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 <scsi/sg.h>
  30 #include <asm/unaligned.h>
  31
  32 #include "nvme.h"
  33 #include "fabrics.h"
  34
  35 #define NVME_MINORS             (1U << MINORBITS)
  36
  37 unsigned char admin_timeout = 60;
  38 module_param(admin_timeout, byte, 0644);
  39 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  40 EXPORT_SYMBOL_GPL(admin_timeout);
  41
  42 unsigned char nvme_io_timeout = 30;
  43 module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
  44 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  45 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  46
  47 unsigned char shutdown_timeout = 5;
  48 module_param(shutdown_timeout, byte, 0644);
  49 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  50
  51 unsigned int nvme_max_retries = 5;
  52 module_param_named(max_retries, nvme_max_retries, uint, 0644);
  53 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  54 EXPORT_SYMBOL_GPL(nvme_max_retries);
  55
  56 static int nvme_char_major;
  57 module_param(nvme_char_major, int, 0);
  58
  59 static LIST_HEAD(nvme_ctrl_list);
  60 static DEFINE_SPINLOCK(dev_list_lock);
  61
  62 static struct class *nvme_class;
  63
  64 void nvme_cancel_request(struct request *req, void *data, bool reserved)
  65 {
  66         int status;
  67
  68         if (!blk_mq_request_started(req))
  69                 return;
  70
  71         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
  72                                 "Cancelling I/O %d", req->tag);
  73
  74         status = NVME_SC_ABORT_REQ;
  75         if (blk_queue_dying(req->q))
  76                 status |= NVME_SC_DNR;
  77         blk_mq_complete_request(req, status);
  78 }
  79 EXPORT_SYMBOL_GPL(nvme_cancel_request);
  80
  81 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
  82                 enum nvme_ctrl_state new_state)
  83 {
  84         enum nvme_ctrl_state old_state = ctrl->state;
  85         bool changed = false;
  86
  87         spin_lock_irq(&ctrl->lock);
  88         switch (new_state) {
  89         case NVME_CTRL_LIVE:
  90                 switch (old_state) {
  91                 case NVME_CTRL_NEW:
  92                 case NVME_CTRL_RESETTING:
  93                 case NVME_CTRL_RECONNECTING:
  94                         changed = true;
  95                         /* FALLTHRU */
  96                 default:
  97                         break;
  98                 }
  99                 break;
 100         case NVME_CTRL_RESETTING:
 101                 switch (old_state) {
 102                 case NVME_CTRL_NEW:
 103                 case NVME_CTRL_LIVE:
 104                 case NVME_CTRL_RECONNECTING:
 105                         changed = true;
 106                         /* FALLTHRU */
 107                 default:
 108                         break;
 109                 }
 110                 break;
 111         case NVME_CTRL_RECONNECTING:
 112                 switch (old_state) {
 113                 case NVME_CTRL_LIVE:
 114                         changed = true;
 115                         /* FALLTHRU */
 116                 default:
 117                         break;
 118                 }
 119                 break;
 120         case NVME_CTRL_DELETING:
 121                 switch (old_state) {
 122                 case NVME_CTRL_LIVE:
 123                 case NVME_CTRL_RESETTING:
 124                 case NVME_CTRL_RECONNECTING:
 125                         changed = true;
 126                         /* FALLTHRU */
 127                 default:
 128                         break;
 129                 }
 130                 break;
 131         case NVME_CTRL_DEAD:
 132                 switch (old_state) {
 133                 case NVME_CTRL_DELETING:
 134                         changed = true;
 135                         /* FALLTHRU */
 136                 default:
 137                         break;
 138                 }
 139                 break;
 140         default:
 141                 break;
 142         }
 143         spin_unlock_irq(&ctrl->lock);
 144
 145         if (changed)
 146                 ctrl->state = new_state;
 147
 148         return changed;
 149 }
 150 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 151
 152 static void nvme_free_ns(struct kref *kref)
 153 {
 154         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 155
 156         if (ns->type == NVME_NS_LIGHTNVM)
 157                 nvme_nvm_unregister(ns->queue, ns->disk->disk_name);
 158
 159         spin_lock(&dev_list_lock);
 160         ns->disk->private_data = NULL;
 161         spin_unlock(&dev_list_lock);
 162
 163         put_disk(ns->disk);
 164         ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
 165         nvme_put_ctrl(ns->ctrl);
 166         kfree(ns);
 167 }
 168
 169 static void nvme_put_ns(struct nvme_ns *ns)
 170 {
 171         kref_put(&ns->kref, nvme_free_ns);
 172 }
 173
 174 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
 175 {
 176         struct nvme_ns *ns;
 177
 178         spin_lock(&dev_list_lock);
 179         ns = disk->private_data;
 180         if (ns) {
 181                 if (!kref_get_unless_zero(&ns->kref))
 182                         goto fail;
 183                 if (!try_module_get(ns->ctrl->ops->module))
 184                         goto fail_put_ns;
 185         }
 186         spin_unlock(&dev_list_lock);
 187
 188         return ns;
 189
 190 fail_put_ns:
 191         kref_put(&ns->kref, nvme_free_ns);
 192 fail:
 193         spin_unlock(&dev_list_lock);
 194         return NULL;
 195 }
 196
 197 void nvme_requeue_req(struct request *req)
 198 {
 199         unsigned long flags;
 200
 201         blk_mq_requeue_request(req);
 202         spin_lock_irqsave(req->q->queue_lock, flags);
 203         if (!blk_queue_stopped(req->q))
 204                 blk_mq_kick_requeue_list(req->q);
 205         spin_unlock_irqrestore(req->q->queue_lock, flags);
 206 }
 207 EXPORT_SYMBOL_GPL(nvme_requeue_req);
 208
 209 struct request *nvme_alloc_request(struct request_queue *q,
 210                 struct nvme_command *cmd, unsigned int flags, int qid)
 211 {
 212         struct request *req;
 213
 214         if (qid == NVME_QID_ANY) {
 215                 req = blk_mq_alloc_request(q, nvme_is_write(cmd), flags);
 216         } else {
 217                 req = blk_mq_alloc_request_hctx(q, nvme_is_write(cmd), flags,
 218                                 qid ? qid - 1 : 0);
 219         }
 220         if (IS_ERR(req))
 221                 return req;
 222
 223         req->cmd_type = REQ_TYPE_DRV_PRIV;
 224         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 225         req->cmd = (unsigned char *)cmd;
 226         req->cmd_len = sizeof(struct nvme_command);
 227
 228         return req;
 229 }
 230 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 231
 232 static inline void nvme_setup_flush(struct nvme_ns *ns,
 233                 struct nvme_command *cmnd)
 234 {
 235         memset(cmnd, 0, sizeof(*cmnd));
 236         cmnd->common.opcode = nvme_cmd_flush;
 237         cmnd->common.nsid = cpu_to_le32(ns->ns_id);
 238 }
 239
 240 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 241                 struct nvme_command *cmnd)
 242 {
 243         struct nvme_dsm_range *range;
 244         struct page *page;
 245         int offset;
 246         unsigned int nr_bytes = blk_rq_bytes(req);
 247
 248         range = kmalloc(sizeof(*range), GFP_ATOMIC);
 249         if (!range)
 250                 return BLK_MQ_RQ_QUEUE_BUSY;
 251
 252         range->cattr = cpu_to_le32(0);
 253         range->nlb = cpu_to_le32(nr_bytes >> ns->lba_shift);
 254         range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 255
 256         memset(cmnd, 0, sizeof(*cmnd));
 257         cmnd->dsm.opcode = nvme_cmd_dsm;
 258         cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
 259         cmnd->dsm.nr = 0;
 260         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 261
 262         req->completion_data = range;
 263         page = virt_to_page(range);
 264         offset = offset_in_page(range);
 265         blk_add_request_payload(req, page, offset, sizeof(*range));
 266
 267         /*
 268          * we set __data_len back to the size of the area to be discarded
 269          * on disk. This allows us to report completion on the full amount
 270          * of blocks described by the request.
 271          */
 272         req->__data_len = nr_bytes;
 273
 274         return 0;
 275 }
 276
 277 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
 278                 struct nvme_command *cmnd)
 279 {
 280         u16 control = 0;
 281         u32 dsmgmt = 0;
 282
 283         if (req->cmd_flags & REQ_FUA)
 284                 control |= NVME_RW_FUA;
 285         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 286                 control |= NVME_RW_LR;
 287
 288         if (req->cmd_flags & REQ_RAHEAD)
 289                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 290
 291         memset(cmnd, 0, sizeof(*cmnd));
 292         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 293         cmnd->rw.command_id = req->tag;
 294         cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
 295         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 296         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 297
 298         if (ns->ms) {
 299                 switch (ns->pi_type) {
 300                 case NVME_NS_DPS_PI_TYPE3:
 301                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 302                         break;
 303                 case NVME_NS_DPS_PI_TYPE1:
 304                 case NVME_NS_DPS_PI_TYPE2:
 305                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 306                                         NVME_RW_PRINFO_PRCHK_REF;
 307                         cmnd->rw.reftag = cpu_to_le32(
 308                                         nvme_block_nr(ns, blk_rq_pos(req)));
 309                         break;
 310                 }
 311                 if (!blk_integrity_rq(req))
 312                         control |= NVME_RW_PRINFO_PRACT;
 313         }
 314
 315         cmnd->rw.control = cpu_to_le16(control);
 316         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 317 }
 318
 319 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 320                 struct nvme_command *cmd)
 321 {
 322         int ret = 0;
 323
 324         if (req->cmd_type == REQ_TYPE_DRV_PRIV)
 325                 memcpy(cmd, req->cmd, sizeof(*cmd));
 326         else if (req_op(req) == REQ_OP_FLUSH)
 327                 nvme_setup_flush(ns, cmd);
 328         else if (req_op(req) == REQ_OP_DISCARD)
 329                 ret = nvme_setup_discard(ns, req, cmd);
 330         else
 331                 nvme_setup_rw(ns, req, cmd);
 332
 333         return ret;
 334 }
 335 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 336
 337 /*
 338  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 339  * if the result is positive, it's an NVM Express status code
 340  */
 341 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 342                 struct nvme_completion *cqe, void *buffer, unsigned bufflen,
 343                 unsigned timeout, int qid, int at_head, int flags)
 344 {
 345         struct request *req;
 346         int ret;
 347
 348         req = nvme_alloc_request(q, cmd, flags, qid);
 349         if (IS_ERR(req))
 350                 return PTR_ERR(req);
 351
 352         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 353         req->special = cqe;
 354
 355         if (buffer && bufflen) {
 356                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 357                 if (ret)
 358                         goto out;
 359         }
 360
 361         blk_execute_rq(req->q, NULL, req, at_head);
 362         ret = req->errors;
 363  out:
 364         blk_mq_free_request(req);
 365         return ret;
 366 }
 367 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 368
 369 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 370                 void *buffer, unsigned bufflen)
 371 {
 372         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 373                         NVME_QID_ANY, 0, 0);
 374 }
 375 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 376
 377 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 378                 void __user *ubuffer, unsigned bufflen,
 379                 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
 380                 u32 *result, unsigned timeout)
 381 {
 382         bool write = nvme_is_write(cmd);
 383         struct nvme_completion cqe;
 384         struct nvme_ns *ns = q->queuedata;
 385         struct gendisk *disk = ns ? ns->disk : NULL;
 386         struct request *req;
 387         struct bio *bio = NULL;
 388         void *meta = NULL;
 389         int ret;
 390
 391         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 392         if (IS_ERR(req))
 393                 return PTR_ERR(req);
 394
 395         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 396         req->special = &cqe;
 397
 398         if (ubuffer && bufflen) {
 399                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 400                                 GFP_KERNEL);
 401                 if (ret)
 402                         goto out;
 403                 bio = req->bio;
 404
 405                 if (!disk)
 406                         goto submit;
 407                 bio->bi_bdev = bdget_disk(disk, 0);
 408                 if (!bio->bi_bdev) {
 409                         ret = -ENODEV;
 410                         goto out_unmap;
 411                 }
 412
 413                 if (meta_buffer && meta_len) {
 414                         struct bio_integrity_payload *bip;
 415
 416                         meta = kmalloc(meta_len, GFP_KERNEL);
 417                         if (!meta) {
 418                                 ret = -ENOMEM;
 419                                 goto out_unmap;
 420                         }
 421
 422                         if (write) {
 423                                 if (copy_from_user(meta, meta_buffer,
 424                                                 meta_len)) {
 425                                         ret = -EFAULT;
 426                                         goto out_free_meta;
 427                                 }
 428                         }
 429
 430                         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 431                         if (IS_ERR(bip)) {
 432                                 ret = PTR_ERR(bip);
 433                                 goto out_free_meta;
 434                         }
 435
 436                         bip->bip_iter.bi_size = meta_len;
 437                         bip->bip_iter.bi_sector = meta_seed;
 438
 439                         ret = bio_integrity_add_page(bio, virt_to_page(meta),
 440                                         meta_len, offset_in_page(meta));
 441                         if (ret != meta_len) {
 442                                 ret = -ENOMEM;
 443                                 goto out_free_meta;
 444                         }
 445                 }
 446         }
 447  submit:
 448         blk_execute_rq(req->q, disk, req, 0);
 449         ret = req->errors;
 450         if (result)
 451                 *result = le32_to_cpu(cqe.result);
 452         if (meta && !ret && !write) {
 453                 if (copy_to_user(meta_buffer, meta, meta_len))
 454                         ret = -EFAULT;
 455         }
 456  out_free_meta:
 457         kfree(meta);
 458  out_unmap:
 459         if (bio) {
 460                 if (disk && bio->bi_bdev)
 461                         bdput(bio->bi_bdev);
 462                 blk_rq_unmap_user(bio);
 463         }
 464  out:
 465         blk_mq_free_request(req);
 466         return ret;
 467 }
 468
 469 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 470                 void __user *ubuffer, unsigned bufflen, u32 *result,
 471                 unsigned timeout)
 472 {
 473         return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
 474                         result, timeout);
 475 }
 476
 477 static void nvme_keep_alive_end_io(struct request *rq, int error)
 478 {
 479         struct nvme_ctrl *ctrl = rq->end_io_data;
 480
 481         blk_mq_free_request(rq);
 482
 483         if (error) {
 484                 dev_err(ctrl->device,
 485                         "failed nvme_keep_alive_end_io error=%d\n", error);
 486                 return;
 487         }
 488
 489         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 490 }
 491
 492 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 493 {
 494         struct nvme_command c;
 495         struct request *rq;
 496
 497         memset(&c, 0, sizeof(c));
 498         c.common.opcode = nvme_admin_keep_alive;
 499
 500         rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
 501                         NVME_QID_ANY);
 502         if (IS_ERR(rq))
 503                 return PTR_ERR(rq);
 504
 505         rq->timeout = ctrl->kato * HZ;
 506         rq->end_io_data = ctrl;
 507
 508         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 509
 510         return 0;
 511 }
 512
 513 static void nvme_keep_alive_work(struct work_struct *work)
 514 {
 515         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 516                         struct nvme_ctrl, ka_work);
 517
 518         if (nvme_keep_alive(ctrl)) {
 519                 /* allocation failure, reset the controller */
 520                 dev_err(ctrl->device, "keep-alive failed\n");
 521                 ctrl->ops->reset_ctrl(ctrl);
 522                 return;
 523         }
 524 }
 525
 526 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 527 {
 528         if (unlikely(ctrl->kato == 0))
 529                 return;
 530
 531         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
 532         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 533 }
 534 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
 535
 536 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 537 {
 538         if (unlikely(ctrl->kato == 0))
 539                 return;
 540
 541         cancel_delayed_work_sync(&ctrl->ka_work);
 542 }
 543 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 544
 545 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 546 {
 547         struct nvme_command c = { };
 548         int error;
 549
 550         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 551         c.identify.opcode = nvme_admin_identify;
 552         c.identify.cns = cpu_to_le32(1);
 553
 554         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 555         if (!*id)
 556                 return -ENOMEM;
 557
 558         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 559                         sizeof(struct nvme_id_ctrl));
 560         if (error)
 561                 kfree(*id);
 562         return error;
 563 }
 564
 565 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 566 {
 567         struct nvme_command c = { };
 568
 569         c.identify.opcode = nvme_admin_identify;
 570         c.identify.cns = cpu_to_le32(2);
 571         c.identify.nsid = cpu_to_le32(nsid);
 572         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 573 }
 574
 575 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
 576                 struct nvme_id_ns **id)
 577 {
 578         struct nvme_command c = { };
 579         int error;
 580
 581         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 582         c.identify.opcode = nvme_admin_identify,
 583         c.identify.nsid = cpu_to_le32(nsid),
 584
 585         *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
 586         if (!*id)
 587                 return -ENOMEM;
 588
 589         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 590                         sizeof(struct nvme_id_ns));
 591         if (error)
 592                 kfree(*id);
 593         return error;
 594 }
 595
 596 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 597                                         dma_addr_t dma_addr, u32 *result)
 598 {
 599         struct nvme_command c;
 600         struct nvme_completion cqe;
 601         int ret;
 602
 603         memset(&c, 0, sizeof(c));
 604         c.features.opcode = nvme_admin_get_features;
 605         c.features.nsid = cpu_to_le32(nsid);
 606         c.features.dptr.prp1 = cpu_to_le64(dma_addr);
 607         c.features.fid = cpu_to_le32(fid);
 608
 609         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0,
 610                         NVME_QID_ANY, 0, 0);
 611         if (ret >= 0)
 612                 *result = le32_to_cpu(cqe.result);
 613         return ret;
 614 }
 615
 616 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 617                                         dma_addr_t dma_addr, u32 *result)
 618 {
 619         struct nvme_command c;
 620         struct nvme_completion cqe;
 621         int ret;
 622
 623         memset(&c, 0, sizeof(c));
 624         c.features.opcode = nvme_admin_set_features;
 625         c.features.dptr.prp1 = cpu_to_le64(dma_addr);
 626         c.features.fid = cpu_to_le32(fid);
 627         c.features.dword11 = cpu_to_le32(dword11);
 628
 629         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0,
 630                         NVME_QID_ANY, 0, 0);
 631         if (ret >= 0)
 632                 *result = le32_to_cpu(cqe.result);
 633         return ret;
 634 }
 635
 636 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
 637 {
 638         struct nvme_command c = { };
 639         int error;
 640
 641         c.common.opcode = nvme_admin_get_log_page,
 642         c.common.nsid = cpu_to_le32(0xFFFFFFFF),
 643         c.common.cdw10[0] = cpu_to_le32(
 644                         (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
 645                          NVME_LOG_SMART),
 646
 647         *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
 648         if (!*log)
 649                 return -ENOMEM;
 650
 651         error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
 652                         sizeof(struct nvme_smart_log));
 653         if (error)
 654                 kfree(*log);
 655         return error;
 656 }
 657
 658 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
 659 {
 660         u32 q_count = (*count - 1) | ((*count - 1) << 16);
 661         u32 result;
 662         int status, nr_io_queues;
 663
 664         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
 665                         &result);
 666         if (status < 0)
 667                 return status;
 668
 669         /*
 670          * Degraded controllers might return an error when setting the queue
 671          * count.  We still want to be able to bring them online and offer
 672          * access to the admin queue, as that might be only way to fix them up.
 673          */
 674         if (status > 0) {
 675                 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
 676                 *count = 0;
 677         } else {
 678                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
 679                 *count = min(*count, nr_io_queues);
 680         }
 681
 682         return 0;
 683 }
 684 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
 685
 686 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 687 {
 688         struct nvme_user_io io;
 689         struct nvme_command c;
 690         unsigned length, meta_len;
 691         void __user *metadata;
 692
 693         if (copy_from_user(&io, uio, sizeof(io)))
 694                 return -EFAULT;
 695         if (io.flags)
 696                 return -EINVAL;
 697
 698         switch (io.opcode) {
 699         case nvme_cmd_write:
 700         case nvme_cmd_read:
 701         case nvme_cmd_compare:
 702                 break;
 703         default:
 704                 return -EINVAL;
 705         }
 706
 707         length = (io.nblocks + 1) << ns->lba_shift;
 708         meta_len = (io.nblocks + 1) * ns->ms;
 709         metadata = (void __user *)(uintptr_t)io.metadata;
 710
 711         if (ns->ext) {
 712                 length += meta_len;
 713                 meta_len = 0;
 714         } else if (meta_len) {
 715                 if ((io.metadata & 3) || !io.metadata)
 716                         return -EINVAL;
 717         }
 718
 719         memset(&c, 0, sizeof(c));
 720         c.rw.opcode = io.opcode;
 721         c.rw.flags = io.flags;
 722         c.rw.nsid = cpu_to_le32(ns->ns_id);
 723         c.rw.slba = cpu_to_le64(io.slba);
 724         c.rw.length = cpu_to_le16(io.nblocks);
 725         c.rw.control = cpu_to_le16(io.control);
 726         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
 727         c.rw.reftag = cpu_to_le32(io.reftag);
 728         c.rw.apptag = cpu_to_le16(io.apptag);
 729         c.rw.appmask = cpu_to_le16(io.appmask);
 730
 731         return __nvme_submit_user_cmd(ns->queue, &c,
 732                         (void __user *)(uintptr_t)io.addr, length,
 733                         metadata, meta_len, io.slba, NULL, 0);
 734 }
 735
 736 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 737                         struct nvme_passthru_cmd __user *ucmd)
 738 {
 739         struct nvme_passthru_cmd cmd;
 740         struct nvme_command c;
 741         unsigned timeout = 0;
 742         int status;
 743
 744         if (!capable(CAP_SYS_ADMIN))
 745                 return -EACCES;
 746         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
 747                 return -EFAULT;
 748         if (cmd.flags)
 749                 return -EINVAL;
 750
 751         memset(&c, 0, sizeof(c));
 752         c.common.opcode = cmd.opcode;
 753         c.common.flags = cmd.flags;
 754         c.common.nsid = cpu_to_le32(cmd.nsid);
 755         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
 756         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
 757         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
 758         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
 759         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
 760         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
 761         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
 762         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
 763
 764         if (cmd.timeout_ms)
 765                 timeout = msecs_to_jiffies(cmd.timeout_ms);
 766
 767         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
 768                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
 769                         &cmd.result, timeout);
 770         if (status >= 0) {
 771                 if (put_user(cmd.result, &ucmd->result))
 772                         return -EFAULT;
 773         }
 774
 775         return status;
 776 }
 777
 778 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 779                 unsigned int cmd, unsigned long arg)
 780 {
 781         struct nvme_ns *ns = bdev->bd_disk->private_data;
 782
 783         switch (cmd) {
 784         case NVME_IOCTL_ID:
 785                 force_successful_syscall_return();
 786                 return ns->ns_id;
 787         case NVME_IOCTL_ADMIN_CMD:
 788                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
 789         case NVME_IOCTL_IO_CMD:
 790                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
 791         case NVME_IOCTL_SUBMIT_IO:
 792                 return nvme_submit_io(ns, (void __user *)arg);
 793 #ifdef CONFIG_BLK_DEV_NVME_SCSI
 794         case SG_GET_VERSION_NUM:
 795                 return nvme_sg_get_version_num((void __user *)arg);
 796         case SG_IO:
 797                 return nvme_sg_io(ns, (void __user *)arg);
 798 #endif
 799         default:
 800                 return -ENOTTY;
 801         }
 802 }
 803
 804 #ifdef CONFIG_COMPAT
 805 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 806                         unsigned int cmd, unsigned long arg)
 807 {
 808         switch (cmd) {
 809         case SG_IO:
 810                 return -ENOIOCTLCMD;
 811         }
 812         return nvme_ioctl(bdev, mode, cmd, arg);
 813 }
 814 #else
 815 #define nvme_compat_ioctl       NULL
 816 #endif
 817
 818 static int nvme_open(struct block_device *bdev, fmode_t mode)
 819 {
 820         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
 821 }
 822
 823 static void nvme_release(struct gendisk *disk, fmode_t mode)
 824 {
 825         struct nvme_ns *ns = disk->private_data;
 826
 827         module_put(ns->ctrl->ops->module);
 828         nvme_put_ns(ns);
 829 }
 830
 831 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 832 {
 833         /* some standard values */
 834         geo->heads = 1 << 6;
 835         geo->sectors = 1 << 5;
 836         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
 837         return 0;
 838 }
 839
 840 #ifdef CONFIG_BLK_DEV_INTEGRITY
 841 static void nvme_init_integrity(struct nvme_ns *ns)
 842 {
 843         struct blk_integrity integrity;
 844
 845         memset(&integrity, 0, sizeof(integrity));
 846         switch (ns->pi_type) {
 847         case NVME_NS_DPS_PI_TYPE3:
 848                 integrity.profile = &t10_pi_type3_crc;
 849                 integrity.tag_size = sizeof(u16) + sizeof(u32);
 850                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
 851                 break;
 852         case NVME_NS_DPS_PI_TYPE1:
 853         case NVME_NS_DPS_PI_TYPE2:
 854                 integrity.profile = &t10_pi_type1_crc;
 855                 integrity.tag_size = sizeof(u16);
 856                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
 857                 break;
 858         default:
 859                 integrity.profile = NULL;
 860                 break;
 861         }
 862         integrity.tuple_size = ns->ms;
 863         blk_integrity_register(ns->disk, &integrity);
 864         blk_queue_max_integrity_segments(ns->queue, 1);
 865 }
 866 #else
 867 static void nvme_init_integrity(struct nvme_ns *ns)
 868 {
 869 }
 870 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 871
 872 static void nvme_config_discard(struct nvme_ns *ns)
 873 {
 874         struct nvme_ctrl *ctrl = ns->ctrl;
 875         u32 logical_block_size = queue_logical_block_size(ns->queue);
 876
 877         if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
 878                 ns->queue->limits.discard_zeroes_data = 1;
 879         else
 880                 ns->queue->limits.discard_zeroes_data = 0;
 881
 882         ns->queue->limits.discard_alignment = logical_block_size;
 883         ns->queue->limits.discard_granularity = logical_block_size;
 884         blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
 885         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
 886 }
 887
 888 static int nvme_revalidate_disk(struct gendisk *disk)
 889 {
 890         struct nvme_ns *ns = disk->private_data;
 891         struct nvme_id_ns *id;
 892         u8 lbaf, pi_type;
 893         u16 old_ms;
 894         unsigned short bs;
 895
 896         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
 897                 set_capacity(disk, 0);
 898                 return -ENODEV;
 899         }
 900         if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
 901                 dev_warn(disk_to_dev(ns->disk), "%s: Identify failure\n",
 902                                 __func__);
 903                 return -ENODEV;
 904         }
 905         if (id->ncap == 0) {
 906                 kfree(id);
 907                 return -ENODEV;
 908         }
 909
 910         if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
 911                 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
 912                         dev_warn(disk_to_dev(ns->disk),
 913                                 "%s: LightNVM init failure\n", __func__);
 914                         kfree(id);
 915                         return -ENODEV;
 916                 }
 917                 ns->type = NVME_NS_LIGHTNVM;
 918         }
 919
 920         if (ns->ctrl->vs >= NVME_VS(1, 1))
 921                 memcpy(ns->eui, id->eui64, sizeof(ns->eui));
 922         if (ns->ctrl->vs >= NVME_VS(1, 2))
 923                 memcpy(ns->uuid, id->nguid, sizeof(ns->uuid));
 924
 925         old_ms = ns->ms;
 926         lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
 927         ns->lba_shift = id->lbaf[lbaf].ds;
 928         ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
 929         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
 930
 931         /*
 932          * If identify namespace failed, use default 512 byte block size so
 933          * block layer can use before failing read/write for 0 capacity.
 934          */
 935         if (ns->lba_shift == 0)
 936                 ns->lba_shift = 9;
 937         bs = 1 << ns->lba_shift;
 938         /* XXX: PI implementation requires metadata equal t10 pi tuple size */
 939         pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
 940                                         id->dps & NVME_NS_DPS_PI_MASK : 0;
 941
 942         blk_mq_freeze_queue(disk->queue);
 943         if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
 944                                 ns->ms != old_ms ||
 945                                 bs != queue_logical_block_size(disk->queue) ||
 946                                 (ns->ms && ns->ext)))
 947                 blk_integrity_unregister(disk);
 948
 949         ns->pi_type = pi_type;
 950         blk_queue_logical_block_size(ns->queue, bs);
 951
 952         if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
 953                 nvme_init_integrity(ns);
 954         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
 955                 set_capacity(disk, 0);
 956         else
 957                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
 958
 959         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
 960                 nvme_config_discard(ns);
 961         blk_mq_unfreeze_queue(disk->queue);
 962
 963         kfree(id);
 964         return 0;
 965 }
 966
 967 static char nvme_pr_type(enum pr_type type)
 968 {
 969         switch (type) {
 970         case PR_WRITE_EXCLUSIVE:
 971                 return 1;
 972         case PR_EXCLUSIVE_ACCESS:
 973                 return 2;
 974         case PR_WRITE_EXCLUSIVE_REG_ONLY:
 975                 return 3;
 976         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
 977                 return 4;
 978         case PR_WRITE_EXCLUSIVE_ALL_REGS:
 979                 return 5;
 980         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
 981                 return 6;
 982         default:
 983                 return 0;
 984         }
 985 };
 986
 987 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
 988                                 u64 key, u64 sa_key, u8 op)
 989 {
 990         struct nvme_ns *ns = bdev->bd_disk->private_data;
 991         struct nvme_command c;
 992         u8 data[16] = { 0, };
 993
 994         put_unaligned_le64(key, &data[0]);
 995         put_unaligned_le64(sa_key, &data[8]);
 996
 997         memset(&c, 0, sizeof(c));
 998         c.common.opcode = op;
 999         c.common.nsid = cpu_to_le32(ns->ns_id);
1000         c.common.cdw10[0] = cpu_to_le32(cdw10);
1001
1002         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1003 }
1004
1005 static int nvme_pr_register(struct block_device *bdev, u64 old,
1006                 u64 new, unsigned flags)
1007 {
1008         u32 cdw10;
1009
1010         if (flags & ~PR_FL_IGNORE_KEY)
1011                 return -EOPNOTSUPP;
1012
1013         cdw10 = old ? 2 : 0;
1014         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1015         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1016         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1017 }
1018
1019 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1020                 enum pr_type type, unsigned flags)
1021 {
1022         u32 cdw10;
1023
1024         if (flags & ~PR_FL_IGNORE_KEY)
1025                 return -EOPNOTSUPP;
1026
1027         cdw10 = nvme_pr_type(type) << 8;
1028         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1029         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1030 }
1031
1032 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1033                 enum pr_type type, bool abort)
1034 {
1035         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1036         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1037 }
1038
1039 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1040 {
1041         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1042         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1043 }
1044
1045 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1046 {
1047         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1048         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1049 }
1050
1051 static const struct pr_ops nvme_pr_ops = {
1052         .pr_register    = nvme_pr_register,
1053         .pr_reserve     = nvme_pr_reserve,
1054         .pr_release     = nvme_pr_release,
1055         .pr_preempt     = nvme_pr_preempt,
1056         .pr_clear       = nvme_pr_clear,
1057 };
1058
1059 static const struct block_device_operations nvme_fops = {
1060         .owner          = THIS_MODULE,
1061         .ioctl          = nvme_ioctl,
1062         .compat_ioctl   = nvme_compat_ioctl,
1063         .open           = nvme_open,
1064         .release        = nvme_release,
1065         .getgeo         = nvme_getgeo,
1066         .revalidate_disk= nvme_revalidate_disk,
1067         .pr_ops         = &nvme_pr_ops,
1068 };
1069
1070 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1071 {
1072         unsigned long timeout =
1073                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1074         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1075         int ret;
1076
1077         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1078                 if ((csts & NVME_CSTS_RDY) == bit)
1079                         break;
1080
1081                 msleep(100);
1082                 if (fatal_signal_pending(current))
1083                         return -EINTR;
1084                 if (time_after(jiffies, timeout)) {
1085                         dev_err(ctrl->device,
1086                                 "Device not ready; aborting %s\n", enabled ?
1087                                                 "initialisation" : "reset");
1088                         return -ENODEV;
1089                 }
1090         }
1091
1092         return ret;
1093 }
1094
1095 /*
1096  * If the device has been passed off to us in an enabled state, just clear
1097  * the enabled bit.  The spec says we should set the 'shutdown notification
1098  * bits', but doing so may cause the device to complete commands to the
1099  * admin queue ... and we don't know what memory that might be pointing at!
1100  */
1101 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1102 {
1103         int ret;
1104
1105         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1106         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1107
1108         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1109         if (ret)
1110                 return ret;
1111
1112         /* Checking for ctrl->tagset is a trick to avoid sleeping on module
1113          * load, since we only need the quirk on reset_controller. Notice
1114          * that the HGST device needs this delay only in firmware activation
1115          * procedure; unfortunately we have no (easy) way to verify this.
1116          */
1117         if ((ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY) && ctrl->tagset)
1118                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1119
1120         return nvme_wait_ready(ctrl, cap, false);
1121 }
1122 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1123
1124 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1125 {
1126         /*
1127          * Default to a 4K page size, with the intention to update this
1128          * path in the future to accomodate architectures with differing
1129          * kernel and IO page sizes.
1130          */
1131         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1132         int ret;
1133
1134         if (page_shift < dev_page_min) {
1135                 dev_err(ctrl->device,
1136                         "Minimum device page size %u too large for host (%u)\n",
1137                         1 << dev_page_min, 1 << page_shift);
1138                 return -ENODEV;
1139         }
1140
1141         ctrl->page_size = 1 << page_shift;
1142
1143         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1144         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1145         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1146         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1147         ctrl->ctrl_config |= NVME_CC_ENABLE;
1148
1149         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1150         if (ret)
1151                 return ret;
1152         return nvme_wait_ready(ctrl, cap, true);
1153 }
1154 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1155
1156 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1157 {
1158         unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1159         u32 csts;
1160         int ret;
1161
1162         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1163         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1164
1165         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1166         if (ret)
1167                 return ret;
1168
1169         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1170                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1171                         break;
1172
1173                 msleep(100);
1174                 if (fatal_signal_pending(current))
1175                         return -EINTR;
1176                 if (time_after(jiffies, timeout)) {
1177                         dev_err(ctrl->device,
1178                                 "Device shutdown incomplete; abort shutdown\n");
1179                         return -ENODEV;
1180                 }
1181         }
1182
1183         return ret;
1184 }
1185 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1186
1187 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1188                 struct request_queue *q)
1189 {
1190         bool vwc = false;
1191
1192         if (ctrl->max_hw_sectors) {
1193                 u32 max_segments =
1194                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1195
1196                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1197                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1198         }
1199         if (ctrl->stripe_size)
1200                 blk_queue_chunk_sectors(q, ctrl->stripe_size >> 9);
1201         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1202         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1203                 vwc = true;
1204         blk_queue_write_cache(q, vwc, vwc);
1205 }
1206
1207 /*
1208  * Initialize the cached copies of the Identify data and various controller
1209  * register in our nvme_ctrl structure.  This should be called as soon as
1210  * the admin queue is fully up and running.
1211  */
1212 int nvme_init_identify(struct nvme_ctrl *ctrl)
1213 {
1214         struct nvme_id_ctrl *id;
1215         u64 cap;
1216         int ret, page_shift;
1217         u32 max_hw_sectors;
1218
1219         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1220         if (ret) {
1221                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1222                 return ret;
1223         }
1224
1225         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1226         if (ret) {
1227                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1228                 return ret;
1229         }
1230         page_shift = NVME_CAP_MPSMIN(cap) + 12;
1231
1232         if (ctrl->vs >= NVME_VS(1, 1))
1233                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1234
1235         ret = nvme_identify_ctrl(ctrl, &id);
1236         if (ret) {
1237                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1238                 return -EIO;
1239         }
1240
1241         ctrl->vid = le16_to_cpu(id->vid);
1242         ctrl->oncs = le16_to_cpup(&id->oncs);
1243         atomic_set(&ctrl->abort_limit, id->acl + 1);
1244         ctrl->vwc = id->vwc;
1245         ctrl->cntlid = le16_to_cpup(&id->cntlid);
1246         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1247         memcpy(ctrl->model, id->mn, sizeof(id->mn));
1248         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1249         if (id->mdts)
1250                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1251         else
1252                 max_hw_sectors = UINT_MAX;
1253         ctrl->max_hw_sectors =
1254                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1255
1256         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
1257                 unsigned int max_hw_sectors;
1258
1259                 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
1260                 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
1261                 if (ctrl->max_hw_sectors) {
1262                         ctrl->max_hw_sectors = min(max_hw_sectors,
1263                                                         ctrl->max_hw_sectors);
1264                 } else {
1265                         ctrl->max_hw_sectors = max_hw_sectors;
1266                 }
1267         }
1268
1269         nvme_set_queue_limits(ctrl, ctrl->admin_q);
1270         ctrl->sgls = le32_to_cpu(id->sgls);
1271         ctrl->kas = le16_to_cpu(id->kas);
1272
1273         if (ctrl->ops->is_fabrics) {
1274                 ctrl->icdoff = le16_to_cpu(id->icdoff);
1275                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1276                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1277                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1278
1279                 /*
1280                  * In fabrics we need to verify the cntlid matches the
1281                  * admin connect
1282                  */
1283                 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1284                         ret = -EINVAL;
1285
1286                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1287                         dev_err(ctrl->dev,
1288                                 "keep-alive support is mandatory for fabrics\n");
1289                         ret = -EINVAL;
1290                 }
1291         } else {
1292                 ctrl->cntlid = le16_to_cpu(id->cntlid);
1293         }
1294
1295         kfree(id);
1296         return ret;
1297 }
1298 EXPORT_SYMBOL_GPL(nvme_init_identify);
1299
1300 static int nvme_dev_open(struct inode *inode, struct file *file)
1301 {
1302         struct nvme_ctrl *ctrl;
1303         int instance = iminor(inode);
1304         int ret = -ENODEV;
1305
1306         spin_lock(&dev_list_lock);
1307         list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1308                 if (ctrl->instance != instance)
1309                         continue;
1310
1311                 if (!ctrl->admin_q) {
1312                         ret = -EWOULDBLOCK;
1313                         break;
1314                 }
1315                 if (!kref_get_unless_zero(&ctrl->kref))
1316                         break;
1317                 file->private_data = ctrl;
1318                 ret = 0;
1319                 break;
1320         }
1321         spin_unlock(&dev_list_lock);
1322
1323         return ret;
1324 }
1325
1326 static int nvme_dev_release(struct inode *inode, struct file *file)
1327 {
1328         nvme_put_ctrl(file->private_data);
1329         return 0;
1330 }
1331
1332 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1333 {
1334         struct nvme_ns *ns;
1335         int ret;
1336
1337         mutex_lock(&ctrl->namespaces_mutex);
1338         if (list_empty(&ctrl->namespaces)) {
1339                 ret = -ENOTTY;
1340                 goto out_unlock;
1341         }
1342
1343         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1344         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1345                 dev_warn(ctrl->device,
1346                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1347                 ret = -EINVAL;
1348                 goto out_unlock;
1349         }
1350
1351         dev_warn(ctrl->device,
1352                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1353         kref_get(&ns->kref);
1354         mutex_unlock(&ctrl->namespaces_mutex);
1355
1356         ret = nvme_user_cmd(ctrl, ns, argp);
1357         nvme_put_ns(ns);
1358         return ret;
1359
1360 out_unlock:
1361         mutex_unlock(&ctrl->namespaces_mutex);
1362         return ret;
1363 }
1364
1365 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1366                 unsigned long arg)
1367 {
1368         struct nvme_ctrl *ctrl = file->private_data;
1369         void __user *argp = (void __user *)arg;
1370
1371         switch (cmd) {
1372         case NVME_IOCTL_ADMIN_CMD:
1373                 return nvme_user_cmd(ctrl, NULL, argp);
1374         case NVME_IOCTL_IO_CMD:
1375                 return nvme_dev_user_cmd(ctrl, argp);
1376         case NVME_IOCTL_RESET:
1377                 dev_warn(ctrl->device, "resetting controller\n");
1378                 return ctrl->ops->reset_ctrl(ctrl);
1379         case NVME_IOCTL_SUBSYS_RESET:
1380                 return nvme_reset_subsystem(ctrl);
1381         case NVME_IOCTL_RESCAN:
1382                 nvme_queue_scan(ctrl);
1383                 return 0;
1384         default:
1385                 return -ENOTTY;
1386         }
1387 }
1388
1389 static const struct file_operations nvme_dev_fops = {
1390         .owner          = THIS_MODULE,
1391         .open           = nvme_dev_open,
1392         .release        = nvme_dev_release,
1393         .unlocked_ioctl = nvme_dev_ioctl,
1394         .compat_ioctl   = nvme_dev_ioctl,
1395 };
1396
1397 static ssize_t nvme_sysfs_reset(struct device *dev,
1398                                 struct device_attribute *attr, const char *buf,
1399                                 size_t count)
1400 {
1401         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1402         int ret;
1403
1404         ret = ctrl->ops->reset_ctrl(ctrl);
1405         if (ret < 0)
1406                 return ret;
1407         return count;
1408 }
1409 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1410
1411 static ssize_t nvme_sysfs_rescan(struct device *dev,
1412                                 struct device_attribute *attr, const char *buf,
1413                                 size_t count)
1414 {
1415         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1416
1417         nvme_queue_scan(ctrl);
1418         return count;
1419 }
1420 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1421
1422 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1423                                                                 char *buf)
1424 {
1425         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1426         struct nvme_ctrl *ctrl = ns->ctrl;
1427         int serial_len = sizeof(ctrl->serial);
1428         int model_len = sizeof(ctrl->model);
1429
1430         if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1431                 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1432
1433         if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1434                 return sprintf(buf, "eui.%8phN\n", ns->eui);
1435
1436         while (ctrl->serial[serial_len - 1] == ' ')
1437                 serial_len--;
1438         while (ctrl->model[model_len - 1] == ' ')
1439                 model_len--;
1440
1441         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1442                 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1443 }
1444 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1445
1446 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1447                                                                 char *buf)
1448 {
1449         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1450         return sprintf(buf, "%pU\n", ns->uuid);
1451 }
1452 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1453
1454 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1455                                                                 char *buf)
1456 {
1457         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1458         return sprintf(buf, "%8phd\n", ns->eui);
1459 }
1460 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1461
1462 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1463                                                                 char *buf)
1464 {
1465         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1466         return sprintf(buf, "%d\n", ns->ns_id);
1467 }
1468 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1469
1470 static struct attribute *nvme_ns_attrs[] = {
1471         &dev_attr_wwid.attr,
1472         &dev_attr_uuid.attr,
1473         &dev_attr_eui.attr,
1474         &dev_attr_nsid.attr,
1475         NULL,
1476 };
1477
1478 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1479                 struct attribute *a, int n)
1480 {
1481         struct device *dev = container_of(kobj, struct device, kobj);
1482         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1483
1484         if (a == &dev_attr_uuid.attr) {
1485                 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1486                         return 0;
1487         }
1488         if (a == &dev_attr_eui.attr) {
1489                 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1490                         return 0;
1491         }
1492         return a->mode;
1493 }
1494
1495 static const struct attribute_group nvme_ns_attr_group = {
1496         .attrs          = nvme_ns_attrs,
1497         .is_visible     = nvme_ns_attrs_are_visible,
1498 };
1499
1500 #define nvme_show_str_function(field)                                           \
1501 static ssize_t  field##_show(struct device *dev,                                \
1502                             struct device_attribute *attr, char *buf)           \
1503 {                                                                               \
1504         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1505         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);   \
1506 }                                                                               \
1507 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1508
1509 #define nvme_show_int_function(field)                                           \
1510 static ssize_t  field##_show(struct device *dev,                                \
1511                             struct device_attribute *attr, char *buf)           \
1512 {                                                                               \
1513         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1514         return sprintf(buf, "%d\n", ctrl->field);       \
1515 }                                                                               \
1516 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1517
1518 nvme_show_str_function(model);
1519 nvme_show_str_function(serial);
1520 nvme_show_str_function(firmware_rev);
1521 nvme_show_int_function(cntlid);
1522
1523 static ssize_t nvme_sysfs_delete(struct device *dev,
1524                                 struct device_attribute *attr, const char *buf,
1525                                 size_t count)
1526 {
1527         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1528
1529         if (device_remove_file_self(dev, attr))
1530                 ctrl->ops->delete_ctrl(ctrl);
1531         return count;
1532 }
1533 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1534
1535 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1536                                          struct device_attribute *attr,
1537                                          char *buf)
1538 {
1539         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1540
1541         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1542 }
1543 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1544
1545 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1546                                          struct device_attribute *attr,
1547                                          char *buf)
1548 {
1549         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1550
1551         return snprintf(buf, PAGE_SIZE, "%s\n",
1552                         ctrl->ops->get_subsysnqn(ctrl));
1553 }
1554 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1555
1556 static ssize_t nvme_sysfs_show_address(struct device *dev,
1557                                          struct device_attribute *attr,
1558                                          char *buf)
1559 {
1560         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1561
1562         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1563 }
1564 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1565
1566 static struct attribute *nvme_dev_attrs[] = {
1567         &dev_attr_reset_controller.attr,
1568         &dev_attr_rescan_controller.attr,
1569         &dev_attr_model.attr,
1570         &dev_attr_serial.attr,
1571         &dev_attr_firmware_rev.attr,
1572         &dev_attr_cntlid.attr,
1573         &dev_attr_delete_controller.attr,
1574         &dev_attr_transport.attr,
1575         &dev_attr_subsysnqn.attr,
1576         &dev_attr_address.attr,
1577         NULL
1578 };
1579
1580 #define CHECK_ATTR(ctrl, a, name)               \
1581         if ((a) == &dev_attr_##name.attr &&     \
1582             !(ctrl)->ops->get_##name)           \
1583                 return 0
1584
1585 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1586                 struct attribute *a, int n)
1587 {
1588         struct device *dev = container_of(kobj, struct device, kobj);
1589         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1590
1591         if (a == &dev_attr_delete_controller.attr) {
1592                 if (!ctrl->ops->delete_ctrl)
1593                         return 0;
1594         }
1595
1596         CHECK_ATTR(ctrl, a, subsysnqn);
1597         CHECK_ATTR(ctrl, a, address);
1598
1599         return a->mode;
1600 }
1601
1602 static struct attribute_group nvme_dev_attrs_group = {
1603         .attrs          = nvme_dev_attrs,
1604         .is_visible     = nvme_dev_attrs_are_visible,
1605 };
1606
1607 static const struct attribute_group *nvme_dev_attr_groups[] = {
1608         &nvme_dev_attrs_group,
1609         NULL,
1610 };
1611
1612 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1613 {
1614         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1615         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1616
1617         return nsa->ns_id - nsb->ns_id;
1618 }
1619
1620 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1621 {
1622         struct nvme_ns *ns, *ret = NULL;
1623
1624         mutex_lock(&ctrl->namespaces_mutex);
1625         list_for_each_entry(ns, &ctrl->namespaces, list) {
1626                 if (ns->ns_id == nsid) {
1627                         kref_get(&ns->kref);
1628                         ret = ns;
1629                         break;
1630                 }
1631                 if (ns->ns_id > nsid)
1632                         break;
1633         }
1634         mutex_unlock(&ctrl->namespaces_mutex);
1635         return ret;
1636 }
1637
1638 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1639 {
1640         struct nvme_ns *ns;
1641         struct gendisk *disk;
1642         int node = dev_to_node(ctrl->dev);
1643
1644         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1645         if (!ns)
1646                 return;
1647
1648         ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1649         if (ns->instance < 0)
1650                 goto out_free_ns;
1651
1652         ns->queue = blk_mq_init_queue(ctrl->tagset);
1653         if (IS_ERR(ns->queue))
1654                 goto out_release_instance;
1655         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1656         ns->queue->queuedata = ns;
1657         ns->ctrl = ctrl;
1658
1659         disk = alloc_disk_node(0, node);
1660         if (!disk)
1661                 goto out_free_queue;
1662
1663         kref_init(&ns->kref);
1664         ns->ns_id = nsid;
1665         ns->disk = disk;
1666         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1667
1668
1669         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1670         nvme_set_queue_limits(ctrl, ns->queue);
1671
1672         disk->fops = &nvme_fops;
1673         disk->private_data = ns;
1674         disk->queue = ns->queue;
1675         disk->flags = GENHD_FL_EXT_DEVT;
1676         sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1677
1678         if (nvme_revalidate_disk(ns->disk))
1679                 goto out_free_disk;
1680
1681         mutex_lock(&ctrl->namespaces_mutex);
1682         list_add_tail(&ns->list, &ctrl->namespaces);
1683         mutex_unlock(&ctrl->namespaces_mutex);
1684
1685         kref_get(&ctrl->kref);
1686         if (ns->type == NVME_NS_LIGHTNVM)
1687                 return;
1688
1689         device_add_disk(ctrl->device, ns->disk);
1690         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1691                                         &nvme_ns_attr_group))
1692                 pr_warn("%s: failed to create sysfs group for identification\n",
1693                         ns->disk->disk_name);
1694         return;
1695  out_free_disk:
1696         kfree(disk);
1697  out_free_queue:
1698         blk_cleanup_queue(ns->queue);
1699  out_release_instance:
1700         ida_simple_remove(&ctrl->ns_ida, ns->instance);
1701  out_free_ns:
1702         kfree(ns);
1703 }
1704
1705 static void nvme_ns_remove(struct nvme_ns *ns)
1706 {
1707         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1708                 return;
1709
1710         if (ns->disk->flags & GENHD_FL_UP) {
1711                 if (blk_get_integrity(ns->disk))
1712                         blk_integrity_unregister(ns->disk);
1713                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1714                                         &nvme_ns_attr_group);
1715                 del_gendisk(ns->disk);
1716                 blk_mq_abort_requeue_list(ns->queue);
1717                 blk_cleanup_queue(ns->queue);
1718         }
1719
1720         mutex_lock(&ns->ctrl->namespaces_mutex);
1721         list_del_init(&ns->list);
1722         mutex_unlock(&ns->ctrl->namespaces_mutex);
1723
1724         nvme_put_ns(ns);
1725 }
1726
1727 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1728 {
1729         struct nvme_ns *ns;
1730
1731         ns = nvme_find_get_ns(ctrl, nsid);
1732         if (ns) {
1733                 if (revalidate_disk(ns->disk))
1734                         nvme_ns_remove(ns);
1735                 nvme_put_ns(ns);
1736         } else
1737                 nvme_alloc_ns(ctrl, nsid);
1738 }
1739
1740 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
1741                                         unsigned nsid)
1742 {
1743         struct nvme_ns *ns, *next;
1744
1745         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1746                 if (ns->ns_id > nsid)
1747                         nvme_ns_remove(ns);
1748         }
1749 }
1750
1751 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1752 {
1753         struct nvme_ns *ns;
1754         __le32 *ns_list;
1755         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1756         int ret = 0;
1757
1758         ns_list = kzalloc(0x1000, GFP_KERNEL);
1759         if (!ns_list)
1760                 return -ENOMEM;
1761
1762         for (i = 0; i < num_lists; i++) {
1763                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1764                 if (ret)
1765                         goto free;
1766
1767                 for (j = 0; j < min(nn, 1024U); j++) {
1768                         nsid = le32_to_cpu(ns_list[j]);
1769                         if (!nsid)
1770                                 goto out;
1771
1772                         nvme_validate_ns(ctrl, nsid);
1773
1774                         while (++prev < nsid) {
1775                                 ns = nvme_find_get_ns(ctrl, prev);
1776                                 if (ns) {
1777                                         nvme_ns_remove(ns);
1778                                         nvme_put_ns(ns);
1779                                 }
1780                         }
1781                 }
1782                 nn -= j;
1783         }
1784  out:
1785         nvme_remove_invalid_namespaces(ctrl, prev);
1786  free:
1787         kfree(ns_list);
1788         return ret;
1789 }
1790
1791 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
1792 {
1793         unsigned i;
1794
1795         for (i = 1; i <= nn; i++)
1796                 nvme_validate_ns(ctrl, i);
1797
1798         nvme_remove_invalid_namespaces(ctrl, nn);
1799 }
1800
1801 static void nvme_scan_work(struct work_struct *work)
1802 {
1803         struct nvme_ctrl *ctrl =
1804                 container_of(work, struct nvme_ctrl, scan_work);
1805         struct nvme_id_ctrl *id;
1806         unsigned nn;
1807
1808         if (ctrl->state != NVME_CTRL_LIVE)
1809                 return;
1810
1811         if (nvme_identify_ctrl(ctrl, &id))
1812                 return;
1813
1814         nn = le32_to_cpu(id->nn);
1815         if (ctrl->vs >= NVME_VS(1, 1) &&
1816             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1817                 if (!nvme_scan_ns_list(ctrl, nn))
1818                         goto done;
1819         }
1820         nvme_scan_ns_sequential(ctrl, nn);
1821  done:
1822         mutex_lock(&ctrl->namespaces_mutex);
1823         list_sort(NULL, &ctrl->namespaces, ns_cmp);
1824         mutex_unlock(&ctrl->namespaces_mutex);
1825         kfree(id);
1826
1827         if (ctrl->ops->post_scan)
1828                 ctrl->ops->post_scan(ctrl);
1829 }
1830
1831 void nvme_queue_scan(struct nvme_ctrl *ctrl)
1832 {
1833         /*
1834          * Do not queue new scan work when a controller is reset during
1835          * removal.
1836          */
1837         if (ctrl->state == NVME_CTRL_LIVE)
1838                 schedule_work(&ctrl->scan_work);
1839 }
1840 EXPORT_SYMBOL_GPL(nvme_queue_scan);
1841
1842 /*
1843  * This function iterates the namespace list unlocked to allow recovery from
1844  * controller failure. It is up to the caller to ensure the namespace list is
1845  * not modified by scan work while this function is executing.
1846  */
1847 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1848 {
1849         struct nvme_ns *ns, *next;
1850
1851         /*
1852          * The dead states indicates the controller was not gracefully
1853          * disconnected. In that case, we won't be able to flush any data while
1854          * removing the namespaces' disks; fail all the queues now to avoid
1855          * potentially having to clean up the failed sync later.
1856          */
1857         if (ctrl->state == NVME_CTRL_DEAD)
1858                 nvme_kill_queues(ctrl);
1859
1860         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1861                 nvme_ns_remove(ns);
1862 }
1863 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1864
1865 static void nvme_async_event_work(struct work_struct *work)
1866 {
1867         struct nvme_ctrl *ctrl =
1868                 container_of(work, struct nvme_ctrl, async_event_work);
1869
1870         spin_lock_irq(&ctrl->lock);
1871         while (ctrl->event_limit > 0) {
1872                 int aer_idx = --ctrl->event_limit;
1873
1874                 spin_unlock_irq(&ctrl->lock);
1875                 ctrl->ops->submit_async_event(ctrl, aer_idx);
1876                 spin_lock_irq(&ctrl->lock);
1877         }
1878         spin_unlock_irq(&ctrl->lock);
1879 }
1880
1881 void nvme_complete_async_event(struct nvme_ctrl *ctrl,
1882                 struct nvme_completion *cqe)
1883 {
1884         u16 status = le16_to_cpu(cqe->status) >> 1;
1885         u32 result = le32_to_cpu(cqe->result);
1886
1887         if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ) {
1888                 ++ctrl->event_limit;
1889                 schedule_work(&ctrl->async_event_work);
1890         }
1891
1892         if (status != NVME_SC_SUCCESS)
1893                 return;
1894
1895         switch (result & 0xff07) {
1896         case NVME_AER_NOTICE_NS_CHANGED:
1897                 dev_info(ctrl->device, "rescanning\n");
1898                 nvme_queue_scan(ctrl);
1899                 break;
1900         default:
1901                 dev_warn(ctrl->device, "async event result %08x\n", result);
1902         }
1903 }
1904 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
1905
1906 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
1907 {
1908         ctrl->event_limit = NVME_NR_AERS;
1909         schedule_work(&ctrl->async_event_work);
1910 }
1911 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
1912
1913 static DEFINE_IDA(nvme_instance_ida);
1914
1915 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1916 {
1917         int instance, error;
1918
1919         do {
1920                 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1921                         return -ENODEV;
1922
1923                 spin_lock(&dev_list_lock);
1924                 error = ida_get_new(&nvme_instance_ida, &instance);
1925                 spin_unlock(&dev_list_lock);
1926         } while (error == -EAGAIN);
1927
1928         if (error)
1929                 return -ENODEV;
1930
1931         ctrl->instance = instance;
1932         return 0;
1933 }
1934
1935 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1936 {
1937         spin_lock(&dev_list_lock);
1938         ida_remove(&nvme_instance_ida, ctrl->instance);
1939         spin_unlock(&dev_list_lock);
1940 }
1941
1942 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1943 {
1944         flush_work(&ctrl->async_event_work);
1945         flush_work(&ctrl->scan_work);
1946         nvme_remove_namespaces(ctrl);
1947
1948         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1949
1950         spin_lock(&dev_list_lock);
1951         list_del(&ctrl->node);
1952         spin_unlock(&dev_list_lock);
1953 }
1954 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
1955
1956 static void nvme_free_ctrl(struct kref *kref)
1957 {
1958         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1959
1960         put_device(ctrl->device);
1961         nvme_release_instance(ctrl);
1962         ida_destroy(&ctrl->ns_ida);
1963
1964         ctrl->ops->free_ctrl(ctrl);
1965 }
1966
1967 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1968 {
1969         kref_put(&ctrl->kref, nvme_free_ctrl);
1970 }
1971 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
1972
1973 /*
1974  * Initialize a NVMe controller structures.  This needs to be called during
1975  * earliest initialization so that we have the initialized structured around
1976  * during probing.
1977  */
1978 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1979                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1980 {
1981         int ret;
1982
1983         ctrl->state = NVME_CTRL_NEW;
1984         spin_lock_init(&ctrl->lock);
1985         INIT_LIST_HEAD(&ctrl->namespaces);
1986         mutex_init(&ctrl->namespaces_mutex);
1987         kref_init(&ctrl->kref);
1988         ctrl->dev = dev;
1989         ctrl->ops = ops;
1990         ctrl->quirks = quirks;
1991         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
1992         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
1993
1994         ret = nvme_set_instance(ctrl);
1995         if (ret)
1996                 goto out;
1997
1998         ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
1999                                 MKDEV(nvme_char_major, ctrl->instance),
2000                                 ctrl, nvme_dev_attr_groups,
2001                                 "nvme%d", ctrl->instance);
2002         if (IS_ERR(ctrl->device)) {
2003                 ret = PTR_ERR(ctrl->device);
2004                 goto out_release_instance;
2005         }
2006         get_device(ctrl->device);
2007         ida_init(&ctrl->ns_ida);
2008
2009         spin_lock(&dev_list_lock);
2010         list_add_tail(&ctrl->node, &nvme_ctrl_list);
2011         spin_unlock(&dev_list_lock);
2012
2013         return 0;
2014 out_release_instance:
2015         nvme_release_instance(ctrl);
2016 out:
2017         return ret;
2018 }
2019 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2020
2021 /**
2022  * nvme_kill_queues(): Ends all namespace queues
2023  * @ctrl: the dead controller that needs to end
2024  *
2025  * Call this function when the driver determines it is unable to get the
2026  * controller in a state capable of servicing IO.
2027  */
2028 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2029 {
2030         struct nvme_ns *ns;
2031
2032         mutex_lock(&ctrl->namespaces_mutex);
2033         list_for_each_entry(ns, &ctrl->namespaces, list) {
2034                 /*
2035                  * Revalidating a dead namespace sets capacity to 0. This will
2036                  * end buffered writers dirtying pages that can't be synced.
2037                  */
2038                 if (!test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2039                         revalidate_disk(ns->disk);
2040
2041                 blk_set_queue_dying(ns->queue);
2042                 blk_mq_abort_requeue_list(ns->queue);
2043                 blk_mq_start_stopped_hw_queues(ns->queue, true);
2044         }
2045         mutex_unlock(&ctrl->namespaces_mutex);
2046 }
2047 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2048
2049 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2050 {
2051         struct nvme_ns *ns;
2052
2053         mutex_lock(&ctrl->namespaces_mutex);
2054         list_for_each_entry(ns, &ctrl->namespaces, list) {
2055                 spin_lock_irq(ns->queue->queue_lock);
2056                 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
2057                 spin_unlock_irq(ns->queue->queue_lock);
2058
2059                 blk_mq_cancel_requeue_work(ns->queue);
2060                 blk_mq_stop_hw_queues(ns->queue);
2061         }
2062         mutex_unlock(&ctrl->namespaces_mutex);
2063 }
2064 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2065
2066 void nvme_start_queues(struct nvme_ctrl *ctrl)
2067 {
2068         struct nvme_ns *ns;
2069
2070         mutex_lock(&ctrl->namespaces_mutex);
2071         list_for_each_entry(ns, &ctrl->namespaces, list) {
2072                 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
2073                 blk_mq_start_stopped_hw_queues(ns->queue, true);
2074                 blk_mq_kick_requeue_list(ns->queue);
2075         }
2076         mutex_unlock(&ctrl->namespaces_mutex);
2077 }
2078 EXPORT_SYMBOL_GPL(nvme_start_queues);
2079
2080 int __init nvme_core_init(void)
2081 {
2082         int result;
2083
2084         result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2085                                                         &nvme_dev_fops);
2086         if (result < 0)
2087                 return result;
2088         else if (result > 0)
2089                 nvme_char_major = result;
2090
2091         nvme_class = class_create(THIS_MODULE, "nvme");
2092         if (IS_ERR(nvme_class)) {
2093                 result = PTR_ERR(nvme_class);
2094                 goto unregister_chrdev;
2095         }
2096
2097         return 0;
2098
2099  unregister_chrdev:
2100         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2101         return result;
2102 }
2103
2104 void nvme_core_exit(void)
2105 {
2106         class_destroy(nvme_class);
2107         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2108 }
2109
2110 MODULE_LICENSE("GPL");
2111 MODULE_VERSION("1.0");
2112 module_init(nvme_core_init);
2113 module_exit(nvme_core_exit);