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 = 100000;
  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 int nvme_error_status(struct request *req)
  74 {
  75         switch (nvme_req(req)->status & 0x7ff) {
  76         case NVME_SC_SUCCESS:
  77                 return 0;
  78         case NVME_SC_CAP_EXCEEDED:
  79                 return -ENOSPC;
  80         default:
  81                 return -EIO;
  82
  83         /*
  84          * XXX: these errors are a nasty side-band protocol to
  85          * drivers/md/dm-mpath.c:noretry_error() that aren't documented
  86          * anywhere..
  87          */
  88         case NVME_SC_CMD_SEQ_ERROR:
  89                 return -EILSEQ;
  90         case NVME_SC_ONCS_NOT_SUPPORTED:
  91                 return -EOPNOTSUPP;
  92         case NVME_SC_WRITE_FAULT:
  93         case NVME_SC_READ_ERROR:
  94         case NVME_SC_UNWRITTEN_BLOCK:
  95                 return -ENODATA;
  96         }
  97 }
  98
  99 static inline bool nvme_req_needs_retry(struct request *req)
 100 {
 101         if (blk_noretry_request(req))
 102                 return false;
 103         if (nvme_req(req)->status & NVME_SC_DNR)
 104                 return false;
 105         if (jiffies - req->start_time >= req->timeout)
 106                 return false;
 107         if (nvme_req(req)->retries >= nvme_max_retries)
 108                 return false;
 109         return true;
 110 }
 111
 112 void nvme_complete_rq(struct request *req)
 113 {
 114         if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) {
 115                 nvme_req(req)->retries++;
 116                 blk_mq_requeue_request(req, !blk_mq_queue_stopped(req->q));
 117                 return;
 118         }
 119
 120         blk_mq_end_request(req, nvme_error_status(req));
 121 }
 122 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 123
 124 void nvme_cancel_request(struct request *req, void *data, bool reserved)
 125 {
 126         int status;
 127
 128         if (!blk_mq_request_started(req))
 129                 return;
 130
 131         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 132                                 "Cancelling I/O %d", req->tag);
 133
 134         status = NVME_SC_ABORT_REQ;
 135         if (blk_queue_dying(req->q))
 136                 status |= NVME_SC_DNR;
 137         nvme_req(req)->status = status;
 138         blk_mq_complete_request(req);
 139
 140 }
 141 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 142
 143 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 144                 enum nvme_ctrl_state new_state)
 145 {
 146         enum nvme_ctrl_state old_state;
 147         bool changed = false;
 148
 149         spin_lock_irq(&ctrl->lock);
 150
 151         old_state = ctrl->state;
 152         switch (new_state) {
 153         case NVME_CTRL_LIVE:
 154                 switch (old_state) {
 155                 case NVME_CTRL_NEW:
 156                 case NVME_CTRL_RESETTING:
 157                 case NVME_CTRL_RECONNECTING:
 158                         changed = true;
 159                         /* FALLTHRU */
 160                 default:
 161                         break;
 162                 }
 163                 break;
 164         case NVME_CTRL_RESETTING:
 165                 switch (old_state) {
 166                 case NVME_CTRL_NEW:
 167                 case NVME_CTRL_LIVE:
 168                 case NVME_CTRL_RECONNECTING:
 169                         changed = true;
 170                         /* FALLTHRU */
 171                 default:
 172                         break;
 173                 }
 174                 break;
 175         case NVME_CTRL_RECONNECTING:
 176                 switch (old_state) {
 177                 case NVME_CTRL_LIVE:
 178                         changed = true;
 179                         /* FALLTHRU */
 180                 default:
 181                         break;
 182                 }
 183                 break;
 184         case NVME_CTRL_DELETING:
 185                 switch (old_state) {
 186                 case NVME_CTRL_LIVE:
 187                 case NVME_CTRL_RESETTING:
 188                 case NVME_CTRL_RECONNECTING:
 189                         changed = true;
 190                         /* FALLTHRU */
 191                 default:
 192                         break;
 193                 }
 194                 break;
 195         case NVME_CTRL_DEAD:
 196                 switch (old_state) {
 197                 case NVME_CTRL_DELETING:
 198                         changed = true;
 199                         /* FALLTHRU */
 200                 default:
 201                         break;
 202                 }
 203                 break;
 204         default:
 205                 break;
 206         }
 207
 208         if (changed)
 209                 ctrl->state = new_state;
 210
 211         spin_unlock_irq(&ctrl->lock);
 212
 213         return changed;
 214 }
 215 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 216
 217 static void nvme_free_ns(struct kref *kref)
 218 {
 219         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 220
 221         if (ns->ndev)
 222                 nvme_nvm_unregister(ns);
 223
 224         if (ns->disk) {
 225                 spin_lock(&dev_list_lock);
 226                 ns->disk->private_data = NULL;
 227                 spin_unlock(&dev_list_lock);
 228         }
 229
 230         put_disk(ns->disk);
 231         ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
 232         nvme_put_ctrl(ns->ctrl);
 233         kfree(ns);
 234 }
 235
 236 static void nvme_put_ns(struct nvme_ns *ns)
 237 {
 238         kref_put(&ns->kref, nvme_free_ns);
 239 }
 240
 241 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
 242 {
 243         struct nvme_ns *ns;
 244
 245         spin_lock(&dev_list_lock);
 246         ns = disk->private_data;
 247         if (ns) {
 248                 if (!kref_get_unless_zero(&ns->kref))
 249                         goto fail;
 250                 if (!try_module_get(ns->ctrl->ops->module))
 251                         goto fail_put_ns;
 252         }
 253         spin_unlock(&dev_list_lock);
 254
 255         return ns;
 256
 257 fail_put_ns:
 258         kref_put(&ns->kref, nvme_free_ns);
 259 fail:
 260         spin_unlock(&dev_list_lock);
 261         return NULL;
 262 }
 263
 264 struct request *nvme_alloc_request(struct request_queue *q,
 265                 struct nvme_command *cmd, unsigned int flags, int qid)
 266 {
 267         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 268         struct request *req;
 269
 270         if (qid == NVME_QID_ANY) {
 271                 req = blk_mq_alloc_request(q, op, flags);
 272         } else {
 273                 req = blk_mq_alloc_request_hctx(q, op, flags,
 274                                 qid ? qid - 1 : 0);
 275         }
 276         if (IS_ERR(req))
 277                 return req;
 278
 279         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 280         nvme_req(req)->cmd = cmd;
 281
 282         return req;
 283 }
 284 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 285
 286 static inline void nvme_setup_flush(struct nvme_ns *ns,
 287                 struct nvme_command *cmnd)
 288 {
 289         memset(cmnd, 0, sizeof(*cmnd));
 290         cmnd->common.opcode = nvme_cmd_flush;
 291         cmnd->common.nsid = cpu_to_le32(ns->ns_id);
 292 }
 293
 294 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 295                 struct nvme_command *cmnd)
 296 {
 297         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 298         struct nvme_dsm_range *range;
 299         struct bio *bio;
 300
 301         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 302         if (!range)
 303                 return BLK_MQ_RQ_QUEUE_BUSY;
 304
 305         __rq_for_each_bio(bio, req) {
 306                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 307                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 308
 309                 range[n].cattr = cpu_to_le32(0);
 310                 range[n].nlb = cpu_to_le32(nlb);
 311                 range[n].slba = cpu_to_le64(slba);
 312                 n++;
 313         }
 314
 315         if (WARN_ON_ONCE(n != segments)) {
 316                 kfree(range);
 317                 return BLK_MQ_RQ_QUEUE_ERROR;
 318         }
 319
 320         memset(cmnd, 0, sizeof(*cmnd));
 321         cmnd->dsm.opcode = nvme_cmd_dsm;
 322         cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
 323         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 324         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 325
 326         req->special_vec.bv_page = virt_to_page(range);
 327         req->special_vec.bv_offset = offset_in_page(range);
 328         req->special_vec.bv_len = sizeof(*range) * segments;
 329         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 330
 331         return BLK_MQ_RQ_QUEUE_OK;
 332 }
 333
 334 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
 335                 struct nvme_command *cmnd)
 336 {
 337         u16 control = 0;
 338         u32 dsmgmt = 0;
 339
 340         if (req->cmd_flags & REQ_FUA)
 341                 control |= NVME_RW_FUA;
 342         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 343                 control |= NVME_RW_LR;
 344
 345         if (req->cmd_flags & REQ_RAHEAD)
 346                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 347
 348         memset(cmnd, 0, sizeof(*cmnd));
 349         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 350         cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
 351         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 352         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 353
 354         if (ns->ms) {
 355                 switch (ns->pi_type) {
 356                 case NVME_NS_DPS_PI_TYPE3:
 357                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 358                         break;
 359                 case NVME_NS_DPS_PI_TYPE1:
 360                 case NVME_NS_DPS_PI_TYPE2:
 361                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 362                                         NVME_RW_PRINFO_PRCHK_REF;
 363                         cmnd->rw.reftag = cpu_to_le32(
 364                                         nvme_block_nr(ns, blk_rq_pos(req)));
 365                         break;
 366                 }
 367                 if (!blk_integrity_rq(req))
 368                         control |= NVME_RW_PRINFO_PRACT;
 369         }
 370
 371         cmnd->rw.control = cpu_to_le16(control);
 372         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 373 }
 374
 375 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 376                 struct nvme_command *cmd)
 377 {
 378         int ret = BLK_MQ_RQ_QUEUE_OK;
 379
 380         if (!(req->rq_flags & RQF_DONTPREP)) {
 381                 nvme_req(req)->retries = 0;
 382                 nvme_req(req)->flags = 0;
 383                 req->rq_flags |= RQF_DONTPREP;
 384         }
 385
 386         switch (req_op(req)) {
 387         case REQ_OP_DRV_IN:
 388         case REQ_OP_DRV_OUT:
 389                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 390                 break;
 391         case REQ_OP_FLUSH:
 392                 nvme_setup_flush(ns, cmd);
 393                 break;
 394         case REQ_OP_WRITE_ZEROES:
 395                 /* currently only aliased to deallocate for a few ctrls: */
 396         case REQ_OP_DISCARD:
 397                 ret = nvme_setup_discard(ns, req, cmd);
 398                 break;
 399         case REQ_OP_READ:
 400         case REQ_OP_WRITE:
 401                 nvme_setup_rw(ns, req, cmd);
 402                 break;
 403         default:
 404                 WARN_ON_ONCE(1);
 405                 return BLK_MQ_RQ_QUEUE_ERROR;
 406         }
 407
 408         cmd->common.command_id = req->tag;
 409         return ret;
 410 }
 411 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 412
 413 /*
 414  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 415  * if the result is positive, it's an NVM Express status code
 416  */
 417 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 418                 union nvme_result *result, void *buffer, unsigned bufflen,
 419                 unsigned timeout, int qid, int at_head, int flags)
 420 {
 421         struct request *req;
 422         int ret;
 423
 424         req = nvme_alloc_request(q, cmd, flags, qid);
 425         if (IS_ERR(req))
 426                 return PTR_ERR(req);
 427
 428         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 429
 430         if (buffer && bufflen) {
 431                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 432                 if (ret)
 433                         goto out;
 434         }
 435
 436         blk_execute_rq(req->q, NULL, req, at_head);
 437         if (result)
 438                 *result = nvme_req(req)->result;
 439         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 440                 ret = -EINTR;
 441         else
 442                 ret = nvme_req(req)->status;
 443  out:
 444         blk_mq_free_request(req);
 445         return ret;
 446 }
 447 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 448
 449 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 450                 void *buffer, unsigned bufflen)
 451 {
 452         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 453                         NVME_QID_ANY, 0, 0);
 454 }
 455 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 456
 457 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 458                 void __user *ubuffer, unsigned bufflen,
 459                 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
 460                 u32 *result, unsigned timeout)
 461 {
 462         bool write = nvme_is_write(cmd);
 463         struct nvme_ns *ns = q->queuedata;
 464         struct gendisk *disk = ns ? ns->disk : NULL;
 465         struct request *req;
 466         struct bio *bio = NULL;
 467         void *meta = NULL;
 468         int ret;
 469
 470         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 471         if (IS_ERR(req))
 472                 return PTR_ERR(req);
 473
 474         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 475
 476         if (ubuffer && bufflen) {
 477                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 478                                 GFP_KERNEL);
 479                 if (ret)
 480                         goto out;
 481                 bio = req->bio;
 482
 483                 if (!disk)
 484                         goto submit;
 485                 bio->bi_bdev = bdget_disk(disk, 0);
 486                 if (!bio->bi_bdev) {
 487                         ret = -ENODEV;
 488                         goto out_unmap;
 489                 }
 490
 491                 if (meta_buffer && meta_len) {
 492                         struct bio_integrity_payload *bip;
 493
 494                         meta = kmalloc(meta_len, GFP_KERNEL);
 495                         if (!meta) {
 496                                 ret = -ENOMEM;
 497                                 goto out_unmap;
 498                         }
 499
 500                         if (write) {
 501                                 if (copy_from_user(meta, meta_buffer,
 502                                                 meta_len)) {
 503                                         ret = -EFAULT;
 504                                         goto out_free_meta;
 505                                 }
 506                         }
 507
 508                         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 509                         if (IS_ERR(bip)) {
 510                                 ret = PTR_ERR(bip);
 511                                 goto out_free_meta;
 512                         }
 513
 514                         bip->bip_iter.bi_size = meta_len;
 515                         bip->bip_iter.bi_sector = meta_seed;
 516
 517                         ret = bio_integrity_add_page(bio, virt_to_page(meta),
 518                                         meta_len, offset_in_page(meta));
 519                         if (ret != meta_len) {
 520                                 ret = -ENOMEM;
 521                                 goto out_free_meta;
 522                         }
 523                 }
 524         }
 525  submit:
 526         blk_execute_rq(req->q, disk, req, 0);
 527         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 528                 ret = -EINTR;
 529         else
 530                 ret = nvme_req(req)->status;
 531         if (result)
 532                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 533         if (meta && !ret && !write) {
 534                 if (copy_to_user(meta_buffer, meta, meta_len))
 535                         ret = -EFAULT;
 536         }
 537  out_free_meta:
 538         kfree(meta);
 539  out_unmap:
 540         if (bio) {
 541                 if (disk && bio->bi_bdev)
 542                         bdput(bio->bi_bdev);
 543                 blk_rq_unmap_user(bio);
 544         }
 545  out:
 546         blk_mq_free_request(req);
 547         return ret;
 548 }
 549
 550 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 551                 void __user *ubuffer, unsigned bufflen, u32 *result,
 552                 unsigned timeout)
 553 {
 554         return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
 555                         result, timeout);
 556 }
 557
 558 static void nvme_keep_alive_end_io(struct request *rq, int error)
 559 {
 560         struct nvme_ctrl *ctrl = rq->end_io_data;
 561
 562         blk_mq_free_request(rq);
 563
 564         if (error) {
 565                 dev_err(ctrl->device,
 566                         "failed nvme_keep_alive_end_io error=%d\n", error);
 567                 return;
 568         }
 569
 570         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 571 }
 572
 573 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 574 {
 575         struct nvme_command c;
 576         struct request *rq;
 577
 578         memset(&c, 0, sizeof(c));
 579         c.common.opcode = nvme_admin_keep_alive;
 580
 581         rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
 582                         NVME_QID_ANY);
 583         if (IS_ERR(rq))
 584                 return PTR_ERR(rq);
 585
 586         rq->timeout = ctrl->kato * HZ;
 587         rq->end_io_data = ctrl;
 588
 589         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 590
 591         return 0;
 592 }
 593
 594 static void nvme_keep_alive_work(struct work_struct *work)
 595 {
 596         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 597                         struct nvme_ctrl, ka_work);
 598
 599         if (nvme_keep_alive(ctrl)) {
 600                 /* allocation failure, reset the controller */
 601                 dev_err(ctrl->device, "keep-alive failed\n");
 602                 ctrl->ops->reset_ctrl(ctrl);
 603                 return;
 604         }
 605 }
 606
 607 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 608 {
 609         if (unlikely(ctrl->kato == 0))
 610                 return;
 611
 612         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
 613         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 614 }
 615 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
 616
 617 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 618 {
 619         if (unlikely(ctrl->kato == 0))
 620                 return;
 621
 622         cancel_delayed_work_sync(&ctrl->ka_work);
 623 }
 624 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 625
 626 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 627 {
 628         struct nvme_command c = { };
 629         int error;
 630
 631         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 632         c.identify.opcode = nvme_admin_identify;
 633         c.identify.cns = NVME_ID_CNS_CTRL;
 634
 635         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 636         if (!*id)
 637                 return -ENOMEM;
 638
 639         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 640                         sizeof(struct nvme_id_ctrl));
 641         if (error)
 642                 kfree(*id);
 643         return error;
 644 }
 645
 646 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 647 {
 648         struct nvme_command c = { };
 649
 650         c.identify.opcode = nvme_admin_identify;
 651         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
 652         c.identify.nsid = cpu_to_le32(nsid);
 653         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 654 }
 655
 656 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
 657                 struct nvme_id_ns **id)
 658 {
 659         struct nvme_command c = { };
 660         int error;
 661
 662         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 663         c.identify.opcode = nvme_admin_identify;
 664         c.identify.nsid = cpu_to_le32(nsid);
 665         c.identify.cns = NVME_ID_CNS_NS;
 666
 667         *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
 668         if (!*id)
 669                 return -ENOMEM;
 670
 671         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 672                         sizeof(struct nvme_id_ns));
 673         if (error)
 674                 kfree(*id);
 675         return error;
 676 }
 677
 678 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 679                       void *buffer, size_t buflen, u32 *result)
 680 {
 681         struct nvme_command c;
 682         union nvme_result res;
 683         int ret;
 684
 685         memset(&c, 0, sizeof(c));
 686         c.features.opcode = nvme_admin_get_features;
 687         c.features.nsid = cpu_to_le32(nsid);
 688         c.features.fid = cpu_to_le32(fid);
 689
 690         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, buffer, buflen, 0,
 691                         NVME_QID_ANY, 0, 0);
 692         if (ret >= 0 && result)
 693                 *result = le32_to_cpu(res.u32);
 694         return ret;
 695 }
 696
 697 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 698                       void *buffer, size_t buflen, u32 *result)
 699 {
 700         struct nvme_command c;
 701         union nvme_result res;
 702         int ret;
 703
 704         memset(&c, 0, sizeof(c));
 705         c.features.opcode = nvme_admin_set_features;
 706         c.features.fid = cpu_to_le32(fid);
 707         c.features.dword11 = cpu_to_le32(dword11);
 708
 709         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
 710                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
 711         if (ret >= 0 && result)
 712                 *result = le32_to_cpu(res.u32);
 713         return ret;
 714 }
 715
 716 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
 717 {
 718         struct nvme_command c = { };
 719         int error;
 720
 721         c.common.opcode = nvme_admin_get_log_page,
 722         c.common.nsid = cpu_to_le32(0xFFFFFFFF),
 723         c.common.cdw10[0] = cpu_to_le32(
 724                         (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
 725                          NVME_LOG_SMART),
 726
 727         *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
 728         if (!*log)
 729                 return -ENOMEM;
 730
 731         error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
 732                         sizeof(struct nvme_smart_log));
 733         if (error)
 734                 kfree(*log);
 735         return error;
 736 }
 737
 738 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
 739 {
 740         u32 q_count = (*count - 1) | ((*count - 1) << 16);
 741         u32 result;
 742         int status, nr_io_queues;
 743
 744         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
 745                         &result);
 746         if (status < 0)
 747                 return status;
 748
 749         /*
 750          * Degraded controllers might return an error when setting the queue
 751          * count.  We still want to be able to bring them online and offer
 752          * access to the admin queue, as that might be only way to fix them up.
 753          */
 754         if (status > 0) {
 755                 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
 756                 *count = 0;
 757         } else {
 758                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
 759                 *count = min(*count, nr_io_queues);
 760         }
 761
 762         return 0;
 763 }
 764 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
 765
 766 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 767 {
 768         struct nvme_user_io io;
 769         struct nvme_command c;
 770         unsigned length, meta_len;
 771         void __user *metadata;
 772
 773         if (copy_from_user(&io, uio, sizeof(io)))
 774                 return -EFAULT;
 775         if (io.flags)
 776                 return -EINVAL;
 777
 778         switch (io.opcode) {
 779         case nvme_cmd_write:
 780         case nvme_cmd_read:
 781         case nvme_cmd_compare:
 782                 break;
 783         default:
 784                 return -EINVAL;
 785         }
 786
 787         length = (io.nblocks + 1) << ns->lba_shift;
 788         meta_len = (io.nblocks + 1) * ns->ms;
 789         metadata = (void __user *)(uintptr_t)io.metadata;
 790
 791         if (ns->ext) {
 792                 length += meta_len;
 793                 meta_len = 0;
 794         } else if (meta_len) {
 795                 if ((io.metadata & 3) || !io.metadata)
 796                         return -EINVAL;
 797         }
 798
 799         memset(&c, 0, sizeof(c));
 800         c.rw.opcode = io.opcode;
 801         c.rw.flags = io.flags;
 802         c.rw.nsid = cpu_to_le32(ns->ns_id);
 803         c.rw.slba = cpu_to_le64(io.slba);
 804         c.rw.length = cpu_to_le16(io.nblocks);
 805         c.rw.control = cpu_to_le16(io.control);
 806         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
 807         c.rw.reftag = cpu_to_le32(io.reftag);
 808         c.rw.apptag = cpu_to_le16(io.apptag);
 809         c.rw.appmask = cpu_to_le16(io.appmask);
 810
 811         return __nvme_submit_user_cmd(ns->queue, &c,
 812                         (void __user *)(uintptr_t)io.addr, length,
 813                         metadata, meta_len, io.slba, NULL, 0);
 814 }
 815
 816 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 817                         struct nvme_passthru_cmd __user *ucmd)
 818 {
 819         struct nvme_passthru_cmd cmd;
 820         struct nvme_command c;
 821         unsigned timeout = 0;
 822         int status;
 823
 824         if (!capable(CAP_SYS_ADMIN))
 825                 return -EACCES;
 826         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
 827                 return -EFAULT;
 828         if (cmd.flags)
 829                 return -EINVAL;
 830
 831         memset(&c, 0, sizeof(c));
 832         c.common.opcode = cmd.opcode;
 833         c.common.flags = cmd.flags;
 834         c.common.nsid = cpu_to_le32(cmd.nsid);
 835         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
 836         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
 837         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
 838         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
 839         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
 840         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
 841         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
 842         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
 843
 844         if (cmd.timeout_ms)
 845                 timeout = msecs_to_jiffies(cmd.timeout_ms);
 846
 847         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
 848                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
 849                         &cmd.result, timeout);
 850         if (status >= 0) {
 851                 if (put_user(cmd.result, &ucmd->result))
 852                         return -EFAULT;
 853         }
 854
 855         return status;
 856 }
 857
 858 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 859                 unsigned int cmd, unsigned long arg)
 860 {
 861         struct nvme_ns *ns = bdev->bd_disk->private_data;
 862
 863         switch (cmd) {
 864         case NVME_IOCTL_ID:
 865                 force_successful_syscall_return();
 866                 return ns->ns_id;
 867         case NVME_IOCTL_ADMIN_CMD:
 868                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
 869         case NVME_IOCTL_IO_CMD:
 870                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
 871         case NVME_IOCTL_SUBMIT_IO:
 872                 return nvme_submit_io(ns, (void __user *)arg);
 873 #ifdef CONFIG_BLK_DEV_NVME_SCSI
 874         case SG_GET_VERSION_NUM:
 875                 return nvme_sg_get_version_num((void __user *)arg);
 876         case SG_IO:
 877                 return nvme_sg_io(ns, (void __user *)arg);
 878 #endif
 879         default:
 880 #ifdef CONFIG_NVM
 881                 if (ns->ndev)
 882                         return nvme_nvm_ioctl(ns, cmd, arg);
 883 #endif
 884                 if (is_sed_ioctl(cmd))
 885                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
 886                                          (void __user *) arg);
 887                 return -ENOTTY;
 888         }
 889 }
 890
 891 #ifdef CONFIG_COMPAT
 892 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 893                         unsigned int cmd, unsigned long arg)
 894 {
 895         switch (cmd) {
 896         case SG_IO:
 897                 return -ENOIOCTLCMD;
 898         }
 899         return nvme_ioctl(bdev, mode, cmd, arg);
 900 }
 901 #else
 902 #define nvme_compat_ioctl       NULL
 903 #endif
 904
 905 static int nvme_open(struct block_device *bdev, fmode_t mode)
 906 {
 907         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
 908 }
 909
 910 static void nvme_release(struct gendisk *disk, fmode_t mode)
 911 {
 912         struct nvme_ns *ns = disk->private_data;
 913
 914         module_put(ns->ctrl->ops->module);
 915         nvme_put_ns(ns);
 916 }
 917
 918 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 919 {
 920         /* some standard values */
 921         geo->heads = 1 << 6;
 922         geo->sectors = 1 << 5;
 923         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
 924         return 0;
 925 }
 926
 927 #ifdef CONFIG_BLK_DEV_INTEGRITY
 928 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
 929                 u16 bs)
 930 {
 931         struct nvme_ns *ns = disk->private_data;
 932         u16 old_ms = ns->ms;
 933         u8 pi_type = 0;
 934
 935         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
 936         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
 937
 938         /* PI implementation requires metadata equal t10 pi tuple size */
 939         if (ns->ms == sizeof(struct t10_pi_tuple))
 940                 pi_type = id->dps & NVME_NS_DPS_PI_MASK;
 941
 942         if (blk_get_integrity(disk) &&
 943             (ns->pi_type != pi_type || ns->ms != old_ms ||
 944              bs != queue_logical_block_size(disk->queue) ||
 945              (ns->ms && ns->ext)))
 946                 blk_integrity_unregister(disk);
 947
 948         ns->pi_type = pi_type;
 949 }
 950
 951 static void nvme_init_integrity(struct nvme_ns *ns)
 952 {
 953         struct blk_integrity integrity;
 954
 955         memset(&integrity, 0, sizeof(integrity));
 956         switch (ns->pi_type) {
 957         case NVME_NS_DPS_PI_TYPE3:
 958                 integrity.profile = &t10_pi_type3_crc;
 959                 integrity.tag_size = sizeof(u16) + sizeof(u32);
 960                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
 961                 break;
 962         case NVME_NS_DPS_PI_TYPE1:
 963         case NVME_NS_DPS_PI_TYPE2:
 964                 integrity.profile = &t10_pi_type1_crc;
 965                 integrity.tag_size = sizeof(u16);
 966                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
 967                 break;
 968         default:
 969                 integrity.profile = NULL;
 970                 break;
 971         }
 972         integrity.tuple_size = ns->ms;
 973         blk_integrity_register(ns->disk, &integrity);
 974         blk_queue_max_integrity_segments(ns->queue, 1);
 975 }
 976 #else
 977 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
 978                 u16 bs)
 979 {
 980 }
 981 static void nvme_init_integrity(struct nvme_ns *ns)
 982 {
 983 }
 984 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 985
 986 static void nvme_config_discard(struct nvme_ns *ns)
 987 {
 988         struct nvme_ctrl *ctrl = ns->ctrl;
 989         u32 logical_block_size = queue_logical_block_size(ns->queue);
 990
 991         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
 992                         NVME_DSM_MAX_RANGES);
 993
 994         ns->queue->limits.discard_alignment = logical_block_size;
 995         ns->queue->limits.discard_granularity = logical_block_size;
 996         blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
 997         blk_queue_max_discard_segments(ns->queue, NVME_DSM_MAX_RANGES);
 998         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
 999
1000         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1001                 blk_queue_max_write_zeroes_sectors(ns->queue, UINT_MAX);
1002 }
1003
1004 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
1005 {
1006         if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
1007                 dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
1008                 return -ENODEV;
1009         }
1010
1011         if ((*id)->ncap == 0) {
1012                 kfree(*id);
1013                 return -ENODEV;
1014         }
1015
1016         if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
1017                 memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
1018         if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
1019                 memcpy(ns->uuid, (*id)->nguid, sizeof(ns->uuid));
1020
1021         return 0;
1022 }
1023
1024 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1025 {
1026         struct nvme_ns *ns = disk->private_data;
1027         u16 bs;
1028
1029         /*
1030          * If identify namespace failed, use default 512 byte block size so
1031          * block layer can use before failing read/write for 0 capacity.
1032          */
1033         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1034         if (ns->lba_shift == 0)
1035                 ns->lba_shift = 9;
1036         bs = 1 << ns->lba_shift;
1037
1038         blk_mq_freeze_queue(disk->queue);
1039
1040         if (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED)
1041                 nvme_prep_integrity(disk, id, bs);
1042         blk_queue_logical_block_size(ns->queue, bs);
1043         if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
1044                 nvme_init_integrity(ns);
1045         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
1046                 set_capacity(disk, 0);
1047         else
1048                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
1049
1050         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1051                 nvme_config_discard(ns);
1052         blk_mq_unfreeze_queue(disk->queue);
1053 }
1054
1055 static int nvme_revalidate_disk(struct gendisk *disk)
1056 {
1057         struct nvme_ns *ns = disk->private_data;
1058         struct nvme_id_ns *id = NULL;
1059         int ret;
1060
1061         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1062                 set_capacity(disk, 0);
1063                 return -ENODEV;
1064         }
1065
1066         ret = nvme_revalidate_ns(ns, &id);
1067         if (ret)
1068                 return ret;
1069
1070         __nvme_revalidate_disk(disk, id);
1071         kfree(id);
1072
1073         return 0;
1074 }
1075
1076 static char nvme_pr_type(enum pr_type type)
1077 {
1078         switch (type) {
1079         case PR_WRITE_EXCLUSIVE:
1080                 return 1;
1081         case PR_EXCLUSIVE_ACCESS:
1082                 return 2;
1083         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1084                 return 3;
1085         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1086                 return 4;
1087         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1088                 return 5;
1089         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1090                 return 6;
1091         default:
1092                 return 0;
1093         }
1094 };
1095
1096 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1097                                 u64 key, u64 sa_key, u8 op)
1098 {
1099         struct nvme_ns *ns = bdev->bd_disk->private_data;
1100         struct nvme_command c;
1101         u8 data[16] = { 0, };
1102
1103         put_unaligned_le64(key, &data[0]);
1104         put_unaligned_le64(sa_key, &data[8]);
1105
1106         memset(&c, 0, sizeof(c));
1107         c.common.opcode = op;
1108         c.common.nsid = cpu_to_le32(ns->ns_id);
1109         c.common.cdw10[0] = cpu_to_le32(cdw10);
1110
1111         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1112 }
1113
1114 static int nvme_pr_register(struct block_device *bdev, u64 old,
1115                 u64 new, unsigned flags)
1116 {
1117         u32 cdw10;
1118
1119         if (flags & ~PR_FL_IGNORE_KEY)
1120                 return -EOPNOTSUPP;
1121
1122         cdw10 = old ? 2 : 0;
1123         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1124         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1125         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1126 }
1127
1128 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1129                 enum pr_type type, unsigned flags)
1130 {
1131         u32 cdw10;
1132
1133         if (flags & ~PR_FL_IGNORE_KEY)
1134                 return -EOPNOTSUPP;
1135
1136         cdw10 = nvme_pr_type(type) << 8;
1137         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1138         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1139 }
1140
1141 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1142                 enum pr_type type, bool abort)
1143 {
1144         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1145         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1146 }
1147
1148 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1149 {
1150         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1151         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1152 }
1153
1154 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1155 {
1156         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1157         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1158 }
1159
1160 static const struct pr_ops nvme_pr_ops = {
1161         .pr_register    = nvme_pr_register,
1162         .pr_reserve     = nvme_pr_reserve,
1163         .pr_release     = nvme_pr_release,
1164         .pr_preempt     = nvme_pr_preempt,
1165         .pr_clear       = nvme_pr_clear,
1166 };
1167
1168 #ifdef CONFIG_BLK_SED_OPAL
1169 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1170                 bool send)
1171 {
1172         struct nvme_ctrl *ctrl = data;
1173         struct nvme_command cmd;
1174
1175         memset(&cmd, 0, sizeof(cmd));
1176         if (send)
1177                 cmd.common.opcode = nvme_admin_security_send;
1178         else
1179                 cmd.common.opcode = nvme_admin_security_recv;
1180         cmd.common.nsid = 0;
1181         cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1182         cmd.common.cdw10[1] = cpu_to_le32(len);
1183
1184         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1185                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1186 }
1187 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1188 #endif /* CONFIG_BLK_SED_OPAL */
1189
1190 static const struct block_device_operations nvme_fops = {
1191         .owner          = THIS_MODULE,
1192         .ioctl          = nvme_ioctl,
1193         .compat_ioctl   = nvme_compat_ioctl,
1194         .open           = nvme_open,
1195         .release        = nvme_release,
1196         .getgeo         = nvme_getgeo,
1197         .revalidate_disk= nvme_revalidate_disk,
1198         .pr_ops         = &nvme_pr_ops,
1199 };
1200
1201 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1202 {
1203         unsigned long timeout =
1204                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1205         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1206         int ret;
1207
1208         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1209                 if (csts == ~0)
1210                         return -ENODEV;
1211                 if ((csts & NVME_CSTS_RDY) == bit)
1212                         break;
1213
1214                 msleep(100);
1215                 if (fatal_signal_pending(current))
1216                         return -EINTR;
1217                 if (time_after(jiffies, timeout)) {
1218                         dev_err(ctrl->device,
1219                                 "Device not ready; aborting %s\n", enabled ?
1220                                                 "initialisation" : "reset");
1221                         return -ENODEV;
1222                 }
1223         }
1224
1225         return ret;
1226 }
1227
1228 /*
1229  * If the device has been passed off to us in an enabled state, just clear
1230  * the enabled bit.  The spec says we should set the 'shutdown notification
1231  * bits', but doing so may cause the device to complete commands to the
1232  * admin queue ... and we don't know what memory that might be pointing at!
1233  */
1234 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1235 {
1236         int ret;
1237
1238         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1239         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1240
1241         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1242         if (ret)
1243                 return ret;
1244
1245         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1246                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1247
1248         return nvme_wait_ready(ctrl, cap, false);
1249 }
1250 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1251
1252 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1253 {
1254         /*
1255          * Default to a 4K page size, with the intention to update this
1256          * path in the future to accomodate architectures with differing
1257          * kernel and IO page sizes.
1258          */
1259         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1260         int ret;
1261
1262         if (page_shift < dev_page_min) {
1263                 dev_err(ctrl->device,
1264                         "Minimum device page size %u too large for host (%u)\n",
1265                         1 << dev_page_min, 1 << page_shift);
1266                 return -ENODEV;
1267         }
1268
1269         ctrl->page_size = 1 << page_shift;
1270
1271         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1272         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1273         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1274         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1275         ctrl->ctrl_config |= NVME_CC_ENABLE;
1276
1277         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1278         if (ret)
1279                 return ret;
1280         return nvme_wait_ready(ctrl, cap, true);
1281 }
1282 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1283
1284 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1285 {
1286         unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1287         u32 csts;
1288         int ret;
1289
1290         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1291         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1292
1293         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1294         if (ret)
1295                 return ret;
1296
1297         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1298                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1299                         break;
1300
1301                 msleep(100);
1302                 if (fatal_signal_pending(current))
1303                         return -EINTR;
1304                 if (time_after(jiffies, timeout)) {
1305                         dev_err(ctrl->device,
1306                                 "Device shutdown incomplete; abort shutdown\n");
1307                         return -ENODEV;
1308                 }
1309         }
1310
1311         return ret;
1312 }
1313 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1314
1315 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1316                 struct request_queue *q)
1317 {
1318         bool vwc = false;
1319
1320         if (ctrl->max_hw_sectors) {
1321                 u32 max_segments =
1322                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1323
1324                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1325                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1326         }
1327         if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1328                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1329         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1330         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1331                 vwc = true;
1332         blk_queue_write_cache(q, vwc, vwc);
1333 }
1334
1335 static void nvme_configure_apst(struct nvme_ctrl *ctrl)
1336 {
1337         /*
1338          * APST (Autonomous Power State Transition) lets us program a
1339          * table of power state transitions that the controller will
1340          * perform automatically.  We configure it with a simple
1341          * heuristic: we are willing to spend at most 2% of the time
1342          * transitioning between power states.  Therefore, when running
1343          * in any given state, we will enter the next lower-power
1344          * non-operational state after waiting 50 * (enlat + exlat)
1345          * microseconds, as long as that state's exit latency is under
1346          * the requested maximum latency.
1347          *
1348          * We will not autonomously enter any non-operational state for
1349          * which the total latency exceeds ps_max_latency_us.  Users
1350          * can set ps_max_latency_us to zero to turn off APST.
1351          */
1352
1353         unsigned apste;
1354         struct nvme_feat_auto_pst *table;
1355         u64 max_lat_us = 0;
1356         int max_ps = -1;
1357         int ret;
1358
1359         /*
1360          * If APST isn't supported or if we haven't been initialized yet,
1361          * then don't do anything.
1362          */
1363         if (!ctrl->apsta)
1364                 return;
1365
1366         if (ctrl->npss > 31) {
1367                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1368                 return;
1369         }
1370
1371         table = kzalloc(sizeof(*table), GFP_KERNEL);
1372         if (!table)
1373                 return;
1374
1375         if (ctrl->ps_max_latency_us == 0) {
1376                 /* Turn off APST. */
1377                 apste = 0;
1378                 dev_dbg(ctrl->device, "APST disabled\n");
1379         } else {
1380                 __le64 target = cpu_to_le64(0);
1381                 int state;
1382
1383                 /*
1384                  * Walk through all states from lowest- to highest-power.
1385                  * According to the spec, lower-numbered states use more
1386                  * power.  NPSS, despite the name, is the index of the
1387                  * lowest-power state, not the number of states.
1388                  */
1389                 for (state = (int)ctrl->npss; state >= 0; state--) {
1390                         u64 total_latency_us, exit_latency_us, transition_ms;
1391
1392                         if (target)
1393                                 table->entries[state] = target;
1394
1395                         /*
1396                          * Don't allow transitions to the deepest state
1397                          * if it's quirked off.
1398                          */
1399                         if (state == ctrl->npss &&
1400                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1401                                 continue;
1402
1403                         /*
1404                          * Is this state a useful non-operational state for
1405                          * higher-power states to autonomously transition to?
1406                          */
1407                         if (!(ctrl->psd[state].flags &
1408                               NVME_PS_FLAGS_NON_OP_STATE))
1409                                 continue;
1410
1411                         exit_latency_us =
1412                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1413                         if (exit_latency_us > ctrl->ps_max_latency_us)
1414                                 continue;
1415
1416                         total_latency_us =
1417                                 exit_latency_us +
1418                                 le32_to_cpu(ctrl->psd[state].entry_lat);
1419
1420                         /*
1421                          * This state is good.  Use it as the APST idle
1422                          * target for higher power states.
1423                          */
1424                         transition_ms = total_latency_us + 19;
1425                         do_div(transition_ms, 20);
1426                         if (transition_ms > (1 << 24) - 1)
1427                                 transition_ms = (1 << 24) - 1;
1428
1429                         target = cpu_to_le64((state << 3) |
1430                                              (transition_ms << 8));
1431
1432                         if (max_ps == -1)
1433                                 max_ps = state;
1434
1435                         if (total_latency_us > max_lat_us)
1436                                 max_lat_us = total_latency_us;
1437                 }
1438
1439                 apste = 1;
1440
1441                 if (max_ps == -1) {
1442                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1443                 } else {
1444                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1445                                 max_ps, max_lat_us, (int)sizeof(*table), table);
1446                 }
1447         }
1448
1449         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1450                                 table, sizeof(*table), NULL);
1451         if (ret)
1452                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1453
1454         kfree(table);
1455 }
1456
1457 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1458 {
1459         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1460         u64 latency;
1461
1462         switch (val) {
1463         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1464         case PM_QOS_LATENCY_ANY:
1465                 latency = U64_MAX;
1466                 break;
1467
1468         default:
1469                 latency = val;
1470         }
1471
1472         if (ctrl->ps_max_latency_us != latency) {
1473                 ctrl->ps_max_latency_us = latency;
1474                 nvme_configure_apst(ctrl);
1475         }
1476 }
1477
1478 struct nvme_core_quirk_entry {
1479         /*
1480          * NVMe model and firmware strings are padded with spaces.  For
1481          * simplicity, strings in the quirk table are padded with NULLs
1482          * instead.
1483          */
1484         u16 vid;
1485         const char *mn;
1486         const char *fr;
1487         unsigned long quirks;
1488 };
1489
1490 static const struct nvme_core_quirk_entry core_quirks[] = {
1491         {
1492                 /*
1493                  * This Toshiba device seems to die using any APST states.  See:
1494                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1495                  */
1496                 .vid = 0x1179,
1497                 .mn = "THNSF5256GPUK TOSHIBA",
1498                 .quirks = NVME_QUIRK_NO_APST,
1499         }
1500 };
1501
1502 /* match is null-terminated but idstr is space-padded. */
1503 static bool string_matches(const char *idstr, const char *match, size_t len)
1504 {
1505         size_t matchlen;
1506
1507         if (!match)
1508                 return true;
1509
1510         matchlen = strlen(match);
1511         WARN_ON_ONCE(matchlen > len);
1512
1513         if (memcmp(idstr, match, matchlen))
1514                 return false;
1515
1516         for (; matchlen < len; matchlen++)
1517                 if (idstr[matchlen] != ' ')
1518                         return false;
1519
1520         return true;
1521 }
1522
1523 static bool quirk_matches(const struct nvme_id_ctrl *id,
1524                           const struct nvme_core_quirk_entry *q)
1525 {
1526         return q->vid == le16_to_cpu(id->vid) &&
1527                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1528                 string_matches(id->fr, q->fr, sizeof(id->fr));
1529 }
1530
1531 /*
1532  * Initialize the cached copies of the Identify data and various controller
1533  * register in our nvme_ctrl structure.  This should be called as soon as
1534  * the admin queue is fully up and running.
1535  */
1536 int nvme_init_identify(struct nvme_ctrl *ctrl)
1537 {
1538         struct nvme_id_ctrl *id;
1539         u64 cap;
1540         int ret, page_shift;
1541         u32 max_hw_sectors;
1542         u8 prev_apsta;
1543
1544         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1545         if (ret) {
1546                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1547                 return ret;
1548         }
1549
1550         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1551         if (ret) {
1552                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1553                 return ret;
1554         }
1555         page_shift = NVME_CAP_MPSMIN(cap) + 12;
1556
1557         if (ctrl->vs >= NVME_VS(1, 1, 0))
1558                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1559
1560         ret = nvme_identify_ctrl(ctrl, &id);
1561         if (ret) {
1562                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1563                 return -EIO;
1564         }
1565
1566         if (!ctrl->identified) {
1567                 /*
1568                  * Check for quirks.  Quirk can depend on firmware version,
1569                  * so, in principle, the set of quirks present can change
1570                  * across a reset.  As a possible future enhancement, we
1571                  * could re-scan for quirks every time we reinitialize
1572                  * the device, but we'd have to make sure that the driver
1573                  * behaves intelligently if the quirks change.
1574                  */
1575
1576                 int i;
1577
1578                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
1579                         if (quirk_matches(id, &core_quirks[i]))
1580                                 ctrl->quirks |= core_quirks[i].quirks;
1581                 }
1582         }
1583
1584         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
1585                 dev_warn(ctrl->dev, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
1586                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
1587         }
1588
1589         ctrl->oacs = le16_to_cpu(id->oacs);
1590         ctrl->vid = le16_to_cpu(id->vid);
1591         ctrl->oncs = le16_to_cpup(&id->oncs);
1592         atomic_set(&ctrl->abort_limit, id->acl + 1);
1593         ctrl->vwc = id->vwc;
1594         ctrl->cntlid = le16_to_cpup(&id->cntlid);
1595         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1596         memcpy(ctrl->model, id->mn, sizeof(id->mn));
1597         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1598         if (id->mdts)
1599                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1600         else
1601                 max_hw_sectors = UINT_MAX;
1602         ctrl->max_hw_sectors =
1603                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1604
1605         nvme_set_queue_limits(ctrl, ctrl->admin_q);
1606         ctrl->sgls = le32_to_cpu(id->sgls);
1607         ctrl->kas = le16_to_cpu(id->kas);
1608
1609         ctrl->npss = id->npss;
1610         prev_apsta = ctrl->apsta;
1611         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
1612                 if (force_apst && id->apsta) {
1613                         dev_warn(ctrl->dev, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
1614                         ctrl->apsta = 1;
1615                 } else {
1616                         ctrl->apsta = 0;
1617                 }
1618         } else {
1619                 ctrl->apsta = id->apsta;
1620         }
1621         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
1622
1623         if (ctrl->ops->flags & NVME_F_FABRICS) {
1624                 ctrl->icdoff = le16_to_cpu(id->icdoff);
1625                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1626                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1627                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1628
1629                 /*
1630                  * In fabrics we need to verify the cntlid matches the
1631                  * admin connect
1632                  */
1633                 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1634                         ret = -EINVAL;
1635
1636                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1637                         dev_err(ctrl->dev,
1638                                 "keep-alive support is mandatory for fabrics\n");
1639                         ret = -EINVAL;
1640                 }
1641         } else {
1642                 ctrl->cntlid = le16_to_cpu(id->cntlid);
1643         }
1644
1645         kfree(id);
1646
1647         if (ctrl->apsta && !prev_apsta)
1648                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
1649         else if (!ctrl->apsta && prev_apsta)
1650                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
1651
1652         nvme_configure_apst(ctrl);
1653
1654         ctrl->identified = true;
1655
1656         return ret;
1657 }
1658 EXPORT_SYMBOL_GPL(nvme_init_identify);
1659
1660 static int nvme_dev_open(struct inode *inode, struct file *file)
1661 {
1662         struct nvme_ctrl *ctrl;
1663         int instance = iminor(inode);
1664         int ret = -ENODEV;
1665
1666         spin_lock(&dev_list_lock);
1667         list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1668                 if (ctrl->instance != instance)
1669                         continue;
1670
1671                 if (!ctrl->admin_q) {
1672                         ret = -EWOULDBLOCK;
1673                         break;
1674                 }
1675                 if (!kref_get_unless_zero(&ctrl->kref))
1676                         break;
1677                 file->private_data = ctrl;
1678                 ret = 0;
1679                 break;
1680         }
1681         spin_unlock(&dev_list_lock);
1682
1683         return ret;
1684 }
1685
1686 static int nvme_dev_release(struct inode *inode, struct file *file)
1687 {
1688         nvme_put_ctrl(file->private_data);
1689         return 0;
1690 }
1691
1692 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1693 {
1694         struct nvme_ns *ns;
1695         int ret;
1696
1697         mutex_lock(&ctrl->namespaces_mutex);
1698         if (list_empty(&ctrl->namespaces)) {
1699                 ret = -ENOTTY;
1700                 goto out_unlock;
1701         }
1702
1703         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1704         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1705                 dev_warn(ctrl->device,
1706                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1707                 ret = -EINVAL;
1708                 goto out_unlock;
1709         }
1710
1711         dev_warn(ctrl->device,
1712                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1713         kref_get(&ns->kref);
1714         mutex_unlock(&ctrl->namespaces_mutex);
1715
1716         ret = nvme_user_cmd(ctrl, ns, argp);
1717         nvme_put_ns(ns);
1718         return ret;
1719
1720 out_unlock:
1721         mutex_unlock(&ctrl->namespaces_mutex);
1722         return ret;
1723 }
1724
1725 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1726                 unsigned long arg)
1727 {
1728         struct nvme_ctrl *ctrl = file->private_data;
1729         void __user *argp = (void __user *)arg;
1730
1731         switch (cmd) {
1732         case NVME_IOCTL_ADMIN_CMD:
1733                 return nvme_user_cmd(ctrl, NULL, argp);
1734         case NVME_IOCTL_IO_CMD:
1735                 return nvme_dev_user_cmd(ctrl, argp);
1736         case NVME_IOCTL_RESET:
1737                 dev_warn(ctrl->device, "resetting controller\n");
1738                 return ctrl->ops->reset_ctrl(ctrl);
1739         case NVME_IOCTL_SUBSYS_RESET:
1740                 return nvme_reset_subsystem(ctrl);
1741         case NVME_IOCTL_RESCAN:
1742                 nvme_queue_scan(ctrl);
1743                 return 0;
1744         default:
1745                 return -ENOTTY;
1746         }
1747 }
1748
1749 static const struct file_operations nvme_dev_fops = {
1750         .owner          = THIS_MODULE,
1751         .open           = nvme_dev_open,
1752         .release        = nvme_dev_release,
1753         .unlocked_ioctl = nvme_dev_ioctl,
1754         .compat_ioctl   = nvme_dev_ioctl,
1755 };
1756
1757 static ssize_t nvme_sysfs_reset(struct device *dev,
1758                                 struct device_attribute *attr, const char *buf,
1759                                 size_t count)
1760 {
1761         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1762         int ret;
1763
1764         ret = ctrl->ops->reset_ctrl(ctrl);
1765         if (ret < 0)
1766                 return ret;
1767         return count;
1768 }
1769 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1770
1771 static ssize_t nvme_sysfs_rescan(struct device *dev,
1772                                 struct device_attribute *attr, const char *buf,
1773                                 size_t count)
1774 {
1775         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1776
1777         nvme_queue_scan(ctrl);
1778         return count;
1779 }
1780 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1781
1782 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1783                                                                 char *buf)
1784 {
1785         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1786         struct nvme_ctrl *ctrl = ns->ctrl;
1787         int serial_len = sizeof(ctrl->serial);
1788         int model_len = sizeof(ctrl->model);
1789
1790         if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1791                 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1792
1793         if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1794                 return sprintf(buf, "eui.%8phN\n", ns->eui);
1795
1796         while (ctrl->serial[serial_len - 1] == ' ')
1797                 serial_len--;
1798         while (ctrl->model[model_len - 1] == ' ')
1799                 model_len--;
1800
1801         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1802                 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1803 }
1804 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1805
1806 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1807                                                                 char *buf)
1808 {
1809         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1810         return sprintf(buf, "%pU\n", ns->uuid);
1811 }
1812 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1813
1814 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1815                                                                 char *buf)
1816 {
1817         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1818         return sprintf(buf, "%8phd\n", ns->eui);
1819 }
1820 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1821
1822 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1823                                                                 char *buf)
1824 {
1825         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1826         return sprintf(buf, "%d\n", ns->ns_id);
1827 }
1828 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1829
1830 static struct attribute *nvme_ns_attrs[] = {
1831         &dev_attr_wwid.attr,
1832         &dev_attr_uuid.attr,
1833         &dev_attr_eui.attr,
1834         &dev_attr_nsid.attr,
1835         NULL,
1836 };
1837
1838 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1839                 struct attribute *a, int n)
1840 {
1841         struct device *dev = container_of(kobj, struct device, kobj);
1842         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1843
1844         if (a == &dev_attr_uuid.attr) {
1845                 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1846                         return 0;
1847         }
1848         if (a == &dev_attr_eui.attr) {
1849                 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1850                         return 0;
1851         }
1852         return a->mode;
1853 }
1854
1855 static const struct attribute_group nvme_ns_attr_group = {
1856         .attrs          = nvme_ns_attrs,
1857         .is_visible     = nvme_ns_attrs_are_visible,
1858 };
1859
1860 #define nvme_show_str_function(field)                                           \
1861 static ssize_t  field##_show(struct device *dev,                                \
1862                             struct device_attribute *attr, char *buf)           \
1863 {                                                                               \
1864         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1865         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);   \
1866 }                                                                               \
1867 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1868
1869 #define nvme_show_int_function(field)                                           \
1870 static ssize_t  field##_show(struct device *dev,                                \
1871                             struct device_attribute *attr, char *buf)           \
1872 {                                                                               \
1873         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1874         return sprintf(buf, "%d\n", ctrl->field);       \
1875 }                                                                               \
1876 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1877
1878 nvme_show_str_function(model);
1879 nvme_show_str_function(serial);
1880 nvme_show_str_function(firmware_rev);
1881 nvme_show_int_function(cntlid);
1882
1883 static ssize_t nvme_sysfs_delete(struct device *dev,
1884                                 struct device_attribute *attr, const char *buf,
1885                                 size_t count)
1886 {
1887         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1888
1889         if (device_remove_file_self(dev, attr))
1890                 ctrl->ops->delete_ctrl(ctrl);
1891         return count;
1892 }
1893 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1894
1895 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1896                                          struct device_attribute *attr,
1897                                          char *buf)
1898 {
1899         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1900
1901         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1902 }
1903 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1904
1905 static ssize_t nvme_sysfs_show_state(struct device *dev,
1906                                      struct device_attribute *attr,
1907                                      char *buf)
1908 {
1909         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1910         static const char *const state_name[] = {
1911                 [NVME_CTRL_NEW]         = "new",
1912                 [NVME_CTRL_LIVE]        = "live",
1913                 [NVME_CTRL_RESETTING]   = "resetting",
1914                 [NVME_CTRL_RECONNECTING]= "reconnecting",
1915                 [NVME_CTRL_DELETING]    = "deleting",
1916                 [NVME_CTRL_DEAD]        = "dead",
1917         };
1918
1919         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
1920             state_name[ctrl->state])
1921                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
1922
1923         return sprintf(buf, "unknown state\n");
1924 }
1925
1926 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
1927
1928 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1929                                          struct device_attribute *attr,
1930                                          char *buf)
1931 {
1932         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1933
1934         return snprintf(buf, PAGE_SIZE, "%s\n",
1935                         ctrl->ops->get_subsysnqn(ctrl));
1936 }
1937 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1938
1939 static ssize_t nvme_sysfs_show_address(struct device *dev,
1940                                          struct device_attribute *attr,
1941                                          char *buf)
1942 {
1943         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1944
1945         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1946 }
1947 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1948
1949 static struct attribute *nvme_dev_attrs[] = {
1950         &dev_attr_reset_controller.attr,
1951         &dev_attr_rescan_controller.attr,
1952         &dev_attr_model.attr,
1953         &dev_attr_serial.attr,
1954         &dev_attr_firmware_rev.attr,
1955         &dev_attr_cntlid.attr,
1956         &dev_attr_delete_controller.attr,
1957         &dev_attr_transport.attr,
1958         &dev_attr_subsysnqn.attr,
1959         &dev_attr_address.attr,
1960         &dev_attr_state.attr,
1961         NULL
1962 };
1963
1964 #define CHECK_ATTR(ctrl, a, name)               \
1965         if ((a) == &dev_attr_##name.attr &&     \
1966             !(ctrl)->ops->get_##name)           \
1967                 return 0
1968
1969 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1970                 struct attribute *a, int n)
1971 {
1972         struct device *dev = container_of(kobj, struct device, kobj);
1973         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1974
1975         if (a == &dev_attr_delete_controller.attr) {
1976                 if (!ctrl->ops->delete_ctrl)
1977                         return 0;
1978         }
1979
1980         CHECK_ATTR(ctrl, a, subsysnqn);
1981         CHECK_ATTR(ctrl, a, address);
1982
1983         return a->mode;
1984 }
1985
1986 static struct attribute_group nvme_dev_attrs_group = {
1987         .attrs          = nvme_dev_attrs,
1988         .is_visible     = nvme_dev_attrs_are_visible,
1989 };
1990
1991 static const struct attribute_group *nvme_dev_attr_groups[] = {
1992         &nvme_dev_attrs_group,
1993         NULL,
1994 };
1995
1996 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1997 {
1998         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1999         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2000
2001         return nsa->ns_id - nsb->ns_id;
2002 }
2003
2004 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2005 {
2006         struct nvme_ns *ns, *ret = NULL;
2007
2008         mutex_lock(&ctrl->namespaces_mutex);
2009         list_for_each_entry(ns, &ctrl->namespaces, list) {
2010                 if (ns->ns_id == nsid) {
2011                         kref_get(&ns->kref);
2012                         ret = ns;
2013                         break;
2014                 }
2015                 if (ns->ns_id > nsid)
2016                         break;
2017         }
2018         mutex_unlock(&ctrl->namespaces_mutex);
2019         return ret;
2020 }
2021
2022 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2023 {
2024         struct nvme_ns *ns;
2025         struct gendisk *disk;
2026         struct nvme_id_ns *id;
2027         char disk_name[DISK_NAME_LEN];
2028         int node = dev_to_node(ctrl->dev);
2029
2030         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2031         if (!ns)
2032                 return;
2033
2034         ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
2035         if (ns->instance < 0)
2036                 goto out_free_ns;
2037
2038         ns->queue = blk_mq_init_queue(ctrl->tagset);
2039         if (IS_ERR(ns->queue))
2040                 goto out_release_instance;
2041         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2042         ns->queue->queuedata = ns;
2043         ns->ctrl = ctrl;
2044
2045         kref_init(&ns->kref);
2046         ns->ns_id = nsid;
2047         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2048
2049         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2050         nvme_set_queue_limits(ctrl, ns->queue);
2051
2052         sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
2053
2054         if (nvme_revalidate_ns(ns, &id))
2055                 goto out_free_queue;
2056
2057         if (nvme_nvm_ns_supported(ns, id) &&
2058                                 nvme_nvm_register(ns, disk_name, node)) {
2059                 dev_warn(ctrl->dev, "%s: LightNVM init failure\n", __func__);
2060                 goto out_free_id;
2061         }
2062
2063         disk = alloc_disk_node(0, node);
2064         if (!disk)
2065                 goto out_free_id;
2066
2067         disk->fops = &nvme_fops;
2068         disk->private_data = ns;
2069         disk->queue = ns->queue;
2070         disk->flags = GENHD_FL_EXT_DEVT;
2071         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
2072         ns->disk = disk;
2073
2074         __nvme_revalidate_disk(disk, id);
2075
2076         mutex_lock(&ctrl->namespaces_mutex);
2077         list_add_tail(&ns->list, &ctrl->namespaces);
2078         mutex_unlock(&ctrl->namespaces_mutex);
2079
2080         kref_get(&ctrl->kref);
2081
2082         kfree(id);
2083
2084         device_add_disk(ctrl->device, ns->disk);
2085         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
2086                                         &nvme_ns_attr_group))
2087                 pr_warn("%s: failed to create sysfs group for identification\n",
2088                         ns->disk->disk_name);
2089         if (ns->ndev && nvme_nvm_register_sysfs(ns))
2090                 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
2091                         ns->disk->disk_name);
2092         return;
2093  out_free_id:
2094         kfree(id);
2095  out_free_queue:
2096         blk_cleanup_queue(ns->queue);
2097  out_release_instance:
2098         ida_simple_remove(&ctrl->ns_ida, ns->instance);
2099  out_free_ns:
2100         kfree(ns);
2101 }
2102
2103 static void nvme_ns_remove(struct nvme_ns *ns)
2104 {
2105         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
2106                 return;
2107
2108         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
2109                 if (blk_get_integrity(ns->disk))
2110                         blk_integrity_unregister(ns->disk);
2111                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
2112                                         &nvme_ns_attr_group);
2113                 if (ns->ndev)
2114                         nvme_nvm_unregister_sysfs(ns);
2115                 del_gendisk(ns->disk);
2116                 blk_cleanup_queue(ns->queue);
2117         }
2118
2119         mutex_lock(&ns->ctrl->namespaces_mutex);
2120         list_del_init(&ns->list);
2121         mutex_unlock(&ns->ctrl->namespaces_mutex);
2122
2123         nvme_put_ns(ns);
2124 }
2125
2126 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2127 {
2128         struct nvme_ns *ns;
2129
2130         ns = nvme_find_get_ns(ctrl, nsid);
2131         if (ns) {
2132                 if (ns->disk && revalidate_disk(ns->disk))
2133                         nvme_ns_remove(ns);
2134                 nvme_put_ns(ns);
2135         } else
2136                 nvme_alloc_ns(ctrl, nsid);
2137 }
2138
2139 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
2140                                         unsigned nsid)
2141 {
2142         struct nvme_ns *ns, *next;
2143
2144         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
2145                 if (ns->ns_id > nsid)
2146                         nvme_ns_remove(ns);
2147         }
2148 }
2149
2150 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
2151 {
2152         struct nvme_ns *ns;
2153         __le32 *ns_list;
2154         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
2155         int ret = 0;
2156
2157         ns_list = kzalloc(0x1000, GFP_KERNEL);
2158         if (!ns_list)
2159                 return -ENOMEM;
2160
2161         for (i = 0; i < num_lists; i++) {
2162                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
2163                 if (ret)
2164                         goto free;
2165
2166                 for (j = 0; j < min(nn, 1024U); j++) {
2167                         nsid = le32_to_cpu(ns_list[j]);
2168                         if (!nsid)
2169                                 goto out;
2170
2171                         nvme_validate_ns(ctrl, nsid);
2172
2173                         while (++prev < nsid) {
2174                                 ns = nvme_find_get_ns(ctrl, prev);
2175                                 if (ns) {
2176                                         nvme_ns_remove(ns);
2177                                         nvme_put_ns(ns);
2178                                 }
2179                         }
2180                 }
2181                 nn -= j;
2182         }
2183  out:
2184         nvme_remove_invalid_namespaces(ctrl, prev);
2185  free:
2186         kfree(ns_list);
2187         return ret;
2188 }
2189
2190 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
2191 {
2192         unsigned i;
2193
2194         for (i = 1; i <= nn; i++)
2195                 nvme_validate_ns(ctrl, i);
2196
2197         nvme_remove_invalid_namespaces(ctrl, nn);
2198 }
2199
2200 static void nvme_scan_work(struct work_struct *work)
2201 {
2202         struct nvme_ctrl *ctrl =
2203                 container_of(work, struct nvme_ctrl, scan_work);
2204         struct nvme_id_ctrl *id;
2205         unsigned nn;
2206
2207         if (ctrl->state != NVME_CTRL_LIVE)
2208                 return;
2209
2210         if (nvme_identify_ctrl(ctrl, &id))
2211                 return;
2212
2213         nn = le32_to_cpu(id->nn);
2214         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
2215             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
2216                 if (!nvme_scan_ns_list(ctrl, nn))
2217                         goto done;
2218         }
2219         nvme_scan_ns_sequential(ctrl, nn);
2220  done:
2221         mutex_lock(&ctrl->namespaces_mutex);
2222         list_sort(NULL, &ctrl->namespaces, ns_cmp);
2223         mutex_unlock(&ctrl->namespaces_mutex);
2224         kfree(id);
2225 }
2226
2227 void nvme_queue_scan(struct nvme_ctrl *ctrl)
2228 {
2229         /*
2230          * Do not queue new scan work when a controller is reset during
2231          * removal.
2232          */
2233         if (ctrl->state == NVME_CTRL_LIVE)
2234                 schedule_work(&ctrl->scan_work);
2235 }
2236 EXPORT_SYMBOL_GPL(nvme_queue_scan);
2237
2238 /*
2239  * This function iterates the namespace list unlocked to allow recovery from
2240  * controller failure. It is up to the caller to ensure the namespace list is
2241  * not modified by scan work while this function is executing.
2242  */
2243 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
2244 {
2245         struct nvme_ns *ns, *next;
2246
2247         /*
2248          * The dead states indicates the controller was not gracefully
2249          * disconnected. In that case, we won't be able to flush any data while
2250          * removing the namespaces' disks; fail all the queues now to avoid
2251          * potentially having to clean up the failed sync later.
2252          */
2253         if (ctrl->state == NVME_CTRL_DEAD)
2254                 nvme_kill_queues(ctrl);
2255
2256         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
2257                 nvme_ns_remove(ns);
2258 }
2259 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
2260
2261 static void nvme_async_event_work(struct work_struct *work)
2262 {
2263         struct nvme_ctrl *ctrl =
2264                 container_of(work, struct nvme_ctrl, async_event_work);
2265
2266         spin_lock_irq(&ctrl->lock);
2267         while (ctrl->event_limit > 0) {
2268                 int aer_idx = --ctrl->event_limit;
2269
2270                 spin_unlock_irq(&ctrl->lock);
2271                 ctrl->ops->submit_async_event(ctrl, aer_idx);
2272                 spin_lock_irq(&ctrl->lock);
2273         }
2274         spin_unlock_irq(&ctrl->lock);
2275 }
2276
2277 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
2278                 union nvme_result *res)
2279 {
2280         u32 result = le32_to_cpu(res->u32);
2281         bool done = true;
2282
2283         switch (le16_to_cpu(status) >> 1) {
2284         case NVME_SC_SUCCESS:
2285                 done = false;
2286                 /*FALLTHRU*/
2287         case NVME_SC_ABORT_REQ:
2288                 ++ctrl->event_limit;
2289                 schedule_work(&ctrl->async_event_work);
2290                 break;
2291         default:
2292                 break;
2293         }
2294
2295         if (done)
2296                 return;
2297
2298         switch (result & 0xff07) {
2299         case NVME_AER_NOTICE_NS_CHANGED:
2300                 dev_info(ctrl->device, "rescanning\n");
2301                 nvme_queue_scan(ctrl);
2302                 break;
2303         default:
2304                 dev_warn(ctrl->device, "async event result %08x\n", result);
2305         }
2306 }
2307 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
2308
2309 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
2310 {
2311         ctrl->event_limit = NVME_NR_AERS;
2312         schedule_work(&ctrl->async_event_work);
2313 }
2314 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
2315
2316 static DEFINE_IDA(nvme_instance_ida);
2317
2318 static int nvme_set_instance(struct nvme_ctrl *ctrl)
2319 {
2320         int instance, error;
2321
2322         do {
2323                 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2324                         return -ENODEV;
2325
2326                 spin_lock(&dev_list_lock);
2327                 error = ida_get_new(&nvme_instance_ida, &instance);
2328                 spin_unlock(&dev_list_lock);
2329         } while (error == -EAGAIN);
2330
2331         if (error)
2332                 return -ENODEV;
2333
2334         ctrl->instance = instance;
2335         return 0;
2336 }
2337
2338 static void nvme_release_instance(struct nvme_ctrl *ctrl)
2339 {
2340         spin_lock(&dev_list_lock);
2341         ida_remove(&nvme_instance_ida, ctrl->instance);
2342         spin_unlock(&dev_list_lock);
2343 }
2344
2345 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
2346 {
2347         flush_work(&ctrl->async_event_work);
2348         flush_work(&ctrl->scan_work);
2349         nvme_remove_namespaces(ctrl);
2350
2351         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
2352
2353         spin_lock(&dev_list_lock);
2354         list_del(&ctrl->node);
2355         spin_unlock(&dev_list_lock);
2356 }
2357 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
2358
2359 static void nvme_free_ctrl(struct kref *kref)
2360 {
2361         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
2362
2363         put_device(ctrl->device);
2364         nvme_release_instance(ctrl);
2365         ida_destroy(&ctrl->ns_ida);
2366
2367         ctrl->ops->free_ctrl(ctrl);
2368 }
2369
2370 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
2371 {
2372         kref_put(&ctrl->kref, nvme_free_ctrl);
2373 }
2374 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2375
2376 /*
2377  * Initialize a NVMe controller structures.  This needs to be called during
2378  * earliest initialization so that we have the initialized structured around
2379  * during probing.
2380  */
2381 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2382                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
2383 {
2384         int ret;
2385
2386         ctrl->state = NVME_CTRL_NEW;
2387         spin_lock_init(&ctrl->lock);
2388         INIT_LIST_HEAD(&ctrl->namespaces);
2389         mutex_init(&ctrl->namespaces_mutex);
2390         kref_init(&ctrl->kref);
2391         ctrl->dev = dev;
2392         ctrl->ops = ops;
2393         ctrl->quirks = quirks;
2394         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2395         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2396
2397         ret = nvme_set_instance(ctrl);
2398         if (ret)
2399                 goto out;
2400
2401         ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2402                                 MKDEV(nvme_char_major, ctrl->instance),
2403                                 ctrl, nvme_dev_attr_groups,
2404                                 "nvme%d", ctrl->instance);
2405         if (IS_ERR(ctrl->device)) {
2406                 ret = PTR_ERR(ctrl->device);
2407                 goto out_release_instance;
2408         }
2409         get_device(ctrl->device);
2410         ida_init(&ctrl->ns_ida);
2411
2412         spin_lock(&dev_list_lock);
2413         list_add_tail(&ctrl->node, &nvme_ctrl_list);
2414         spin_unlock(&dev_list_lock);
2415
2416         /*
2417          * Initialize latency tolerance controls.  The sysfs files won't
2418          * be visible to userspace unless the device actually supports APST.
2419          */
2420         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
2421         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
2422                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
2423
2424         return 0;
2425 out_release_instance:
2426         nvme_release_instance(ctrl);
2427 out:
2428         return ret;
2429 }
2430 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2431
2432 /**
2433  * nvme_kill_queues(): Ends all namespace queues
2434  * @ctrl: the dead controller that needs to end
2435  *
2436  * Call this function when the driver determines it is unable to get the
2437  * controller in a state capable of servicing IO.
2438  */
2439 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2440 {
2441         struct nvme_ns *ns;
2442
2443         mutex_lock(&ctrl->namespaces_mutex);
2444
2445         /* Forcibly start all queues to avoid having stuck requests */
2446         blk_mq_start_hw_queues(ctrl->admin_q);
2447
2448         list_for_each_entry(ns, &ctrl->namespaces, list) {
2449                 /*
2450                  * Revalidating a dead namespace sets capacity to 0. This will
2451                  * end buffered writers dirtying pages that can't be synced.
2452                  */
2453                 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2454                         continue;
2455                 revalidate_disk(ns->disk);
2456                 blk_set_queue_dying(ns->queue);
2457
2458                 /*
2459                  * Forcibly start all queues to avoid having stuck requests.
2460                  * Note that we must ensure the queues are not stopped
2461                  * when the final removal happens.
2462                  */
2463                 blk_mq_start_hw_queues(ns->queue);
2464
2465                 /* draining requests in requeue list */
2466                 blk_mq_kick_requeue_list(ns->queue);
2467         }
2468         mutex_unlock(&ctrl->namespaces_mutex);
2469 }
2470 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2471
2472 void nvme_unfreeze(struct nvme_ctrl *ctrl)
2473 {
2474         struct nvme_ns *ns;
2475
2476         mutex_lock(&ctrl->namespaces_mutex);
2477         list_for_each_entry(ns, &ctrl->namespaces, list)
2478                 blk_mq_unfreeze_queue(ns->queue);
2479         mutex_unlock(&ctrl->namespaces_mutex);
2480 }
2481 EXPORT_SYMBOL_GPL(nvme_unfreeze);
2482
2483 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
2484 {
2485         struct nvme_ns *ns;
2486
2487         mutex_lock(&ctrl->namespaces_mutex);
2488         list_for_each_entry(ns, &ctrl->namespaces, list) {
2489                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
2490                 if (timeout <= 0)
2491                         break;
2492         }
2493         mutex_unlock(&ctrl->namespaces_mutex);
2494 }
2495 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
2496
2497 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
2498 {
2499         struct nvme_ns *ns;
2500
2501         mutex_lock(&ctrl->namespaces_mutex);
2502         list_for_each_entry(ns, &ctrl->namespaces, list)
2503                 blk_mq_freeze_queue_wait(ns->queue);
2504         mutex_unlock(&ctrl->namespaces_mutex);
2505 }
2506 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
2507
2508 void nvme_start_freeze(struct nvme_ctrl *ctrl)
2509 {
2510         struct nvme_ns *ns;
2511
2512         mutex_lock(&ctrl->namespaces_mutex);
2513         list_for_each_entry(ns, &ctrl->namespaces, list)
2514                 blk_freeze_queue_start(ns->queue);
2515         mutex_unlock(&ctrl->namespaces_mutex);
2516 }
2517 EXPORT_SYMBOL_GPL(nvme_start_freeze);
2518
2519 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2520 {
2521         struct nvme_ns *ns;
2522
2523         mutex_lock(&ctrl->namespaces_mutex);
2524         list_for_each_entry(ns, &ctrl->namespaces, list)
2525                 blk_mq_quiesce_queue(ns->queue);
2526         mutex_unlock(&ctrl->namespaces_mutex);
2527 }
2528 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2529
2530 void nvme_start_queues(struct nvme_ctrl *ctrl)
2531 {
2532         struct nvme_ns *ns;
2533
2534         mutex_lock(&ctrl->namespaces_mutex);
2535         list_for_each_entry(ns, &ctrl->namespaces, list) {
2536                 blk_mq_start_stopped_hw_queues(ns->queue, true);
2537                 blk_mq_kick_requeue_list(ns->queue);
2538         }
2539         mutex_unlock(&ctrl->namespaces_mutex);
2540 }
2541 EXPORT_SYMBOL_GPL(nvme_start_queues);
2542
2543 int __init nvme_core_init(void)
2544 {
2545         int result;
2546
2547         result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2548                                                         &nvme_dev_fops);
2549         if (result < 0)
2550                 return result;
2551         else if (result > 0)
2552                 nvme_char_major = result;
2553
2554         nvme_class = class_create(THIS_MODULE, "nvme");
2555         if (IS_ERR(nvme_class)) {
2556                 result = PTR_ERR(nvme_class);
2557                 goto unregister_chrdev;
2558         }
2559
2560         return 0;
2561
2562  unregister_chrdev:
2563         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2564         return result;
2565 }
2566
2567 void nvme_core_exit(void)
2568 {
2569         class_destroy(nvme_class);
2570         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2571 }
2572
2573 MODULE_LICENSE("GPL");
2574 MODULE_VERSION("1.0");
2575 module_init(nvme_core_init);
2576 module_exit(nvme_core_exit);