift;
1531         u32 max_hw_sectors;
1532         u8 prev_apsta;
1533
1534         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1535         if (ret) {
1536                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1537                 return ret;
1538         }
1539
1540         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1541         if (ret) {
1542                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1543                 return ret;
1544         }
1545         page_shift = NVME_CAP_MPSMIN(cap) + 12;
1546
1547         if (ctrl->vs >= NVME_VS(1, 1, 0))
1548                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1549
1550         ret = nvme_identify_ctrl(ctrl, &id);
1551         if (ret) {
1552                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1553                 return -EIO;
1554         }
1555
1556         if (!ctrl->identified) {
1557                 /*
1558                  * Check for quirks.  Quirk can depend on firmware version,
1559                  * so, in principle, the set of quirks present can change
1560                  * across a reset.  As a possible future enhancement, we
1561                  * could re-scan for quirks every time we reinitialize
1562                  * the device, but we'd have to make sure that the driver
1563                  * behaves intelligently if the quirks change.
1564                  */
1565
1566                 int i;
1567
1568                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
1569                         if (quirk_matches(id, &core_quirks[i]))
1570                                 ctrl->quirks |= core_quirks[i].quirks;
1571                 }
1572         }
1573
1574         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
1575                 dev_warn(ctrl->dev, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
1576                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
1577         }
1578
1579         ctrl->oacs = le16_to_cpu(id->oacs);
1580         ctrl->vid = le16_to_cpu(id->vid);
1581         ctrl->oncs = le16_to_cpup(&id->oncs);
1582         atomic_set(&ctrl->abort_limit, id->acl + 1);
1583         ctrl->vwc = id->vwc;
1584         ctrl->cntlid = le16_to_cpup(&id->cntlid);
1585         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1586         memcpy(ctrl->model, id->mn, sizeof(id->mn));
1587         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1588         if (id->mdts)
1589                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1590         else
1591                 max_hw_sectors = UINT_MAX;
1592         ctrl->max_hw_sectors =
1593                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1594
1595         nvme_set_queue_limits(ctrl, ctrl->admin_q);
1596         ctrl->sgls = le32_to_cpu(id->sgls);
1597         ctrl->kas = le16_to_cpu(id->kas);
1598
1599         ctrl->npss = id->npss;
1600         prev_apsta = ctrl->apsta;
1601         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
1602                 if (force_apst && id->apsta) {
1603                         dev_warn(ctrl->dev, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
1604                         ctrl->apsta = 1;
1605                 } else {
1606                         ctrl->apsta = 0;
1607                 }
1608         } else {
1609                 ctrl->apsta = id->apsta;
1610         }
1611         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
1612
1613         if (ctrl->ops->flags & NVME_F_FABRICS) {
1614                 ctrl->icdoff = le16_to_cpu(id->icdoff);
1615                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1616                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1617                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1618
1619                 /*
1620                  * In fabrics we need to verify the cntlid matches the
1621                  * admin connect
1622                  */
1623                 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1624                         ret = -EINVAL;
1625
1626                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1627                         dev_err(ctrl->dev,
1628                                 "keep-alive support is mandatory for fabrics\n");
1629                         ret = -EINVAL;
1630                 }
1631         } else {
1632                 ctrl->cntlid = le16_to_cpu(id->cntlid);
1633         }
1634
1635         kfree(id);
1636
1637         if (ctrl->apsta && !prev_apsta)
1638                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
1639         else if (!ctrl->apsta && prev_apsta)
1640                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
1641
1642         nvme_configure_apst(ctrl);
1643
1644         ctrl->identified = true;
1645
1646         return ret;
1647 }
1648 EXPORT_SYMBOL_GPL(nvme_init_identify);
1649
1650 static int nvme_dev_open(struct inode *inode, struct file *file)
1651 {
1652         struct nvme_ctrl *ctrl;
1653         int instance = iminor(inode);
1654         int ret = -ENODEV;
1655
1656         spin_lock(&dev_list_lock);
1657         list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1658                 if (ctrl->instance != instance)
1659                         continue;
1660
1661                 if (!ctrl->admin_q) {
1662                         ret = -EWOULDBLOCK;
1663                         break;
1664                 }
1665                 if (!kref_get_unless_zero(&ctrl->kref))
1666                         break;
1667                 file->private_data = ctrl;
1668                 ret = 0;
1669                 break;
1670         }
1671         spin_unlock(&dev_list_lock);
1672
1673         return ret;
1674 }
1675
1676 static int nvme_dev_release(struct inode *inode, struct file *file)
1677 {
1678         nvme_put_ctrl(file->private_data);
1679         return 0;
1680 }
1681
1682 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1683 {
1684         struct nvme_ns *ns;
1685         int ret;
1686
1687         mutex_lock(&ctrl->namespaces_mutex);
1688         if (list_empty(&ctrl->namespaces)) {
1689                 ret = -ENOTTY;
1690                 goto out_unlock;
1691         }
1692
1693         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1694         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1695                 dev_warn(ctrl->device,
1696                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1697                 ret = -EINVAL;
1698                 goto out_unlock;
1699         }
1700
1701         dev_warn(ctrl->device,
1702                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1703         kref_get(&ns->kref);
1704         mutex_unlock(&ctrl->namespaces_mutex);
1705
1706         ret = nvme_user_cmd(ctrl, ns, argp);
1707         nvme_put_ns(ns);
1708         return ret;
1709
1710 out_unlock:
1711         mutex_unlock(&ctrl->namespaces_mutex);
1712         return ret;
1713 }
1714
1715 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1716                 unsigned long arg)
1717 {
1718         struct nvme_ctrl *ctrl = file->private_data;
1719         void __user *argp = (void __user *)arg;
1720
1721         switch (cmd) {
1722         case NVME_IOCTL_ADMIN_CMD:
1723                 return nvme_user_cmd(ctrl, NULL, argp);
1724         case NVME_IOCTL_IO_CMD:
1725                 return nvme_dev_user_cmd(ctrl, argp);
1726         case NVME_IOCTL_RESET:
1727                 dev_warn(ctrl->device, "resetting controller\n");
1728                 return ctrl->ops->reset_ctrl(ctrl);
1729         case NVME_IOCTL_SUBSYS_RESET:
1730                 return nvme_reset_subsystem(ctrl);
1731         case NVME_IOCTL_RESCAN:
1732                 nvme_queue_scan(ctrl);
1733                 return 0;
1734         default:
1735                 return -ENOTTY;
1736         }
1737 }
1738
1739 static const struct file_operations nvme_dev_fops = {
1740         .owner          = THIS_MODULE,
1741         .open           = nvme_dev_open,
1742         .release        = nvme_dev_release,
1743         .unlocked_ioctl = nvme_dev_ioctl,
1744         .compat_ioctl   = nvme_dev_ioctl,
1745 };
1746
1747 static ssize_t nvme_sysfs_reset(struct device *dev,
1748                                 struct device_attribute *attr, const char *buf,
1749                                 size_t count)
1750 {
1751         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1752         int ret;
1753
1754         ret = ctrl->ops->reset_ctrl(ctrl);
1755         if (ret < 0)
1756                 return ret;
1757         return count;
1758 }
1759 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1760
1761 static ssize_t nvme_sysfs_rescan(struct device *dev,
1762                                 struct device_attribute *attr, const char *buf,
1763                                 size_t count)
1764 {
1765         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1766
1767         nvme_queue_scan(ctrl);
1768         return count;
1769 }
1770 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1771
1772 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1773                                                                 char *buf)
1774 {
1775         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1776         struct nvme_ctrl *ctrl = ns->ctrl;
1777         int serial_len = sizeof(ctrl->serial);
1778         int model_len = sizeof(ctrl->model);
1779
1780         if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1781                 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1782
1783         if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1784                 return sprintf(buf, "eui.%8phN\n", ns->eui);
1785
1786         while (ctrl->serial[serial_len - 1] == ' ')
1787                 serial_len--;
1788         while (ctrl->model[model_len - 1] == ' ')
1789                 model_len--;
1790
1791         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1792                 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1793 }
1794 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1795
1796 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1797                                                                 char *buf)
1798 {
1799         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1800         return sprintf(buf, "%pU\n", ns->uuid);
1801 }
1802 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1803
1804 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1805                                                                 char *buf)
1806 {
1807         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1808         return sprintf(buf, "%8phd\n", ns->eui);
1809 }
1810 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1811
1812 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1813                                                                 char *buf)
1814 {
1815         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1816         return sprintf(buf, "%d\n", ns->ns_id);
1817 }
1818 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1819
1820 static struct attribute *nvme_ns_attrs[] = {
1821         &dev_attr_wwid.attr,
1822         &dev_attr_uuid.attr,
1823         &dev_attr_eui.attr,
1824         &dev_attr_nsid.attr,
1825         NULL,
1826 };
1827
1828 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1829                 struct attribute *a, int n)
1830 {
1831         struct device *dev = container_of(kobj, struct device, kobj);
1832         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1833
1834         if (a == &dev_attr_uuid.attr) {
1835                 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1836                         return 0;
1837         }
1838         if (a == &dev_attr_eui.attr) {
1839                 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1840                         return 0;
1841         }
1842         return a->mode;
1843 }
1844
1845 static const struct attribute_group nvme_ns_attr_group = {
1846         .attrs          = nvme_ns_attrs,
1847         .is_visible     = nvme_ns_attrs_are_visible,
1848 };
1849
1850 #define nvme_show_str_function(field)                                           \
1851 static ssize_t  field##_show(struct device *dev,                                \
1852                             struct device_attribute *attr, char *buf)           \
1853 {                                                                               \
1854         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1855         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);   \
1856 }                                                                               \
1857 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1858
1859 #define nvme_show_int_function(field)                                           \
1860 static ssize_t  field##_show(struct device *dev,                                \
1861                             struct device_attribute *attr, char *buf)           \
1862 {                                                                               \
1863         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1864         return sprintf(buf, "%d\n", ctrl->field);       \
1865 }                                                                               \
1866 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1867
1868 nvme_show_str_function(model);
1869 nvme_show_str_function(serial);
1870 nvme_show_str_function(firmware_rev);
1871 nvme_show_int_function(cntlid);
1872
1873 static ssize_t nvme_sysfs_delete(struct device *dev,
1874                                 struct device_attribute *attr, const char *buf,
1875                                 size_t count)
1876 {
1877         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1878
1879         if (device_remove_file_self(dev, attr))
1880                 ctrl->ops->delete_ctrl(ctrl);
1881         return count;
1882 }
1883 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1884
1885 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1886                                          struct device_attribute *attr,
1887                                          char *buf)
1888 {
1889         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1890
1891         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1892 }
1893 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1894
1895 static ssize_t nvme_sysfs_show_state(struct device *dev,
1896                                      struct device_attribute *attr,
1897                                      char *buf)
1898 {
1899         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1900         static const char *const state_name[] = {
1901                 [NVME_CTRL_NEW]         = "new",
1902                 [NVME_CTRL_LIVE]        = "live",
1903                 [NVME_CTRL_RESETTING]   = "resetting",
1904                 [NVME_CTRL_RECONNECTING]= "reconnecting",
1905                 [NVME_CTRL_DELETING]    = "deleting",
1906                 [NVME_CTRL_DEAD]        = "dead",
1907         };
1908
1909         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
1910             state_name[ctrl->state])
1911                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
1912
1913         return sprintf(buf, "unknown state\n");
1914 }
1915
1916 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
1917
1918 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1919                                          struct device_attribute *attr,
1920                                          char *buf)
1921 {
1922         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1923
1924         return snprintf(buf, PAGE_SIZE, "%s\n",
1925                         ctrl->ops->get_subsysnqn(ctrl));
1926 }
1927 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1928
1929 static ssize_t nvme_sysfs_show_address(struct device *dev,
1930                                          struct device_attribute *attr,
1931                                          char *buf)
1932 {
1933         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1934
1935         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1936 }
1937 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1938
1939 static struct attribute *nvme_dev_attrs[] = {
1940         &dev_attr_reset_controller.attr,
1941         &dev_attr_rescan_controller.attr,
1942         &dev_attr_model.attr,
1943         &dev_attr_serial.attr,
1944         &dev_attr_firmware_rev.attr,
1945         &dev_attr_cntlid.attr,
1946         &dev_attr_delete_controller.attr,
1947         &dev_attr_transport.attr,
1948         &dev_attr_subsysnqn.attr,
1949         &dev_attr_address.attr,
1950         &dev_attr_state.attr,
1951         NULL
1952 };
1953
1954 #define CHECK_ATTR(ctrl, a, name)               \
1955         if ((a) == &dev_attr_##name.attr &&     \
1956             !(ctrl)->ops->get_##name)           \
1957                 return 0
1958
1959 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1960                 struct attribute *a, int n)
1961 {
1962         struct device *dev = container_of(kobj, struct device, kobj);
1963         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1964
1965         if (a == &dev_attr_delete_controller.attr) {
1966                 if (!ctrl->ops->delete_ctrl)
1967                         return 0;
1968         }
1969
1970         CHECK_ATTR(ctrl, a, subsysnqn);
1971         CHECK_ATTR(ctrl, a, address);
1972
1973         return a->mode;
1974 }
1975
1976 static struct attribute_group nvme_dev_attrs_group = {
1977         .attrs          = nvme_dev_attrs,
1978         .is_visible     = nvme_dev_attrs_are_visible,
1979 };
1980
1981 static const struct attribute_group *nvme_dev_attr_groups[] = {
1982         &nvme_dev_attrs_group,
1983         NULL,
1984 };
1985
1986 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1987 {
1988         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1989         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1990
1991         return nsa->ns_id - nsb->ns_id;
1992 }
1993
1994 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1995 {
1996         struct nvme_ns *ns, *ret = NULL;
1997
1998         mutex_lock(&ctrl->namespaces_mutex);
1999         list_for_each_entry(ns, &ctrl->namespaces, list) {
2000                 if (ns->ns_id == nsid) {
2001                         kref_get(&ns->kref);
2002                         ret = ns;
2003                         break;
2004                 }
2005                 if (ns->ns_id > nsid)
2006                         break;
2007         }
2008         mutex_unlock(&ctrl->namespaces_mutex);
2009         return ret;
2010 }
2011
2012 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2013 {
2014         struct nvme_ns *ns;
2015         struct gendisk *disk;
2016         struct nvme_id_ns *id;
2017         char disk_name[DISK_NAME_LEN];
2018         int node = dev_to_node(ctrl->dev);
2019
2020         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2021         if (!ns)
2022                 return;
2023
2024         ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
2025         if (ns->instance < 0)
2026                 goto out_free_ns;
2027
2028         ns->queue = blk_mq_init_queue(ctrl->tagset);
2029         if (IS_ERR(ns->queue))
2030                 goto out_release_instance;
2031         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2032         ns->queue->queuedata = ns;
2033         ns->ctrl = ctrl;
2034
2035         kref_init(&ns->kref);
2036         ns->ns_id = nsid;
2037         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2038
2039         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2040         nvme_set_queue_limits(ctrl, ns->queue);
2041
2042         sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
2043
2044         if (nvme_revalidate_ns(ns, &id))
2045                 goto out_free_queue;
2046
2047         if (nvme_nvm_ns_supported(ns, id) &&
2048                                 nvme_nvm_register(ns, disk_name, node)) {
2049                 dev_warn(ctrl->dev, "%s: LightNVM init failure\n", __func__);
2050                 goto out_free_id;
2051         }
2052
2053         disk = alloc_disk_node(0, node);
2054         if (!disk)
2055                 goto out_free_id;
2056
2057         disk->fops = &nvme_fops;
2058         disk->private_data = ns;
2059         disk->queue = ns->queue;
2060         disk->flags = GENHD_FL_EXT_DEVT;
2061         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
2062         ns->disk = disk;
2063
2064         __nvme_revalidate_disk(disk, id);
2065
2066         mutex_lock(&ctrl->namespaces_mutex);
2067         list_add_tail(&ns->list, &ctrl->namespaces);
2068         mutex_unlock(&ctrl->namespaces_mutex);
2069
2070         kref_get(&ctrl->kref);
2071
2072         kfree(id);
2073
2074         device_add_disk(ctrl->device, ns->disk);
2075         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
2076                                         &nvme_ns_attr_group))
2077                 pr_warn("%s: failed to create sysfs group for identification\n",
2078                         ns->disk->disk_name);
2079         if (ns->ndev && nvme_nvm_register_sysfs(ns))
2080                 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
2081                         ns->disk->disk_name);
2082         return;
2083  out_free_id:
2084         kfree(id);
2085  out_free_queue:
2086         blk_cleanup_queue(ns->queue);
2087  out_release_instance:
2088         ida_simple_remove(&ctrl->ns_ida, ns->instance);
2089  out_free_ns:
2090         kfree(ns);
2091 }
2092
2093 static void nvme_ns_remove(struct nvme_ns *ns)
2094 {
2095         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
2096                 return;
2097
2098         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
2099                 if (blk_get_integrity(ns->disk))
2100                         blk_integrity_unregister(ns->disk);
2101                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
2102                                         &nvme_ns_attr_group);
2103                 if (ns->ndev)
2104                         nvme_nvm_unregister_sysfs(ns);
2105                 del_gendisk(ns->disk);
2106                 blk_cleanup_queue(ns->queue);
2107         }
2108
2109         mutex_lock(&ns->ctrl->namespaces_mutex);
2110         list_del_init(&ns->list);
2111         mutex_unlock(&ns->ctrl->namespaces_mutex);
2112
2113         nvme_put_ns(ns);
2114 }
2115
2116 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2117 {
2118         struct nvme_ns *ns;
2119
2120         ns = nvme_find_get_ns(ctrl, nsid);
2121         if (ns) {
2122                 if (ns->disk && revalidate_disk(ns->disk))
2123                         nvme_ns_remove(ns);
2124                 nvme_put_ns(ns);
2125         } else
2126                 nvme_alloc_ns(ctrl, nsid);
2127 }
2128
2129 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
2130                                         unsigned nsid)
2131 {
2132         struct nvme_ns *ns, *next;
2133
2134         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
2135                 if (ns->ns_id > nsid)
2136                         nvme_ns_remove(ns);
2137         }
2138 }
2139
2140 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
2141 {
2142         struct nvme_ns *ns;
2143         __le32 *ns_list;
2144         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
2145         int ret = 0;
2146
2147         ns_list = kzalloc(0x1000, GFP_KERNEL);
2148         if (!ns_list)
2149                 return -ENOMEM;
2150
2151         for (i = 0; i < num_lists; i++) {
2152                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
2153                 if (ret)
2154                         goto free;
2155
2156                 for (j = 0; j < min(nn, 1024U); j++) {
2157                         nsid = le32_to_cpu(ns_list[j]);
2158                         if (!nsid)
2159                                 goto out;
2160
2161                         nvme_validate_ns(ctrl, nsid);
2162
2163                         while (++prev < nsid) {
2164                                 ns = nvme_find_get_ns(ctrl, prev);
2165                                 if (ns) {
2166                                         nvme_ns_remove(ns);
2167                                         nvme_put_ns(ns);
2168                                 }
2169                         }
2170                 }
2171                 nn -= j;
2172         }
2173  out:
2174         nvme_remove_invalid_namespaces(ctrl, prev);
2175  free:
2176         kfree(ns_list);
2177         return ret;
2178 }
2179
2180 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
2181 {
2182         unsigned i;
2183
2184         for (i = 1; i <= nn; i++)
2185                 nvme_validate_ns(ctrl, i);
2186
2187         nvme_remove_invalid_namespaces(ctrl, nn);
2188 }
2189
2190 static void nvme_scan_work(struct work_struct *work)
2191 {
2192         struct nvme_ctrl *ctrl =
2193                 container_of(work, struct nvme_ctrl, scan_work);
2194         struct nvme_id_ctrl *id;
2195         unsigned nn;
2196
2197         if (ctrl->state != NVME_CTRL_LIVE)
2198                 return;
2199
2200         if (nvme_identify_ctrl(ctrl, &id))
2201                 return;
2202
2203         nn = le32_to_cpu(id->nn);
2204         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
2205             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
2206                 if (!nvme_scan_ns_list(ctrl, nn))
2207                         goto done;
2208         }
2209         nvme_scan_ns_sequential(ctrl, nn);
2210  done:
2211         mutex_lock(&ctrl->namespaces_mutex);
2212         list_sort(NULL, &ctrl->namespaces, ns_cmp);
2213         mutex_unlock(&ctrl->namespaces_mutex);
2214         kfree(id);
2215 }
2216
2217 void nvme_queue_scan(struct nvme_ctrl *ctrl)
2218 {
2219         /*
2220          * Do not queue new scan work when a controller is reset during
2221          * removal.
2222          */
2223         if (ctrl->state == NVME_CTRL_LIVE)
2224                 schedule_work(&ctrl->scan_work);
2225 }
2226 EXPORT_SYMBOL_GPL(nvme_queue_scan);
2227
2228 /*
2229  * This function iterates the namespace list unlocked to allow recovery from
2230  * controller failure. It is up to the caller to ensure the namespace list is
2231  * not modified by scan work while this function is executing.
2232  */
2233 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
2234 {
2235         struct nvme_ns *ns, *next;
2236
2237         /*
2238          * The dead states indicates the controller was not gracefully
2239          * disconnected. In that case, we won't be able to flush any data while
2240          * removing the namespaces' disks; fail all the queues now to avoid
2241          * potentially having to clean up the failed sync later.
2242          */
2243         if (ctrl->state == NVME_CTRL_DEAD)
2244                 nvme_kill_queues(ctrl);
2245
2246         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
2247                 nvme_ns_remove(ns);
2248 }
2249 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
2250
2251 static void nvme_async_event_work(struct work_struct *work)
2252 {
2253         struct nvme_ctrl *ctrl =
2254                 container_of(work, struct nvme_ctrl, async_event_work);
2255
2256         spin_lock_irq(&ctrl->lock);
2257         while (ctrl->event_limit > 0) {
2258                 int aer_idx = --ctrl->event_limit;
2259
2260                 spin_unlock_irq(&ctrl->lock);
2261                 ctrl->ops->submit_async_event(ctrl, aer_idx);
2262                 spin_lock_irq(&ctrl->lock);
2263         }
2264         spin_unlock_irq(&ctrl->lock);
2265 }
2266
2267 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
2268                 union nvme_result *res)
2269 {
2270         u32 result = le32_to_cpu(res->u32);
2271         bool done = true;
2272
2273         switch (le16_to_cpu(status) >> 1) {
2274         case NVME_SC_SUCCESS:
2275                 done = false;
2276                 /*FALLTHRU*/
2277         case NVME_SC_ABORT_REQ:
2278                 ++ctrl->event_limit;
2279                 schedule_work(&ctrl->async_event_work);
2280                 break;
2281         default:
2282                 break;
2283         }
2284
2285         if (done)
2286                 return;
2287
2288         switch (result & 0xff07) {
2289         case NVME_AER_NOTICE_NS_CHANGED:
2290                 dev_info(ctrl->device, "rescanning\n");
2291                 nvme_queue_scan(ctrl);
2292                 break;
2293         default:
2294                 dev_warn(ctrl->device, "async event result %08x\n", result);
2295         }
2296 }
2297 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
2298
2299 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
2300 {
2301         ctrl->event_limit = NVME_NR_AERS;
2302         schedule_work(&ctrl->async_event_work);
2303 }
2304 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
2305
2306 static DEFINE_IDA(nvme_instance_ida);
2307
2308 static int nvme_set_instance(struct nvme_ctrl *ctrl)
2309 {
2310         int instance, error;
2311
2312         do {
2313                 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2314                         return -ENODEV;
2315
2316                 spin_lock(&dev_list_lock);
2317                 error = ida_get_new(&nvme_instance_ida, &instance);
2318                 spin_unlock(&dev_list_lock);
2319         } while (error == -EAGAIN);
2320
2321         if (error)
2322                 return -ENODEV;
2323
2324         ctrl->instance = instance;
2325         return 0;
2326 }
2327
2328 static void nvme_release_instance(struct nvme_ctrl *ctrl)
2329 {
2330         spin_lock(&dev_list_lock);
2331         ida_remove(&nvme_instance_ida, ctrl->instance);
2332         spin_unlock(&dev_list_lock);
2333 }
2334
2335 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
2336 {
2337         flush_work(&ctrl->async_event_work);
2338         flush_work(&ctrl->scan_work);
2339         nvme_remove_namespaces(ctrl);
2340
2341         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
2342
2343         spin_lock(&dev_list_lock);
2344         list_del(&ctrl->node);
2345         spin_unlock(&dev_list_lock);
2346 }
2347 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
2348
2349 static void nvme_free_ctrl(struct kref *kref)
2350 {
2351         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
2352
2353         put_device(ctrl->device);
2354         nvme_release_instance(ctrl);
2355         ida_destroy(&ctrl->ns_ida);
2356
2357         ctrl->ops->free_ctrl(ctrl);
2358 }
2359
2360 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
2361 {
2362         kref_put(&ctrl->kref, nvme_free_ctrl);
2363 }
2364 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2365
2366 /*
2367  * Initialize a NVMe controller structures.  This needs to be called during
2368  * earliest initialization so that we have the initialized structured around
2369  * during probing.
2370  */
2371 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2372                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
2373 {
2374         int ret;
2375
2376         ctrl->state = NVME_CTRL_NEW;
2377         spin_lock_init(&ctrl->lock);
2378         INIT_LIST_HEAD(&ctrl->namespaces);
2379         mutex_init(&ctrl->namespaces_mutex);
2380         kref_init(&ctrl->kref);
2381         ctrl->dev = dev;
2382         ctrl->ops = ops;
2383         ctrl->quirks = quirks;
2384         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2385         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2386
2387         ret = nvme_set_instance(ctrl);
2388         if (ret)
2389                 goto out;
2390
2391         ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2392                                 MKDEV(nvme_char_major, ctrl->instance),
2393                                 ctrl, nvme_dev_attr_groups,
2394                                 "nvme%d", ctrl->instance);
2395         if (IS_ERR(ctrl->device)) {
2396                 ret = PTR_ERR(ctrl->device);
2397                 goto out_release_instance;
2398         }
2399         get_device(ctrl->device);
2400         ida_init(&ctrl->ns_ida);
2401
2402         spin_lock(&dev_list_lock);
2403         list_add_tail(&ctrl->node, &nvme_ctrl_list);
2404         spin_unlock(&dev_list_lock);
2405
2406         /*
2407          * Initialize latency tolerance controls.  The sysfs files won't
2408          * be visible to userspace unless the device actually supports APST.
2409          */
2410         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
2411         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
2412                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
2413
2414         return 0;
2415 out_release_instance:
2416         nvme_release_instance(ctrl);
2417 out:
2418         return ret;
2419 }
2420 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2421
2422 /**
2423  * nvme_kill_queues(): Ends all namespace queues
2424  * @ctrl: the dead controller that needs to end
2425  *
2426  * Call this function when the driver determines it is unable to get the
2427  * controller in a state capable of servicing IO.
2428  */
2429 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2430 {
2431         struct nvme_ns *ns;
2432
2433         mutex_lock(&ctrl->namespaces_mutex);
2434         list_for_each_entry(ns, &ctrl->namespaces, list) {
2435                 /*
2436                  * Revalidating a dead namespace sets capacity to 0. This will
2437                  * end buffered writers dirtying pages that can't be synced.
2438                  */
2439                 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2440                         continue;
2441                 revalidate_disk(ns->disk);
2442                 blk_set_queue_dying(ns->queue);
2443
2444                 /*
2445                  * Forcibly start all queues to avoid having stuck requests.
2446                  * Note that we must ensure the queues are not stopped
2447                  * when the final removal happens.
2448                  */
2449                 blk_mq_start_hw_queues(ns->queue);
2450
2451                 /* draining requests in requeue list */
2452                 blk_mq_kick_requeue_list(ns->queue);
2453         }
2454         mutex_unlock(&ctrl->namespaces_mutex);
2455 }
2456 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2457
2458 void nvme_unfreeze(struct nvme_ctrl *ctrl)
2459 {
2460         struct nvme_ns *ns;
2461
2462         mutex_lock(&ctrl->namespaces_mutex);
2463         list_for_each_entry(ns, &ctrl->namespaces, list)
2464                 blk_mq_unfreeze_queue(ns->queue);
2465         mutex_unlock(&ctrl->namespaces_mutex);
2466 }
2467 EXPORT_SYMBOL_GPL(nvme_unfreeze);
2468
2469 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
2470 {
2471         struct nvme_ns *ns;
2472
2473         mutex_lock(&ctrl->namespaces_mutex);
2474         list_for_each_entry(ns, &ctrl->namespaces, list) {
2475                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
2476                 if (timeout <= 0)
2477                         break;
2478         }
2479         mutex_unlock(&ctrl->namespaces_mutex);
2480 }
2481 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
2482
2483 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
2484 {
2485         struct nvme_ns *ns;
2486
2487         mutex_lock(&ctrl->namespaces_mutex);
2488         list_for_each_entry(ns, &ctrl->namespaces, list)
2489                 blk_mq_freeze_queue_wait(ns->queue);
2490         mutex_unlock(&ctrl->namespaces_mutex);
2491 }
2492 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
2493
2494 void nvme_start_freeze(struct nvme_ctrl *ctrl)
2495 {
2496         struct nvme_ns *ns;
2497
2498         mutex_lock(&ctrl->namespaces_mutex);
2499         list_for_each_entry(ns, &ctrl->namespaces, list)
2500                 blk_freeze_queue_start(ns->queue);
2501         mutex_unlock(&ctrl->namespaces_mutex);
2502 }
2503 EXPORT_SYMBOL_GPL(nvme_start_freeze);
2504
2505 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2506 {
2507         struct nvme_ns *ns;
2508
2509         mutex_lock(&ctrl->namespaces_mutex);
2510         list_for_each_entry(ns, &ctrl->namespaces, list)
2511                 blk_mq_quiesce_queue(ns->queue);
2512         mutex_unlock(&ctrl->namespaces_mutex);
2513 }
2514 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2515
2516 void nvme_start_queues(struct nvme_ctrl *ctrl)
2517 {
2518         struct nvme_ns *ns;
2519
2520         mutex_lock(&ctrl->namespaces_mutex);
2521         list_for_each_entry(ns, &ctrl->namespaces, list) {
2522                 blk_mq_start_stopped_hw_queues(ns->queue, true);
2523                 blk_mq_kick_requeue_list(ns->queue);
2524         }
2525         mutex_unlock(&ctrl->namespaces_mutex);
2526 }
2527 EXPORT_SYMBOL_GPL(nvme_start_queues);
2528
2529 int __init nvme_core_init(void)
2530 {
2531         int result;
2532
2533         result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2534                                                         &nvme_dev_fops);
2535         if (result < 0)
2536                 return result;
2537         else if (result > 0)
2538                 nvme_char_major = result;
2539
2540         nvme_class = class_create(THIS_MODULE, "nvme");
2541         if (IS_ERR(nvme_class)) {
2542                 result = PTR_ERR(nvme_class);
2543                 goto unregister_chrdev;
2544         }
2545
2546         return 0;
2547
2548  unregister_chrdev:
2549         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2550         return result;
2551 }
2552
2553 void nvme_core_exit(void)
2554 {
2555         class_destroy(nvme_class);
2556         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2557 }
2558
2559 MODULE_LICENSE("GPL");
2560 MODULE_VERSION("1.0");
2561 module_init(nvme_core_init);
2562 module_exit(nvme_core_exit);