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 static 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 struct workqueue_struct *nvme_wq;
  69 EXPORT_SYMBOL_GPL(nvme_wq);
  70
  71 static LIST_HEAD(nvme_ctrl_list);
  72 static DEFINE_SPINLOCK(dev_list_lock);
  73
  74 static struct class *nvme_class;
  75
  76 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
  77 {
  78         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
  79                 return -EBUSY;
  80         if (!queue_work(nvme_wq, &ctrl->reset_work))
  81                 return -EBUSY;
  82         return 0;
  83 }
  84 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
  85
  86 static int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
  87 {
  88         int ret;
  89
  90         ret = nvme_reset_ctrl(ctrl);
  91         if (!ret)
  92                 flush_work(&ctrl->reset_work);
  93         return ret;
  94 }
  95
  96 static blk_status_t nvme_error_status(struct request *req)
  97 {
  98         switch (nvme_req(req)->status & 0x7ff) {
  99         case NVME_SC_SUCCESS:
 100                 return BLK_STS_OK;
 101         case NVME_SC_CAP_EXCEEDED:
 102                 return BLK_STS_NOSPC;
 103         case NVME_SC_ONCS_NOT_SUPPORTED:
 104                 return BLK_STS_NOTSUPP;
 105         case NVME_SC_WRITE_FAULT:
 106         case NVME_SC_READ_ERROR:
 107         case NVME_SC_UNWRITTEN_BLOCK:
 108                 return BLK_STS_MEDIUM;
 109         default:
 110                 return BLK_STS_IOERR;
 111         }
 112 }
 113
 114 static inline bool nvme_req_needs_retry(struct request *req)
 115 {
 116         if (blk_noretry_request(req))
 117                 return false;
 118         if (nvme_req(req)->status & NVME_SC_DNR)
 119                 return false;
 120         if (jiffies - req->start_time >= req->timeout)
 121                 return false;
 122         if (nvme_req(req)->retries >= nvme_max_retries)
 123                 return false;
 124         return true;
 125 }
 126
 127 void nvme_complete_rq(struct request *req)
 128 {
 129         if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) {
 130                 nvme_req(req)->retries++;
 131                 blk_mq_requeue_request(req, !blk_mq_queue_stopped(req->q));
 132                 return;
 133         }
 134
 135         blk_mq_end_request(req, nvme_error_status(req));
 136 }
 137 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 138
 139 void nvme_cancel_request(struct request *req, void *data, bool reserved)
 140 {
 141         int status;
 142
 143         if (!blk_mq_request_started(req))
 144                 return;
 145
 146         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 147                                 "Cancelling I/O %d", req->tag);
 148
 149         status = NVME_SC_ABORT_REQ;
 150         if (blk_queue_dying(req->q))
 151                 status |= NVME_SC_DNR;
 152         nvme_req(req)->status = status;
 153         blk_mq_complete_request(req);
 154
 155 }
 156 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 157
 158 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 159                 enum nvme_ctrl_state new_state)
 160 {
 161         enum nvme_ctrl_state old_state;
 162         bool changed = false;
 163
 164         spin_lock_irq(&ctrl->lock);
 165
 166         old_state = ctrl->state;
 167         switch (new_state) {
 168         case NVME_CTRL_LIVE:
 169                 switch (old_state) {
 170                 case NVME_CTRL_NEW:
 171                 case NVME_CTRL_RESETTING:
 172                 case NVME_CTRL_RECONNECTING:
 173                         changed = true;
 174                         /* FALLTHRU */
 175                 default:
 176                         break;
 177                 }
 178                 break;
 179         case NVME_CTRL_RESETTING:
 180                 switch (old_state) {
 181                 case NVME_CTRL_NEW:
 182                 case NVME_CTRL_LIVE:
 183                         changed = true;
 184                         /* FALLTHRU */
 185                 default:
 186                         break;
 187                 }
 188                 break;
 189         case NVME_CTRL_RECONNECTING:
 190                 switch (old_state) {
 191                 case NVME_CTRL_LIVE:
 192                         changed = true;
 193                         /* FALLTHRU */
 194                 default:
 195                         break;
 196                 }
 197                 break;
 198         case NVME_CTRL_DELETING:
 199                 switch (old_state) {
 200                 case NVME_CTRL_LIVE:
 201                 case NVME_CTRL_RESETTING:
 202                 case NVME_CTRL_RECONNECTING:
 203                         changed = true;
 204                         /* FALLTHRU */
 205                 default:
 206                         break;
 207                 }
 208                 break;
 209         case NVME_CTRL_DEAD:
 210                 switch (old_state) {
 211                 case NVME_CTRL_DELETING:
 212                         changed = true;
 213                         /* FALLTHRU */
 214                 default:
 215                         break;
 216                 }
 217                 break;
 218         default:
 219                 break;
 220         }
 221
 222         if (changed)
 223                 ctrl->state = new_state;
 224
 225         spin_unlock_irq(&ctrl->lock);
 226
 227         return changed;
 228 }
 229 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 230
 231 static void nvme_free_ns(struct kref *kref)
 232 {
 233         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 234
 235         if (ns->ndev)
 236                 nvme_nvm_unregister(ns);
 237
 238         if (ns->disk) {
 239                 spin_lock(&dev_list_lock);
 240                 ns->disk->private_data = NULL;
 241                 spin_unlock(&dev_list_lock);
 242         }
 243
 244         put_disk(ns->disk);
 245         ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
 246         nvme_put_ctrl(ns->ctrl);
 247         kfree(ns);
 248 }
 249
 250 static void nvme_put_ns(struct nvme_ns *ns)
 251 {
 252         kref_put(&ns->kref, nvme_free_ns);
 253 }
 254
 255 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
 256 {
 257         struct nvme_ns *ns;
 258
 259         spin_lock(&dev_list_lock);
 260         ns = disk->private_data;
 261         if (ns) {
 262                 if (!kref_get_unless_zero(&ns->kref))
 263                         goto fail;
 264                 if (!try_module_get(ns->ctrl->ops->module))
 265                         goto fail_put_ns;
 266         }
 267         spin_unlock(&dev_list_lock);
 268
 269         return ns;
 270
 271 fail_put_ns:
 272         kref_put(&ns->kref, nvme_free_ns);
 273 fail:
 274         spin_unlock(&dev_list_lock);
 275         return NULL;
 276 }
 277
 278 struct request *nvme_alloc_request(struct request_queue *q,
 279                 struct nvme_command *cmd, unsigned int flags, int qid)
 280 {
 281         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 282         struct request *req;
 283
 284         if (qid == NVME_QID_ANY) {
 285                 req = blk_mq_alloc_request(q, op, flags);
 286         } else {
 287                 req = blk_mq_alloc_request_hctx(q, op, flags,
 288                                 qid ? qid - 1 : 0);
 289         }
 290         if (IS_ERR(req))
 291                 return req;
 292
 293         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 294         nvme_req(req)->cmd = cmd;
 295
 296         return req;
 297 }
 298 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 299
 300 static inline void nvme_setup_flush(struct nvme_ns *ns,
 301                 struct nvme_command *cmnd)
 302 {
 303         memset(cmnd, 0, sizeof(*cmnd));
 304         cmnd->common.opcode = nvme_cmd_flush;
 305         cmnd->common.nsid = cpu_to_le32(ns->ns_id);
 306 }
 307
 308 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 309                 struct nvme_command *cmnd)
 310 {
 311         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 312         struct nvme_dsm_range *range;
 313         struct bio *bio;
 314
 315         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 316         if (!range)
 317                 return BLK_STS_RESOURCE;
 318
 319         __rq_for_each_bio(bio, req) {
 320                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 321                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 322
 323                 range[n].cattr = cpu_to_le32(0);
 324                 range[n].nlb = cpu_to_le32(nlb);
 325                 range[n].slba = cpu_to_le64(slba);
 326                 n++;
 327         }
 328
 329         if (WARN_ON_ONCE(n != segments)) {
 330                 kfree(range);
 331                 return BLK_STS_IOERR;
 332         }
 333
 334         memset(cmnd, 0, sizeof(*cmnd));
 335         cmnd->dsm.opcode = nvme_cmd_dsm;
 336         cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
 337         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 338         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 339
 340         req->special_vec.bv_page = virt_to_page(range);
 341         req->special_vec.bv_offset = offset_in_page(range);
 342         req->special_vec.bv_len = sizeof(*range) * segments;
 343         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 344
 345         return BLK_STS_OK;
 346 }
 347
 348 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 349                 struct request *req, struct nvme_command *cmnd)
 350 {
 351         u16 control = 0;
 352         u32 dsmgmt = 0;
 353
 354         /*
 355          * If formated with metadata, require the block layer provide a buffer
 356          * unless this namespace is formated such that the metadata can be
 357          * stripped/generated by the controller with PRACT=1.
 358          */
 359         if (ns && ns->ms && (!ns->pi_type || ns->ms != 8) &&
 360             !blk_integrity_rq(req) && !blk_rq_is_passthrough(req))
 361                 return BLK_STS_NOTSUPP;
 362
 363         if (req->cmd_flags & REQ_FUA)
 364                 control |= NVME_RW_FUA;
 365         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 366                 control |= NVME_RW_LR;
 367
 368         if (req->cmd_flags & REQ_RAHEAD)
 369                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 370
 371         memset(cmnd, 0, sizeof(*cmnd));
 372         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 373         cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
 374         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 375         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 376
 377         if (ns->ms) {
 378                 switch (ns->pi_type) {
 379                 case NVME_NS_DPS_PI_TYPE3:
 380                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 381                         break;
 382                 case NVME_NS_DPS_PI_TYPE1:
 383                 case NVME_NS_DPS_PI_TYPE2:
 384                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 385                                         NVME_RW_PRINFO_PRCHK_REF;
 386                         cmnd->rw.reftag = cpu_to_le32(
 387                                         nvme_block_nr(ns, blk_rq_pos(req)));
 388                         break;
 389                 }
 390                 if (!blk_integrity_rq(req))
 391                         control |= NVME_RW_PRINFO_PRACT;
 392         }
 393
 394         cmnd->rw.control = cpu_to_le16(control);
 395         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 396         return 0;
 397 }
 398
 399 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 400                 struct nvme_command *cmd)
 401 {
 402         blk_status_t ret = BLK_STS_OK;
 403
 404         if (!(req->rq_flags & RQF_DONTPREP)) {
 405                 nvme_req(req)->retries = 0;
 406                 nvme_req(req)->flags = 0;
 407                 req->rq_flags |= RQF_DONTPREP;
 408         }
 409
 410         switch (req_op(req)) {
 411         case REQ_OP_DRV_IN:
 412         case REQ_OP_DRV_OUT:
 413                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 414                 break;
 415         case REQ_OP_FLUSH:
 416                 nvme_setup_flush(ns, cmd);
 417                 break;
 418         case REQ_OP_WRITE_ZEROES:
 419                 /* currently only aliased to deallocate for a few ctrls: */
 420         case REQ_OP_DISCARD:
 421                 ret = nvme_setup_discard(ns, req, cmd);
 422                 break;
 423         case REQ_OP_READ:
 424         case REQ_OP_WRITE:
 425                 ret = nvme_setup_rw(ns, req, cmd);
 426                 break;
 427         default:
 428                 WARN_ON_ONCE(1);
 429                 return BLK_STS_IOERR;
 430         }
 431
 432         cmd->common.command_id = req->tag;
 433         return ret;
 434 }
 435 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 436
 437 /*
 438  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 439  * if the result is positive, it's an NVM Express status code
 440  */
 441 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 442                 union nvme_result *result, void *buffer, unsigned bufflen,
 443                 unsigned timeout, int qid, int at_head, int flags)
 444 {
 445         struct request *req;
 446         int ret;
 447
 448         req = nvme_alloc_request(q, cmd, flags, qid);
 449         if (IS_ERR(req))
 450                 return PTR_ERR(req);
 451
 452         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 453
 454         if (buffer && bufflen) {
 455                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 456                 if (ret)
 457                         goto out;
 458         }
 459
 460         blk_execute_rq(req->q, NULL, req, at_head);
 461         if (result)
 462                 *result = nvme_req(req)->result;
 463         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 464                 ret = -EINTR;
 465         else
 466                 ret = nvme_req(req)->status;
 467  out:
 468         blk_mq_free_request(req);
 469         return ret;
 470 }
 471 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 472
 473 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 474                 void *buffer, unsigned bufflen)
 475 {
 476         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 477                         NVME_QID_ANY, 0, 0);
 478 }
 479 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 480
 481 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 482                 void __user *ubuffer, unsigned bufflen,
 483                 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
 484                 u32 *result, unsigned timeout)
 485 {
 486         bool write = nvme_is_write(cmd);
 487         struct nvme_ns *ns = q->queuedata;
 488         struct gendisk *disk = ns ? ns->disk : NULL;
 489         struct request *req;
 490         struct bio *bio = NULL;
 491         void *meta = NULL;
 492         int ret;
 493
 494         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 495         if (IS_ERR(req))
 496                 return PTR_ERR(req);
 497
 498         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 499
 500         if (ubuffer && bufflen) {
 501                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 502                                 GFP_KERNEL);
 503                 if (ret)
 504                         goto out;
 505                 bio = req->bio;
 506
 507                 if (!disk)
 508                         goto submit;
 509                 bio->bi_bdev = bdget_disk(disk, 0);
 510                 if (!bio->bi_bdev) {
 511                         ret = -ENODEV;
 512                         goto out_unmap;
 513                 }
 514
 515                 if (meta_buffer && meta_len) {
 516                         struct bio_integrity_payload *bip;
 517
 518                         meta = kmalloc(meta_len, GFP_KERNEL);
 519                         if (!meta) {
 520                                 ret = -ENOMEM;
 521                                 goto out_unmap;
 522                         }
 523
 524                         if (write) {
 525                                 if (copy_from_user(meta, meta_buffer,
 526                                                 meta_len)) {
 527                                         ret = -EFAULT;
 528                                         goto out_free_meta;
 529                                 }
 530                         }
 531
 532                         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 533                         if (IS_ERR(bip)) {
 534                                 ret = PTR_ERR(bip);
 535                                 goto out_free_meta;
 536                         }
 537
 538                         bip->bip_iter.bi_size = meta_len;
 539                         bip->bip_iter.bi_sector = meta_seed;
 540
 541                         ret = bio_integrity_add_page(bio, virt_to_page(meta),
 542                                         meta_len, offset_in_page(meta));
 543                         if (ret != meta_len) {
 544                                 ret = -ENOMEM;
 545                                 goto out_free_meta;
 546                         }
 547                 }
 548         }
 549  submit:
 550         blk_execute_rq(req->q, disk, req, 0);
 551         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 552                 ret = -EINTR;
 553         else
 554                 ret = nvme_req(req)->status;
 555         if (result)
 556                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 557         if (meta && !ret && !write) {
 558                 if (copy_to_user(meta_buffer, meta, meta_len))
 559                         ret = -EFAULT;
 560         }
 561  out_free_meta:
 562         kfree(meta);
 563  out_unmap:
 564         if (bio) {
 565                 if (disk && bio->bi_bdev)
 566                         bdput(bio->bi_bdev);
 567                 blk_rq_unmap_user(bio);
 568         }
 569  out:
 570         blk_mq_free_request(req);
 571         return ret;
 572 }
 573
 574 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 575                 void __user *ubuffer, unsigned bufflen, u32 *result,
 576                 unsigned timeout)
 577 {
 578         return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
 579                         result, timeout);
 580 }
 581
 582 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 583 {
 584         struct nvme_ctrl *ctrl = rq->end_io_data;
 585
 586         blk_mq_free_request(rq);
 587
 588         if (status) {
 589                 dev_err(ctrl->device,
 590                         "failed nvme_keep_alive_end_io error=%d\n",
 591                                 status);
 592                 return;
 593         }
 594
 595         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 596 }
 597
 598 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 599 {
 600         struct nvme_command c;
 601         struct request *rq;
 602
 603         memset(&c, 0, sizeof(c));
 604         c.common.opcode = nvme_admin_keep_alive;
 605
 606         rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
 607                         NVME_QID_ANY);
 608         if (IS_ERR(rq))
 609                 return PTR_ERR(rq);
 610
 611         rq->timeout = ctrl->kato * HZ;
 612         rq->end_io_data = ctrl;
 613
 614         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 615
 616         return 0;
 617 }
 618
 619 static void nvme_keep_alive_work(struct work_struct *work)
 620 {
 621         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 622                         struct nvme_ctrl, ka_work);
 623
 624         if (nvme_keep_alive(ctrl)) {
 625                 /* allocation failure, reset the controller */
 626                 dev_err(ctrl->device, "keep-alive failed\n");
 627                 nvme_reset_ctrl_sync(ctrl);
 628                 return;
 629         }
 630 }
 631
 632 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 633 {
 634         if (unlikely(ctrl->kato == 0))
 635                 return;
 636
 637         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
 638         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 639 }
 640 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
 641
 642 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 643 {
 644         if (unlikely(ctrl->kato == 0))
 645                 return;
 646
 647         cancel_delayed_work_sync(&ctrl->ka_work);
 648 }
 649 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 650
 651 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 652 {
 653         struct nvme_command c = { };
 654         int error;
 655
 656         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 657         c.identify.opcode = nvme_admin_identify;
 658         c.identify.cns = NVME_ID_CNS_CTRL;
 659
 660         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 661         if (!*id)
 662                 return -ENOMEM;
 663
 664         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 665                         sizeof(struct nvme_id_ctrl));
 666         if (error)
 667                 kfree(*id);
 668         return error;
 669 }
 670
 671 static int nvme_identify_ns_descs(struct nvme_ns *ns, unsigned nsid)
 672 {
 673         struct nvme_command c = { };
 674         int status;
 675         void *data;
 676         int pos;
 677         int len;
 678
 679         c.identify.opcode = nvme_admin_identify;
 680         c.identify.nsid = cpu_to_le32(nsid);
 681         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
 682
 683         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
 684         if (!data)
 685                 return -ENOMEM;
 686
 687         status = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, data,
 688                                       NVME_IDENTIFY_DATA_SIZE);
 689         if (status)
 690                 goto free_data;
 691
 692         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
 693                 struct nvme_ns_id_desc *cur = data + pos;
 694
 695                 if (cur->nidl == 0)
 696                         break;
 697
 698                 switch (cur->nidt) {
 699                 case NVME_NIDT_EUI64:
 700                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
 701                                 dev_warn(ns->ctrl->device,
 702                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
 703                                          cur->nidl);
 704                                 goto free_data;
 705                         }
 706                         len = NVME_NIDT_EUI64_LEN;
 707                         memcpy(ns->eui, data + pos + sizeof(*cur), len);
 708                         break;
 709                 case NVME_NIDT_NGUID:
 710                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
 711                                 dev_warn(ns->ctrl->device,
 712                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
 713                                          cur->nidl);
 714                                 goto free_data;
 715                         }
 716                         len = NVME_NIDT_NGUID_LEN;
 717                         memcpy(ns->nguid, data + pos + sizeof(*cur), len);
 718                         break;
 719                 case NVME_NIDT_UUID:
 720                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
 721                                 dev_warn(ns->ctrl->device,
 722                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
 723                                          cur->nidl);
 724                                 goto free_data;
 725                         }
 726                         len = NVME_NIDT_UUID_LEN;
 727                         uuid_copy(&ns->uuid, data + pos + sizeof(*cur));
 728                         break;
 729                 default:
 730                         /* Skip unnkown types */
 731                         len = cur->nidl;
 732                         break;
 733                 }
 734
 735                 len += sizeof(*cur);
 736         }
 737 free_data:
 738         kfree(data);
 739         return status;
 740 }
 741
 742 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 743 {
 744         struct nvme_command c = { };
 745
 746         c.identify.opcode = nvme_admin_identify;
 747         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
 748         c.identify.nsid = cpu_to_le32(nsid);
 749         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 750 }
 751
 752 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
 753                 struct nvme_id_ns **id)
 754 {
 755         struct nvme_command c = { };
 756         int error;
 757
 758         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 759         c.identify.opcode = nvme_admin_identify;
 760         c.identify.nsid = cpu_to_le32(nsid);
 761         c.identify.cns = NVME_ID_CNS_NS;
 762
 763         *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
 764         if (!*id)
 765                 return -ENOMEM;
 766
 767         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 768                         sizeof(struct nvme_id_ns));
 769         if (error)
 770                 kfree(*id);
 771         return error;
 772 }
 773
 774 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 775                       void *buffer, size_t buflen, u32 *result)
 776 {
 777         struct nvme_command c;
 778         union nvme_result res;
 779         int ret;
 780
 781         memset(&c, 0, sizeof(c));
 782         c.features.opcode = nvme_admin_get_features;
 783         c.features.nsid = cpu_to_le32(nsid);
 784         c.features.fid = cpu_to_le32(fid);
 785
 786         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, buffer, buflen, 0,
 787                         NVME_QID_ANY, 0, 0);
 788         if (ret >= 0 && result)
 789                 *result = le32_to_cpu(res.u32);
 790         return ret;
 791 }
 792
 793 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 794                       void *buffer, size_t buflen, u32 *result)
 795 {
 796         struct nvme_command c;
 797         union nvme_result res;
 798         int ret;
 799
 800         memset(&c, 0, sizeof(c));
 801         c.features.opcode = nvme_admin_set_features;
 802         c.features.fid = cpu_to_le32(fid);
 803         c.features.dword11 = cpu_to_le32(dword11);
 804
 805         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
 806                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
 807         if (ret >= 0 && result)
 808                 *result = le32_to_cpu(res.u32);
 809         return ret;
 810 }
 811
 812 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
 813 {
 814         struct nvme_command c = { };
 815         int error;
 816
 817         c.common.opcode = nvme_admin_get_log_page,
 818         c.common.nsid = cpu_to_le32(0xFFFFFFFF),
 819         c.common.cdw10[0] = cpu_to_le32(
 820                         (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
 821                          NVME_LOG_SMART),
 822
 823         *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
 824         if (!*log)
 825                 return -ENOMEM;
 826
 827         error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
 828                         sizeof(struct nvme_smart_log));
 829         if (error)
 830                 kfree(*log);
 831         return error;
 832 }
 833
 834 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
 835 {
 836         u32 q_count = (*count - 1) | ((*count - 1) << 16);
 837         u32 result;
 838         int status, nr_io_queues;
 839
 840         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
 841                         &result);
 842         if (status < 0)
 843                 return status;
 844
 845         /*
 846          * Degraded controllers might return an error when setting the queue
 847          * count.  We still want to be able to bring them online and offer
 848          * access to the admin queue, as that might be only way to fix them up.
 849          */
 850         if (status > 0) {
 851                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
 852                 *count = 0;
 853         } else {
 854                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
 855                 *count = min(*count, nr_io_queues);
 856         }
 857
 858         return 0;
 859 }
 860 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
 861
 862 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 863 {
 864         struct nvme_user_io io;
 865         struct nvme_command c;
 866         unsigned length, meta_len;
 867         void __user *metadata;
 868
 869         if (copy_from_user(&io, uio, sizeof(io)))
 870                 return -EFAULT;
 871         if (io.flags)
 872                 return -EINVAL;
 873
 874         switch (io.opcode) {
 875         case nvme_cmd_write:
 876         case nvme_cmd_read:
 877         case nvme_cmd_compare:
 878                 break;
 879         default:
 880                 return -EINVAL;
 881         }
 882
 883         length = (io.nblocks + 1) << ns->lba_shift;
 884         meta_len = (io.nblocks + 1) * ns->ms;
 885         metadata = (void __user *)(uintptr_t)io.metadata;
 886
 887         if (ns->ext) {
 888                 length += meta_len;
 889                 meta_len = 0;
 890         } else if (meta_len) {
 891                 if ((io.metadata & 3) || !io.metadata)
 892                         return -EINVAL;
 893         }
 894
 895         memset(&c, 0, sizeof(c));
 896         c.rw.opcode = io.opcode;
 897         c.rw.flags = io.flags;
 898         c.rw.nsid = cpu_to_le32(ns->ns_id);
 899         c.rw.slba = cpu_to_le64(io.slba);
 900         c.rw.length = cpu_to_le16(io.nblocks);
 901         c.rw.control = cpu_to_le16(io.control);
 902         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
 903         c.rw.reftag = cpu_to_le32(io.reftag);
 904         c.rw.apptag = cpu_to_le16(io.apptag);
 905         c.rw.appmask = cpu_to_le16(io.appmask);
 906
 907         return __nvme_submit_user_cmd(ns->queue, &c,
 908                         (void __user *)(uintptr_t)io.addr, length,
 909                         metadata, meta_len, io.slba, NULL, 0);
 910 }
 911
 912 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 913                         struct nvme_passthru_cmd __user *ucmd)
 914 {
 915         struct nvme_passthru_cmd cmd;
 916         struct nvme_command c;
 917         unsigned timeout = 0;
 918         int status;
 919
 920         if (!capable(CAP_SYS_ADMIN))
 921                 return -EACCES;
 922         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
 923                 return -EFAULT;
 924         if (cmd.flags)
 925                 return -EINVAL;
 926
 927         memset(&c, 0, sizeof(c));
 928         c.common.opcode = cmd.opcode;
 929         c.common.flags = cmd.flags;
 930         c.common.nsid = cpu_to_le32(cmd.nsid);
 931         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
 932         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
 933         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
 934         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
 935         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
 936         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
 937         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
 938         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
 939
 940         if (cmd.timeout_ms)
 941                 timeout = msecs_to_jiffies(cmd.timeout_ms);
 942
 943         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
 944                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
 945                         &cmd.result, timeout);
 946         if (status >= 0) {
 947                 if (put_user(cmd.result, &ucmd->result))
 948                         return -EFAULT;
 949         }
 950
 951         return status;
 952 }
 953
 954 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 955                 unsigned int cmd, unsigned long arg)
 956 {
 957         struct nvme_ns *ns = bdev->bd_disk->private_data;
 958
 959         switch (cmd) {
 960         case NVME_IOCTL_ID:
 961                 force_successful_syscall_return();
 962                 return ns->ns_id;
 963         case NVME_IOCTL_ADMIN_CMD:
 964                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
 965         case NVME_IOCTL_IO_CMD:
 966                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
 967         case NVME_IOCTL_SUBMIT_IO:
 968                 return nvme_submit_io(ns, (void __user *)arg);
 969 #ifdef CONFIG_BLK_DEV_NVME_SCSI
 970         case SG_GET_VERSION_NUM:
 971                 return nvme_sg_get_version_num((void __user *)arg);
 972         case SG_IO:
 973                 return nvme_sg_io(ns, (void __user *)arg);
 974 #endif
 975         default:
 976 #ifdef CONFIG_NVM
 977                 if (ns->ndev)
 978                         return nvme_nvm_ioctl(ns, cmd, arg);
 979 #endif
 980                 if (is_sed_ioctl(cmd))
 981                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
 982                                          (void __user *) arg);
 983                 return -ENOTTY;
 984         }
 985 }
 986
 987 #ifdef CONFIG_COMPAT
 988 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 989                         unsigned int cmd, unsigned long arg)
 990 {
 991         switch (cmd) {
 992         case SG_IO:
 993                 return -ENOIOCTLCMD;
 994         }
 995         return nvme_ioctl(bdev, mode, cmd, arg);
 996 }
 997 #else
 998 #define nvme_compat_ioctl       NULL
 999 #endif
1000
1001 static int nvme_open(struct block_device *bdev, fmode_t mode)
1002 {
1003         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
1004 }
1005
1006 static void nvme_release(struct gendisk *disk, fmode_t mode)
1007 {
1008         struct nvme_ns *ns = disk->private_data;
1009
1010         module_put(ns->ctrl->ops->module);
1011         nvme_put_ns(ns);
1012 }
1013
1014 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1015 {
1016         /* some standard values */
1017         geo->heads = 1 << 6;
1018         geo->sectors = 1 << 5;
1019         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1020         return 0;
1021 }
1022
1023 #ifdef CONFIG_BLK_DEV_INTEGRITY
1024 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
1025                 u16 bs)
1026 {
1027         struct nvme_ns *ns = disk->private_data;
1028         u16 old_ms = ns->ms;
1029         u8 pi_type = 0;
1030
1031         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1032         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1033
1034         /* PI implementation requires metadata equal t10 pi tuple size */
1035         if (ns->ms == sizeof(struct t10_pi_tuple))
1036                 pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1037
1038         if (blk_get_integrity(disk) &&
1039             (ns->pi_type != pi_type || ns->ms != old_ms ||
1040              bs != queue_logical_block_size(disk->queue) ||
1041              (ns->ms && ns->ext)))
1042                 blk_integrity_unregister(disk);
1043
1044         ns->pi_type = pi_type;
1045 }
1046
1047 static void nvme_init_integrity(struct nvme_ns *ns)
1048 {
1049         struct blk_integrity integrity;
1050
1051         memset(&integrity, 0, sizeof(integrity));
1052         switch (ns->pi_type) {
1053         case NVME_NS_DPS_PI_TYPE3:
1054                 integrity.profile = &t10_pi_type3_crc;
1055                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1056                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1057                 break;
1058         case NVME_NS_DPS_PI_TYPE1:
1059         case NVME_NS_DPS_PI_TYPE2:
1060                 integrity.profile = &t10_pi_type1_crc;
1061                 integrity.tag_size = sizeof(u16);
1062                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1063                 break;
1064         default:
1065                 integrity.profile = NULL;
1066                 break;
1067         }
1068         integrity.tuple_size = ns->ms;
1069         blk_integrity_register(ns->disk, &integrity);
1070         blk_queue_max_integrity_segments(ns->queue, 1);
1071 }
1072 #else
1073 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
1074                 u16 bs)
1075 {
1076 }
1077 static void nvme_init_integrity(struct nvme_ns *ns)
1078 {
1079 }
1080 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1081
1082 static void nvme_config_discard(struct nvme_ns *ns)
1083 {
1084         struct nvme_ctrl *ctrl = ns->ctrl;
1085         u32 logical_block_size = queue_logical_block_size(ns->queue);
1086
1087         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1088                         NVME_DSM_MAX_RANGES);
1089
1090         ns->queue->limits.discard_alignment = logical_block_size;
1091         ns->queue->limits.discard_granularity = logical_block_size;
1092         blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
1093         blk_queue_max_discard_segments(ns->queue, NVME_DSM_MAX_RANGES);
1094         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
1095
1096         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1097                 blk_queue_max_write_zeroes_sectors(ns->queue, UINT_MAX);
1098 }
1099
1100 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
1101 {
1102         if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
1103                 dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
1104                 return -ENODEV;
1105         }
1106
1107         if ((*id)->ncap == 0) {
1108                 kfree(*id);
1109                 return -ENODEV;
1110         }
1111
1112         if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
1113                 memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
1114         if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
1115                 memcpy(ns->nguid, (*id)->nguid, sizeof(ns->nguid));
1116         if (ns->ctrl->vs >= NVME_VS(1, 3, 0)) {
1117                  /* Don't treat error as fatal we potentially
1118                   * already have a NGUID or EUI-64
1119                   */
1120                 if (nvme_identify_ns_descs(ns, ns->ns_id))
1121                         dev_warn(ns->ctrl->device,
1122                                  "%s: Identify Descriptors failed\n", __func__);
1123         }
1124
1125         return 0;
1126 }
1127
1128 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1129 {
1130         struct nvme_ns *ns = disk->private_data;
1131         u16 bs;
1132
1133         /*
1134          * If identify namespace failed, use default 512 byte block size so
1135          * block layer can use before failing read/write for 0 capacity.
1136          */
1137         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1138         if (ns->lba_shift == 0)
1139                 ns->lba_shift = 9;
1140         bs = 1 << ns->lba_shift;
1141
1142         blk_mq_freeze_queue(disk->queue);
1143
1144         if (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED)
1145                 nvme_prep_integrity(disk, id, bs);
1146         blk_queue_logical_block_size(ns->queue, bs);
1147         if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
1148                 nvme_init_integrity(ns);
1149         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
1150                 set_capacity(disk, 0);
1151         else
1152                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
1153
1154         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1155                 nvme_config_discard(ns);
1156         blk_mq_unfreeze_queue(disk->queue);
1157 }
1158
1159 static int nvme_revalidate_disk(struct gendisk *disk)
1160 {
1161         struct nvme_ns *ns = disk->private_data;
1162         struct nvme_id_ns *id = NULL;
1163         int ret;
1164
1165         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1166                 set_capacity(disk, 0);
1167                 return -ENODEV;
1168         }
1169
1170         ret = nvme_revalidate_ns(ns, &id);
1171         if (ret)
1172                 return ret;
1173
1174         __nvme_revalidate_disk(disk, id);
1175         kfree(id);
1176
1177         return 0;
1178 }
1179
1180 static char nvme_pr_type(enum pr_type type)
1181 {
1182         switch (type) {
1183         case PR_WRITE_EXCLUSIVE:
1184                 return 1;
1185         case PR_EXCLUSIVE_ACCESS:
1186                 return 2;
1187         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1188                 return 3;
1189         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1190                 return 4;
1191         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1192                 return 5;
1193         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1194                 return 6;
1195         default:
1196                 return 0;
1197         }
1198 };
1199
1200 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1201                                 u64 key, u64 sa_key, u8 op)
1202 {
1203         struct nvme_ns *ns = bdev->bd_disk->private_data;
1204         struct nvme_command c;
1205         u8 data[16] = { 0, };
1206
1207         put_unaligned_le64(key, &data[0]);
1208         put_unaligned_le64(sa_key, &data[8]);
1209
1210         memset(&c, 0, sizeof(c));
1211         c.common.opcode = op;
1212         c.common.nsid = cpu_to_le32(ns->ns_id);
1213         c.common.cdw10[0] = cpu_to_le32(cdw10);
1214
1215         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1216 }
1217
1218 static int nvme_pr_register(struct block_device *bdev, u64 old,
1219                 u64 new, unsigned flags)
1220 {
1221         u32 cdw10;
1222
1223         if (flags & ~PR_FL_IGNORE_KEY)
1224                 return -EOPNOTSUPP;
1225
1226         cdw10 = old ? 2 : 0;
1227         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1228         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1229         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1230 }
1231
1232 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1233                 enum pr_type type, unsigned flags)
1234 {
1235         u32 cdw10;
1236
1237         if (flags & ~PR_FL_IGNORE_KEY)
1238                 return -EOPNOTSUPP;
1239
1240         cdw10 = nvme_pr_type(type) << 8;
1241         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1242         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1243 }
1244
1245 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1246                 enum pr_type type, bool abort)
1247 {
1248         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1249         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1250 }
1251
1252 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1253 {
1254         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1255         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1256 }
1257
1258 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1259 {
1260         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1261         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1262 }
1263
1264 static const struct pr_ops nvme_pr_ops = {
1265         .pr_register    = nvme_pr_register,
1266         .pr_reserve     = nvme_pr_reserve,
1267         .pr_release     = nvme_pr_release,
1268         .pr_preempt     = nvme_pr_preempt,
1269         .pr_clear       = nvme_pr_clear,
1270 };
1271
1272 #ifdef CONFIG_BLK_SED_OPAL
1273 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1274                 bool send)
1275 {
1276         struct nvme_ctrl *ctrl = data;
1277         struct nvme_command cmd;
1278
1279         memset(&cmd, 0, sizeof(cmd));
1280         if (send)
1281                 cmd.common.opcode = nvme_admin_security_send;
1282         else
1283                 cmd.common.opcode = nvme_admin_security_recv;
1284         cmd.common.nsid = 0;
1285         cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1286         cmd.common.cdw10[1] = cpu_to_le32(len);
1287
1288         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1289                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1290 }
1291 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1292 #endif /* CONFIG_BLK_SED_OPAL */
1293
1294 static const struct block_device_operations nvme_fops = {
1295         .owner          = THIS_MODULE,
1296         .ioctl          = nvme_ioctl,
1297         .compat_ioctl   = nvme_compat_ioctl,
1298         .open           = nvme_open,
1299         .release        = nvme_release,
1300         .getgeo         = nvme_getgeo,
1301         .revalidate_disk= nvme_revalidate_disk,
1302         .pr_ops         = &nvme_pr_ops,
1303 };
1304
1305 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1306 {
1307         unsigned long timeout =
1308                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1309         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1310         int ret;
1311
1312         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1313                 if (csts == ~0)
1314                         return -ENODEV;
1315                 if ((csts & NVME_CSTS_RDY) == bit)
1316                         break;
1317
1318                 msleep(100);
1319                 if (fatal_signal_pending(current))
1320                         return -EINTR;
1321                 if (time_after(jiffies, timeout)) {
1322                         dev_err(ctrl->device,
1323                                 "Device not ready; aborting %s\n", enabled ?
1324                                                 "initialisation" : "reset");
1325                         return -ENODEV;
1326                 }
1327         }
1328
1329         return ret;
1330 }
1331
1332 /*
1333  * If the device has been passed off to us in an enabled state, just clear
1334  * the enabled bit.  The spec says we should set the 'shutdown notification
1335  * bits', but doing so may cause the device to complete commands to the
1336  * admin queue ... and we don't know what memory that might be pointing at!
1337  */
1338 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1339 {
1340         int ret;
1341
1342         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1343         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1344
1345         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1346         if (ret)
1347                 return ret;
1348
1349         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1350                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1351
1352         return nvme_wait_ready(ctrl, cap, false);
1353 }
1354 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1355
1356 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1357 {
1358         /*
1359          * Default to a 4K page size, with the intention to update this
1360          * path in the future to accomodate architectures with differing
1361          * kernel and IO page sizes.
1362          */
1363         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1364         int ret;
1365
1366         if (page_shift < dev_page_min) {
1367                 dev_err(ctrl->device,
1368                         "Minimum device page size %u too large for host (%u)\n",
1369                         1 << dev_page_min, 1 << page_shift);
1370                 return -ENODEV;
1371         }
1372
1373         ctrl->page_size = 1 << page_shift;
1374
1375         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1376         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1377         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1378         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1379         ctrl->ctrl_config |= NVME_CC_ENABLE;
1380
1381         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1382         if (ret)
1383                 return ret;
1384         return nvme_wait_ready(ctrl, cap, true);
1385 }
1386 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1387
1388 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1389 {
1390         unsigned long timeout = jiffies + (shutdown_timeout * HZ);
1391         u32 csts;
1392         int ret;
1393
1394         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1395         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1396
1397         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1398         if (ret)
1399                 return ret;
1400
1401         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1402                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1403                         break;
1404
1405                 msleep(100);
1406                 if (fatal_signal_pending(current))
1407                         return -EINTR;
1408                 if (time_after(jiffies, timeout)) {
1409                         dev_err(ctrl->device,
1410                                 "Device shutdown incomplete; abort shutdown\n");
1411                         return -ENODEV;
1412                 }
1413         }
1414
1415         return ret;
1416 }
1417 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1418
1419 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1420                 struct request_queue *q)
1421 {
1422         bool vwc = false;
1423
1424         if (ctrl->max_hw_sectors) {
1425                 u32 max_segments =
1426                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1427
1428                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1429                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1430         }
1431         if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1432                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1433         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1434         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1435                 vwc = true;
1436         blk_queue_write_cache(q, vwc, vwc);
1437 }
1438
1439 static void nvme_configure_apst(struct nvme_ctrl *ctrl)
1440 {
1441         /*
1442          * APST (Autonomous Power State Transition) lets us program a
1443          * table of power state transitions that the controller will
1444          * perform automatically.  We configure it with a simple
1445          * heuristic: we are willing to spend at most 2% of the time
1446          * transitioning between power states.  Therefore, when running
1447          * in any given state, we will enter the next lower-power
1448          * non-operational state after waiting 50 * (enlat + exlat)
1449          * microseconds, as long as that state's exit latency is under
1450          * the requested maximum latency.
1451          *
1452          * We will not autonomously enter any non-operational state for
1453          * which the total latency exceeds ps_max_latency_us.  Users
1454          * can set ps_max_latency_us to zero to turn off APST.
1455          */
1456
1457         unsigned apste;
1458         struct nvme_feat_auto_pst *table;
1459         u64 max_lat_us = 0;
1460         int max_ps = -1;
1461         int ret;
1462
1463         /*
1464          * If APST isn't supported or if we haven't been initialized yet,
1465          * then don't do anything.
1466          */
1467         if (!ctrl->apsta)
1468                 return;
1469
1470         if (ctrl->npss > 31) {
1471                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1472                 return;
1473         }
1474
1475         table = kzalloc(sizeof(*table), GFP_KERNEL);
1476         if (!table)
1477                 return;
1478
1479         if (ctrl->ps_max_latency_us == 0) {
1480                 /* Turn off APST. */
1481                 apste = 0;
1482                 dev_dbg(ctrl->device, "APST disabled\n");
1483         } else {
1484                 __le64 target = cpu_to_le64(0);
1485                 int state;
1486
1487                 /*
1488                  * Walk through all states from lowest- to highest-power.
1489                  * According to the spec, lower-numbered states use more
1490                  * power.  NPSS, despite the name, is the index of the
1491                  * lowest-power state, not the number of states.
1492                  */
1493                 for (state = (int)ctrl->npss; state >= 0; state--) {
1494                         u64 total_latency_us, exit_latency_us, transition_ms;
1495
1496                         if (target)
1497                                 table->entries[state] = target;
1498
1499                         /*
1500                          * Don't allow transitions to the deepest state
1501                          * if it's quirked off.
1502                          */
1503                         if (state == ctrl->npss &&
1504                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1505                                 continue;
1506
1507                         /*
1508                          * Is this state a useful non-operational state for
1509                          * higher-power states to autonomously transition to?
1510                          */
1511                         if (!(ctrl->psd[state].flags &
1512                               NVME_PS_FLAGS_NON_OP_STATE))
1513                                 continue;
1514
1515                         exit_latency_us =
1516                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1517                         if (exit_latency_us > ctrl->ps_max_latency_us)
1518                                 continue;
1519
1520                         total_latency_us =
1521                                 exit_latency_us +
1522                                 le32_to_cpu(ctrl->psd[state].entry_lat);
1523
1524                         /*
1525                          * This state is good.  Use it as the APST idle
1526                          * target for higher power states.
1527                          */
1528                         transition_ms = total_latency_us + 19;
1529                         do_div(transition_ms, 20);
1530                         if (transition_ms > (1 << 24) - 1)
1531                                 transition_ms = (1 << 24) - 1;
1532
1533                         target = cpu_to_le64((state << 3) |
1534                                              (transition_ms << 8));
1535
1536                         if (max_ps == -1)
1537                                 max_ps = state;
1538
1539                         if (total_latency_us > max_lat_us)
1540                                 max_lat_us = total_latency_us;
1541                 }
1542
1543                 apste = 1;
1544
1545                 if (max_ps == -1) {
1546                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1547                 } else {
1548                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1549                                 max_ps, max_lat_us, (int)sizeof(*table), table);
1550                 }
1551         }
1552
1553         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1554                                 table, sizeof(*table), NULL);
1555         if (ret)
1556                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1557
1558         kfree(table);
1559 }
1560
1561 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1562 {
1563         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1564         u64 latency;
1565
1566         switch (val) {
1567         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1568         case PM_QOS_LATENCY_ANY:
1569                 latency = U64_MAX;
1570                 break;
1571
1572         default:
1573                 latency = val;
1574         }
1575
1576         if (ctrl->ps_max_latency_us != latency) {
1577                 ctrl->ps_max_latency_us = latency;
1578                 nvme_configure_apst(ctrl);
1579         }
1580 }
1581
1582 struct nvme_core_quirk_entry {
1583         /*
1584          * NVMe model and firmware strings are padded with spaces.  For
1585          * simplicity, strings in the quirk table are padded with NULLs
1586          * instead.
1587          */
1588         u16 vid;
1589         const char *mn;
1590         const char *fr;
1591         unsigned long quirks;
1592 };
1593
1594 static const struct nvme_core_quirk_entry core_quirks[] = {
1595         {
1596                 /*
1597                  * This Toshiba device seems to die using any APST states.  See:
1598                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1599                  */
1600                 .vid = 0x1179,
1601                 .mn = "THNSF5256GPUK TOSHIBA",
1602                 .quirks = NVME_QUIRK_NO_APST,
1603         }
1604 };
1605
1606 /* match is null-terminated but idstr is space-padded. */
1607 static bool string_matches(const char *idstr, const char *match, size_t len)
1608 {
1609         size_t matchlen;
1610
1611         if (!match)
1612                 return true;
1613
1614         matchlen = strlen(match);
1615         WARN_ON_ONCE(matchlen > len);
1616
1617         if (memcmp(idstr, match, matchlen))
1618                 return false;
1619
1620         for (; matchlen < len; matchlen++)
1621                 if (idstr[matchlen] != ' ')
1622                         return false;
1623
1624         return true;
1625 }
1626
1627 static bool quirk_matches(const struct nvme_id_ctrl *id,
1628                           const struct nvme_core_quirk_entry *q)
1629 {
1630         return q->vid == le16_to_cpu(id->vid) &&
1631                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1632                 string_matches(id->fr, q->fr, sizeof(id->fr));
1633 }
1634
1635 /*
1636  * Initialize the cached copies of the Identify data and various controller
1637  * register in our nvme_ctrl structure.  This should be called as soon as
1638  * the admin queue is fully up and running.
1639  */
1640 int nvme_init_identify(struct nvme_ctrl *ctrl)
1641 {
1642         struct nvme_id_ctrl *id;
1643         u64 cap;
1644         int ret, page_shift;
1645         u32 max_hw_sectors;
1646         u8 prev_apsta;
1647
1648         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1649         if (ret) {
1650                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1651                 return ret;
1652         }
1653
1654         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1655         if (ret) {
1656                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1657                 return ret;
1658         }
1659         page_shift = NVME_CAP_MPSMIN(cap) + 12;
1660
1661         if (ctrl->vs >= NVME_VS(1, 1, 0))
1662                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1663
1664         ret = nvme_identify_ctrl(ctrl, &id);
1665         if (ret) {
1666                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1667                 return -EIO;
1668         }
1669
1670         if (!ctrl->identified) {
1671                 /*
1672                  * Check for quirks.  Quirk can depend on firmware version,
1673                  * so, in principle, the set of quirks present can change
1674                  * across a reset.  As a possible future enhancement, we
1675                  * could re-scan for quirks every time we reinitialize
1676                  * the device, but we'd have to make sure that the driver
1677                  * behaves intelligently if the quirks change.
1678                  */
1679
1680                 int i;
1681
1682                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
1683                         if (quirk_matches(id, &core_quirks[i]))
1684                                 ctrl->quirks |= core_quirks[i].quirks;
1685                 }
1686         }
1687
1688         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
1689                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
1690                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
1691         }
1692
1693         ctrl->oacs = le16_to_cpu(id->oacs);
1694         ctrl->vid = le16_to_cpu(id->vid);
1695         ctrl->oncs = le16_to_cpup(&id->oncs);
1696         atomic_set(&ctrl->abort_limit, id->acl + 1);
1697         ctrl->vwc = id->vwc;
1698         ctrl->cntlid = le16_to_cpup(&id->cntlid);
1699         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1700         memcpy(ctrl->model, id->mn, sizeof(id->mn));
1701         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1702         if (id->mdts)
1703                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1704         else
1705                 max_hw_sectors = UINT_MAX;
1706         ctrl->max_hw_sectors =
1707                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1708
1709         nvme_set_queue_limits(ctrl, ctrl->admin_q);
1710         ctrl->sgls = le32_to_cpu(id->sgls);
1711         ctrl->kas = le16_to_cpu(id->kas);
1712
1713         ctrl->npss = id->npss;
1714         prev_apsta = ctrl->apsta;
1715         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
1716                 if (force_apst && id->apsta) {
1717                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
1718                         ctrl->apsta = 1;
1719                 } else {
1720                         ctrl->apsta = 0;
1721                 }
1722         } else {
1723                 ctrl->apsta = id->apsta;
1724         }
1725         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
1726
1727         if (ctrl->ops->flags & NVME_F_FABRICS) {
1728                 ctrl->icdoff = le16_to_cpu(id->icdoff);
1729                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1730                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1731                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1732
1733                 /*
1734                  * In fabrics we need to verify the cntlid matches the
1735                  * admin connect
1736                  */
1737                 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1738                         ret = -EINVAL;
1739
1740                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1741                         dev_err(ctrl->device,
1742                                 "keep-alive support is mandatory for fabrics\n");
1743                         ret = -EINVAL;
1744                 }
1745         } else {
1746                 ctrl->cntlid = le16_to_cpu(id->cntlid);
1747                 ctrl->hmpre = le32_to_cpu(id->hmpre);
1748                 ctrl->hmmin = le32_to_cpu(id->hmmin);
1749         }
1750
1751         kfree(id);
1752
1753         if (ctrl->apsta && !prev_apsta)
1754                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
1755         else if (!ctrl->apsta && prev_apsta)
1756                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
1757
1758         nvme_configure_apst(ctrl);
1759
1760         ctrl->identified = true;
1761
1762         return ret;
1763 }
1764 EXPORT_SYMBOL_GPL(nvme_init_identify);
1765
1766 static int nvme_dev_open(struct inode *inode, struct file *file)
1767 {
1768         struct nvme_ctrl *ctrl;
1769         int instance = iminor(inode);
1770         int ret = -ENODEV;
1771
1772         spin_lock(&dev_list_lock);
1773         list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1774                 if (ctrl->instance != instance)
1775                         continue;
1776
1777                 if (!ctrl->admin_q) {
1778                         ret = -EWOULDBLOCK;
1779                         break;
1780                 }
1781                 if (!kref_get_unless_zero(&ctrl->kref))
1782                         break;
1783                 file->private_data = ctrl;
1784                 ret = 0;
1785                 break;
1786         }
1787         spin_unlock(&dev_list_lock);
1788
1789         return ret;
1790 }
1791
1792 static int nvme_dev_release(struct inode *inode, struct file *file)
1793 {
1794         nvme_put_ctrl(file->private_data);
1795         return 0;
1796 }
1797
1798 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1799 {
1800         struct nvme_ns *ns;
1801         int ret;
1802
1803         mutex_lock(&ctrl->namespaces_mutex);
1804         if (list_empty(&ctrl->namespaces)) {
1805                 ret = -ENOTTY;
1806                 goto out_unlock;
1807         }
1808
1809         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1810         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1811                 dev_warn(ctrl->device,
1812                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1813                 ret = -EINVAL;
1814                 goto out_unlock;
1815         }
1816
1817         dev_warn(ctrl->device,
1818                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1819         kref_get(&ns->kref);
1820         mutex_unlock(&ctrl->namespaces_mutex);
1821
1822         ret = nvme_user_cmd(ctrl, ns, argp);
1823         nvme_put_ns(ns);
1824         return ret;
1825
1826 out_unlock:
1827         mutex_unlock(&ctrl->namespaces_mutex);
1828         return ret;
1829 }
1830
1831 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1832                 unsigned long arg)
1833 {
1834         struct nvme_ctrl *ctrl = file->private_data;
1835         void __user *argp = (void __user *)arg;
1836
1837         switch (cmd) {
1838         case NVME_IOCTL_ADMIN_CMD:
1839                 return nvme_user_cmd(ctrl, NULL, argp);
1840         case NVME_IOCTL_IO_CMD:
1841                 return nvme_dev_user_cmd(ctrl, argp);
1842         case NVME_IOCTL_RESET:
1843                 dev_warn(ctrl->device, "resetting controller\n");
1844                 return nvme_reset_ctrl_sync(ctrl);
1845         case NVME_IOCTL_SUBSYS_RESET:
1846                 return nvme_reset_subsystem(ctrl);
1847         case NVME_IOCTL_RESCAN:
1848                 nvme_queue_scan(ctrl);
1849                 return 0;
1850         default:
1851                 return -ENOTTY;
1852         }
1853 }
1854
1855 static const struct file_operations nvme_dev_fops = {
1856         .owner          = THIS_MODULE,
1857         .open           = nvme_dev_open,
1858         .release        = nvme_dev_release,
1859         .unlocked_ioctl = nvme_dev_ioctl,
1860         .compat_ioctl   = nvme_dev_ioctl,
1861 };
1862
1863 static ssize_t nvme_sysfs_reset(struct device *dev,
1864                                 struct device_attribute *attr, const char *buf,
1865                                 size_t count)
1866 {
1867         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1868         int ret;
1869
1870         ret = nvme_reset_ctrl_sync(ctrl);
1871         if (ret < 0)
1872                 return ret;
1873         return count;
1874 }
1875 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1876
1877 static ssize_t nvme_sysfs_rescan(struct device *dev,
1878                                 struct device_attribute *attr, const char *buf,
1879                                 size_t count)
1880 {
1881         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1882
1883         nvme_queue_scan(ctrl);
1884         return count;
1885 }
1886 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1887
1888 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1889                                                                 char *buf)
1890 {
1891         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1892         struct nvme_ctrl *ctrl = ns->ctrl;
1893         int serial_len = sizeof(ctrl->serial);
1894         int model_len = sizeof(ctrl->model);
1895
1896         if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
1897                 return sprintf(buf, "eui.%16phN\n", ns->nguid);
1898
1899         if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1900                 return sprintf(buf, "eui.%8phN\n", ns->eui);
1901
1902         while (ctrl->serial[serial_len - 1] == ' ')
1903                 serial_len--;
1904         while (ctrl->model[model_len - 1] == ' ')
1905                 model_len--;
1906
1907         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1908                 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1909 }
1910 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1911
1912 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
1913                           char *buf)
1914 {
1915         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1916         return sprintf(buf, "%pU\n", ns->nguid);
1917 }
1918 static DEVICE_ATTR(nguid, S_IRUGO, nguid_show, NULL);
1919
1920 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1921                                                                 char *buf)
1922 {
1923         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1924
1925         /* For backward compatibility expose the NGUID to userspace if
1926          * we have no UUID set
1927          */
1928         if (uuid_is_null(&ns->uuid)) {
1929                 printk_ratelimited(KERN_WARNING
1930                                    "No UUID available providing old NGUID\n");
1931                 return sprintf(buf, "%pU\n", ns->nguid);
1932         }
1933         return sprintf(buf, "%pU\n", &ns->uuid);
1934 }
1935 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1936
1937 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1938                                                                 char *buf)
1939 {
1940         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1941         return sprintf(buf, "%8phd\n", ns->eui);
1942 }
1943 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1944
1945 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1946                                                                 char *buf)
1947 {
1948         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1949         return sprintf(buf, "%d\n", ns->ns_id);
1950 }
1951 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1952
1953 static struct attribute *nvme_ns_attrs[] = {
1954         &dev_attr_wwid.attr,
1955         &dev_attr_uuid.attr,
1956         &dev_attr_nguid.attr,
1957         &dev_attr_eui.attr,
1958         &dev_attr_nsid.attr,
1959         NULL,
1960 };
1961
1962 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1963                 struct attribute *a, int n)
1964 {
1965         struct device *dev = container_of(kobj, struct device, kobj);
1966         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1967
1968         if (a == &dev_attr_uuid.attr) {
1969                 if (uuid_is_null(&ns->uuid) ||
1970                     !memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
1971                         return 0;
1972         }
1973         if (a == &dev_attr_nguid.attr) {
1974                 if (!memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
1975                         return 0;
1976         }
1977         if (a == &dev_attr_eui.attr) {
1978                 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1979                         return 0;
1980         }
1981         return a->mode;
1982 }
1983
1984 static const struct attribute_group nvme_ns_attr_group = {
1985         .attrs          = nvme_ns_attrs,
1986         .is_visible     = nvme_ns_attrs_are_visible,
1987 };
1988
1989 #define nvme_show_str_function(field)                                           \
1990 static ssize_t  field##_show(struct device *dev,                                \
1991                             struct device_attribute *attr, char *buf)           \
1992 {                                                                               \
1993         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1994         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);   \
1995 }                                                                               \
1996 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1997
1998 #define nvme_show_int_function(field)                                           \
1999 static ssize_t  field##_show(struct device *dev,                                \
2000                             struct device_attribute *attr, char *buf)           \
2001 {                                                                               \
2002         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2003         return sprintf(buf, "%d\n", ctrl->field);       \
2004 }                                                                               \
2005 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2006
2007 nvme_show_str_function(model);
2008 nvme_show_str_function(serial);
2009 nvme_show_str_function(firmware_rev);
2010 nvme_show_int_function(cntlid);
2011
2012 static ssize_t nvme_sysfs_delete(struct device *dev,
2013                                 struct device_attribute *attr, const char *buf,
2014                                 size_t count)
2015 {
2016         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2017
2018         if (device_remove_file_self(dev, attr))
2019                 ctrl->ops->delete_ctrl(ctrl);
2020         return count;
2021 }
2022 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2023
2024 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2025                                          struct device_attribute *attr,
2026                                          char *buf)
2027 {
2028         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2029
2030         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2031 }
2032 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2033
2034 static ssize_t nvme_sysfs_show_state(struct device *dev,
2035                                      struct device_attribute *attr,
2036                                      char *buf)
2037 {
2038         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2039         static const char *const state_name[] = {
2040                 [NVME_CTRL_NEW]         = "new",
2041                 [NVME_CTRL_LIVE]        = "live",
2042                 [NVME_CTRL_RESETTING]   = "resetting",
2043                 [NVME_CTRL_RECONNECTING]= "reconnecting",
2044                 [NVME_CTRL_DELETING]    = "deleting",
2045                 [NVME_CTRL_DEAD]        = "dead",
2046         };
2047
2048         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2049             state_name[ctrl->state])
2050                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2051
2052         return sprintf(buf, "unknown state\n");
2053 }
2054
2055 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2056
2057 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2058                                          struct device_attribute *attr,
2059                                          char *buf)
2060 {
2061         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2062
2063         return snprintf(buf, PAGE_SIZE, "%s\n",
2064                         ctrl->ops->get_subsysnqn(ctrl));
2065 }
2066 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2067
2068 static ssize_t nvme_sysfs_show_address(struct device *dev,
2069                                          struct device_attribute *attr,
2070                                          char *buf)
2071 {
2072         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2073
2074         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2075 }
2076 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2077
2078 static struct attribute *nvme_dev_attrs[] = {
2079         &dev_attr_reset_controller.attr,
2080         &dev_attr_rescan_controller.attr,
2081         &dev_attr_model.attr,
2082         &dev_attr_serial.attr,
2083         &dev_attr_firmware_rev.attr,
2084         &dev_attr_cntlid.attr,
2085         &dev_attr_delete_controller.attr,
2086         &dev_attr_transport.attr,
2087         &dev_attr_subsysnqn.attr,
2088         &dev_attr_address.attr,
2089         &dev_attr_state.attr,
2090         NULL
2091 };
2092
2093 #define CHECK_ATTR(ctrl, a, name)               \
2094         if ((a) == &dev_attr_##name.attr &&     \
2095             !(ctrl)->ops->get_##name)           \
2096                 return 0
2097
2098 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2099                 struct attribute *a, int n)
2100 {
2101         struct device *dev = container_of(kobj, struct device, kobj);
2102         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2103
2104         if (a == &dev_attr_delete_controller.attr) {
2105                 if (!ctrl->ops->delete_ctrl)
2106                         return 0;
2107         }
2108
2109         CHECK_ATTR(ctrl, a, subsysnqn);
2110         CHECK_ATTR(ctrl, a, address);
2111
2112         return a->mode;
2113 }
2114
2115 static struct attribute_group nvme_dev_attrs_group = {
2116         .attrs          = nvme_dev_attrs,
2117         .is_visible     = nvme_dev_attrs_are_visible,
2118 };
2119
2120 static const struct attribute_group *nvme_dev_attr_groups[] = {
2121         &nvme_dev_attrs_group,
2122         NULL,
2123 };
2124
2125 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2126 {
2127         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2128         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2129
2130         return nsa->ns_id - nsb->ns_id;
2131 }
2132
2133 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2134 {
2135         struct nvme_ns *ns, *ret = NULL;
2136
2137         mutex_lock(&ctrl->namespaces_mutex);
2138         list_for_each_entry(ns, &ctrl->namespaces, list) {
2139                 if (ns->ns_id == nsid) {
2140                         kref_get(&ns->kref);
2141                         ret = ns;
2142                         break;
2143                 }
2144                 if (ns->ns_id > nsid)
2145                         break;
2146         }
2147         mutex_unlock(&ctrl->namespaces_mutex);
2148         return ret;
2149 }
2150
2151 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2152 {
2153         struct nvme_ns *ns;
2154         struct gendisk *disk;
2155         struct nvme_id_ns *id;
2156         char disk_name[DISK_NAME_LEN];
2157         int node = dev_to_node(ctrl->dev);
2158
2159         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2160         if (!ns)
2161                 return;
2162
2163         ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
2164         if (ns->instance < 0)
2165                 goto out_free_ns;
2166
2167         ns->queue = blk_mq_init_queue(ctrl->tagset);
2168         if (IS_ERR(ns->queue))
2169                 goto out_release_instance;
2170         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2171         ns->queue->queuedata = ns;
2172         ns->ctrl = ctrl;
2173
2174         kref_init(&ns->kref);
2175         ns->ns_id = nsid;
2176         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2177
2178         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2179         nvme_set_queue_limits(ctrl, ns->queue);
2180
2181         sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
2182
2183         if (nvme_revalidate_ns(ns, &id))
2184                 goto out_free_queue;
2185
2186         if (nvme_nvm_ns_supported(ns, id) &&
2187                                 nvme_nvm_register(ns, disk_name, node)) {
2188                 dev_warn(ctrl->device, "%s: LightNVM init failure\n", __func__);
2189                 goto out_free_id;
2190         }
2191
2192         disk = alloc_disk_node(0, node);
2193         if (!disk)
2194                 goto out_free_id;
2195
2196         disk->fops = &nvme_fops;
2197         disk->private_data = ns;
2198         disk->queue = ns->queue;
2199         disk->flags = GENHD_FL_EXT_DEVT;
2200         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
2201         ns->disk = disk;
2202
2203         __nvme_revalidate_disk(disk, id);
2204
2205         mutex_lock(&ctrl->namespaces_mutex);
2206         list_add_tail(&ns->list, &ctrl->namespaces);
2207         mutex_unlock(&ctrl->namespaces_mutex);
2208
2209         kref_get(&ctrl->kref);
2210
2211         kfree(id);
2212
2213         device_add_disk(ctrl->device, ns->disk);
2214         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
2215                                         &nvme_ns_attr_group))
2216                 pr_warn("%s: failed to create sysfs group for identification\n",
2217                         ns->disk->disk_name);
2218         if (ns->ndev && nvme_nvm_register_sysfs(ns))
2219                 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
2220                         ns->disk->disk_name);
2221         return;
2222  out_free_id:
2223         kfree(id);
2224  out_free_queue:
2225         blk_cleanup_queue(ns->queue);
2226  out_release_instance:
2227         ida_simple_remove(&ctrl->ns_ida, ns->instance);
2228  out_free_ns:
2229         kfree(ns);
2230 }
2231
2232 static void nvme_ns_remove(struct nvme_ns *ns)
2233 {
2234         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
2235                 return;
2236
2237         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
2238                 if (blk_get_integrity(ns->disk))
2239                         blk_integrity_unregister(ns->disk);
2240                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
2241                                         &nvme_ns_attr_group);
2242                 if (ns->ndev)
2243                         nvme_nvm_unregister_sysfs(ns);
2244                 del_gendisk(ns->disk);
2245                 blk_cleanup_queue(ns->queue);
2246         }
2247
2248         mutex_lock(&ns->ctrl->namespaces_mutex);
2249         list_del_init(&ns->list);
2250         mutex_unlock(&ns->ctrl->namespaces_mutex);
2251
2252         nvme_put_ns(ns);
2253 }
2254
2255 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2256 {
2257         struct nvme_ns *ns;
2258
2259         ns = nvme_find_get_ns(ctrl, nsid);
2260         if (ns) {
2261                 if (ns->disk && revalidate_disk(ns->disk))
2262                         nvme_ns_remove(ns);
2263                 nvme_put_ns(ns);
2264         } else
2265                 nvme_alloc_ns(ctrl, nsid);
2266 }
2267
2268 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
2269                                         unsigned nsid)
2270 {
2271         struct nvme_ns *ns, *next;
2272
2273         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
2274                 if (ns->ns_id > nsid)
2275                         nvme_ns_remove(ns);
2276         }
2277 }
2278
2279 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
2280 {
2281         struct nvme_ns *ns;
2282         __le32 *ns_list;
2283         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
2284         int ret = 0;
2285
2286         ns_list = kzalloc(0x1000, GFP_KERNEL);
2287         if (!ns_list)
2288                 return -ENOMEM;
2289
2290         for (i = 0; i < num_lists; i++) {
2291                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
2292                 if (ret)
2293                         goto free;
2294
2295                 for (j = 0; j < min(nn, 1024U); j++) {
2296                         nsid = le32_to_cpu(ns_list[j]);
2297                         if (!nsid)
2298                                 goto out;
2299
2300                         nvme_validate_ns(ctrl, nsid);
2301
2302                         while (++prev < nsid) {
2303                                 ns = nvme_find_get_ns(ctrl, prev);
2304                                 if (ns) {
2305                                         nvme_ns_remove(ns);
2306                                         nvme_put_ns(ns);
2307                                 }
2308                         }
2309                 }
2310                 nn -= j;
2311         }
2312  out:
2313         nvme_remove_invalid_namespaces(ctrl, prev);
2314  free:
2315         kfree(ns_list);
2316         return ret;
2317 }
2318
2319 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
2320 {
2321         unsigned i;
2322
2323         for (i = 1; i <= nn; i++)
2324                 nvme_validate_ns(ctrl, i);
2325
2326         nvme_remove_invalid_namespaces(ctrl, nn);
2327 }
2328
2329 static void nvme_scan_work(struct work_struct *work)
2330 {
2331         struct nvme_ctrl *ctrl =
2332                 container_of(work, struct nvme_ctrl, scan_work);
2333         struct nvme_id_ctrl *id;
2334         unsigned nn;
2335
2336         if (ctrl->state != NVME_CTRL_LIVE)
2337                 return;
2338
2339         if (nvme_identify_ctrl(ctrl, &id))
2340                 return;
2341
2342         nn = le32_to_cpu(id->nn);
2343         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
2344             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
2345                 if (!nvme_scan_ns_list(ctrl, nn))
2346                         goto done;
2347         }
2348         nvme_scan_ns_sequential(ctrl, nn);
2349  done:
2350         mutex_lock(&ctrl->namespaces_mutex);
2351         list_sort(NULL, &ctrl->namespaces, ns_cmp);
2352         mutex_unlock(&ctrl->namespaces_mutex);
2353         kfree(id);
2354 }
2355
2356 void nvme_queue_scan(struct nvme_ctrl *ctrl)
2357 {
2358         /*
2359          * Do not queue new scan work when a controller is reset during
2360          * removal.
2361          */
2362         if (ctrl->state == NVME_CTRL_LIVE)
2363                 queue_work(nvme_wq, &ctrl->scan_work);
2364 }
2365 EXPORT_SYMBOL_GPL(nvme_queue_scan);
2366
2367 /*
2368  * This function iterates the namespace list unlocked to allow recovery from
2369  * controller failure. It is up to the caller to ensure the namespace list is
2370  * not modified by scan work while this function is executing.
2371  */
2372 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
2373 {
2374         struct nvme_ns *ns, *next;
2375
2376         /*
2377          * The dead states indicates the controller was not gracefully
2378          * disconnected. In that case, we won't be able to flush any data while
2379          * removing the namespaces' disks; fail all the queues now to avoid
2380          * potentially having to clean up the failed sync later.
2381          */
2382         if (ctrl->state == NVME_CTRL_DEAD)
2383                 nvme_kill_queues(ctrl);
2384
2385         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
2386                 nvme_ns_remove(ns);
2387 }
2388 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
2389
2390 static void nvme_async_event_work(struct work_struct *work)
2391 {
2392         struct nvme_ctrl *ctrl =
2393                 container_of(work, struct nvme_ctrl, async_event_work);
2394
2395         spin_lock_irq(&ctrl->lock);
2396         while (ctrl->event_limit > 0) {
2397                 int aer_idx = --ctrl->event_limit;
2398
2399                 spin_unlock_irq(&ctrl->lock);
2400                 ctrl->ops->submit_async_event(ctrl, aer_idx);
2401                 spin_lock_irq(&ctrl->lock);
2402         }
2403         spin_unlock_irq(&ctrl->lock);
2404 }
2405
2406 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
2407                 union nvme_result *res)
2408 {
2409         u32 result = le32_to_cpu(res->u32);
2410         bool done = true;
2411
2412         switch (le16_to_cpu(status) >> 1) {
2413         case NVME_SC_SUCCESS:
2414                 done = false;
2415                 /*FALLTHRU*/
2416         case NVME_SC_ABORT_REQ:
2417                 ++ctrl->event_limit;
2418                 queue_work(nvme_wq, &ctrl->async_event_work);
2419                 break;
2420         default:
2421                 break;
2422         }
2423
2424         if (done)
2425                 return;
2426
2427         switch (result & 0xff07) {
2428         case NVME_AER_NOTICE_NS_CHANGED:
2429                 dev_info(ctrl->device, "rescanning\n");
2430                 nvme_queue_scan(ctrl);
2431                 break;
2432         default:
2433                 dev_warn(ctrl->device, "async event result %08x\n", result);
2434         }
2435 }
2436 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
2437
2438 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
2439 {
2440         ctrl->event_limit = NVME_NR_AERS;
2441         queue_work(nvme_wq, &ctrl->async_event_work);
2442 }
2443 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
2444
2445 static DEFINE_IDA(nvme_instance_ida);
2446
2447 static int nvme_set_instance(struct nvme_ctrl *ctrl)
2448 {
2449         int instance, error;
2450
2451         do {
2452                 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2453                         return -ENODEV;
2454
2455                 spin_lock(&dev_list_lock);
2456                 error = ida_get_new(&nvme_instance_ida, &instance);
2457                 spin_unlock(&dev_list_lock);
2458         } while (error == -EAGAIN);
2459
2460         if (error)
2461                 return -ENODEV;
2462
2463         ctrl->instance = instance;
2464         return 0;
2465 }
2466
2467 static void nvme_release_instance(struct nvme_ctrl *ctrl)
2468 {
2469         spin_lock(&dev_list_lock);
2470         ida_remove(&nvme_instance_ida, ctrl->instance);
2471         spin_unlock(&dev_list_lock);
2472 }
2473
2474 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
2475 {
2476         flush_work(&ctrl->async_event_work);
2477         flush_work(&ctrl->scan_work);
2478         nvme_remove_namespaces(ctrl);
2479
2480         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
2481
2482         spin_lock(&dev_list_lock);
2483         list_del(&ctrl->node);
2484         spin_unlock(&dev_list_lock);
2485 }
2486 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
2487
2488 static void nvme_free_ctrl(struct kref *kref)
2489 {
2490         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
2491
2492         put_device(ctrl->device);
2493         nvme_release_instance(ctrl);
2494         ida_destroy(&ctrl->ns_ida);
2495
2496         ctrl->ops->free_ctrl(ctrl);
2497 }
2498
2499 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
2500 {
2501         kref_put(&ctrl->kref, nvme_free_ctrl);
2502 }
2503 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2504
2505 /*
2506  * Initialize a NVMe controller structures.  This needs to be called during
2507  * earliest initialization so that we have the initialized structured around
2508  * during probing.
2509  */
2510 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2511                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
2512 {
2513         int ret;
2514
2515         ctrl->state = NVME_CTRL_NEW;
2516         spin_lock_init(&ctrl->lock);
2517         INIT_LIST_HEAD(&ctrl->namespaces);
2518         mutex_init(&ctrl->namespaces_mutex);
2519         kref_init(&ctrl->kref);
2520         ctrl->dev = dev;
2521         ctrl->ops = ops;
2522         ctrl->quirks = quirks;
2523         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2524         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2525
2526         ret = nvme_set_instance(ctrl);
2527         if (ret)
2528                 goto out;
2529
2530         ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2531                                 MKDEV(nvme_char_major, ctrl->instance),
2532                                 ctrl, nvme_dev_attr_groups,
2533                                 "nvme%d", ctrl->instance);
2534         if (IS_ERR(ctrl->device)) {
2535                 ret = PTR_ERR(ctrl->device);
2536                 goto out_release_instance;
2537         }
2538         get_device(ctrl->device);
2539         ida_init(&ctrl->ns_ida);
2540
2541         spin_lock(&dev_list_lock);
2542         list_add_tail(&ctrl->node, &nvme_ctrl_list);
2543         spin_unlock(&dev_list_lock);
2544
2545         /*
2546          * Initialize latency tolerance controls.  The sysfs files won't
2547          * be visible to userspace unless the device actually supports APST.
2548          */
2549         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
2550         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
2551                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
2552
2553         return 0;
2554 out_release_instance:
2555         nvme_release_instance(ctrl);
2556 out:
2557         return ret;
2558 }
2559 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2560
2561 /**
2562  * nvme_kill_queues(): Ends all namespace queues
2563  * @ctrl: the dead controller that needs to end
2564  *
2565  * Call this function when the driver determines it is unable to get the
2566  * controller in a state capable of servicing IO.
2567  */
2568 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2569 {
2570         struct nvme_ns *ns;
2571
2572         mutex_lock(&ctrl->namespaces_mutex);
2573
2574         /* Forcibly start all queues to avoid having stuck requests */
2575         blk_mq_start_hw_queues(ctrl->admin_q);
2576
2577         list_for_each_entry(ns, &ctrl->namespaces, list) {
2578                 /*
2579                  * Revalidating a dead namespace sets capacity to 0. This will
2580                  * end buffered writers dirtying pages that can't be synced.
2581                  */
2582                 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2583                         continue;
2584                 revalidate_disk(ns->disk);
2585                 blk_set_queue_dying(ns->queue);
2586
2587                 /*
2588                  * Forcibly start all queues to avoid having stuck requests.
2589                  * Note that we must ensure the queues are not stopped
2590                  * when the final removal happens.
2591                  */
2592                 blk_mq_start_hw_queues(ns->queue);
2593
2594                 /* draining requests in requeue list */
2595                 blk_mq_kick_requeue_list(ns->queue);
2596         }
2597         mutex_unlock(&ctrl->namespaces_mutex);
2598 }
2599 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2600
2601 void nvme_unfreeze(struct nvme_ctrl *ctrl)
2602 {
2603         struct nvme_ns *ns;
2604
2605         mutex_lock(&ctrl->namespaces_mutex);
2606         list_for_each_entry(ns, &ctrl->namespaces, list)
2607                 blk_mq_unfreeze_queue(ns->queue);
2608         mutex_unlock(&ctrl->namespaces_mutex);
2609 }
2610 EXPORT_SYMBOL_GPL(nvme_unfreeze);
2611
2612 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
2613 {
2614         struct nvme_ns *ns;
2615
2616         mutex_lock(&ctrl->namespaces_mutex);
2617         list_for_each_entry(ns, &ctrl->namespaces, list) {
2618                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
2619                 if (timeout <= 0)
2620                         break;
2621         }
2622         mutex_unlock(&ctrl->namespaces_mutex);
2623 }
2624 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
2625
2626 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
2627 {
2628         struct nvme_ns *ns;
2629
2630         mutex_lock(&ctrl->namespaces_mutex);
2631         list_for_each_entry(ns, &ctrl->namespaces, list)
2632                 blk_mq_freeze_queue_wait(ns->queue);
2633         mutex_unlock(&ctrl->namespaces_mutex);
2634 }
2635 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
2636
2637 void nvme_start_freeze(struct nvme_ctrl *ctrl)
2638 {
2639         struct nvme_ns *ns;
2640
2641         mutex_lock(&ctrl->namespaces_mutex);
2642         list_for_each_entry(ns, &ctrl->namespaces, list)
2643                 blk_freeze_queue_start(ns->queue);
2644         mutex_unlock(&ctrl->namespaces_mutex);
2645 }
2646 EXPORT_SYMBOL_GPL(nvme_start_freeze);
2647
2648 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2649 {
2650         struct nvme_ns *ns;
2651
2652         mutex_lock(&ctrl->namespaces_mutex);
2653         list_for_each_entry(ns, &ctrl->namespaces, list)
2654                 blk_mq_quiesce_queue(ns->queue);
2655         mutex_unlock(&ctrl->namespaces_mutex);
2656 }
2657 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2658
2659 void nvme_start_queues(struct nvme_ctrl *ctrl)
2660 {
2661         struct nvme_ns *ns;
2662
2663         mutex_lock(&ctrl->namespaces_mutex);
2664         list_for_each_entry(ns, &ctrl->namespaces, list) {
2665                 blk_mq_start_stopped_hw_queues(ns->queue, true);
2666                 blk_mq_kick_requeue_list(ns->queue);
2667         }
2668         mutex_unlock(&ctrl->namespaces_mutex);
2669 }
2670 EXPORT_SYMBOL_GPL(nvme_start_queues);
2671
2672 int __init nvme_core_init(void)
2673 {
2674         int result;
2675
2676         nvme_wq = alloc_workqueue("nvme-wq",
2677                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
2678         if (!nvme_wq)
2679                 return -ENOMEM;
2680
2681         result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2682                                                         &nvme_dev_fops);
2683         if (result < 0)
2684                 goto destroy_wq;
2685         else if (result > 0)
2686                 nvme_char_major = result;
2687
2688         nvme_class = class_create(THIS_MODULE, "nvme");
2689         if (IS_ERR(nvme_class)) {
2690                 result = PTR_ERR(nvme_class);
2691                 goto unregister_chrdev;
2692         }
2693
2694         return 0;
2695
2696 unregister_chrdev:
2697         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2698 destroy_wq:
2699         destroy_workqueue(nvme_wq);
2700         return result;
2701 }
2702
2703 void nvme_core_exit(void)
2704 {
2705         class_destroy(nvme_class);
2706         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2707         destroy_workqueue(nvme_wq);
2708 }
2709
2710 MODULE_LICENSE("GPL");
2711 MODULE_VERSION("1.0");
2712 module_init(nvme_core_init);
2713 module_exit(nvme_core_exit);