root/drivers/nvme/host/core.c

/* [<][>][^][v][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. nvme_set_queue_dying
  2. nvme_queue_scan
  3. nvme_try_sched_reset
  4. nvme_reset_ctrl
  5. nvme_reset_ctrl_sync
  6. nvme_do_delete_ctrl
  7. nvme_delete_ctrl_work
  8. nvme_delete_ctrl
  9. nvme_delete_ctrl_sync
  10. nvme_ns_has_pi
  11. nvme_error_status
  12. nvme_req_needs_retry
  13. nvme_retry_req
  14. nvme_complete_rq
  15. nvme_cancel_request
  16. nvme_change_ctrl_state
  17. nvme_state_terminal
  18. nvme_wait_reset
  19. nvme_free_ns_head
  20. nvme_put_ns_head
  21. nvme_free_ns
  22. nvme_put_ns
  23. nvme_clear_nvme_request
  24. nvme_alloc_request
  25. nvme_toggle_streams
  26. nvme_disable_streams
  27. nvme_enable_streams
  28. nvme_get_stream_params
  29. nvme_configure_directives
  30. nvme_assign_write_stream
  31. nvme_setup_flush
  32. nvme_setup_discard
  33. nvme_setup_write_zeroes
  34. nvme_setup_rw
  35. nvme_cleanup_cmd
  36. nvme_setup_cmd
  37. nvme_end_sync_rq
  38. nvme_execute_rq_polled
  39. __nvme_submit_sync_cmd
  40. nvme_submit_sync_cmd
  41. nvme_add_user_metadata
  42. nvme_submit_user_cmd
  43. nvme_keep_alive_end_io
  44. nvme_keep_alive
  45. nvme_keep_alive_work
  46. nvme_start_keep_alive
  47. nvme_stop_keep_alive
  48. nvme_identify_ctrl
  49. nvme_identify_ns_descs
  50. nvme_identify_ns_list
  51. nvme_identify_ns
  52. nvme_features
  53. nvme_set_features
  54. nvme_get_features
  55. nvme_set_queue_count
  56. nvme_enable_aen
  57. nvme_to_user_ptr
  58. nvme_submit_io
  59. nvme_known_admin_effects
  60. nvme_passthru_start
  61. nvme_update_formats
  62. nvme_passthru_end
  63. nvme_user_cmd
  64. nvme_user_cmd64
  65. nvme_get_ns_from_disk
  66. nvme_put_ns_from_disk
  67. is_ctrl_ioctl
  68. nvme_handle_ctrl_ioctl
  69. nvme_ioctl
  70. nvme_open
  71. nvme_release
  72. nvme_getgeo
  73. nvme_init_integrity
  74. nvme_init_integrity
  75. nvme_set_chunk_size
  76. nvme_config_discard
  77. nvme_config_write_zeroes
  78. nvme_report_ns_ids
  79. nvme_ns_ids_valid
  80. nvme_ns_ids_equal
  81. nvme_update_disk_info
  82. __nvme_revalidate_disk
  83. nvme_revalidate_disk
  84. nvme_pr_type
  85. nvme_pr_command
  86. nvme_pr_register
  87. nvme_pr_reserve
  88. nvme_pr_preempt
  89. nvme_pr_clear
  90. nvme_pr_release
  91. nvme_sec_submit
  92. nvme_ns_head_open
  93. nvme_ns_head_release
  94. nvme_wait_ready
  95. nvme_disable_ctrl
  96. nvme_enable_ctrl
  97. nvme_shutdown_ctrl
  98. nvme_set_queue_limits
  99. nvme_configure_timestamp
  100. nvme_configure_acre
  101. nvme_configure_apst
  102. nvme_set_latency_tolerance
  103. string_matches
  104. quirk_matches
  105. nvme_init_subnqn
  106. nvme_release_subsystem
  107. nvme_destroy_subsystem
  108. nvme_put_subsystem
  109. __nvme_find_get_subsystem
  110. nvme_subsys_show_nqn
  111. nvme_validate_cntlid
  112. nvme_init_subsystem
  113. nvme_get_log
  114. nvme_get_effects_log
  115. nvme_init_identify
  116. nvme_dev_open
  117. nvme_dev_user_cmd
  118. nvme_dev_ioctl
  119. nvme_sysfs_reset
  120. nvme_sysfs_rescan
  121. dev_to_ns_head
  122. wwid_show
  123. nguid_show
  124. uuid_show
  125. eui_show
  126. nsid_show
  127. nvme_ns_id_attrs_are_visible
  128. nvme_sysfs_delete
  129. nvme_sysfs_show_transport
  130. nvme_sysfs_show_state
  131. nvme_sysfs_show_subsysnqn
  132. nvme_sysfs_show_address
  133. nvme_dev_attrs_are_visible
  134. __nvme_find_ns_head
  135. __nvme_check_ids
  136. nvme_alloc_ns_head
  137. nvme_init_ns_head
  138. ns_cmp
  139. nvme_find_get_ns
  140. nvme_setup_streams_ns
  141. nvme_alloc_ns
  142. nvme_ns_remove
  143. nvme_validate_ns
  144. nvme_remove_invalid_namespaces
  145. nvme_scan_ns_list
  146. nvme_scan_ns_sequential
  147. nvme_clear_changed_ns_log
  148. nvme_scan_work
  149. nvme_remove_namespaces
  150. nvme_class_uevent
  151. nvme_aen_uevent
  152. nvme_async_event_work
  153. nvme_ctrl_pp_status
  154. nvme_get_fw_slot_info
  155. nvme_fw_act_work
  156. nvme_handle_aen_notice
  157. nvme_complete_async_event
  158. nvme_stop_ctrl
  159. nvme_start_ctrl
  160. nvme_uninit_ctrl
  161. nvme_free_ctrl
  162. nvme_init_ctrl
  163. nvme_kill_queues
  164. nvme_unfreeze
  165. nvme_wait_freeze_timeout
  166. nvme_wait_freeze
  167. nvme_start_freeze
  168. nvme_stop_queues
  169. nvme_start_queues
  170. nvme_sync_queues
  171. _nvme_check_size
  172. nvme_core_init
  173. nvme_core_exit

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * NVM Express device driver
   4  * Copyright (c) 2011-2014, Intel Corporation.
   5  */
   6 
   7 #include <linux/blkdev.h>
   8 #include <linux/blk-mq.h>
   9 #include <linux/compat.h>
  10 #include <linux/delay.h>
  11 #include <linux/errno.h>
  12 #include <linux/hdreg.h>
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/list_sort.h>
  17 #include <linux/slab.h>
  18 #include <linux/types.h>
  19 #include <linux/pr.h>
  20 #include <linux/ptrace.h>
  21 #include <linux/nvme_ioctl.h>
  22 #include <linux/t10-pi.h>
  23 #include <linux/pm_qos.h>
  24 #include <asm/unaligned.h>
  25 
  26 #include "nvme.h"
  27 #include "fabrics.h"
  28 
  29 #define CREATE_TRACE_POINTS
  30 #include "trace.h"
  31 
  32 #define NVME_MINORS             (1U << MINORBITS)
  33 
  34 unsigned int admin_timeout = 60;
  35 module_param(admin_timeout, uint, 0644);
  36 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  37 EXPORT_SYMBOL_GPL(admin_timeout);
  38 
  39 unsigned int nvme_io_timeout = 30;
  40 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  41 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  42 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  43 
  44 static unsigned char shutdown_timeout = 5;
  45 module_param(shutdown_timeout, byte, 0644);
  46 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  47 
  48 static u8 nvme_max_retries = 5;
  49 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  50 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  51 
  52 static unsigned long default_ps_max_latency_us = 100000;
  53 module_param(default_ps_max_latency_us, ulong, 0644);
  54 MODULE_PARM_DESC(default_ps_max_latency_us,
  55                  "max power saving latency for new devices; use PM QOS to change per device");
  56 
  57 static bool force_apst;
  58 module_param(force_apst, bool, 0644);
  59 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  60 
  61 static bool streams;
  62 module_param(streams, bool, 0644);
  63 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  64 
  65 /*
  66  * nvme_wq - hosts nvme related works that are not reset or delete
  67  * nvme_reset_wq - hosts nvme reset works
  68  * nvme_delete_wq - hosts nvme delete works
  69  *
  70  * nvme_wq will host works such as scan, aen handling, fw activation,
  71  * keep-alive, periodic reconnects etc. nvme_reset_wq
  72  * runs reset works which also flush works hosted on nvme_wq for
  73  * serialization purposes. nvme_delete_wq host controller deletion
  74  * works which flush reset works for serialization.
  75  */
  76 struct workqueue_struct *nvme_wq;
  77 EXPORT_SYMBOL_GPL(nvme_wq);
  78 
  79 struct workqueue_struct *nvme_reset_wq;
  80 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  81 
  82 struct workqueue_struct *nvme_delete_wq;
  83 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  84 
  85 static LIST_HEAD(nvme_subsystems);
  86 static DEFINE_MUTEX(nvme_subsystems_lock);
  87 
  88 static DEFINE_IDA(nvme_instance_ida);
  89 static dev_t nvme_chr_devt;
  90 static struct class *nvme_class;
  91 static struct class *nvme_subsys_class;
  92 
  93 static int nvme_revalidate_disk(struct gendisk *disk);
  94 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
  95 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
  96                                            unsigned nsid);
  97 
  98 static void nvme_set_queue_dying(struct nvme_ns *ns)
  99 {
 100         /*
 101          * Revalidating a dead namespace sets capacity to 0. This will end
 102          * buffered writers dirtying pages that can't be synced.
 103          */
 104         if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 105                 return;
 106         blk_set_queue_dying(ns->queue);
 107         /* Forcibly unquiesce queues to avoid blocking dispatch */
 108         blk_mq_unquiesce_queue(ns->queue);
 109         /*
 110          * Revalidate after unblocking dispatchers that may be holding bd_butex
 111          */
 112         revalidate_disk(ns->disk);
 113 }
 114 
 115 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 116 {
 117         /*
 118          * Only new queue scan work when admin and IO queues are both alive
 119          */
 120         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 121                 queue_work(nvme_wq, &ctrl->scan_work);
 122 }
 123 
 124 /*
 125  * Use this function to proceed with scheduling reset_work for a controller
 126  * that had previously been set to the resetting state. This is intended for
 127  * code paths that can't be interrupted by other reset attempts. A hot removal
 128  * may prevent this from succeeding.
 129  */
 130 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 131 {
 132         if (ctrl->state != NVME_CTRL_RESETTING)
 133                 return -EBUSY;
 134         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 135                 return -EBUSY;
 136         return 0;
 137 }
 138 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 139 
 140 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 141 {
 142         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 143                 return -EBUSY;
 144         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 145                 return -EBUSY;
 146         return 0;
 147 }
 148 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 149 
 150 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 151 {
 152         int ret;
 153 
 154         ret = nvme_reset_ctrl(ctrl);
 155         if (!ret) {
 156                 flush_work(&ctrl->reset_work);
 157                 if (ctrl->state != NVME_CTRL_LIVE)
 158                         ret = -ENETRESET;
 159         }
 160 
 161         return ret;
 162 }
 163 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 164 
 165 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 166 {
 167         dev_info(ctrl->device,
 168                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 169 
 170         flush_work(&ctrl->reset_work);
 171         nvme_stop_ctrl(ctrl);
 172         nvme_remove_namespaces(ctrl);
 173         ctrl->ops->delete_ctrl(ctrl);
 174         nvme_uninit_ctrl(ctrl);
 175         nvme_put_ctrl(ctrl);
 176 }
 177 
 178 static void nvme_delete_ctrl_work(struct work_struct *work)
 179 {
 180         struct nvme_ctrl *ctrl =
 181                 container_of(work, struct nvme_ctrl, delete_work);
 182 
 183         nvme_do_delete_ctrl(ctrl);
 184 }
 185 
 186 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 187 {
 188         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 189                 return -EBUSY;
 190         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 191                 return -EBUSY;
 192         return 0;
 193 }
 194 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 195 
 196 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 197 {
 198         int ret = 0;
 199 
 200         /*
 201          * Keep a reference until nvme_do_delete_ctrl() complete,
 202          * since ->delete_ctrl can free the controller.
 203          */
 204         nvme_get_ctrl(ctrl);
 205         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 206                 ret = -EBUSY;
 207         if (!ret)
 208                 nvme_do_delete_ctrl(ctrl);
 209         nvme_put_ctrl(ctrl);
 210         return ret;
 211 }
 212 
 213 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 214 {
 215         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 216 }
 217 
 218 static blk_status_t nvme_error_status(u16 status)
 219 {
 220         switch (status & 0x7ff) {
 221         case NVME_SC_SUCCESS:
 222                 return BLK_STS_OK;
 223         case NVME_SC_CAP_EXCEEDED:
 224                 return BLK_STS_NOSPC;
 225         case NVME_SC_LBA_RANGE:
 226                 return BLK_STS_TARGET;
 227         case NVME_SC_BAD_ATTRIBUTES:
 228         case NVME_SC_ONCS_NOT_SUPPORTED:
 229         case NVME_SC_INVALID_OPCODE:
 230         case NVME_SC_INVALID_FIELD:
 231         case NVME_SC_INVALID_NS:
 232                 return BLK_STS_NOTSUPP;
 233         case NVME_SC_WRITE_FAULT:
 234         case NVME_SC_READ_ERROR:
 235         case NVME_SC_UNWRITTEN_BLOCK:
 236         case NVME_SC_ACCESS_DENIED:
 237         case NVME_SC_READ_ONLY:
 238         case NVME_SC_COMPARE_FAILED:
 239                 return BLK_STS_MEDIUM;
 240         case NVME_SC_GUARD_CHECK:
 241         case NVME_SC_APPTAG_CHECK:
 242         case NVME_SC_REFTAG_CHECK:
 243         case NVME_SC_INVALID_PI:
 244                 return BLK_STS_PROTECTION;
 245         case NVME_SC_RESERVATION_CONFLICT:
 246                 return BLK_STS_NEXUS;
 247         case NVME_SC_HOST_PATH_ERROR:
 248                 return BLK_STS_TRANSPORT;
 249         default:
 250                 return BLK_STS_IOERR;
 251         }
 252 }
 253 
 254 static inline bool nvme_req_needs_retry(struct request *req)
 255 {
 256         if (blk_noretry_request(req))
 257                 return false;
 258         if (nvme_req(req)->status & NVME_SC_DNR)
 259                 return false;
 260         if (nvme_req(req)->retries >= nvme_max_retries)
 261                 return false;
 262         return true;
 263 }
 264 
 265 static void nvme_retry_req(struct request *req)
 266 {
 267         struct nvme_ns *ns = req->q->queuedata;
 268         unsigned long delay = 0;
 269         u16 crd;
 270 
 271         /* The mask and shift result must be <= 3 */
 272         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 273         if (ns && crd)
 274                 delay = ns->ctrl->crdt[crd - 1] * 100;
 275 
 276         nvme_req(req)->retries++;
 277         blk_mq_requeue_request(req, false);
 278         blk_mq_delay_kick_requeue_list(req->q, delay);
 279 }
 280 
 281 void nvme_complete_rq(struct request *req)
 282 {
 283         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 284 
 285         trace_nvme_complete_rq(req);
 286 
 287         if (nvme_req(req)->ctrl->kas)
 288                 nvme_req(req)->ctrl->comp_seen = true;
 289 
 290         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 291                 if ((req->cmd_flags & REQ_NVME_MPATH) &&
 292                     blk_path_error(status)) {
 293                         nvme_failover_req(req);
 294                         return;
 295                 }
 296 
 297                 if (!blk_queue_dying(req->q)) {
 298                         nvme_retry_req(req);
 299                         return;
 300                 }
 301         }
 302 
 303         nvme_trace_bio_complete(req, status);
 304         blk_mq_end_request(req, status);
 305 }
 306 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 307 
 308 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 309 {
 310         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 311                                 "Cancelling I/O %d", req->tag);
 312 
 313         /* don't abort one completed request */
 314         if (blk_mq_request_completed(req))
 315                 return true;
 316 
 317         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 318         blk_mq_complete_request(req);
 319         return true;
 320 }
 321 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 322 
 323 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 324                 enum nvme_ctrl_state new_state)
 325 {
 326         enum nvme_ctrl_state old_state;
 327         unsigned long flags;
 328         bool changed = false;
 329 
 330         spin_lock_irqsave(&ctrl->lock, flags);
 331 
 332         old_state = ctrl->state;
 333         switch (new_state) {
 334         case NVME_CTRL_LIVE:
 335                 switch (old_state) {
 336                 case NVME_CTRL_NEW:
 337                 case NVME_CTRL_RESETTING:
 338                 case NVME_CTRL_CONNECTING:
 339                         changed = true;
 340                         /* FALLTHRU */
 341                 default:
 342                         break;
 343                 }
 344                 break;
 345         case NVME_CTRL_RESETTING:
 346                 switch (old_state) {
 347                 case NVME_CTRL_NEW:
 348                 case NVME_CTRL_LIVE:
 349                         changed = true;
 350                         /* FALLTHRU */
 351                 default:
 352                         break;
 353                 }
 354                 break;
 355         case NVME_CTRL_CONNECTING:
 356                 switch (old_state) {
 357                 case NVME_CTRL_NEW:
 358                 case NVME_CTRL_RESETTING:
 359                         changed = true;
 360                         /* FALLTHRU */
 361                 default:
 362                         break;
 363                 }
 364                 break;
 365         case NVME_CTRL_DELETING:
 366                 switch (old_state) {
 367                 case NVME_CTRL_LIVE:
 368                 case NVME_CTRL_RESETTING:
 369                 case NVME_CTRL_CONNECTING:
 370                         changed = true;
 371                         /* FALLTHRU */
 372                 default:
 373                         break;
 374                 }
 375                 break;
 376         case NVME_CTRL_DEAD:
 377                 switch (old_state) {
 378                 case NVME_CTRL_DELETING:
 379                         changed = true;
 380                         /* FALLTHRU */
 381                 default:
 382                         break;
 383                 }
 384                 break;
 385         default:
 386                 break;
 387         }
 388 
 389         if (changed) {
 390                 ctrl->state = new_state;
 391                 wake_up_all(&ctrl->state_wq);
 392         }
 393 
 394         spin_unlock_irqrestore(&ctrl->lock, flags);
 395         if (changed && ctrl->state == NVME_CTRL_LIVE)
 396                 nvme_kick_requeue_lists(ctrl);
 397         return changed;
 398 }
 399 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 400 
 401 /*
 402  * Returns true for sink states that can't ever transition back to live.
 403  */
 404 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 405 {
 406         switch (ctrl->state) {
 407         case NVME_CTRL_NEW:
 408         case NVME_CTRL_LIVE:
 409         case NVME_CTRL_RESETTING:
 410         case NVME_CTRL_CONNECTING:
 411                 return false;
 412         case NVME_CTRL_DELETING:
 413         case NVME_CTRL_DEAD:
 414                 return true;
 415         default:
 416                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 417                 return true;
 418         }
 419 }
 420 
 421 /*
 422  * Waits for the controller state to be resetting, or returns false if it is
 423  * not possible to ever transition to that state.
 424  */
 425 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 426 {
 427         wait_event(ctrl->state_wq,
 428                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 429                    nvme_state_terminal(ctrl));
 430         return ctrl->state == NVME_CTRL_RESETTING;
 431 }
 432 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 433 
 434 static void nvme_free_ns_head(struct kref *ref)
 435 {
 436         struct nvme_ns_head *head =
 437                 container_of(ref, struct nvme_ns_head, ref);
 438 
 439         nvme_mpath_remove_disk(head);
 440         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 441         list_del_init(&head->entry);
 442         cleanup_srcu_struct(&head->srcu);
 443         nvme_put_subsystem(head->subsys);
 444         kfree(head);
 445 }
 446 
 447 static void nvme_put_ns_head(struct nvme_ns_head *head)
 448 {
 449         kref_put(&head->ref, nvme_free_ns_head);
 450 }
 451 
 452 static void nvme_free_ns(struct kref *kref)
 453 {
 454         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 455 
 456         if (ns->ndev)
 457                 nvme_nvm_unregister(ns);
 458 
 459         put_disk(ns->disk);
 460         nvme_put_ns_head(ns->head);
 461         nvme_put_ctrl(ns->ctrl);
 462         kfree(ns);
 463 }
 464 
 465 static void nvme_put_ns(struct nvme_ns *ns)
 466 {
 467         kref_put(&ns->kref, nvme_free_ns);
 468 }
 469 
 470 static inline void nvme_clear_nvme_request(struct request *req)
 471 {
 472         if (!(req->rq_flags & RQF_DONTPREP)) {
 473                 nvme_req(req)->retries = 0;
 474                 nvme_req(req)->flags = 0;
 475                 req->rq_flags |= RQF_DONTPREP;
 476         }
 477 }
 478 
 479 struct request *nvme_alloc_request(struct request_queue *q,
 480                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 481 {
 482         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 483         struct request *req;
 484 
 485         if (qid == NVME_QID_ANY) {
 486                 req = blk_mq_alloc_request(q, op, flags);
 487         } else {
 488                 req = blk_mq_alloc_request_hctx(q, op, flags,
 489                                 qid ? qid - 1 : 0);
 490         }
 491         if (IS_ERR(req))
 492                 return req;
 493 
 494         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 495         nvme_clear_nvme_request(req);
 496         nvme_req(req)->cmd = cmd;
 497 
 498         return req;
 499 }
 500 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 501 
 502 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 503 {
 504         struct nvme_command c;
 505 
 506         memset(&c, 0, sizeof(c));
 507 
 508         c.directive.opcode = nvme_admin_directive_send;
 509         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 510         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 511         c.directive.dtype = NVME_DIR_IDENTIFY;
 512         c.directive.tdtype = NVME_DIR_STREAMS;
 513         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 514 
 515         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 516 }
 517 
 518 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 519 {
 520         return nvme_toggle_streams(ctrl, false);
 521 }
 522 
 523 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 524 {
 525         return nvme_toggle_streams(ctrl, true);
 526 }
 527 
 528 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 529                                   struct streams_directive_params *s, u32 nsid)
 530 {
 531         struct nvme_command c;
 532 
 533         memset(&c, 0, sizeof(c));
 534         memset(s, 0, sizeof(*s));
 535 
 536         c.directive.opcode = nvme_admin_directive_recv;
 537         c.directive.nsid = cpu_to_le32(nsid);
 538         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
 539         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 540         c.directive.dtype = NVME_DIR_STREAMS;
 541 
 542         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 543 }
 544 
 545 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 546 {
 547         struct streams_directive_params s;
 548         int ret;
 549 
 550         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 551                 return 0;
 552         if (!streams)
 553                 return 0;
 554 
 555         ret = nvme_enable_streams(ctrl);
 556         if (ret)
 557                 return ret;
 558 
 559         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 560         if (ret)
 561                 return ret;
 562 
 563         ctrl->nssa = le16_to_cpu(s.nssa);
 564         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 565                 dev_info(ctrl->device, "too few streams (%u) available\n",
 566                                         ctrl->nssa);
 567                 nvme_disable_streams(ctrl);
 568                 return 0;
 569         }
 570 
 571         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 572         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 573         return 0;
 574 }
 575 
 576 /*
 577  * Check if 'req' has a write hint associated with it. If it does, assign
 578  * a valid namespace stream to the write.
 579  */
 580 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 581                                      struct request *req, u16 *control,
 582                                      u32 *dsmgmt)
 583 {
 584         enum rw_hint streamid = req->write_hint;
 585 
 586         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 587                 streamid = 0;
 588         else {
 589                 streamid--;
 590                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 591                         return;
 592 
 593                 *control |= NVME_RW_DTYPE_STREAMS;
 594                 *dsmgmt |= streamid << 16;
 595         }
 596 
 597         if (streamid < ARRAY_SIZE(req->q->write_hints))
 598                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 599 }
 600 
 601 static inline void nvme_setup_flush(struct nvme_ns *ns,
 602                 struct nvme_command *cmnd)
 603 {
 604         cmnd->common.opcode = nvme_cmd_flush;
 605         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 606 }
 607 
 608 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 609                 struct nvme_command *cmnd)
 610 {
 611         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 612         struct nvme_dsm_range *range;
 613         struct bio *bio;
 614 
 615         /*
 616          * Some devices do not consider the DSM 'Number of Ranges' field when
 617          * determining how much data to DMA. Always allocate memory for maximum
 618          * number of segments to prevent device reading beyond end of buffer.
 619          */
 620         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 621 
 622         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 623         if (!range) {
 624                 /*
 625                  * If we fail allocation our range, fallback to the controller
 626                  * discard page. If that's also busy, it's safe to return
 627                  * busy, as we know we can make progress once that's freed.
 628                  */
 629                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 630                         return BLK_STS_RESOURCE;
 631 
 632                 range = page_address(ns->ctrl->discard_page);
 633         }
 634 
 635         __rq_for_each_bio(bio, req) {
 636                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 637                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 638 
 639                 if (n < segments) {
 640                         range[n].cattr = cpu_to_le32(0);
 641                         range[n].nlb = cpu_to_le32(nlb);
 642                         range[n].slba = cpu_to_le64(slba);
 643                 }
 644                 n++;
 645         }
 646 
 647         if (WARN_ON_ONCE(n != segments)) {
 648                 if (virt_to_page(range) == ns->ctrl->discard_page)
 649                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 650                 else
 651                         kfree(range);
 652                 return BLK_STS_IOERR;
 653         }
 654 
 655         cmnd->dsm.opcode = nvme_cmd_dsm;
 656         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 657         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 658         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 659 
 660         req->special_vec.bv_page = virt_to_page(range);
 661         req->special_vec.bv_offset = offset_in_page(range);
 662         req->special_vec.bv_len = alloc_size;
 663         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 664 
 665         return BLK_STS_OK;
 666 }
 667 
 668 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 669                 struct request *req, struct nvme_command *cmnd)
 670 {
 671         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 672                 return nvme_setup_discard(ns, req, cmnd);
 673 
 674         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 675         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 676         cmnd->write_zeroes.slba =
 677                 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 678         cmnd->write_zeroes.length =
 679                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 680         cmnd->write_zeroes.control = 0;
 681         return BLK_STS_OK;
 682 }
 683 
 684 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 685                 struct request *req, struct nvme_command *cmnd)
 686 {
 687         struct nvme_ctrl *ctrl = ns->ctrl;
 688         u16 control = 0;
 689         u32 dsmgmt = 0;
 690 
 691         if (req->cmd_flags & REQ_FUA)
 692                 control |= NVME_RW_FUA;
 693         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 694                 control |= NVME_RW_LR;
 695 
 696         if (req->cmd_flags & REQ_RAHEAD)
 697                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 698 
 699         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 700         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 701         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 702         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 703 
 704         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 705                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 706 
 707         if (ns->ms) {
 708                 /*
 709                  * If formated with metadata, the block layer always provides a
 710                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 711                  * we enable the PRACT bit for protection information or set the
 712                  * namespace capacity to zero to prevent any I/O.
 713                  */
 714                 if (!blk_integrity_rq(req)) {
 715                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 716                                 return BLK_STS_NOTSUPP;
 717                         control |= NVME_RW_PRINFO_PRACT;
 718                 }
 719 
 720                 switch (ns->pi_type) {
 721                 case NVME_NS_DPS_PI_TYPE3:
 722                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 723                         break;
 724                 case NVME_NS_DPS_PI_TYPE1:
 725                 case NVME_NS_DPS_PI_TYPE2:
 726                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 727                                         NVME_RW_PRINFO_PRCHK_REF;
 728                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 729                         break;
 730                 }
 731         }
 732 
 733         cmnd->rw.control = cpu_to_le16(control);
 734         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 735         return 0;
 736 }
 737 
 738 void nvme_cleanup_cmd(struct request *req)
 739 {
 740         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 741                 struct nvme_ns *ns = req->rq_disk->private_data;
 742                 struct page *page = req->special_vec.bv_page;
 743 
 744                 if (page == ns->ctrl->discard_page)
 745                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 746                 else
 747                         kfree(page_address(page) + req->special_vec.bv_offset);
 748         }
 749 }
 750 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 751 
 752 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 753                 struct nvme_command *cmd)
 754 {
 755         blk_status_t ret = BLK_STS_OK;
 756 
 757         nvme_clear_nvme_request(req);
 758 
 759         memset(cmd, 0, sizeof(*cmd));
 760         switch (req_op(req)) {
 761         case REQ_OP_DRV_IN:
 762         case REQ_OP_DRV_OUT:
 763                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 764                 break;
 765         case REQ_OP_FLUSH:
 766                 nvme_setup_flush(ns, cmd);
 767                 break;
 768         case REQ_OP_WRITE_ZEROES:
 769                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 770                 break;
 771         case REQ_OP_DISCARD:
 772                 ret = nvme_setup_discard(ns, req, cmd);
 773                 break;
 774         case REQ_OP_READ:
 775         case REQ_OP_WRITE:
 776                 ret = nvme_setup_rw(ns, req, cmd);
 777                 break;
 778         default:
 779                 WARN_ON_ONCE(1);
 780                 return BLK_STS_IOERR;
 781         }
 782 
 783         cmd->common.command_id = req->tag;
 784         trace_nvme_setup_cmd(req, cmd);
 785         return ret;
 786 }
 787 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 788 
 789 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 790 {
 791         struct completion *waiting = rq->end_io_data;
 792 
 793         rq->end_io_data = NULL;
 794         complete(waiting);
 795 }
 796 
 797 static void nvme_execute_rq_polled(struct request_queue *q,
 798                 struct gendisk *bd_disk, struct request *rq, int at_head)
 799 {
 800         DECLARE_COMPLETION_ONSTACK(wait);
 801 
 802         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
 803 
 804         rq->cmd_flags |= REQ_HIPRI;
 805         rq->end_io_data = &wait;
 806         blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
 807 
 808         while (!completion_done(&wait)) {
 809                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
 810                 cond_resched();
 811         }
 812 }
 813 
 814 /*
 815  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 816  * if the result is positive, it's an NVM Express status code
 817  */
 818 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 819                 union nvme_result *result, void *buffer, unsigned bufflen,
 820                 unsigned timeout, int qid, int at_head,
 821                 blk_mq_req_flags_t flags, bool poll)
 822 {
 823         struct request *req;
 824         int ret;
 825 
 826         req = nvme_alloc_request(q, cmd, flags, qid);
 827         if (IS_ERR(req))
 828                 return PTR_ERR(req);
 829 
 830         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 831 
 832         if (buffer && bufflen) {
 833                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 834                 if (ret)
 835                         goto out;
 836         }
 837 
 838         if (poll)
 839                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
 840         else
 841                 blk_execute_rq(req->q, NULL, req, at_head);
 842         if (result)
 843                 *result = nvme_req(req)->result;
 844         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 845                 ret = -EINTR;
 846         else
 847                 ret = nvme_req(req)->status;
 848  out:
 849         blk_mq_free_request(req);
 850         return ret;
 851 }
 852 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 853 
 854 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 855                 void *buffer, unsigned bufflen)
 856 {
 857         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 858                         NVME_QID_ANY, 0, 0, false);
 859 }
 860 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 861 
 862 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 863                 unsigned len, u32 seed, bool write)
 864 {
 865         struct bio_integrity_payload *bip;
 866         int ret = -ENOMEM;
 867         void *buf;
 868 
 869         buf = kmalloc(len, GFP_KERNEL);
 870         if (!buf)
 871                 goto out;
 872 
 873         ret = -EFAULT;
 874         if (write && copy_from_user(buf, ubuf, len))
 875                 goto out_free_meta;
 876 
 877         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 878         if (IS_ERR(bip)) {
 879                 ret = PTR_ERR(bip);
 880                 goto out_free_meta;
 881         }
 882 
 883         bip->bip_iter.bi_size = len;
 884         bip->bip_iter.bi_sector = seed;
 885         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 886                         offset_in_page(buf));
 887         if (ret == len)
 888                 return buf;
 889         ret = -ENOMEM;
 890 out_free_meta:
 891         kfree(buf);
 892 out:
 893         return ERR_PTR(ret);
 894 }
 895 
 896 static int nvme_submit_user_cmd(struct request_queue *q,
 897                 struct nvme_command *cmd, void __user *ubuffer,
 898                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 899                 u32 meta_seed, u64 *result, unsigned timeout)
 900 {
 901         bool write = nvme_is_write(cmd);
 902         struct nvme_ns *ns = q->queuedata;
 903         struct gendisk *disk = ns ? ns->disk : NULL;
 904         struct request *req;
 905         struct bio *bio = NULL;
 906         void *meta = NULL;
 907         int ret;
 908 
 909         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 910         if (IS_ERR(req))
 911                 return PTR_ERR(req);
 912 
 913         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 914         nvme_req(req)->flags |= NVME_REQ_USERCMD;
 915 
 916         if (ubuffer && bufflen) {
 917                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 918                                 GFP_KERNEL);
 919                 if (ret)
 920                         goto out;
 921                 bio = req->bio;
 922                 bio->bi_disk = disk;
 923                 if (disk && meta_buffer && meta_len) {
 924                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 925                                         meta_seed, write);
 926                         if (IS_ERR(meta)) {
 927                                 ret = PTR_ERR(meta);
 928                                 goto out_unmap;
 929                         }
 930                         req->cmd_flags |= REQ_INTEGRITY;
 931                 }
 932         }
 933 
 934         blk_execute_rq(req->q, disk, req, 0);
 935         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 936                 ret = -EINTR;
 937         else
 938                 ret = nvme_req(req)->status;
 939         if (result)
 940                 *result = le64_to_cpu(nvme_req(req)->result.u64);
 941         if (meta && !ret && !write) {
 942                 if (copy_to_user(meta_buffer, meta, meta_len))
 943                         ret = -EFAULT;
 944         }
 945         kfree(meta);
 946  out_unmap:
 947         if (bio)
 948                 blk_rq_unmap_user(bio);
 949  out:
 950         blk_mq_free_request(req);
 951         return ret;
 952 }
 953 
 954 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 955 {
 956         struct nvme_ctrl *ctrl = rq->end_io_data;
 957         unsigned long flags;
 958         bool startka = false;
 959 
 960         blk_mq_free_request(rq);
 961 
 962         if (status) {
 963                 dev_err(ctrl->device,
 964                         "failed nvme_keep_alive_end_io error=%d\n",
 965                                 status);
 966                 return;
 967         }
 968 
 969         ctrl->comp_seen = false;
 970         spin_lock_irqsave(&ctrl->lock, flags);
 971         if (ctrl->state == NVME_CTRL_LIVE ||
 972             ctrl->state == NVME_CTRL_CONNECTING)
 973                 startka = true;
 974         spin_unlock_irqrestore(&ctrl->lock, flags);
 975         if (startka)
 976                 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
 977 }
 978 
 979 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 980 {
 981         struct request *rq;
 982 
 983         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
 984                         NVME_QID_ANY);
 985         if (IS_ERR(rq))
 986                 return PTR_ERR(rq);
 987 
 988         rq->timeout = ctrl->kato * HZ;
 989         rq->end_io_data = ctrl;
 990 
 991         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 992 
 993         return 0;
 994 }
 995 
 996 static void nvme_keep_alive_work(struct work_struct *work)
 997 {
 998         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 999                         struct nvme_ctrl, ka_work);
1000         bool comp_seen = ctrl->comp_seen;
1001 
1002         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1003                 dev_dbg(ctrl->device,
1004                         "reschedule traffic based keep-alive timer\n");
1005                 ctrl->comp_seen = false;
1006                 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1007                 return;
1008         }
1009 
1010         if (nvme_keep_alive(ctrl)) {
1011                 /* allocation failure, reset the controller */
1012                 dev_err(ctrl->device, "keep-alive failed\n");
1013                 nvme_reset_ctrl(ctrl);
1014                 return;
1015         }
1016 }
1017 
1018 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1019 {
1020         if (unlikely(ctrl->kato == 0))
1021                 return;
1022 
1023         queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1024 }
1025 
1026 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1027 {
1028         if (unlikely(ctrl->kato == 0))
1029                 return;
1030 
1031         cancel_delayed_work_sync(&ctrl->ka_work);
1032 }
1033 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1034 
1035 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1036 {
1037         struct nvme_command c = { };
1038         int error;
1039 
1040         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1041         c.identify.opcode = nvme_admin_identify;
1042         c.identify.cns = NVME_ID_CNS_CTRL;
1043 
1044         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1045         if (!*id)
1046                 return -ENOMEM;
1047 
1048         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1049                         sizeof(struct nvme_id_ctrl));
1050         if (error)
1051                 kfree(*id);
1052         return error;
1053 }
1054 
1055 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1056                 struct nvme_ns_ids *ids)
1057 {
1058         struct nvme_command c = { };
1059         int status;
1060         void *data;
1061         int pos;
1062         int len;
1063 
1064         c.identify.opcode = nvme_admin_identify;
1065         c.identify.nsid = cpu_to_le32(nsid);
1066         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1067 
1068         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1069         if (!data)
1070                 return -ENOMEM;
1071 
1072         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1073                                       NVME_IDENTIFY_DATA_SIZE);
1074         if (status) {
1075                 dev_warn(ctrl->device,
1076                         "Identify Descriptors failed (%d)\n", status);
1077                  /*
1078                   * Don't treat an error as fatal, as we potentially already
1079                   * have a NGUID or EUI-64.
1080                   */
1081                 if (status > 0 && !(status & NVME_SC_DNR))
1082                         status = 0;
1083                 goto free_data;
1084         }
1085 
1086         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1087                 struct nvme_ns_id_desc *cur = data + pos;
1088 
1089                 if (cur->nidl == 0)
1090                         break;
1091 
1092                 switch (cur->nidt) {
1093                 case NVME_NIDT_EUI64:
1094                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1095                                 dev_warn(ctrl->device,
1096                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1097                                          cur->nidl);
1098                                 goto free_data;
1099                         }
1100                         len = NVME_NIDT_EUI64_LEN;
1101                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1102                         break;
1103                 case NVME_NIDT_NGUID:
1104                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1105                                 dev_warn(ctrl->device,
1106                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1107                                          cur->nidl);
1108                                 goto free_data;
1109                         }
1110                         len = NVME_NIDT_NGUID_LEN;
1111                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1112                         break;
1113                 case NVME_NIDT_UUID:
1114                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
1115                                 dev_warn(ctrl->device,
1116                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1117                                          cur->nidl);
1118                                 goto free_data;
1119                         }
1120                         len = NVME_NIDT_UUID_LEN;
1121                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1122                         break;
1123                 default:
1124                         /* Skip unknown types */
1125                         len = cur->nidl;
1126                         break;
1127                 }
1128 
1129                 len += sizeof(*cur);
1130         }
1131 free_data:
1132         kfree(data);
1133         return status;
1134 }
1135 
1136 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1137 {
1138         struct nvme_command c = { };
1139 
1140         c.identify.opcode = nvme_admin_identify;
1141         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1142         c.identify.nsid = cpu_to_le32(nsid);
1143         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1144                                     NVME_IDENTIFY_DATA_SIZE);
1145 }
1146 
1147 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1148                 unsigned nsid, struct nvme_id_ns **id)
1149 {
1150         struct nvme_command c = { };
1151         int error;
1152 
1153         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1154         c.identify.opcode = nvme_admin_identify;
1155         c.identify.nsid = cpu_to_le32(nsid);
1156         c.identify.cns = NVME_ID_CNS_NS;
1157 
1158         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1159         if (!*id)
1160                 return -ENOMEM;
1161 
1162         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1163         if (error) {
1164                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1165                 kfree(*id);
1166         }
1167 
1168         return error;
1169 }
1170 
1171 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1172                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1173 {
1174         union nvme_result res = { 0 };
1175         struct nvme_command c;
1176         int ret;
1177 
1178         memset(&c, 0, sizeof(c));
1179         c.features.opcode = op;
1180         c.features.fid = cpu_to_le32(fid);
1181         c.features.dword11 = cpu_to_le32(dword11);
1182 
1183         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1184                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1185         if (ret >= 0 && result)
1186                 *result = le32_to_cpu(res.u32);
1187         return ret;
1188 }
1189 
1190 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1191                       unsigned int dword11, void *buffer, size_t buflen,
1192                       u32 *result)
1193 {
1194         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1195                              buflen, result);
1196 }
1197 EXPORT_SYMBOL_GPL(nvme_set_features);
1198 
1199 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1200                       unsigned int dword11, void *buffer, size_t buflen,
1201                       u32 *result)
1202 {
1203         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1204                              buflen, result);
1205 }
1206 EXPORT_SYMBOL_GPL(nvme_get_features);
1207 
1208 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1209 {
1210         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1211         u32 result;
1212         int status, nr_io_queues;
1213 
1214         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1215                         &result);
1216         if (status < 0)
1217                 return status;
1218 
1219         /*
1220          * Degraded controllers might return an error when setting the queue
1221          * count.  We still want to be able to bring them online and offer
1222          * access to the admin queue, as that might be only way to fix them up.
1223          */
1224         if (status > 0) {
1225                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1226                 *count = 0;
1227         } else {
1228                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1229                 *count = min(*count, nr_io_queues);
1230         }
1231 
1232         return 0;
1233 }
1234 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1235 
1236 #define NVME_AEN_SUPPORTED \
1237         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1238          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1239 
1240 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1241 {
1242         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1243         int status;
1244 
1245         if (!supported_aens)
1246                 return;
1247 
1248         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1249                         NULL, 0, &result);
1250         if (status)
1251                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1252                          supported_aens);
1253 
1254         queue_work(nvme_wq, &ctrl->async_event_work);
1255 }
1256 
1257 /*
1258  * Convert integer values from ioctl structures to user pointers, silently
1259  * ignoring the upper bits in the compat case to match behaviour of 32-bit
1260  * kernels.
1261  */
1262 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1263 {
1264         if (in_compat_syscall())
1265                 ptrval = (compat_uptr_t)ptrval;
1266         return (void __user *)ptrval;
1267 }
1268 
1269 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1270 {
1271         struct nvme_user_io io;
1272         struct nvme_command c;
1273         unsigned length, meta_len;
1274         void __user *metadata;
1275 
1276         if (copy_from_user(&io, uio, sizeof(io)))
1277                 return -EFAULT;
1278         if (io.flags)
1279                 return -EINVAL;
1280 
1281         switch (io.opcode) {
1282         case nvme_cmd_write:
1283         case nvme_cmd_read:
1284         case nvme_cmd_compare:
1285                 break;
1286         default:
1287                 return -EINVAL;
1288         }
1289 
1290         length = (io.nblocks + 1) << ns->lba_shift;
1291         meta_len = (io.nblocks + 1) * ns->ms;
1292         metadata = nvme_to_user_ptr(io.metadata);
1293 
1294         if (ns->ext) {
1295                 length += meta_len;
1296                 meta_len = 0;
1297         } else if (meta_len) {
1298                 if ((io.metadata & 3) || !io.metadata)
1299                         return -EINVAL;
1300         }
1301 
1302         memset(&c, 0, sizeof(c));
1303         c.rw.opcode = io.opcode;
1304         c.rw.flags = io.flags;
1305         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1306         c.rw.slba = cpu_to_le64(io.slba);
1307         c.rw.length = cpu_to_le16(io.nblocks);
1308         c.rw.control = cpu_to_le16(io.control);
1309         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1310         c.rw.reftag = cpu_to_le32(io.reftag);
1311         c.rw.apptag = cpu_to_le16(io.apptag);
1312         c.rw.appmask = cpu_to_le16(io.appmask);
1313 
1314         return nvme_submit_user_cmd(ns->queue, &c,
1315                         nvme_to_user_ptr(io.addr), length,
1316                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1317 }
1318 
1319 static u32 nvme_known_admin_effects(u8 opcode)
1320 {
1321         switch (opcode) {
1322         case nvme_admin_format_nvm:
1323                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1324                                         NVME_CMD_EFFECTS_CSE_MASK;
1325         case nvme_admin_sanitize_nvm:
1326                 return NVME_CMD_EFFECTS_CSE_MASK;
1327         default:
1328                 break;
1329         }
1330         return 0;
1331 }
1332 
1333 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1334                                                                 u8 opcode)
1335 {
1336         u32 effects = 0;
1337 
1338         if (ns) {
1339                 if (ctrl->effects)
1340                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1341                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1342                         dev_warn(ctrl->device,
1343                                  "IO command:%02x has unhandled effects:%08x\n",
1344                                  opcode, effects);
1345                 return 0;
1346         }
1347 
1348         if (ctrl->effects)
1349                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1350         effects |= nvme_known_admin_effects(opcode);
1351 
1352         /*
1353          * For simplicity, IO to all namespaces is quiesced even if the command
1354          * effects say only one namespace is affected.
1355          */
1356         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1357                 mutex_lock(&ctrl->scan_lock);
1358                 mutex_lock(&ctrl->subsys->lock);
1359                 nvme_mpath_start_freeze(ctrl->subsys);
1360                 nvme_mpath_wait_freeze(ctrl->subsys);
1361                 nvme_start_freeze(ctrl);
1362                 nvme_wait_freeze(ctrl);
1363         }
1364         return effects;
1365 }
1366 
1367 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1368 {
1369         struct nvme_ns *ns;
1370 
1371         down_read(&ctrl->namespaces_rwsem);
1372         list_for_each_entry(ns, &ctrl->namespaces, list)
1373                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1374                         nvme_set_queue_dying(ns);
1375         up_read(&ctrl->namespaces_rwsem);
1376 }
1377 
1378 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1379 {
1380         /*
1381          * Revalidate LBA changes prior to unfreezing. This is necessary to
1382          * prevent memory corruption if a logical block size was changed by
1383          * this command.
1384          */
1385         if (effects & NVME_CMD_EFFECTS_LBCC)
1386                 nvme_update_formats(ctrl);
1387         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1388                 nvme_unfreeze(ctrl);
1389                 nvme_mpath_unfreeze(ctrl->subsys);
1390                 mutex_unlock(&ctrl->subsys->lock);
1391                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1392                 mutex_unlock(&ctrl->scan_lock);
1393         }
1394         if (effects & NVME_CMD_EFFECTS_CCC)
1395                 nvme_init_identify(ctrl);
1396         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1397                 nvme_queue_scan(ctrl);
1398 }
1399 
1400 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1401                         struct nvme_passthru_cmd __user *ucmd)
1402 {
1403         struct nvme_passthru_cmd cmd;
1404         struct nvme_command c;
1405         unsigned timeout = 0;
1406         u32 effects;
1407         u64 result;
1408         int status;
1409 
1410         if (!capable(CAP_SYS_ADMIN))
1411                 return -EACCES;
1412         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1413                 return -EFAULT;
1414         if (cmd.flags)
1415                 return -EINVAL;
1416 
1417         memset(&c, 0, sizeof(c));
1418         c.common.opcode = cmd.opcode;
1419         c.common.flags = cmd.flags;
1420         c.common.nsid = cpu_to_le32(cmd.nsid);
1421         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1422         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1423         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1424         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1425         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1426         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1427         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1428         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1429 
1430         if (cmd.timeout_ms)
1431                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1432 
1433         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1434         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1435                         nvme_to_user_ptr(cmd.addr), cmd.data_len,
1436                         nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1437                         0, &result, timeout);
1438         nvme_passthru_end(ctrl, effects);
1439 
1440         if (status >= 0) {
1441                 if (put_user(result, &ucmd->result))
1442                         return -EFAULT;
1443         }
1444 
1445         return status;
1446 }
1447 
1448 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1449                         struct nvme_passthru_cmd64 __user *ucmd)
1450 {
1451         struct nvme_passthru_cmd64 cmd;
1452         struct nvme_command c;
1453         unsigned timeout = 0;
1454         u32 effects;
1455         int status;
1456 
1457         if (!capable(CAP_SYS_ADMIN))
1458                 return -EACCES;
1459         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1460                 return -EFAULT;
1461         if (cmd.flags)
1462                 return -EINVAL;
1463 
1464         memset(&c, 0, sizeof(c));
1465         c.common.opcode = cmd.opcode;
1466         c.common.flags = cmd.flags;
1467         c.common.nsid = cpu_to_le32(cmd.nsid);
1468         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1469         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1470         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1471         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1472         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1473         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1474         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1475         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1476 
1477         if (cmd.timeout_ms)
1478                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1479 
1480         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1481         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1482                         nvme_to_user_ptr(cmd.addr), cmd.data_len,
1483                         nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1484                         0, &cmd.result, timeout);
1485         nvme_passthru_end(ctrl, effects);
1486 
1487         if (status >= 0) {
1488                 if (put_user(cmd.result, &ucmd->result))
1489                         return -EFAULT;
1490         }
1491 
1492         return status;
1493 }
1494 
1495 /*
1496  * Issue ioctl requests on the first available path.  Note that unlike normal
1497  * block layer requests we will not retry failed request on another controller.
1498  */
1499 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1500                 struct nvme_ns_head **head, int *srcu_idx)
1501 {
1502 #ifdef CONFIG_NVME_MULTIPATH
1503         if (disk->fops == &nvme_ns_head_ops) {
1504                 struct nvme_ns *ns;
1505 
1506                 *head = disk->private_data;
1507                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1508                 ns = nvme_find_path(*head);
1509                 if (!ns)
1510                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1511                 return ns;
1512         }
1513 #endif
1514         *head = NULL;
1515         *srcu_idx = -1;
1516         return disk->private_data;
1517 }
1518 
1519 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1520 {
1521         if (head)
1522                 srcu_read_unlock(&head->srcu, idx);
1523 }
1524 
1525 static bool is_ctrl_ioctl(unsigned int cmd)
1526 {
1527         if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1528                 return true;
1529         if (is_sed_ioctl(cmd))
1530                 return true;
1531         return false;
1532 }
1533 
1534 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1535                                   void __user *argp,
1536                                   struct nvme_ns_head *head,
1537                                   int srcu_idx)
1538 {
1539         struct nvme_ctrl *ctrl = ns->ctrl;
1540         int ret;
1541 
1542         nvme_get_ctrl(ns->ctrl);
1543         nvme_put_ns_from_disk(head, srcu_idx);
1544 
1545         switch (cmd) {
1546         case NVME_IOCTL_ADMIN_CMD:
1547                 ret = nvme_user_cmd(ctrl, NULL, argp);
1548                 break;
1549         case NVME_IOCTL_ADMIN64_CMD:
1550                 ret = nvme_user_cmd64(ctrl, NULL, argp);
1551                 break;
1552         default:
1553                 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1554                 break;
1555         }
1556         nvme_put_ctrl(ctrl);
1557         return ret;
1558 }
1559 
1560 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1561                 unsigned int cmd, unsigned long arg)
1562 {
1563         struct nvme_ns_head *head = NULL;
1564         void __user *argp = (void __user *)arg;
1565         struct nvme_ns *ns;
1566         int srcu_idx, ret;
1567 
1568         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1569         if (unlikely(!ns))
1570                 return -EWOULDBLOCK;
1571 
1572         /*
1573          * Handle ioctls that apply to the controller instead of the namespace
1574          * seperately and drop the ns SRCU reference early.  This avoids a
1575          * deadlock when deleting namespaces using the passthrough interface.
1576          */
1577         if (is_ctrl_ioctl(cmd))
1578                 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1579 
1580         switch (cmd) {
1581         case NVME_IOCTL_ID:
1582                 force_successful_syscall_return();
1583                 ret = ns->head->ns_id;
1584                 break;
1585         case NVME_IOCTL_IO_CMD:
1586                 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1587                 break;
1588         case NVME_IOCTL_SUBMIT_IO:
1589                 ret = nvme_submit_io(ns, argp);
1590                 break;
1591         case NVME_IOCTL_IO64_CMD:
1592                 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1593                 break;
1594         default:
1595                 if (ns->ndev)
1596                         ret = nvme_nvm_ioctl(ns, cmd, arg);
1597                 else
1598                         ret = -ENOTTY;
1599         }
1600 
1601         nvme_put_ns_from_disk(head, srcu_idx);
1602         return ret;
1603 }
1604 
1605 static int nvme_open(struct block_device *bdev, fmode_t mode)
1606 {
1607         struct nvme_ns *ns = bdev->bd_disk->private_data;
1608 
1609 #ifdef CONFIG_NVME_MULTIPATH
1610         /* should never be called due to GENHD_FL_HIDDEN */
1611         if (WARN_ON_ONCE(ns->head->disk))
1612                 goto fail;
1613 #endif
1614         if (!kref_get_unless_zero(&ns->kref))
1615                 goto fail;
1616         if (!try_module_get(ns->ctrl->ops->module))
1617                 goto fail_put_ns;
1618 
1619         return 0;
1620 
1621 fail_put_ns:
1622         nvme_put_ns(ns);
1623 fail:
1624         return -ENXIO;
1625 }
1626 
1627 static void nvme_release(struct gendisk *disk, fmode_t mode)
1628 {
1629         struct nvme_ns *ns = disk->private_data;
1630 
1631         module_put(ns->ctrl->ops->module);
1632         nvme_put_ns(ns);
1633 }
1634 
1635 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1636 {
1637         /* some standard values */
1638         geo->heads = 1 << 6;
1639         geo->sectors = 1 << 5;
1640         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1641         return 0;
1642 }
1643 
1644 #ifdef CONFIG_BLK_DEV_INTEGRITY
1645 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1646 {
1647         struct blk_integrity integrity;
1648 
1649         memset(&integrity, 0, sizeof(integrity));
1650         switch (pi_type) {
1651         case NVME_NS_DPS_PI_TYPE3:
1652                 integrity.profile = &t10_pi_type3_crc;
1653                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1654                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1655                 break;
1656         case NVME_NS_DPS_PI_TYPE1:
1657         case NVME_NS_DPS_PI_TYPE2:
1658                 integrity.profile = &t10_pi_type1_crc;
1659                 integrity.tag_size = sizeof(u16);
1660                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1661                 break;
1662         default:
1663                 integrity.profile = NULL;
1664                 break;
1665         }
1666         integrity.tuple_size = ms;
1667         blk_integrity_register(disk, &integrity);
1668         blk_queue_max_integrity_segments(disk->queue, 1);
1669 }
1670 #else
1671 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1672 {
1673 }
1674 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1675 
1676 static void nvme_set_chunk_size(struct nvme_ns *ns)
1677 {
1678         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1679         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1680 }
1681 
1682 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1683 {
1684         struct nvme_ctrl *ctrl = ns->ctrl;
1685         struct request_queue *queue = disk->queue;
1686         u32 size = queue_logical_block_size(queue);
1687 
1688         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1689                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1690                 return;
1691         }
1692 
1693         if (ctrl->nr_streams && ns->sws && ns->sgs)
1694                 size *= ns->sws * ns->sgs;
1695 
1696         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1697                         NVME_DSM_MAX_RANGES);
1698 
1699         queue->limits.discard_alignment = 0;
1700         queue->limits.discard_granularity = size;
1701 
1702         /* If discard is already enabled, don't reset queue limits */
1703         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1704                 return;
1705 
1706         blk_queue_max_discard_sectors(queue, UINT_MAX);
1707         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1708 
1709         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1710                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1711 }
1712 
1713 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1714 {
1715         u32 max_sectors;
1716         unsigned short bs = 1 << ns->lba_shift;
1717 
1718         if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1719             (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1720                 return;
1721         /*
1722          * Even though NVMe spec explicitly states that MDTS is not
1723          * applicable to the write-zeroes:- "The restriction does not apply to
1724          * commands that do not transfer data between the host and the
1725          * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1726          * In order to be more cautious use controller's max_hw_sectors value
1727          * to configure the maximum sectors for the write-zeroes which is
1728          * configured based on the controller's MDTS field in the
1729          * nvme_init_identify() if available.
1730          */
1731         if (ns->ctrl->max_hw_sectors == UINT_MAX)
1732                 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1733         else
1734                 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1735 
1736         blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1737 }
1738 
1739 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1740                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1741 {
1742         memset(ids, 0, sizeof(*ids));
1743 
1744         if (ctrl->vs >= NVME_VS(1, 1, 0))
1745                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1746         if (ctrl->vs >= NVME_VS(1, 2, 0))
1747                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1748         if (ctrl->vs >= NVME_VS(1, 3, 0))
1749                 return nvme_identify_ns_descs(ctrl, nsid, ids);
1750         return 0;
1751 }
1752 
1753 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1754 {
1755         return !uuid_is_null(&ids->uuid) ||
1756                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1757                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1758 }
1759 
1760 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1761 {
1762         return uuid_equal(&a->uuid, &b->uuid) &&
1763                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1764                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1765 }
1766 
1767 static void nvme_update_disk_info(struct gendisk *disk,
1768                 struct nvme_ns *ns, struct nvme_id_ns *id)
1769 {
1770         sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1771         unsigned short bs = 1 << ns->lba_shift;
1772         u32 atomic_bs, phys_bs, io_opt;
1773 
1774         if (ns->lba_shift > PAGE_SHIFT) {
1775                 /* unsupported block size, set capacity to 0 later */
1776                 bs = (1 << 9);
1777         }
1778         blk_mq_freeze_queue(disk->queue);
1779         blk_integrity_unregister(disk);
1780 
1781         if (id->nabo == 0) {
1782                 /*
1783                  * Bit 1 indicates whether NAWUPF is defined for this namespace
1784                  * and whether it should be used instead of AWUPF. If NAWUPF ==
1785                  * 0 then AWUPF must be used instead.
1786                  */
1787                 if (id->nsfeat & (1 << 1) && id->nawupf)
1788                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1789                 else
1790                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1791         } else {
1792                 atomic_bs = bs;
1793         }
1794         phys_bs = bs;
1795         io_opt = bs;
1796         if (id->nsfeat & (1 << 4)) {
1797                 /* NPWG = Namespace Preferred Write Granularity */
1798                 phys_bs *= 1 + le16_to_cpu(id->npwg);
1799                 /* NOWS = Namespace Optimal Write Size */
1800                 io_opt *= 1 + le16_to_cpu(id->nows);
1801         }
1802 
1803         blk_queue_logical_block_size(disk->queue, bs);
1804         /*
1805          * Linux filesystems assume writing a single physical block is
1806          * an atomic operation. Hence limit the physical block size to the
1807          * value of the Atomic Write Unit Power Fail parameter.
1808          */
1809         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1810         blk_queue_io_min(disk->queue, phys_bs);
1811         blk_queue_io_opt(disk->queue, io_opt);
1812 
1813         if (ns->ms && !ns->ext &&
1814             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1815                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1816         if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1817             ns->lba_shift > PAGE_SHIFT)
1818                 capacity = 0;
1819 
1820         set_capacity(disk, capacity);
1821 
1822         nvme_config_discard(disk, ns);
1823         nvme_config_write_zeroes(disk, ns);
1824 
1825         if (id->nsattr & (1 << 0))
1826                 set_disk_ro(disk, true);
1827         else
1828                 set_disk_ro(disk, false);
1829 
1830         blk_mq_unfreeze_queue(disk->queue);
1831 }
1832 
1833 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1834 {
1835         struct nvme_ns *ns = disk->private_data;
1836 
1837         /*
1838          * If identify namespace failed, use default 512 byte block size so
1839          * block layer can use before failing read/write for 0 capacity.
1840          */
1841         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1842         if (ns->lba_shift == 0)
1843                 ns->lba_shift = 9;
1844         ns->noiob = le16_to_cpu(id->noiob);
1845         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1846         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1847         /* the PI implementation requires metadata equal t10 pi tuple size */
1848         if (ns->ms == sizeof(struct t10_pi_tuple))
1849                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1850         else
1851                 ns->pi_type = 0;
1852 
1853         if (ns->noiob)
1854                 nvme_set_chunk_size(ns);
1855         nvme_update_disk_info(disk, ns, id);
1856 #ifdef CONFIG_NVME_MULTIPATH
1857         if (ns->head->disk) {
1858                 nvme_update_disk_info(ns->head->disk, ns, id);
1859                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1860                 revalidate_disk(ns->head->disk);
1861         }
1862 #endif
1863 }
1864 
1865 static int nvme_revalidate_disk(struct gendisk *disk)
1866 {
1867         struct nvme_ns *ns = disk->private_data;
1868         struct nvme_ctrl *ctrl = ns->ctrl;
1869         struct nvme_id_ns *id;
1870         struct nvme_ns_ids ids;
1871         int ret = 0;
1872 
1873         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1874                 set_capacity(disk, 0);
1875                 return -ENODEV;
1876         }
1877 
1878         ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1879         if (ret)
1880                 goto out;
1881 
1882         if (id->ncap == 0) {
1883                 ret = -ENODEV;
1884                 goto free_id;
1885         }
1886 
1887         __nvme_revalidate_disk(disk, id);
1888         ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1889         if (ret)
1890                 goto free_id;
1891 
1892         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1893                 dev_err(ctrl->device,
1894                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1895                 ret = -ENODEV;
1896         }
1897 
1898 free_id:
1899         kfree(id);
1900 out:
1901         /*
1902          * Only fail the function if we got a fatal error back from the
1903          * device, otherwise ignore the error and just move on.
1904          */
1905         if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1906                 ret = 0;
1907         else if (ret > 0)
1908                 ret = blk_status_to_errno(nvme_error_status(ret));
1909         return ret;
1910 }
1911 
1912 static char nvme_pr_type(enum pr_type type)
1913 {
1914         switch (type) {
1915         case PR_WRITE_EXCLUSIVE:
1916                 return 1;
1917         case PR_EXCLUSIVE_ACCESS:
1918                 return 2;
1919         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1920                 return 3;
1921         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1922                 return 4;
1923         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1924                 return 5;
1925         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1926                 return 6;
1927         default:
1928                 return 0;
1929         }
1930 };
1931 
1932 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1933                                 u64 key, u64 sa_key, u8 op)
1934 {
1935         struct nvme_ns_head *head = NULL;
1936         struct nvme_ns *ns;
1937         struct nvme_command c;
1938         int srcu_idx, ret;
1939         u8 data[16] = { 0, };
1940 
1941         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1942         if (unlikely(!ns))
1943                 return -EWOULDBLOCK;
1944 
1945         put_unaligned_le64(key, &data[0]);
1946         put_unaligned_le64(sa_key, &data[8]);
1947 
1948         memset(&c, 0, sizeof(c));
1949         c.common.opcode = op;
1950         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1951         c.common.cdw10 = cpu_to_le32(cdw10);
1952 
1953         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1954         nvme_put_ns_from_disk(head, srcu_idx);
1955         return ret;
1956 }
1957 
1958 static int nvme_pr_register(struct block_device *bdev, u64 old,
1959                 u64 new, unsigned flags)
1960 {
1961         u32 cdw10;
1962 
1963         if (flags & ~PR_FL_IGNORE_KEY)
1964                 return -EOPNOTSUPP;
1965 
1966         cdw10 = old ? 2 : 0;
1967         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1968         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1969         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1970 }
1971 
1972 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1973                 enum pr_type type, unsigned flags)
1974 {
1975         u32 cdw10;
1976 
1977         if (flags & ~PR_FL_IGNORE_KEY)
1978                 return -EOPNOTSUPP;
1979 
1980         cdw10 = nvme_pr_type(type) << 8;
1981         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1982         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1983 }
1984 
1985 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1986                 enum pr_type type, bool abort)
1987 {
1988         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1989         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1990 }
1991 
1992 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1993 {
1994         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1995         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1996 }
1997 
1998 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1999 {
2000         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2001         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2002 }
2003 
2004 static const struct pr_ops nvme_pr_ops = {
2005         .pr_register    = nvme_pr_register,
2006         .pr_reserve     = nvme_pr_reserve,
2007         .pr_release     = nvme_pr_release,
2008         .pr_preempt     = nvme_pr_preempt,
2009         .pr_clear       = nvme_pr_clear,
2010 };
2011 
2012 #ifdef CONFIG_BLK_SED_OPAL
2013 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2014                 bool send)
2015 {
2016         struct nvme_ctrl *ctrl = data;
2017         struct nvme_command cmd;
2018 
2019         memset(&cmd, 0, sizeof(cmd));
2020         if (send)
2021                 cmd.common.opcode = nvme_admin_security_send;
2022         else
2023                 cmd.common.opcode = nvme_admin_security_recv;
2024         cmd.common.nsid = 0;
2025         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2026         cmd.common.cdw11 = cpu_to_le32(len);
2027 
2028         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2029                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2030 }
2031 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2032 #endif /* CONFIG_BLK_SED_OPAL */
2033 
2034 static const struct block_device_operations nvme_fops = {
2035         .owner          = THIS_MODULE,
2036         .ioctl          = nvme_ioctl,
2037         .compat_ioctl   = nvme_ioctl,
2038         .open           = nvme_open,
2039         .release        = nvme_release,
2040         .getgeo         = nvme_getgeo,
2041         .revalidate_disk= nvme_revalidate_disk,
2042         .pr_ops         = &nvme_pr_ops,
2043 };
2044 
2045 #ifdef CONFIG_NVME_MULTIPATH
2046 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2047 {
2048         struct nvme_ns_head *head = bdev->bd_disk->private_data;
2049 
2050         if (!kref_get_unless_zero(&head->ref))
2051                 return -ENXIO;
2052         return 0;
2053 }
2054 
2055 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2056 {
2057         nvme_put_ns_head(disk->private_data);
2058 }
2059 
2060 const struct block_device_operations nvme_ns_head_ops = {
2061         .owner          = THIS_MODULE,
2062         .open           = nvme_ns_head_open,
2063         .release        = nvme_ns_head_release,
2064         .ioctl          = nvme_ioctl,
2065         .compat_ioctl   = nvme_ioctl,
2066         .getgeo         = nvme_getgeo,
2067         .pr_ops         = &nvme_pr_ops,
2068 };
2069 #endif /* CONFIG_NVME_MULTIPATH */
2070 
2071 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2072 {
2073         unsigned long timeout =
2074                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2075         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2076         int ret;
2077 
2078         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2079                 if (csts == ~0)
2080                         return -ENODEV;
2081                 if ((csts & NVME_CSTS_RDY) == bit)
2082                         break;
2083 
2084                 msleep(100);
2085                 if (fatal_signal_pending(current))
2086                         return -EINTR;
2087                 if (time_after(jiffies, timeout)) {
2088                         dev_err(ctrl->device,
2089                                 "Device not ready; aborting %s\n", enabled ?
2090                                                 "initialisation" : "reset");
2091                         return -ENODEV;
2092                 }
2093         }
2094 
2095         return ret;
2096 }
2097 
2098 /*
2099  * If the device has been passed off to us in an enabled state, just clear
2100  * the enabled bit.  The spec says we should set the 'shutdown notification
2101  * bits', but doing so may cause the device to complete commands to the
2102  * admin queue ... and we don't know what memory that might be pointing at!
2103  */
2104 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2105 {
2106         int ret;
2107 
2108         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2109         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2110 
2111         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2112         if (ret)
2113                 return ret;
2114 
2115         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2116                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2117 
2118         return nvme_wait_ready(ctrl, ctrl->cap, false);
2119 }
2120 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2121 
2122 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2123 {
2124         /*
2125          * Default to a 4K page size, with the intention to update this
2126          * path in the future to accomodate architectures with differing
2127          * kernel and IO page sizes.
2128          */
2129         unsigned dev_page_min, page_shift = 12;
2130         int ret;
2131 
2132         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2133         if (ret) {
2134                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2135                 return ret;
2136         }
2137         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2138 
2139         if (page_shift < dev_page_min) {
2140                 dev_err(ctrl->device,
2141                         "Minimum device page size %u too large for host (%u)\n",
2142                         1 << dev_page_min, 1 << page_shift);
2143                 return -ENODEV;
2144         }
2145 
2146         ctrl->page_size = 1 << page_shift;
2147 
2148         ctrl->ctrl_config = NVME_CC_CSS_NVM;
2149         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2150         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2151         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2152         ctrl->ctrl_config |= NVME_CC_ENABLE;
2153 
2154         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2155         if (ret)
2156                 return ret;
2157         return nvme_wait_ready(ctrl, ctrl->cap, true);
2158 }
2159 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2160 
2161 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2162 {
2163         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2164         u32 csts;
2165         int ret;
2166 
2167         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2168         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2169 
2170         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2171         if (ret)
2172                 return ret;
2173 
2174         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2175                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2176                         break;
2177 
2178                 msleep(100);
2179                 if (fatal_signal_pending(current))
2180                         return -EINTR;
2181                 if (time_after(jiffies, timeout)) {
2182                         dev_err(ctrl->device,
2183                                 "Device shutdown incomplete; abort shutdown\n");
2184                         return -ENODEV;
2185                 }
2186         }
2187 
2188         return ret;
2189 }
2190 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2191 
2192 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2193                 struct request_queue *q)
2194 {
2195         bool vwc = false;
2196 
2197         if (ctrl->max_hw_sectors) {
2198                 u32 max_segments =
2199                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2200 
2201                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2202                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2203                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2204         }
2205         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2206             is_power_of_2(ctrl->max_hw_sectors))
2207                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2208         blk_queue_virt_boundary(q, ctrl->page_size - 1);
2209         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2210                 vwc = true;
2211         blk_queue_write_cache(q, vwc, vwc);
2212 }
2213 
2214 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2215 {
2216         __le64 ts;
2217         int ret;
2218 
2219         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2220                 return 0;
2221 
2222         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2223         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2224                         NULL);
2225         if (ret)
2226                 dev_warn_once(ctrl->device,
2227                         "could not set timestamp (%d)\n", ret);
2228         return ret;
2229 }
2230 
2231 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2232 {
2233         struct nvme_feat_host_behavior *host;
2234         int ret;
2235 
2236         /* Don't bother enabling the feature if retry delay is not reported */
2237         if (!ctrl->crdt[0])
2238                 return 0;
2239 
2240         host = kzalloc(sizeof(*host), GFP_KERNEL);
2241         if (!host)
2242                 return 0;
2243 
2244         host->acre = NVME_ENABLE_ACRE;
2245         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2246                                 host, sizeof(*host), NULL);
2247         kfree(host);
2248         return ret;
2249 }
2250 
2251 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2252 {
2253         /*
2254          * APST (Autonomous Power State Transition) lets us program a
2255          * table of power state transitions that the controller will
2256          * perform automatically.  We configure it with a simple
2257          * heuristic: we are willing to spend at most 2% of the time
2258          * transitioning between power states.  Therefore, when running
2259          * in any given state, we will enter the next lower-power
2260          * non-operational state after waiting 50 * (enlat + exlat)
2261          * microseconds, as long as that state's exit latency is under
2262          * the requested maximum latency.
2263          *
2264          * We will not autonomously enter any non-operational state for
2265          * which the total latency exceeds ps_max_latency_us.  Users
2266          * can set ps_max_latency_us to zero to turn off APST.
2267          */
2268 
2269         unsigned apste;
2270         struct nvme_feat_auto_pst *table;
2271         u64 max_lat_us = 0;
2272         int max_ps = -1;
2273         int ret;
2274 
2275         /*
2276          * If APST isn't supported or if we haven't been initialized yet,
2277          * then don't do anything.
2278          */
2279         if (!ctrl->apsta)
2280                 return 0;
2281 
2282         if (ctrl->npss > 31) {
2283                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2284                 return 0;
2285         }
2286 
2287         table = kzalloc(sizeof(*table), GFP_KERNEL);
2288         if (!table)
2289                 return 0;
2290 
2291         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2292                 /* Turn off APST. */
2293                 apste = 0;
2294                 dev_dbg(ctrl->device, "APST disabled\n");
2295         } else {
2296                 __le64 target = cpu_to_le64(0);
2297                 int state;
2298 
2299                 /*
2300                  * Walk through all states from lowest- to highest-power.
2301                  * According to the spec, lower-numbered states use more
2302                  * power.  NPSS, despite the name, is the index of the
2303                  * lowest-power state, not the number of states.
2304                  */
2305                 for (state = (int)ctrl->npss; state >= 0; state--) {
2306                         u64 total_latency_us, exit_latency_us, transition_ms;
2307 
2308                         if (target)
2309                                 table->entries[state] = target;
2310 
2311                         /*
2312                          * Don't allow transitions to the deepest state
2313                          * if it's quirked off.
2314                          */
2315                         if (state == ctrl->npss &&
2316                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2317                                 continue;
2318 
2319                         /*
2320                          * Is this state a useful non-operational state for
2321                          * higher-power states to autonomously transition to?
2322                          */
2323                         if (!(ctrl->psd[state].flags &
2324                               NVME_PS_FLAGS_NON_OP_STATE))
2325                                 continue;
2326 
2327                         exit_latency_us =
2328                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2329                         if (exit_latency_us > ctrl->ps_max_latency_us)
2330                                 continue;
2331 
2332                         total_latency_us =
2333                                 exit_latency_us +
2334                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2335 
2336                         /*
2337                          * This state is good.  Use it as the APST idle
2338                          * target for higher power states.
2339                          */
2340                         transition_ms = total_latency_us + 19;
2341                         do_div(transition_ms, 20);
2342                         if (transition_ms > (1 << 24) - 1)
2343                                 transition_ms = (1 << 24) - 1;
2344 
2345                         target = cpu_to_le64((state << 3) |
2346                                              (transition_ms << 8));
2347 
2348                         if (max_ps == -1)
2349                                 max_ps = state;
2350 
2351                         if (total_latency_us > max_lat_us)
2352                                 max_lat_us = total_latency_us;
2353                 }
2354 
2355                 apste = 1;
2356 
2357                 if (max_ps == -1) {
2358                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2359                 } else {
2360                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2361                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2362                 }
2363         }
2364 
2365         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2366                                 table, sizeof(*table), NULL);
2367         if (ret)
2368                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2369 
2370         kfree(table);
2371         return ret;
2372 }
2373 
2374 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2375 {
2376         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2377         u64 latency;
2378 
2379         switch (val) {
2380         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2381         case PM_QOS_LATENCY_ANY:
2382                 latency = U64_MAX;
2383                 break;
2384 
2385         default:
2386                 latency = val;
2387         }
2388 
2389         if (ctrl->ps_max_latency_us != latency) {
2390                 ctrl->ps_max_latency_us = latency;
2391                 nvme_configure_apst(ctrl);
2392         }
2393 }
2394 
2395 struct nvme_core_quirk_entry {
2396         /*
2397          * NVMe model and firmware strings are padded with spaces.  For
2398          * simplicity, strings in the quirk table are padded with NULLs
2399          * instead.
2400          */
2401         u16 vid;
2402         const char *mn;
2403         const char *fr;
2404         unsigned long quirks;
2405 };
2406 
2407 static const struct nvme_core_quirk_entry core_quirks[] = {
2408         {
2409                 /*
2410                  * This Toshiba device seems to die using any APST states.  See:
2411                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2412                  */
2413                 .vid = 0x1179,
2414                 .mn = "THNSF5256GPUK TOSHIBA",
2415                 .quirks = NVME_QUIRK_NO_APST,
2416         },
2417         {
2418                 /*
2419                  * This LiteON CL1-3D*-Q11 firmware version has a race
2420                  * condition associated with actions related to suspend to idle
2421                  * LiteON has resolved the problem in future firmware
2422                  */
2423                 .vid = 0x14a4,
2424                 .fr = "22301111",
2425                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2426         }
2427 };
2428 
2429 /* match is null-terminated but idstr is space-padded. */
2430 static bool string_matches(const char *idstr, const char *match, size_t len)
2431 {
2432         size_t matchlen;
2433 
2434         if (!match)
2435                 return true;
2436 
2437         matchlen = strlen(match);
2438         WARN_ON_ONCE(matchlen > len);
2439 
2440         if (memcmp(idstr, match, matchlen))
2441                 return false;
2442 
2443         for (; matchlen < len; matchlen++)
2444                 if (idstr[matchlen] != ' ')
2445                         return false;
2446 
2447         return true;
2448 }
2449 
2450 static bool quirk_matches(const struct nvme_id_ctrl *id,
2451                           const struct nvme_core_quirk_entry *q)
2452 {
2453         return q->vid == le16_to_cpu(id->vid) &&
2454                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2455                 string_matches(id->fr, q->fr, sizeof(id->fr));
2456 }
2457 
2458 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2459                 struct nvme_id_ctrl *id)
2460 {
2461         size_t nqnlen;
2462         int off;
2463 
2464         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2465                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2466                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2467                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2468                         return;
2469                 }
2470 
2471                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2472                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2473         }
2474 
2475         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2476         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2477                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2478                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2479         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2480         off += sizeof(id->sn);
2481         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2482         off += sizeof(id->mn);
2483         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2484 }
2485 
2486 static void nvme_release_subsystem(struct device *dev)
2487 {
2488         struct nvme_subsystem *subsys =
2489                 container_of(dev, struct nvme_subsystem, dev);
2490 
2491         if (subsys->instance >= 0)
2492                 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2493         kfree(subsys);
2494 }
2495 
2496 static void nvme_destroy_subsystem(struct kref *ref)
2497 {
2498         struct nvme_subsystem *subsys =
2499                         container_of(ref, struct nvme_subsystem, ref);
2500 
2501         mutex_lock(&nvme_subsystems_lock);
2502         list_del(&subsys->entry);
2503         mutex_unlock(&nvme_subsystems_lock);
2504 
2505         ida_destroy(&subsys->ns_ida);
2506         device_del(&subsys->dev);
2507         put_device(&subsys->dev);
2508 }
2509 
2510 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2511 {
2512         kref_put(&subsys->ref, nvme_destroy_subsystem);
2513 }
2514 
2515 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2516 {
2517         struct nvme_subsystem *subsys;
2518 
2519         lockdep_assert_held(&nvme_subsystems_lock);
2520 
2521         /*
2522          * Fail matches for discovery subsystems. This results
2523          * in each discovery controller bound to a unique subsystem.
2524          * This avoids issues with validating controller values
2525          * that can only be true when there is a single unique subsystem.
2526          * There may be multiple and completely independent entities
2527          * that provide discovery controllers.
2528          */
2529         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2530                 return NULL;
2531 
2532         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2533                 if (strcmp(subsys->subnqn, subsysnqn))
2534                         continue;
2535                 if (!kref_get_unless_zero(&subsys->ref))
2536                         continue;
2537                 return subsys;
2538         }
2539 
2540         return NULL;
2541 }
2542 
2543 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2544         struct device_attribute subsys_attr_##_name = \
2545                 __ATTR(_name, _mode, _show, NULL)
2546 
2547 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2548                                     struct device_attribute *attr,
2549                                     char *buf)
2550 {
2551         struct nvme_subsystem *subsys =
2552                 container_of(dev, struct nvme_subsystem, dev);
2553 
2554         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2555 }
2556 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2557 
2558 #define nvme_subsys_show_str_function(field)                            \
2559 static ssize_t subsys_##field##_show(struct device *dev,                \
2560                             struct device_attribute *attr, char *buf)   \
2561 {                                                                       \
2562         struct nvme_subsystem *subsys =                                 \
2563                 container_of(dev, struct nvme_subsystem, dev);          \
2564         return sprintf(buf, "%.*s\n",                                   \
2565                        (int)sizeof(subsys->field), subsys->field);      \
2566 }                                                                       \
2567 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2568 
2569 nvme_subsys_show_str_function(model);
2570 nvme_subsys_show_str_function(serial);
2571 nvme_subsys_show_str_function(firmware_rev);
2572 
2573 static struct attribute *nvme_subsys_attrs[] = {
2574         &subsys_attr_model.attr,
2575         &subsys_attr_serial.attr,
2576         &subsys_attr_firmware_rev.attr,
2577         &subsys_attr_subsysnqn.attr,
2578 #ifdef CONFIG_NVME_MULTIPATH
2579         &subsys_attr_iopolicy.attr,
2580 #endif
2581         NULL,
2582 };
2583 
2584 static struct attribute_group nvme_subsys_attrs_group = {
2585         .attrs = nvme_subsys_attrs,
2586 };
2587 
2588 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2589         &nvme_subsys_attrs_group,
2590         NULL,
2591 };
2592 
2593 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2594                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2595 {
2596         struct nvme_ctrl *tmp;
2597 
2598         lockdep_assert_held(&nvme_subsystems_lock);
2599 
2600         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2601                 if (tmp->state == NVME_CTRL_DELETING ||
2602                     tmp->state == NVME_CTRL_DEAD)
2603                         continue;
2604 
2605                 if (tmp->cntlid == ctrl->cntlid) {
2606                         dev_err(ctrl->device,
2607                                 "Duplicate cntlid %u with %s, rejecting\n",
2608                                 ctrl->cntlid, dev_name(tmp->device));
2609                         return false;
2610                 }
2611 
2612                 if ((id->cmic & (1 << 1)) ||
2613                     (ctrl->opts && ctrl->opts->discovery_nqn))
2614                         continue;
2615 
2616                 dev_err(ctrl->device,
2617                         "Subsystem does not support multiple controllers\n");
2618                 return false;
2619         }
2620 
2621         return true;
2622 }
2623 
2624 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2625 {
2626         struct nvme_subsystem *subsys, *found;
2627         int ret;
2628 
2629         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2630         if (!subsys)
2631                 return -ENOMEM;
2632 
2633         subsys->instance = -1;
2634         mutex_init(&subsys->lock);
2635         kref_init(&subsys->ref);
2636         INIT_LIST_HEAD(&subsys->ctrls);
2637         INIT_LIST_HEAD(&subsys->nsheads);
2638         nvme_init_subnqn(subsys, ctrl, id);
2639         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2640         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2641         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2642         subsys->vendor_id = le16_to_cpu(id->vid);
2643         subsys->cmic = id->cmic;
2644         subsys->awupf = le16_to_cpu(id->awupf);
2645 #ifdef CONFIG_NVME_MULTIPATH
2646         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2647 #endif
2648 
2649         subsys->dev.class = nvme_subsys_class;
2650         subsys->dev.release = nvme_release_subsystem;
2651         subsys->dev.groups = nvme_subsys_attrs_groups;
2652         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2653         device_initialize(&subsys->dev);
2654 
2655         mutex_lock(&nvme_subsystems_lock);
2656         found = __nvme_find_get_subsystem(subsys->subnqn);
2657         if (found) {
2658                 put_device(&subsys->dev);
2659                 subsys = found;
2660 
2661                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2662                         ret = -EINVAL;
2663                         goto out_put_subsystem;
2664                 }
2665         } else {
2666                 ret = device_add(&subsys->dev);
2667                 if (ret) {
2668                         dev_err(ctrl->device,
2669                                 "failed to register subsystem device.\n");
2670                         put_device(&subsys->dev);
2671                         goto out_unlock;
2672                 }
2673                 ida_init(&subsys->ns_ida);
2674                 list_add_tail(&subsys->entry, &nvme_subsystems);
2675         }
2676 
2677         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2678                                 dev_name(ctrl->device));
2679         if (ret) {
2680                 dev_err(ctrl->device,
2681                         "failed to create sysfs link from subsystem.\n");
2682                 goto out_put_subsystem;
2683         }
2684 
2685         if (!found)
2686                 subsys->instance = ctrl->instance;
2687         ctrl->subsys = subsys;
2688         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2689         mutex_unlock(&nvme_subsystems_lock);
2690         return 0;
2691 
2692 out_put_subsystem:
2693         nvme_put_subsystem(subsys);
2694 out_unlock:
2695         mutex_unlock(&nvme_subsystems_lock);
2696         return ret;
2697 }
2698 
2699 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2700                 void *log, size_t size, u64 offset)
2701 {
2702         struct nvme_command c = { };
2703         unsigned long dwlen = size / 4 - 1;
2704 
2705         c.get_log_page.opcode = nvme_admin_get_log_page;
2706         c.get_log_page.nsid = cpu_to_le32(nsid);
2707         c.get_log_page.lid = log_page;
2708         c.get_log_page.lsp = lsp;
2709         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2710         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2711         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2712         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2713 
2714         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2715 }
2716 
2717 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2718 {
2719         int ret;
2720 
2721         if (!ctrl->effects)
2722                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2723 
2724         if (!ctrl->effects)
2725                 return 0;
2726 
2727         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2728                         ctrl->effects, sizeof(*ctrl->effects), 0);
2729         if (ret) {
2730                 kfree(ctrl->effects);
2731                 ctrl->effects = NULL;
2732         }
2733         return ret;
2734 }
2735 
2736 /*
2737  * Initialize the cached copies of the Identify data and various controller
2738  * register in our nvme_ctrl structure.  This should be called as soon as
2739  * the admin queue is fully up and running.
2740  */
2741 int nvme_init_identify(struct nvme_ctrl *ctrl)
2742 {
2743         struct nvme_id_ctrl *id;
2744         int ret, page_shift;
2745         u32 max_hw_sectors;
2746         bool prev_apst_enabled;
2747 
2748         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2749         if (ret) {
2750                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2751                 return ret;
2752         }
2753         page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2754         ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2755 
2756         if (ctrl->vs >= NVME_VS(1, 1, 0))
2757                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2758 
2759         ret = nvme_identify_ctrl(ctrl, &id);
2760         if (ret) {
2761                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2762                 return -EIO;
2763         }
2764 
2765         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2766                 ret = nvme_get_effects_log(ctrl);
2767                 if (ret < 0)
2768                         goto out_free;
2769         }
2770 
2771         if (!(ctrl->ops->flags & NVME_F_FABRICS))
2772                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2773 
2774         if (!ctrl->identified) {
2775                 int i;
2776 
2777                 ret = nvme_init_subsystem(ctrl, id);
2778                 if (ret)
2779                         goto out_free;
2780 
2781                 /*
2782                  * Check for quirks.  Quirk can depend on firmware version,
2783                  * so, in principle, the set of quirks present can change
2784                  * across a reset.  As a possible future enhancement, we
2785                  * could re-scan for quirks every time we reinitialize
2786                  * the device, but we'd have to make sure that the driver
2787                  * behaves intelligently if the quirks change.
2788                  */
2789                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2790                         if (quirk_matches(id, &core_quirks[i]))
2791                                 ctrl->quirks |= core_quirks[i].quirks;
2792                 }
2793         }
2794 
2795         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2796                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2797                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2798         }
2799 
2800         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2801         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2802         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2803 
2804         ctrl->oacs = le16_to_cpu(id->oacs);
2805         ctrl->oncs = le16_to_cpu(id->oncs);
2806         ctrl->mtfa = le16_to_cpu(id->mtfa);
2807         ctrl->oaes = le32_to_cpu(id->oaes);
2808         atomic_set(&ctrl->abort_limit, id->acl + 1);
2809         ctrl->vwc = id->vwc;
2810         if (id->mdts)
2811                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2812         else
2813                 max_hw_sectors = UINT_MAX;
2814         ctrl->max_hw_sectors =
2815                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2816 
2817         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2818         ctrl->sgls = le32_to_cpu(id->sgls);
2819         ctrl->kas = le16_to_cpu(id->kas);
2820         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2821         ctrl->ctratt = le32_to_cpu(id->ctratt);
2822 
2823         if (id->rtd3e) {
2824                 /* us -> s */
2825                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2826 
2827                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2828                                                  shutdown_timeout, 60);
2829 
2830                 if (ctrl->shutdown_timeout != shutdown_timeout)
2831                         dev_info(ctrl->device,
2832                                  "Shutdown timeout set to %u seconds\n",
2833                                  ctrl->shutdown_timeout);
2834         } else
2835                 ctrl->shutdown_timeout = shutdown_timeout;
2836 
2837         ctrl->npss = id->npss;
2838         ctrl->apsta = id->apsta;
2839         prev_apst_enabled = ctrl->apst_enabled;
2840         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2841                 if (force_apst && id->apsta) {
2842                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2843                         ctrl->apst_enabled = true;
2844                 } else {
2845                         ctrl->apst_enabled = false;
2846                 }
2847         } else {
2848                 ctrl->apst_enabled = id->apsta;
2849         }
2850         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2851 
2852         if (ctrl->ops->flags & NVME_F_FABRICS) {
2853                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2854                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2855                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2856                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2857 
2858                 /*
2859                  * In fabrics we need to verify the cntlid matches the
2860                  * admin connect
2861                  */
2862                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2863                         ret = -EINVAL;
2864                         goto out_free;
2865                 }
2866 
2867                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2868                         dev_err(ctrl->device,
2869                                 "keep-alive support is mandatory for fabrics\n");
2870                         ret = -EINVAL;
2871                         goto out_free;
2872                 }
2873         } else {
2874                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2875                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2876                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2877                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2878         }
2879 
2880         ret = nvme_mpath_init(ctrl, id);
2881         kfree(id);
2882 
2883         if (ret < 0)
2884                 return ret;
2885 
2886         if (ctrl->apst_enabled && !prev_apst_enabled)
2887                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2888         else if (!ctrl->apst_enabled && prev_apst_enabled)
2889                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2890 
2891         ret = nvme_configure_apst(ctrl);
2892         if (ret < 0)
2893                 return ret;
2894         
2895         ret = nvme_configure_timestamp(ctrl);
2896         if (ret < 0)
2897                 return ret;
2898 
2899         ret = nvme_configure_directives(ctrl);
2900         if (ret < 0)
2901                 return ret;
2902 
2903         ret = nvme_configure_acre(ctrl);
2904         if (ret < 0)
2905                 return ret;
2906 
2907         ctrl->identified = true;
2908 
2909         return 0;
2910 
2911 out_free:
2912         kfree(id);
2913         return ret;
2914 }
2915 EXPORT_SYMBOL_GPL(nvme_init_identify);
2916 
2917 static int nvme_dev_open(struct inode *inode, struct file *file)
2918 {
2919         struct nvme_ctrl *ctrl =
2920                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2921 
2922         switch (ctrl->state) {
2923         case NVME_CTRL_LIVE:
2924                 break;
2925         default:
2926                 return -EWOULDBLOCK;
2927         }
2928 
2929         file->private_data = ctrl;
2930         return 0;
2931 }
2932 
2933 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2934 {
2935         struct nvme_ns *ns;
2936         int ret;
2937 
2938         down_read(&ctrl->namespaces_rwsem);
2939         if (list_empty(&ctrl->namespaces)) {
2940                 ret = -ENOTTY;
2941                 goto out_unlock;
2942         }
2943 
2944         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2945         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2946                 dev_warn(ctrl->device,
2947                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2948                 ret = -EINVAL;
2949                 goto out_unlock;
2950         }
2951 
2952         dev_warn(ctrl->device,
2953                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2954         kref_get(&ns->kref);
2955         up_read(&ctrl->namespaces_rwsem);
2956 
2957         ret = nvme_user_cmd(ctrl, ns, argp);
2958         nvme_put_ns(ns);
2959         return ret;
2960 
2961 out_unlock:
2962         up_read(&ctrl->namespaces_rwsem);
2963         return ret;
2964 }
2965 
2966 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2967                 unsigned long arg)
2968 {
2969         struct nvme_ctrl *ctrl = file->private_data;
2970         void __user *argp = (void __user *)arg;
2971 
2972         switch (cmd) {
2973         case NVME_IOCTL_ADMIN_CMD:
2974                 return nvme_user_cmd(ctrl, NULL, argp);
2975         case NVME_IOCTL_ADMIN64_CMD:
2976                 return nvme_user_cmd64(ctrl, NULL, argp);
2977         case NVME_IOCTL_IO_CMD:
2978                 return nvme_dev_user_cmd(ctrl, argp);
2979         case NVME_IOCTL_RESET:
2980                 dev_warn(ctrl->device, "resetting controller\n");
2981                 return nvme_reset_ctrl_sync(ctrl);
2982         case NVME_IOCTL_SUBSYS_RESET:
2983                 return nvme_reset_subsystem(ctrl);
2984         case NVME_IOCTL_RESCAN:
2985                 nvme_queue_scan(ctrl);
2986                 return 0;
2987         default:
2988                 return -ENOTTY;
2989         }
2990 }
2991 
2992 static const struct file_operations nvme_dev_fops = {
2993         .owner          = THIS_MODULE,
2994         .open           = nvme_dev_open,
2995         .unlocked_ioctl = nvme_dev_ioctl,
2996         .compat_ioctl   = nvme_dev_ioctl,
2997 };
2998 
2999 static ssize_t nvme_sysfs_reset(struct device *dev,
3000                                 struct device_attribute *attr, const char *buf,
3001                                 size_t count)
3002 {
3003         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3004         int ret;
3005 
3006         ret = nvme_reset_ctrl_sync(ctrl);
3007         if (ret < 0)
3008                 return ret;
3009         return count;
3010 }
3011 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3012 
3013 static ssize_t nvme_sysfs_rescan(struct device *dev,
3014                                 struct device_attribute *attr, const char *buf,
3015                                 size_t count)
3016 {
3017         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3018 
3019         nvme_queue_scan(ctrl);
3020         return count;
3021 }
3022 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3023 
3024 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3025 {
3026         struct gendisk *disk = dev_to_disk(dev);
3027 
3028         if (disk->fops == &nvme_fops)
3029                 return nvme_get_ns_from_dev(dev)->head;
3030         else
3031                 return disk->private_data;
3032 }
3033 
3034 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3035                 char *buf)
3036 {
3037         struct nvme_ns_head *head = dev_to_ns_head(dev);
3038         struct nvme_ns_ids *ids = &head->ids;
3039         struct nvme_subsystem *subsys = head->subsys;
3040         int serial_len = sizeof(subsys->serial);
3041         int model_len = sizeof(subsys->model);
3042 
3043         if (!uuid_is_null(&ids->uuid))
3044                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3045 
3046         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3047                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3048 
3049         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3050                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3051 
3052         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3053                                   subsys->serial[serial_len - 1] == '\0'))
3054                 serial_len--;
3055         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3056                                  subsys->model[model_len - 1] == '\0'))
3057                 model_len--;
3058 
3059         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3060                 serial_len, subsys->serial, model_len, subsys->model,
3061                 head->ns_id);
3062 }
3063 static DEVICE_ATTR_RO(wwid);
3064 
3065 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3066                 char *buf)
3067 {
3068         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3069 }
3070 static DEVICE_ATTR_RO(nguid);
3071 
3072 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3073                 char *buf)
3074 {
3075         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3076 
3077         /* For backward compatibility expose the NGUID to userspace if
3078          * we have no UUID set
3079          */
3080         if (uuid_is_null(&ids->uuid)) {
3081                 printk_ratelimited(KERN_WARNING
3082                                    "No UUID available providing old NGUID\n");
3083                 return sprintf(buf, "%pU\n", ids->nguid);
3084         }
3085         return sprintf(buf, "%pU\n", &ids->uuid);
3086 }
3087 static DEVICE_ATTR_RO(uuid);
3088 
3089 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3090                 char *buf)
3091 {
3092         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3093 }
3094 static DEVICE_ATTR_RO(eui);
3095 
3096 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3097                 char *buf)
3098 {
3099         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3100 }
3101 static DEVICE_ATTR_RO(nsid);
3102 
3103 static struct attribute *nvme_ns_id_attrs[] = {
3104         &dev_attr_wwid.attr,
3105         &dev_attr_uuid.attr,
3106         &dev_attr_nguid.attr,
3107         &dev_attr_eui.attr,
3108         &dev_attr_nsid.attr,
3109 #ifdef CONFIG_NVME_MULTIPATH
3110         &dev_attr_ana_grpid.attr,
3111         &dev_attr_ana_state.attr,
3112 #endif
3113         NULL,
3114 };
3115 
3116 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3117                 struct attribute *a, int n)
3118 {
3119         struct device *dev = container_of(kobj, struct device, kobj);
3120         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3121 
3122         if (a == &dev_attr_uuid.attr) {
3123                 if (uuid_is_null(&ids->uuid) &&
3124                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3125                         return 0;
3126         }
3127         if (a == &dev_attr_nguid.attr) {
3128                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3129                         return 0;
3130         }
3131         if (a == &dev_attr_eui.attr) {
3132                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3133                         return 0;
3134         }
3135 #ifdef CONFIG_NVME_MULTIPATH
3136         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3137                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3138                         return 0;
3139                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3140                         return 0;
3141         }
3142 #endif
3143         return a->mode;
3144 }
3145 
3146 static const struct attribute_group nvme_ns_id_attr_group = {
3147         .attrs          = nvme_ns_id_attrs,
3148         .is_visible     = nvme_ns_id_attrs_are_visible,
3149 };
3150 
3151 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3152         &nvme_ns_id_attr_group,
3153 #ifdef CONFIG_NVM
3154         &nvme_nvm_attr_group,
3155 #endif
3156         NULL,
3157 };
3158 
3159 #define nvme_show_str_function(field)                                           \
3160 static ssize_t  field##_show(struct device *dev,                                \
3161                             struct device_attribute *attr, char *buf)           \
3162 {                                                                               \
3163         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3164         return sprintf(buf, "%.*s\n",                                           \
3165                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3166 }                                                                               \
3167 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3168 
3169 nvme_show_str_function(model);
3170 nvme_show_str_function(serial);
3171 nvme_show_str_function(firmware_rev);
3172 
3173 #define nvme_show_int_function(field)                                           \
3174 static ssize_t  field##_show(struct device *dev,                                \
3175                             struct device_attribute *attr, char *buf)           \
3176 {                                                                               \
3177         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3178         return sprintf(buf, "%d\n", ctrl->field);       \
3179 }                                                                               \
3180 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3181 
3182 nvme_show_int_function(cntlid);
3183 nvme_show_int_function(numa_node);
3184 nvme_show_int_function(queue_count);
3185 nvme_show_int_function(sqsize);
3186 
3187 static ssize_t nvme_sysfs_delete(struct device *dev,
3188                                 struct device_attribute *attr, const char *buf,
3189                                 size_t count)
3190 {
3191         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3192 
3193         if (device_remove_file_self(dev, attr))
3194                 nvme_delete_ctrl_sync(ctrl);
3195         return count;
3196 }
3197 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3198 
3199 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3200                                          struct device_attribute *attr,
3201                                          char *buf)
3202 {
3203         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3204 
3205         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3206 }
3207 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3208 
3209 static ssize_t nvme_sysfs_show_state(struct device *dev,
3210                                      struct device_attribute *attr,
3211                                      char *buf)
3212 {
3213         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3214         static const char *const state_name[] = {
3215                 [NVME_CTRL_NEW]         = "new",
3216                 [NVME_CTRL_LIVE]        = "live",
3217                 [NVME_CTRL_RESETTING]   = "resetting",
3218                 [NVME_CTRL_CONNECTING]  = "connecting",
3219                 [NVME_CTRL_DELETING]    = "deleting",
3220                 [NVME_CTRL_DEAD]        = "dead",
3221         };
3222 
3223         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3224             state_name[ctrl->state])
3225                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3226 
3227         return sprintf(buf, "unknown state\n");
3228 }
3229 
3230 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3231 
3232 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3233                                          struct device_attribute *attr,
3234                                          char *buf)
3235 {
3236         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3237 
3238         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3239 }
3240 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3241 
3242 static ssize_t nvme_sysfs_show_address(struct device *dev,
3243                                          struct device_attribute *attr,
3244                                          char *buf)
3245 {
3246         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3247 
3248         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3249 }
3250 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3251 
3252 static struct attribute *nvme_dev_attrs[] = {
3253         &dev_attr_reset_controller.attr,
3254         &dev_attr_rescan_controller.attr,
3255         &dev_attr_model.attr,
3256         &dev_attr_serial.attr,
3257         &dev_attr_firmware_rev.attr,
3258         &dev_attr_cntlid.attr,
3259         &dev_attr_delete_controller.attr,
3260         &dev_attr_transport.attr,
3261         &dev_attr_subsysnqn.attr,
3262         &dev_attr_address.attr,
3263         &dev_attr_state.attr,
3264         &dev_attr_numa_node.attr,
3265         &dev_attr_queue_count.attr,
3266         &dev_attr_sqsize.attr,
3267         NULL
3268 };
3269 
3270 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3271                 struct attribute *a, int n)
3272 {
3273         struct device *dev = container_of(kobj, struct device, kobj);
3274         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3275 
3276         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3277                 return 0;
3278         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3279                 return 0;
3280 
3281         return a->mode;
3282 }
3283 
3284 static struct attribute_group nvme_dev_attrs_group = {
3285         .attrs          = nvme_dev_attrs,
3286         .is_visible     = nvme_dev_attrs_are_visible,
3287 };
3288 
3289 static const struct attribute_group *nvme_dev_attr_groups[] = {
3290         &nvme_dev_attrs_group,
3291         NULL,
3292 };
3293 
3294 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3295                 unsigned nsid)
3296 {
3297         struct nvme_ns_head *h;
3298 
3299         lockdep_assert_held(&subsys->lock);
3300 
3301         list_for_each_entry(h, &subsys->nsheads, entry) {
3302                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3303                         return h;
3304         }
3305 
3306         return NULL;
3307 }
3308 
3309 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3310                 struct nvme_ns_head *new)
3311 {
3312         struct nvme_ns_head *h;
3313 
3314         lockdep_assert_held(&subsys->lock);
3315 
3316         list_for_each_entry(h, &subsys->nsheads, entry) {
3317                 if (nvme_ns_ids_valid(&new->ids) &&
3318                     !list_empty(&h->list) &&
3319                     nvme_ns_ids_equal(&new->ids, &h->ids))
3320                         return -EINVAL;
3321         }
3322 
3323         return 0;
3324 }
3325 
3326 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3327                 unsigned nsid, struct nvme_id_ns *id)
3328 {
3329         struct nvme_ns_head *head;
3330         size_t size = sizeof(*head);
3331         int ret = -ENOMEM;
3332 
3333 #ifdef CONFIG_NVME_MULTIPATH
3334         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3335 #endif
3336 
3337         head = kzalloc(size, GFP_KERNEL);
3338         if (!head)
3339                 goto out;
3340         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3341         if (ret < 0)
3342                 goto out_free_head;
3343         head->instance = ret;
3344         INIT_LIST_HEAD(&head->list);
3345         ret = init_srcu_struct(&head->srcu);
3346         if (ret)
3347                 goto out_ida_remove;
3348         head->subsys = ctrl->subsys;
3349         head->ns_id = nsid;
3350         kref_init(&head->ref);
3351 
3352         ret = nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3353         if (ret)
3354                 goto out_cleanup_srcu;
3355 
3356         ret = __nvme_check_ids(ctrl->subsys, head);
3357         if (ret) {
3358                 dev_err(ctrl->device,
3359                         "duplicate IDs for nsid %d\n", nsid);
3360                 goto out_cleanup_srcu;
3361         }
3362 
3363         ret = nvme_mpath_alloc_disk(ctrl, head);
3364         if (ret)
3365                 goto out_cleanup_srcu;
3366 
3367         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3368 
3369         kref_get(&ctrl->subsys->ref);
3370 
3371         return head;
3372 out_cleanup_srcu:
3373         cleanup_srcu_struct(&head->srcu);
3374 out_ida_remove:
3375         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3376 out_free_head:
3377         kfree(head);
3378 out:
3379         if (ret > 0)
3380                 ret = blk_status_to_errno(nvme_error_status(ret));
3381         return ERR_PTR(ret);
3382 }
3383 
3384 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3385                 struct nvme_id_ns *id)
3386 {
3387         struct nvme_ctrl *ctrl = ns->ctrl;
3388         bool is_shared = id->nmic & (1 << 0);
3389         struct nvme_ns_head *head = NULL;
3390         int ret = 0;
3391 
3392         mutex_lock(&ctrl->subsys->lock);
3393         if (is_shared)
3394                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3395         if (!head) {
3396                 head = nvme_alloc_ns_head(ctrl, nsid, id);
3397                 if (IS_ERR(head)) {
3398                         ret = PTR_ERR(head);
3399                         goto out_unlock;
3400                 }
3401         } else {
3402                 struct nvme_ns_ids ids;
3403 
3404                 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3405                 if (ret)
3406                         goto out_unlock;
3407 
3408                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3409                         dev_err(ctrl->device,
3410                                 "IDs don't match for shared namespace %d\n",
3411                                         nsid);
3412                         ret = -EINVAL;
3413                         goto out_unlock;
3414                 }
3415         }
3416 
3417         list_add_tail(&ns->siblings, &head->list);
3418         ns->head = head;
3419 
3420 out_unlock:
3421         mutex_unlock(&ctrl->subsys->lock);
3422         if (ret > 0)
3423                 ret = blk_status_to_errno(nvme_error_status(ret));
3424         return ret;
3425 }
3426 
3427 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3428 {
3429         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3430         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3431 
3432         return nsa->head->ns_id - nsb->head->ns_id;
3433 }
3434 
3435 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3436 {
3437         struct nvme_ns *ns, *ret = NULL;
3438 
3439         down_read(&ctrl->namespaces_rwsem);
3440         list_for_each_entry(ns, &ctrl->namespaces, list) {
3441                 if (ns->head->ns_id == nsid) {
3442                         if (!kref_get_unless_zero(&ns->kref))
3443                                 continue;
3444                         ret = ns;
3445                         break;
3446                 }
3447                 if (ns->head->ns_id > nsid)
3448                         break;
3449         }
3450         up_read(&ctrl->namespaces_rwsem);
3451         return ret;
3452 }
3453 
3454 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3455 {
3456         struct streams_directive_params s;
3457         int ret;
3458 
3459         if (!ctrl->nr_streams)
3460                 return 0;
3461 
3462         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3463         if (ret)
3464                 return ret;
3465 
3466         ns->sws = le32_to_cpu(s.sws);
3467         ns->sgs = le16_to_cpu(s.sgs);
3468 
3469         if (ns->sws) {
3470                 unsigned int bs = 1 << ns->lba_shift;
3471 
3472                 blk_queue_io_min(ns->queue, bs * ns->sws);
3473                 if (ns->sgs)
3474                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3475         }
3476 
3477         return 0;
3478 }
3479 
3480 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3481 {
3482         struct nvme_ns *ns;
3483         struct gendisk *disk;
3484         struct nvme_id_ns *id;
3485         char disk_name[DISK_NAME_LEN];
3486         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3487 
3488         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3489         if (!ns)
3490                 return -ENOMEM;
3491 
3492         ns->queue = blk_mq_init_queue(ctrl->tagset);
3493         if (IS_ERR(ns->queue)) {
3494                 ret = PTR_ERR(ns->queue);
3495                 goto out_free_ns;
3496         }
3497 
3498         if (ctrl->opts && ctrl->opts->data_digest)
3499                 ns->queue->backing_dev_info->capabilities
3500                         |= BDI_CAP_STABLE_WRITES;
3501 
3502         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3503         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3504                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3505 
3506         ns->queue->queuedata = ns;
3507         ns->ctrl = ctrl;
3508 
3509         kref_init(&ns->kref);
3510         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3511 
3512         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3513         nvme_set_queue_limits(ctrl, ns->queue);
3514 
3515         ret = nvme_identify_ns(ctrl, nsid, &id);
3516         if (ret)
3517                 goto out_free_queue;
3518 
3519         if (id->ncap == 0) {
3520                 ret = -EINVAL;
3521                 goto out_free_id;
3522         }
3523 
3524         ret = nvme_init_ns_head(ns, nsid, id);
3525         if (ret)
3526                 goto out_free_id;
3527         nvme_setup_streams_ns(ctrl, ns);
3528         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3529 
3530         disk = alloc_disk_node(0, node);
3531         if (!disk) {
3532                 ret = -ENOMEM;
3533                 goto out_unlink_ns;
3534         }
3535 
3536         disk->fops = &nvme_fops;
3537         disk->private_data = ns;
3538         disk->queue = ns->queue;
3539         disk->flags = flags;
3540         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3541         ns->disk = disk;
3542 
3543         __nvme_revalidate_disk(disk, id);
3544 
3545         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3546                 ret = nvme_nvm_register(ns, disk_name, node);
3547                 if (ret) {
3548                         dev_warn(ctrl->device, "LightNVM init failure\n");
3549                         goto out_put_disk;
3550                 }
3551         }
3552 
3553         down_write(&ctrl->namespaces_rwsem);
3554         list_add_tail(&ns->list, &ctrl->namespaces);
3555         up_write(&ctrl->namespaces_rwsem);
3556 
3557         nvme_get_ctrl(ctrl);
3558 
3559         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3560 
3561         nvme_mpath_add_disk(ns, id);
3562         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3563         kfree(id);
3564 
3565         return 0;
3566  out_put_disk:
3567         /* prevent double queue cleanup */
3568         ns->disk->queue = NULL;
3569         put_disk(ns->disk);
3570  out_unlink_ns:
3571         mutex_lock(&ctrl->subsys->lock);
3572         list_del_rcu(&ns->siblings);
3573         mutex_unlock(&ctrl->subsys->lock);
3574         nvme_put_ns_head(ns->head);
3575  out_free_id:
3576         kfree(id);
3577  out_free_queue:
3578         blk_cleanup_queue(ns->queue);
3579  out_free_ns:
3580         kfree(ns);
3581         if (ret > 0)
3582                 ret = blk_status_to_errno(nvme_error_status(ret));
3583         return ret;
3584 }
3585 
3586 static void nvme_ns_remove(struct nvme_ns *ns)
3587 {
3588         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3589                 return;
3590 
3591         nvme_fault_inject_fini(&ns->fault_inject);
3592 
3593         mutex_lock(&ns->ctrl->subsys->lock);
3594         list_del_rcu(&ns->siblings);
3595         mutex_unlock(&ns->ctrl->subsys->lock);
3596         synchronize_rcu(); /* guarantee not available in head->list */
3597         nvme_mpath_clear_current_path(ns);
3598         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3599 
3600         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3601                 del_gendisk(ns->disk);
3602                 blk_cleanup_queue(ns->queue);
3603                 if (blk_get_integrity(ns->disk))
3604                         blk_integrity_unregister(ns->disk);
3605         }
3606 
3607         down_write(&ns->ctrl->namespaces_rwsem);
3608         list_del_init(&ns->list);
3609         up_write(&ns->ctrl->namespaces_rwsem);
3610 
3611         nvme_mpath_check_last_path(ns);
3612         nvme_put_ns(ns);
3613 }
3614 
3615 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3616 {
3617         struct nvme_ns *ns;
3618 
3619         ns = nvme_find_get_ns(ctrl, nsid);
3620         if (ns) {
3621                 if (ns->disk && revalidate_disk(ns->disk))
3622                         nvme_ns_remove(ns);
3623                 nvme_put_ns(ns);
3624         } else
3625                 nvme_alloc_ns(ctrl, nsid);
3626 }
3627 
3628 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3629                                         unsigned nsid)
3630 {
3631         struct nvme_ns *ns, *next;
3632         LIST_HEAD(rm_list);
3633 
3634         down_write(&ctrl->namespaces_rwsem);
3635         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3636                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3637                         list_move_tail(&ns->list, &rm_list);
3638         }
3639         up_write(&ctrl->namespaces_rwsem);
3640 
3641         list_for_each_entry_safe(ns, next, &rm_list, list)
3642                 nvme_ns_remove(ns);
3643 
3644 }
3645 
3646 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3647 {
3648         struct nvme_ns *ns;
3649         __le32 *ns_list;
3650         unsigned i, j, nsid, prev = 0;
3651         unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3652         int ret = 0;
3653 
3654         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3655         if (!ns_list)
3656                 return -ENOMEM;
3657 
3658         for (i = 0; i < num_lists; i++) {
3659                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3660                 if (ret)
3661                         goto free;
3662 
3663                 for (j = 0; j < min(nn, 1024U); j++) {
3664                         nsid = le32_to_cpu(ns_list[j]);
3665                         if (!nsid)
3666                                 goto out;
3667 
3668                         nvme_validate_ns(ctrl, nsid);
3669 
3670                         while (++prev < nsid) {
3671                                 ns = nvme_find_get_ns(ctrl, prev);
3672                                 if (ns) {
3673                                         nvme_ns_remove(ns);
3674                                         nvme_put_ns(ns);
3675                                 }
3676                         }
3677                 }
3678                 nn -= j;
3679         }
3680  out:
3681         nvme_remove_invalid_namespaces(ctrl, prev);
3682  free:
3683         kfree(ns_list);
3684         return ret;
3685 }
3686 
3687 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3688 {
3689         unsigned i;
3690 
3691         for (i = 1; i <= nn; i++)
3692                 nvme_validate_ns(ctrl, i);
3693 
3694         nvme_remove_invalid_namespaces(ctrl, nn);
3695 }
3696 
3697 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3698 {
3699         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3700         __le32 *log;
3701         int error;
3702 
3703         log = kzalloc(log_size, GFP_KERNEL);
3704         if (!log)
3705                 return;
3706 
3707         /*
3708          * We need to read the log to clear the AEN, but we don't want to rely
3709          * on it for the changed namespace information as userspace could have
3710          * raced with us in reading the log page, which could cause us to miss
3711          * updates.
3712          */
3713         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3714                         log_size, 0);
3715         if (error)
3716                 dev_warn(ctrl->device,
3717                         "reading changed ns log failed: %d\n", error);
3718 
3719         kfree(log);
3720 }
3721 
3722 static void nvme_scan_work(struct work_struct *work)
3723 {
3724         struct nvme_ctrl *ctrl =
3725                 container_of(work, struct nvme_ctrl, scan_work);
3726         struct nvme_id_ctrl *id;
3727         unsigned nn;
3728 
3729         /* No tagset on a live ctrl means IO queues could not created */
3730         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3731                 return;
3732 
3733         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3734                 dev_info(ctrl->device, "rescanning namespaces.\n");
3735                 nvme_clear_changed_ns_log(ctrl);
3736         }
3737 
3738         if (nvme_identify_ctrl(ctrl, &id))
3739                 return;
3740 
3741         mutex_lock(&ctrl->scan_lock);
3742         nn = le32_to_cpu(id->nn);
3743         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3744             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3745                 if (!nvme_scan_ns_list(ctrl, nn))
3746                         goto out_free_id;
3747         }
3748         nvme_scan_ns_sequential(ctrl, nn);
3749 out_free_id:
3750         mutex_unlock(&ctrl->scan_lock);
3751         kfree(id);
3752         down_write(&ctrl->namespaces_rwsem);
3753         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3754         up_write(&ctrl->namespaces_rwsem);
3755 }
3756 
3757 /*
3758  * This function iterates the namespace list unlocked to allow recovery from
3759  * controller failure. It is up to the caller to ensure the namespace list is
3760  * not modified by scan work while this function is executing.
3761  */
3762 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3763 {
3764         struct nvme_ns *ns, *next;
3765         LIST_HEAD(ns_list);
3766 
3767         /*
3768          * make sure to requeue I/O to all namespaces as these
3769          * might result from the scan itself and must complete
3770          * for the scan_work to make progress
3771          */
3772         nvme_mpath_clear_ctrl_paths(ctrl);
3773 
3774         /* prevent racing with ns scanning */
3775         flush_work(&ctrl->scan_work);
3776 
3777         /*
3778          * The dead states indicates the controller was not gracefully
3779          * disconnected. In that case, we won't be able to flush any data while
3780          * removing the namespaces' disks; fail all the queues now to avoid
3781          * potentially having to clean up the failed sync later.
3782          */
3783         if (ctrl->state == NVME_CTRL_DEAD)
3784                 nvme_kill_queues(ctrl);
3785 
3786         down_write(&ctrl->namespaces_rwsem);
3787         list_splice_init(&ctrl->namespaces, &ns_list);
3788         up_write(&ctrl->namespaces_rwsem);
3789 
3790         list_for_each_entry_safe(ns, next, &ns_list, list)
3791                 nvme_ns_remove(ns);
3792 }
3793 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3794 
3795 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3796 {
3797         struct nvme_ctrl *ctrl =
3798                 container_of(dev, struct nvme_ctrl, ctrl_device);
3799         struct nvmf_ctrl_options *opts = ctrl->opts;
3800         int ret;
3801 
3802         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3803         if (ret)
3804                 return ret;
3805 
3806         if (opts) {
3807                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3808                 if (ret)
3809                         return ret;
3810 
3811                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3812                                 opts->trsvcid ?: "none");
3813                 if (ret)
3814                         return ret;
3815 
3816                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3817                                 opts->host_traddr ?: "none");
3818         }
3819         return ret;
3820 }
3821 
3822 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3823 {
3824         char *envp[2] = { NULL, NULL };
3825         u32 aen_result = ctrl->aen_result;
3826 
3827         ctrl->aen_result = 0;
3828         if (!aen_result)
3829                 return;
3830 
3831         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3832         if (!envp[0])
3833                 return;
3834         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3835         kfree(envp[0]);
3836 }
3837 
3838 static void nvme_async_event_work(struct work_struct *work)
3839 {
3840         struct nvme_ctrl *ctrl =
3841                 container_of(work, struct nvme_ctrl, async_event_work);
3842 
3843         nvme_aen_uevent(ctrl);
3844         ctrl->ops->submit_async_event(ctrl);
3845 }
3846 
3847 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3848 {
3849 
3850         u32 csts;
3851 
3852         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3853                 return false;
3854 
3855         if (csts == ~0)
3856                 return false;
3857 
3858         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3859 }
3860 
3861 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3862 {
3863         struct nvme_fw_slot_info_log *log;
3864 
3865         log = kmalloc(sizeof(*log), GFP_KERNEL);
3866         if (!log)
3867                 return;
3868 
3869         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, log,
3870                         sizeof(*log), 0))
3871                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3872         kfree(log);
3873 }
3874 
3875 static void nvme_fw_act_work(struct work_struct *work)
3876 {
3877         struct nvme_ctrl *ctrl = container_of(work,
3878                                 struct nvme_ctrl, fw_act_work);
3879         unsigned long fw_act_timeout;
3880 
3881         if (ctrl->mtfa)
3882                 fw_act_timeout = jiffies +
3883                                 msecs_to_jiffies(ctrl->mtfa * 100);
3884         else
3885                 fw_act_timeout = jiffies +
3886                                 msecs_to_jiffies(admin_timeout * 1000);
3887 
3888         nvme_stop_queues(ctrl);
3889         while (nvme_ctrl_pp_status(ctrl)) {
3890                 if (time_after(jiffies, fw_act_timeout)) {
3891                         dev_warn(ctrl->device,
3892                                 "Fw activation timeout, reset controller\n");
3893                         nvme_try_sched_reset(ctrl);
3894                         return;
3895                 }
3896                 msleep(100);
3897         }
3898 
3899         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
3900                 return;
3901 
3902         nvme_start_queues(ctrl);
3903         /* read FW slot information to clear the AER */
3904         nvme_get_fw_slot_info(ctrl);
3905 }
3906 
3907 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3908 {
3909         u32 aer_notice_type = (result & 0xff00) >> 8;
3910 
3911         trace_nvme_async_event(ctrl, aer_notice_type);
3912 
3913         switch (aer_notice_type) {
3914         case NVME_AER_NOTICE_NS_CHANGED:
3915                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3916                 nvme_queue_scan(ctrl);
3917                 break;
3918         case NVME_AER_NOTICE_FW_ACT_STARTING:
3919                 /*
3920                  * We are (ab)using the RESETTING state to prevent subsequent
3921                  * recovery actions from interfering with the controller's
3922                  * firmware activation.
3923                  */
3924                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
3925                         queue_work(nvme_wq, &ctrl->fw_act_work);
3926                 break;
3927 #ifdef CONFIG_NVME_MULTIPATH
3928         case NVME_AER_NOTICE_ANA:
3929                 if (!ctrl->ana_log_buf)
3930                         break;
3931                 queue_work(nvme_wq, &ctrl->ana_work);
3932                 break;
3933 #endif
3934         case NVME_AER_NOTICE_DISC_CHANGED:
3935                 ctrl->aen_result = result;
3936                 break;
3937         default:
3938                 dev_warn(ctrl->device, "async event result %08x\n", result);
3939         }
3940 }
3941 
3942 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3943                 volatile union nvme_result *res)
3944 {
3945         u32 result = le32_to_cpu(res->u32);
3946         u32 aer_type = result & 0x07;
3947 
3948         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3949                 return;
3950 
3951         switch (aer_type) {
3952         case NVME_AER_NOTICE:
3953                 nvme_handle_aen_notice(ctrl, result);
3954                 break;
3955         case NVME_AER_ERROR:
3956         case NVME_AER_SMART:
3957         case NVME_AER_CSS:
3958         case NVME_AER_VS:
3959                 trace_nvme_async_event(ctrl, aer_type);
3960                 ctrl->aen_result = result;
3961                 break;
3962         default:
3963                 break;
3964         }
3965         queue_work(nvme_wq, &ctrl->async_event_work);
3966 }
3967 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3968 
3969 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3970 {
3971         nvme_mpath_stop(ctrl);
3972         nvme_stop_keep_alive(ctrl);
3973         flush_work(&ctrl->async_event_work);
3974         cancel_work_sync(&ctrl->fw_act_work);
3975 }
3976 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3977 
3978 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3979 {
3980         if (ctrl->kato)
3981                 nvme_start_keep_alive(ctrl);
3982 
3983         nvme_enable_aen(ctrl);
3984 
3985         if (ctrl->queue_count > 1) {
3986                 nvme_queue_scan(ctrl);
3987                 nvme_start_queues(ctrl);
3988         }
3989 }
3990 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3991 
3992 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3993 {
3994         nvme_fault_inject_fini(&ctrl->fault_inject);
3995         dev_pm_qos_hide_latency_tolerance(ctrl->device);
3996         cdev_device_del(&ctrl->cdev, ctrl->device);
3997 }
3998 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3999 
4000 static void nvme_free_ctrl(struct device *dev)
4001 {
4002         struct nvme_ctrl *ctrl =
4003                 container_of(dev, struct nvme_ctrl, ctrl_device);
4004         struct nvme_subsystem *subsys = ctrl->subsys;
4005 
4006         if (subsys && ctrl->instance != subsys->instance)
4007                 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4008 
4009         kfree(ctrl->effects);
4010         nvme_mpath_uninit(ctrl);
4011         __free_page(ctrl->discard_page);
4012 
4013         if (subsys) {
4014                 mutex_lock(&nvme_subsystems_lock);
4015                 list_del(&ctrl->subsys_entry);
4016                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4017                 mutex_unlock(&nvme_subsystems_lock);
4018         }
4019 
4020         ctrl->ops->free_ctrl(ctrl);
4021 
4022         if (subsys)
4023                 nvme_put_subsystem(subsys);
4024 }
4025 
4026 /*
4027  * Initialize a NVMe controller structures.  This needs to be called during
4028  * earliest initialization so that we have the initialized structured around
4029  * during probing.
4030  */
4031 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4032                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4033 {
4034         int ret;
4035 
4036         ctrl->state = NVME_CTRL_NEW;
4037         spin_lock_init(&ctrl->lock);
4038         mutex_init(&ctrl->scan_lock);
4039         INIT_LIST_HEAD(&ctrl->namespaces);
4040         init_rwsem(&ctrl->namespaces_rwsem);
4041         ctrl->dev = dev;
4042         ctrl->ops = ops;
4043         ctrl->quirks = quirks;
4044         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4045         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4046         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4047         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4048         init_waitqueue_head(&ctrl->state_wq);
4049 
4050         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4051         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4052         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4053 
4054         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4055                         PAGE_SIZE);
4056         ctrl->discard_page = alloc_page(GFP_KERNEL);
4057         if (!ctrl->discard_page) {
4058                 ret = -ENOMEM;
4059                 goto out;
4060         }
4061 
4062         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4063         if (ret < 0)
4064                 goto out;
4065         ctrl->instance = ret;
4066 
4067         device_initialize(&ctrl->ctrl_device);
4068         ctrl->device = &ctrl->ctrl_device;
4069         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4070         ctrl->device->class = nvme_class;
4071         ctrl->device->parent = ctrl->dev;
4072         ctrl->device->groups = nvme_dev_attr_groups;
4073         ctrl->device->release = nvme_free_ctrl;
4074         dev_set_drvdata(ctrl->device, ctrl);
4075         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4076         if (ret)
4077                 goto out_release_instance;
4078 
4079         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4080         ctrl->cdev.owner = ops->module;
4081         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4082         if (ret)
4083                 goto out_free_name;
4084 
4085         /*
4086          * Initialize latency tolerance controls.  The sysfs files won't
4087          * be visible to userspace unless the device actually supports APST.
4088          */
4089         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4090         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4091                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4092 
4093         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4094 
4095         return 0;
4096 out_free_name:
4097         kfree_const(ctrl->device->kobj.name);
4098 out_release_instance:
4099         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4100 out:
4101         if (ctrl->discard_page)
4102                 __free_page(ctrl->discard_page);
4103         return ret;
4104 }
4105 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4106 
4107 /**
4108  * nvme_kill_queues(): Ends all namespace queues
4109  * @ctrl: the dead controller that needs to end
4110  *
4111  * Call this function when the driver determines it is unable to get the
4112  * controller in a state capable of servicing IO.
4113  */
4114 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4115 {
4116         struct nvme_ns *ns;
4117 
4118         down_read(&ctrl->namespaces_rwsem);
4119 
4120         /* Forcibly unquiesce queues to avoid blocking dispatch */
4121         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4122                 blk_mq_unquiesce_queue(ctrl->admin_q);
4123 
4124         list_for_each_entry(ns, &ctrl->namespaces, list)
4125                 nvme_set_queue_dying(ns);
4126 
4127         up_read(&ctrl->namespaces_rwsem);
4128 }
4129 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4130 
4131 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4132 {
4133         struct nvme_ns *ns;
4134 
4135         down_read(&ctrl->namespaces_rwsem);
4136         list_for_each_entry(ns, &ctrl->namespaces, list)
4137                 blk_mq_unfreeze_queue(ns->queue);
4138         up_read(&ctrl->namespaces_rwsem);
4139 }
4140 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4141 
4142 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4143 {
4144         struct nvme_ns *ns;
4145 
4146         down_read(&ctrl->namespaces_rwsem);
4147         list_for_each_entry(ns, &ctrl->namespaces, list) {
4148                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4149                 if (timeout <= 0)
4150                         break;
4151         }
4152         up_read(&ctrl->namespaces_rwsem);
4153 }
4154 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4155 
4156 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4157 {
4158         struct nvme_ns *ns;
4159 
4160         down_read(&ctrl->namespaces_rwsem);
4161         list_for_each_entry(ns, &ctrl->namespaces, list)
4162                 blk_mq_freeze_queue_wait(ns->queue);
4163         up_read(&ctrl->namespaces_rwsem);
4164 }
4165 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4166 
4167 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4168 {
4169         struct nvme_ns *ns;
4170 
4171         down_read(&ctrl->namespaces_rwsem);
4172         list_for_each_entry(ns, &ctrl->namespaces, list)
4173                 blk_freeze_queue_start(ns->queue);
4174         up_read(&ctrl->namespaces_rwsem);
4175 }
4176 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4177 
4178 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4179 {
4180         struct nvme_ns *ns;
4181 
4182         down_read(&ctrl->namespaces_rwsem);
4183         list_for_each_entry(ns, &ctrl->namespaces, list)
4184                 blk_mq_quiesce_queue(ns->queue);
4185         up_read(&ctrl->namespaces_rwsem);
4186 }
4187 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4188 
4189 void nvme_start_queues(struct nvme_ctrl *ctrl)
4190 {
4191         struct nvme_ns *ns;
4192 
4193         down_read(&ctrl->namespaces_rwsem);
4194         list_for_each_entry(ns, &ctrl->namespaces, list)
4195                 blk_mq_unquiesce_queue(ns->queue);
4196         up_read(&ctrl->namespaces_rwsem);
4197 }
4198 EXPORT_SYMBOL_GPL(nvme_start_queues);
4199 
4200 
4201 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4202 {
4203         struct nvme_ns *ns;
4204 
4205         down_read(&ctrl->namespaces_rwsem);
4206         list_for_each_entry(ns, &ctrl->namespaces, list)
4207                 blk_sync_queue(ns->queue);
4208         up_read(&ctrl->namespaces_rwsem);
4209 
4210         if (ctrl->admin_q)
4211                 blk_sync_queue(ctrl->admin_q);
4212 }
4213 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4214 
4215 /*
4216  * Check we didn't inadvertently grow the command structure sizes:
4217  */
4218 static inline void _nvme_check_size(void)
4219 {
4220         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4221         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4222         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4223         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4224         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4225         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4226         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4227         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4228         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4229         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4230         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4231         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4232         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4233         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4234         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4235         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4236         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4237 }
4238 
4239 
4240 static int __init nvme_core_init(void)
4241 {
4242         int result = -ENOMEM;
4243 
4244         _nvme_check_size();
4245 
4246         nvme_wq = alloc_workqueue("nvme-wq",
4247                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4248         if (!nvme_wq)
4249                 goto out;
4250 
4251         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4252                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4253         if (!nvme_reset_wq)
4254                 goto destroy_wq;
4255 
4256         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4257                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4258         if (!nvme_delete_wq)
4259                 goto destroy_reset_wq;
4260 
4261         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4262         if (result < 0)
4263                 goto destroy_delete_wq;
4264 
4265         nvme_class = class_create(THIS_MODULE, "nvme");
4266         if (IS_ERR(nvme_class)) {
4267                 result = PTR_ERR(nvme_class);
4268                 goto unregister_chrdev;
4269         }
4270         nvme_class->dev_uevent = nvme_class_uevent;
4271 
4272         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4273         if (IS_ERR(nvme_subsys_class)) {
4274                 result = PTR_ERR(nvme_subsys_class);
4275                 goto destroy_class;
4276         }
4277         return 0;
4278 
4279 destroy_class:
4280         class_destroy(nvme_class);
4281 unregister_chrdev:
4282         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4283 destroy_delete_wq:
4284         destroy_workqueue(nvme_delete_wq);
4285 destroy_reset_wq:
4286         destroy_workqueue(nvme_reset_wq);
4287 destroy_wq:
4288         destroy_workqueue(nvme_wq);
4289 out:
4290         return result;
4291 }
4292 
4293 static void __exit nvme_core_exit(void)
4294 {
4295         class_destroy(nvme_subsys_class);
4296         class_destroy(nvme_class);
4297         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4298         destroy_workqueue(nvme_delete_wq);
4299         destroy_workqueue(nvme_reset_wq);
4300         destroy_workqueue(nvme_wq);
4301 }
4302 
4303 MODULE_LICENSE("GPL");
4304 MODULE_VERSION("1.0");
4305 module_init(nvme_core_init);
4306 module_exit(nvme_core_exit);

/* [<][>][^][v][top][bottom][index][help] */