1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2009, Microsoft Corporation. 4 * 5 * Authors: 6 * Haiyang Zhang <haiyangz@microsoft.com> 7 * Hank Janssen <hjanssen@microsoft.com> 8 * K. Y. Srinivasan <kys@microsoft.com> 9 */ 10 11 #include <linux/kernel.h> 12 #include <linux/wait.h> 13 #include <linux/sched.h> 14 #include <linux/completion.h> 15 #include <linux/string.h> 16 #include <linux/mm.h> 17 #include <linux/delay.h> 18 #include <linux/init.h> 19 #include <linux/slab.h> 20 #include <linux/module.h> 21 #include <linux/device.h> 22 #include <linux/hyperv.h> 23 #include <linux/blkdev.h> 24 #include <scsi/scsi.h> 25 #include <scsi/scsi_cmnd.h> 26 #include <scsi/scsi_host.h> 27 #include <scsi/scsi_device.h> 28 #include <scsi/scsi_tcq.h> 29 #include <scsi/scsi_eh.h> 30 #include <scsi/scsi_devinfo.h> 31 #include <scsi/scsi_dbg.h> 32 #include <scsi/scsi_transport_fc.h> 33 #include <scsi/scsi_transport.h> 34 35 /* 36 * All wire protocol details (storage protocol between the guest and the host) 37 * are consolidated here. 38 * 39 * Begin protocol definitions. 40 */ 41 42 /* 43 * Version history: 44 * V1 Beta: 0.1 45 * V1 RC < 2008/1/31: 1.0 46 * V1 RC > 2008/1/31: 2.0 47 * Win7: 4.2 48 * Win8: 5.1 49 * Win8.1: 6.0 50 * Win10: 6.2 51 */ 52 53 #define VMSTOR_PROTO_VERSION(MAJOR_, MINOR_) ((((MAJOR_) & 0xff) << 8) | \ 54 (((MINOR_) & 0xff))) 55 56 #define VMSTOR_PROTO_VERSION_WIN6 VMSTOR_PROTO_VERSION(2, 0) 57 #define VMSTOR_PROTO_VERSION_WIN7 VMSTOR_PROTO_VERSION(4, 2) 58 #define VMSTOR_PROTO_VERSION_WIN8 VMSTOR_PROTO_VERSION(5, 1) 59 #define VMSTOR_PROTO_VERSION_WIN8_1 VMSTOR_PROTO_VERSION(6, 0) 60 #define VMSTOR_PROTO_VERSION_WIN10 VMSTOR_PROTO_VERSION(6, 2) 61 62 /* Packet structure describing virtual storage requests. */ 63 enum vstor_packet_operation { 64 VSTOR_OPERATION_COMPLETE_IO = 1, 65 VSTOR_OPERATION_REMOVE_DEVICE = 2, 66 VSTOR_OPERATION_EXECUTE_SRB = 3, 67 VSTOR_OPERATION_RESET_LUN = 4, 68 VSTOR_OPERATION_RESET_ADAPTER = 5, 69 VSTOR_OPERATION_RESET_BUS = 6, 70 VSTOR_OPERATION_BEGIN_INITIALIZATION = 7, 71 VSTOR_OPERATION_END_INITIALIZATION = 8, 72 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION = 9, 73 VSTOR_OPERATION_QUERY_PROPERTIES = 10, 74 VSTOR_OPERATION_ENUMERATE_BUS = 11, 75 VSTOR_OPERATION_FCHBA_DATA = 12, 76 VSTOR_OPERATION_CREATE_SUB_CHANNELS = 13, 77 VSTOR_OPERATION_MAXIMUM = 13 78 }; 79 80 /* 81 * WWN packet for Fibre Channel HBA 82 */ 83 84 struct hv_fc_wwn_packet { 85 u8 primary_active; 86 u8 reserved1[3]; 87 u8 primary_port_wwn[8]; 88 u8 primary_node_wwn[8]; 89 u8 secondary_port_wwn[8]; 90 u8 secondary_node_wwn[8]; 91 }; 92 93 94 95 /* 96 * SRB Flag Bits 97 */ 98 99 #define SRB_FLAGS_QUEUE_ACTION_ENABLE 0x00000002 100 #define SRB_FLAGS_DISABLE_DISCONNECT 0x00000004 101 #define SRB_FLAGS_DISABLE_SYNCH_TRANSFER 0x00000008 102 #define SRB_FLAGS_BYPASS_FROZEN_QUEUE 0x00000010 103 #define SRB_FLAGS_DISABLE_AUTOSENSE 0x00000020 104 #define SRB_FLAGS_DATA_IN 0x00000040 105 #define SRB_FLAGS_DATA_OUT 0x00000080 106 #define SRB_FLAGS_NO_DATA_TRANSFER 0x00000000 107 #define SRB_FLAGS_UNSPECIFIED_DIRECTION (SRB_FLAGS_DATA_IN | SRB_FLAGS_DATA_OUT) 108 #define SRB_FLAGS_NO_QUEUE_FREEZE 0x00000100 109 #define SRB_FLAGS_ADAPTER_CACHE_ENABLE 0x00000200 110 #define SRB_FLAGS_FREE_SENSE_BUFFER 0x00000400 111 112 /* 113 * This flag indicates the request is part of the workflow for processing a D3. 114 */ 115 #define SRB_FLAGS_D3_PROCESSING 0x00000800 116 #define SRB_FLAGS_IS_ACTIVE 0x00010000 117 #define SRB_FLAGS_ALLOCATED_FROM_ZONE 0x00020000 118 #define SRB_FLAGS_SGLIST_FROM_POOL 0x00040000 119 #define SRB_FLAGS_BYPASS_LOCKED_QUEUE 0x00080000 120 #define SRB_FLAGS_NO_KEEP_AWAKE 0x00100000 121 #define SRB_FLAGS_PORT_DRIVER_ALLOCSENSE 0x00200000 122 #define SRB_FLAGS_PORT_DRIVER_SENSEHASPORT 0x00400000 123 #define SRB_FLAGS_DONT_START_NEXT_PACKET 0x00800000 124 #define SRB_FLAGS_PORT_DRIVER_RESERVED 0x0F000000 125 #define SRB_FLAGS_CLASS_DRIVER_RESERVED 0xF0000000 126 127 #define SP_UNTAGGED ((unsigned char) ~0) 128 #define SRB_SIMPLE_TAG_REQUEST 0x20 129 130 /* 131 * Platform neutral description of a scsi request - 132 * this remains the same across the write regardless of 32/64 bit 133 * note: it's patterned off the SCSI_PASS_THROUGH structure 134 */ 135 #define STORVSC_MAX_CMD_LEN 0x10 136 137 #define POST_WIN7_STORVSC_SENSE_BUFFER_SIZE 0x14 138 #define PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE 0x12 139 140 #define STORVSC_SENSE_BUFFER_SIZE 0x14 141 #define STORVSC_MAX_BUF_LEN_WITH_PADDING 0x14 142 143 /* 144 * Sense buffer size changed in win8; have a run-time 145 * variable to track the size we should use. This value will 146 * likely change during protocol negotiation but it is valid 147 * to start by assuming pre-Win8. 148 */ 149 static int sense_buffer_size = PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE; 150 151 /* 152 * The storage protocol version is determined during the 153 * initial exchange with the host. It will indicate which 154 * storage functionality is available in the host. 155 */ 156 static int vmstor_proto_version; 157 158 #define STORVSC_LOGGING_NONE 0 159 #define STORVSC_LOGGING_ERROR 1 160 #define STORVSC_LOGGING_WARN 2 161 162 static int logging_level = STORVSC_LOGGING_ERROR; 163 module_param(logging_level, int, S_IRUGO|S_IWUSR); 164 MODULE_PARM_DESC(logging_level, 165 "Logging level, 0 - None, 1 - Error (default), 2 - Warning."); 166 167 static inline bool do_logging(int level) 168 { 169 return logging_level >= level; 170 } 171 172 #define storvsc_log(dev, level, fmt, ...) \ 173 do { \ 174 if (do_logging(level)) \ 175 dev_warn(&(dev)->device, fmt, ##__VA_ARGS__); \ 176 } while (0) 177 178 struct vmscsi_win8_extension { 179 /* 180 * The following were added in Windows 8 181 */ 182 u16 reserve; 183 u8 queue_tag; 184 u8 queue_action; 185 u32 srb_flags; 186 u32 time_out_value; 187 u32 queue_sort_ey; 188 } __packed; 189 190 struct vmscsi_request { 191 u16 length; 192 u8 srb_status; 193 u8 scsi_status; 194 195 u8 port_number; 196 u8 path_id; 197 u8 target_id; 198 u8 lun; 199 200 u8 cdb_length; 201 u8 sense_info_length; 202 u8 data_in; 203 u8 reserved; 204 205 u32 data_transfer_length; 206 207 union { 208 u8 cdb[STORVSC_MAX_CMD_LEN]; 209 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE]; 210 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING]; 211 }; 212 /* 213 * The following was added in win8. 214 */ 215 struct vmscsi_win8_extension win8_extension; 216 217 } __attribute((packed)); 218 219 220 /* 221 * The size of the vmscsi_request has changed in win8. The 222 * additional size is because of new elements added to the 223 * structure. These elements are valid only when we are talking 224 * to a win8 host. 225 * Track the correction to size we need to apply. This value 226 * will likely change during protocol negotiation but it is 227 * valid to start by assuming pre-Win8. 228 */ 229 static int vmscsi_size_delta = sizeof(struct vmscsi_win8_extension); 230 231 /* 232 * The list of storage protocols in order of preference. 233 */ 234 struct vmstor_protocol { 235 int protocol_version; 236 int sense_buffer_size; 237 int vmscsi_size_delta; 238 }; 239 240 241 static const struct vmstor_protocol vmstor_protocols[] = { 242 { 243 VMSTOR_PROTO_VERSION_WIN10, 244 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE, 245 0 246 }, 247 { 248 VMSTOR_PROTO_VERSION_WIN8_1, 249 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE, 250 0 251 }, 252 { 253 VMSTOR_PROTO_VERSION_WIN8, 254 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE, 255 0 256 }, 257 { 258 VMSTOR_PROTO_VERSION_WIN7, 259 PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE, 260 sizeof(struct vmscsi_win8_extension), 261 }, 262 { 263 VMSTOR_PROTO_VERSION_WIN6, 264 PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE, 265 sizeof(struct vmscsi_win8_extension), 266 } 267 }; 268 269 270 /* 271 * This structure is sent during the initialization phase to get the different 272 * properties of the channel. 273 */ 274 275 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL 0x1 276 277 struct vmstorage_channel_properties { 278 u32 reserved; 279 u16 max_channel_cnt; 280 u16 reserved1; 281 282 u32 flags; 283 u32 max_transfer_bytes; 284 285 u64 reserved2; 286 } __packed; 287 288 /* This structure is sent during the storage protocol negotiations. */ 289 struct vmstorage_protocol_version { 290 /* Major (MSW) and minor (LSW) version numbers. */ 291 u16 major_minor; 292 293 /* 294 * Revision number is auto-incremented whenever this file is changed 295 * (See FILL_VMSTOR_REVISION macro above). Mismatch does not 296 * definitely indicate incompatibility--but it does indicate mismatched 297 * builds. 298 * This is only used on the windows side. Just set it to 0. 299 */ 300 u16 revision; 301 } __packed; 302 303 /* Channel Property Flags */ 304 #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1 305 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2 306 307 struct vstor_packet { 308 /* Requested operation type */ 309 enum vstor_packet_operation operation; 310 311 /* Flags - see below for values */ 312 u32 flags; 313 314 /* Status of the request returned from the server side. */ 315 u32 status; 316 317 /* Data payload area */ 318 union { 319 /* 320 * Structure used to forward SCSI commands from the 321 * client to the server. 322 */ 323 struct vmscsi_request vm_srb; 324 325 /* Structure used to query channel properties. */ 326 struct vmstorage_channel_properties storage_channel_properties; 327 328 /* Used during version negotiations. */ 329 struct vmstorage_protocol_version version; 330 331 /* Fibre channel address packet */ 332 struct hv_fc_wwn_packet wwn_packet; 333 334 /* Number of sub-channels to create */ 335 u16 sub_channel_count; 336 337 /* This will be the maximum of the union members */ 338 u8 buffer[0x34]; 339 }; 340 } __packed; 341 342 /* 343 * Packet Flags: 344 * 345 * This flag indicates that the server should send back a completion for this 346 * packet. 347 */ 348 349 #define REQUEST_COMPLETION_FLAG 0x1 350 351 /* Matches Windows-end */ 352 enum storvsc_request_type { 353 WRITE_TYPE = 0, 354 READ_TYPE, 355 UNKNOWN_TYPE, 356 }; 357 358 /* 359 * SRB status codes and masks; a subset of the codes used here. 360 */ 361 362 #define SRB_STATUS_AUTOSENSE_VALID 0x80 363 #define SRB_STATUS_QUEUE_FROZEN 0x40 364 #define SRB_STATUS_INVALID_LUN 0x20 365 #define SRB_STATUS_SUCCESS 0x01 366 #define SRB_STATUS_ABORTED 0x02 367 #define SRB_STATUS_ERROR 0x04 368 #define SRB_STATUS_DATA_OVERRUN 0x12 369 370 #define SRB_STATUS(status) \ 371 (status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN)) 372 /* 373 * This is the end of Protocol specific defines. 374 */ 375 376 static int storvsc_ringbuffer_size = (128 * 1024); 377 static u32 max_outstanding_req_per_channel; 378 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth); 379 380 static int storvsc_vcpus_per_sub_channel = 4; 381 382 module_param(storvsc_ringbuffer_size, int, S_IRUGO); 383 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)"); 384 385 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO); 386 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels"); 387 388 static int ring_avail_percent_lowater = 10; 389 module_param(ring_avail_percent_lowater, int, S_IRUGO); 390 MODULE_PARM_DESC(ring_avail_percent_lowater, 391 "Select a channel if available ring size > this in percent"); 392 393 /* 394 * Timeout in seconds for all devices managed by this driver. 395 */ 396 static int storvsc_timeout = 180; 397 398 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 399 static struct scsi_transport_template *fc_transport_template; 400 #endif 401 402 static void storvsc_on_channel_callback(void *context); 403 404 #define STORVSC_MAX_LUNS_PER_TARGET 255 405 #define STORVSC_MAX_TARGETS 2 406 #define STORVSC_MAX_CHANNELS 8 407 408 #define STORVSC_FC_MAX_LUNS_PER_TARGET 255 409 #define STORVSC_FC_MAX_TARGETS 128 410 #define STORVSC_FC_MAX_CHANNELS 8 411 412 #define STORVSC_IDE_MAX_LUNS_PER_TARGET 64 413 #define STORVSC_IDE_MAX_TARGETS 1 414 #define STORVSC_IDE_MAX_CHANNELS 1 415 416 struct storvsc_cmd_request { 417 struct scsi_cmnd *cmd; 418 419 struct hv_device *device; 420 421 /* Synchronize the request/response if needed */ 422 struct completion wait_event; 423 424 struct vmbus_channel_packet_multipage_buffer mpb; 425 struct vmbus_packet_mpb_array *payload; 426 u32 payload_sz; 427 428 struct vstor_packet vstor_packet; 429 }; 430 431 432 /* A storvsc device is a device object that contains a vmbus channel */ 433 struct storvsc_device { 434 struct hv_device *device; 435 436 bool destroy; 437 bool drain_notify; 438 atomic_t num_outstanding_req; 439 struct Scsi_Host *host; 440 441 wait_queue_head_t waiting_to_drain; 442 443 /* 444 * Each unique Port/Path/Target represents 1 channel ie scsi 445 * controller. In reality, the pathid, targetid is always 0 446 * and the port is set by us 447 */ 448 unsigned int port_number; 449 unsigned char path_id; 450 unsigned char target_id; 451 452 /* 453 * Max I/O, the device can support. 454 */ 455 u32 max_transfer_bytes; 456 /* 457 * Number of sub-channels we will open. 458 */ 459 u16 num_sc; 460 struct vmbus_channel **stor_chns; 461 /* 462 * Mask of CPUs bound to subchannels. 463 */ 464 struct cpumask alloced_cpus; 465 /* Used for vsc/vsp channel reset process */ 466 struct storvsc_cmd_request init_request; 467 struct storvsc_cmd_request reset_request; 468 /* 469 * Currently active port and node names for FC devices. 470 */ 471 u64 node_name; 472 u64 port_name; 473 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 474 struct fc_rport *rport; 475 #endif 476 }; 477 478 struct hv_host_device { 479 struct hv_device *dev; 480 unsigned int port; 481 unsigned char path; 482 unsigned char target; 483 struct workqueue_struct *handle_error_wq; 484 struct work_struct host_scan_work; 485 struct Scsi_Host *host; 486 }; 487 488 struct storvsc_scan_work { 489 struct work_struct work; 490 struct Scsi_Host *host; 491 u8 lun; 492 u8 tgt_id; 493 }; 494 495 static void storvsc_device_scan(struct work_struct *work) 496 { 497 struct storvsc_scan_work *wrk; 498 struct scsi_device *sdev; 499 500 wrk = container_of(work, struct storvsc_scan_work, work); 501 502 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun); 503 if (!sdev) 504 goto done; 505 scsi_rescan_device(&sdev->sdev_gendev); 506 scsi_device_put(sdev); 507 508 done: 509 kfree(wrk); 510 } 511 512 static void storvsc_host_scan(struct work_struct *work) 513 { 514 struct Scsi_Host *host; 515 struct scsi_device *sdev; 516 struct hv_host_device *host_device = 517 container_of(work, struct hv_host_device, host_scan_work); 518 519 host = host_device->host; 520 /* 521 * Before scanning the host, first check to see if any of the 522 * currrently known devices have been hot removed. We issue a 523 * "unit ready" command against all currently known devices. 524 * This I/O will result in an error for devices that have been 525 * removed. As part of handling the I/O error, we remove the device. 526 * 527 * When a LUN is added or removed, the host sends us a signal to 528 * scan the host. Thus we are forced to discover the LUNs that 529 * may have been removed this way. 530 */ 531 mutex_lock(&host->scan_mutex); 532 shost_for_each_device(sdev, host) 533 scsi_test_unit_ready(sdev, 1, 1, NULL); 534 mutex_unlock(&host->scan_mutex); 535 /* 536 * Now scan the host to discover LUNs that may have been added. 537 */ 538 scsi_scan_host(host); 539 } 540 541 static void storvsc_remove_lun(struct work_struct *work) 542 { 543 struct storvsc_scan_work *wrk; 544 struct scsi_device *sdev; 545 546 wrk = container_of(work, struct storvsc_scan_work, work); 547 if (!scsi_host_get(wrk->host)) 548 goto done; 549 550 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun); 551 552 if (sdev) { 553 scsi_remove_device(sdev); 554 scsi_device_put(sdev); 555 } 556 scsi_host_put(wrk->host); 557 558 done: 559 kfree(wrk); 560 } 561 562 563 /* 564 * We can get incoming messages from the host that are not in response to 565 * messages that we have sent out. An example of this would be messages 566 * received by the guest to notify dynamic addition/removal of LUNs. To 567 * deal with potential race conditions where the driver may be in the 568 * midst of being unloaded when we might receive an unsolicited message 569 * from the host, we have implemented a mechanism to gurantee sequential 570 * consistency: 571 * 572 * 1) Once the device is marked as being destroyed, we will fail all 573 * outgoing messages. 574 * 2) We permit incoming messages when the device is being destroyed, 575 * only to properly account for messages already sent out. 576 */ 577 578 static inline struct storvsc_device *get_out_stor_device( 579 struct hv_device *device) 580 { 581 struct storvsc_device *stor_device; 582 583 stor_device = hv_get_drvdata(device); 584 585 if (stor_device && stor_device->destroy) 586 stor_device = NULL; 587 588 return stor_device; 589 } 590 591 592 static inline void storvsc_wait_to_drain(struct storvsc_device *dev) 593 { 594 dev->drain_notify = true; 595 wait_event(dev->waiting_to_drain, 596 atomic_read(&dev->num_outstanding_req) == 0); 597 dev->drain_notify = false; 598 } 599 600 static inline struct storvsc_device *get_in_stor_device( 601 struct hv_device *device) 602 { 603 struct storvsc_device *stor_device; 604 605 stor_device = hv_get_drvdata(device); 606 607 if (!stor_device) 608 goto get_in_err; 609 610 /* 611 * If the device is being destroyed; allow incoming 612 * traffic only to cleanup outstanding requests. 613 */ 614 615 if (stor_device->destroy && 616 (atomic_read(&stor_device->num_outstanding_req) == 0)) 617 stor_device = NULL; 618 619 get_in_err: 620 return stor_device; 621 622 } 623 624 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old, 625 u32 new) 626 { 627 struct storvsc_device *stor_device; 628 struct vmbus_channel *cur_chn; 629 bool old_is_alloced = false; 630 struct hv_device *device; 631 unsigned long flags; 632 int cpu; 633 634 device = channel->primary_channel ? 635 channel->primary_channel->device_obj 636 : channel->device_obj; 637 stor_device = get_out_stor_device(device); 638 if (!stor_device) 639 return; 640 641 /* See storvsc_do_io() -> get_og_chn(). */ 642 spin_lock_irqsave(&device->channel->lock, flags); 643 644 /* 645 * Determines if the storvsc device has other channels assigned to 646 * the "old" CPU to update the alloced_cpus mask and the stor_chns 647 * array. 648 */ 649 if (device->channel != channel && device->channel->target_cpu == old) { 650 cur_chn = device->channel; 651 old_is_alloced = true; 652 goto old_is_alloced; 653 } 654 list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) { 655 if (cur_chn == channel) 656 continue; 657 if (cur_chn->target_cpu == old) { 658 old_is_alloced = true; 659 goto old_is_alloced; 660 } 661 } 662 663 old_is_alloced: 664 if (old_is_alloced) 665 WRITE_ONCE(stor_device->stor_chns[old], cur_chn); 666 else 667 cpumask_clear_cpu(old, &stor_device->alloced_cpus); 668 669 /* "Flush" the stor_chns array. */ 670 for_each_possible_cpu(cpu) { 671 if (stor_device->stor_chns[cpu] && !cpumask_test_cpu( 672 cpu, &stor_device->alloced_cpus)) 673 WRITE_ONCE(stor_device->stor_chns[cpu], NULL); 674 } 675 676 WRITE_ONCE(stor_device->stor_chns[new], channel); 677 cpumask_set_cpu(new, &stor_device->alloced_cpus); 678 679 spin_unlock_irqrestore(&device->channel->lock, flags); 680 } 681 682 static void handle_sc_creation(struct vmbus_channel *new_sc) 683 { 684 struct hv_device *device = new_sc->primary_channel->device_obj; 685 struct device *dev = &device->device; 686 struct storvsc_device *stor_device; 687 struct vmstorage_channel_properties props; 688 int ret; 689 690 stor_device = get_out_stor_device(device); 691 if (!stor_device) 692 return; 693 694 memset(&props, 0, sizeof(struct vmstorage_channel_properties)); 695 696 ret = vmbus_open(new_sc, 697 storvsc_ringbuffer_size, 698 storvsc_ringbuffer_size, 699 (void *)&props, 700 sizeof(struct vmstorage_channel_properties), 701 storvsc_on_channel_callback, new_sc); 702 703 /* In case vmbus_open() fails, we don't use the sub-channel. */ 704 if (ret != 0) { 705 dev_err(dev, "Failed to open sub-channel: err=%d\n", ret); 706 return; 707 } 708 709 new_sc->change_target_cpu_callback = storvsc_change_target_cpu; 710 711 /* Add the sub-channel to the array of available channels. */ 712 stor_device->stor_chns[new_sc->target_cpu] = new_sc; 713 cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus); 714 } 715 716 static void handle_multichannel_storage(struct hv_device *device, int max_chns) 717 { 718 struct device *dev = &device->device; 719 struct storvsc_device *stor_device; 720 int num_sc; 721 struct storvsc_cmd_request *request; 722 struct vstor_packet *vstor_packet; 723 int ret, t; 724 725 /* 726 * If the number of CPUs is artificially restricted, such as 727 * with maxcpus=1 on the kernel boot line, Hyper-V could offer 728 * sub-channels >= the number of CPUs. These sub-channels 729 * should not be created. The primary channel is already created 730 * and assigned to one CPU, so check against # CPUs - 1. 731 */ 732 num_sc = min((int)(num_online_cpus() - 1), max_chns); 733 if (!num_sc) 734 return; 735 736 stor_device = get_out_stor_device(device); 737 if (!stor_device) 738 return; 739 740 stor_device->num_sc = num_sc; 741 request = &stor_device->init_request; 742 vstor_packet = &request->vstor_packet; 743 744 /* 745 * Establish a handler for dealing with subchannels. 746 */ 747 vmbus_set_sc_create_callback(device->channel, handle_sc_creation); 748 749 /* 750 * Request the host to create sub-channels. 751 */ 752 memset(request, 0, sizeof(struct storvsc_cmd_request)); 753 init_completion(&request->wait_event); 754 vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS; 755 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 756 vstor_packet->sub_channel_count = num_sc; 757 758 ret = vmbus_sendpacket(device->channel, vstor_packet, 759 (sizeof(struct vstor_packet) - 760 vmscsi_size_delta), 761 (unsigned long)request, 762 VM_PKT_DATA_INBAND, 763 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 764 765 if (ret != 0) { 766 dev_err(dev, "Failed to create sub-channel: err=%d\n", ret); 767 return; 768 } 769 770 t = wait_for_completion_timeout(&request->wait_event, 10*HZ); 771 if (t == 0) { 772 dev_err(dev, "Failed to create sub-channel: timed out\n"); 773 return; 774 } 775 776 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 777 vstor_packet->status != 0) { 778 dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n", 779 vstor_packet->operation, vstor_packet->status); 780 return; 781 } 782 783 /* 784 * We need to do nothing here, because vmbus_process_offer() 785 * invokes channel->sc_creation_callback, which will open and use 786 * the sub-channel(s). 787 */ 788 } 789 790 static void cache_wwn(struct storvsc_device *stor_device, 791 struct vstor_packet *vstor_packet) 792 { 793 /* 794 * Cache the currently active port and node ww names. 795 */ 796 if (vstor_packet->wwn_packet.primary_active) { 797 stor_device->node_name = 798 wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn); 799 stor_device->port_name = 800 wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn); 801 } else { 802 stor_device->node_name = 803 wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn); 804 stor_device->port_name = 805 wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn); 806 } 807 } 808 809 810 static int storvsc_execute_vstor_op(struct hv_device *device, 811 struct storvsc_cmd_request *request, 812 bool status_check) 813 { 814 struct vstor_packet *vstor_packet; 815 int ret, t; 816 817 vstor_packet = &request->vstor_packet; 818 819 init_completion(&request->wait_event); 820 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 821 822 ret = vmbus_sendpacket(device->channel, vstor_packet, 823 (sizeof(struct vstor_packet) - 824 vmscsi_size_delta), 825 (unsigned long)request, 826 VM_PKT_DATA_INBAND, 827 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 828 if (ret != 0) 829 return ret; 830 831 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 832 if (t == 0) 833 return -ETIMEDOUT; 834 835 if (!status_check) 836 return ret; 837 838 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 839 vstor_packet->status != 0) 840 return -EINVAL; 841 842 return ret; 843 } 844 845 static int storvsc_channel_init(struct hv_device *device, bool is_fc) 846 { 847 struct storvsc_device *stor_device; 848 struct storvsc_cmd_request *request; 849 struct vstor_packet *vstor_packet; 850 int ret, i; 851 int max_chns; 852 bool process_sub_channels = false; 853 854 stor_device = get_out_stor_device(device); 855 if (!stor_device) 856 return -ENODEV; 857 858 request = &stor_device->init_request; 859 vstor_packet = &request->vstor_packet; 860 861 /* 862 * Now, initiate the vsc/vsp initialization protocol on the open 863 * channel 864 */ 865 memset(request, 0, sizeof(struct storvsc_cmd_request)); 866 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION; 867 ret = storvsc_execute_vstor_op(device, request, true); 868 if (ret) 869 return ret; 870 /* 871 * Query host supported protocol version. 872 */ 873 874 for (i = 0; i < ARRAY_SIZE(vmstor_protocols); i++) { 875 /* reuse the packet for version range supported */ 876 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 877 vstor_packet->operation = 878 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION; 879 880 vstor_packet->version.major_minor = 881 vmstor_protocols[i].protocol_version; 882 883 /* 884 * The revision number is only used in Windows; set it to 0. 885 */ 886 vstor_packet->version.revision = 0; 887 ret = storvsc_execute_vstor_op(device, request, false); 888 if (ret != 0) 889 return ret; 890 891 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO) 892 return -EINVAL; 893 894 if (vstor_packet->status == 0) { 895 vmstor_proto_version = 896 vmstor_protocols[i].protocol_version; 897 898 sense_buffer_size = 899 vmstor_protocols[i].sense_buffer_size; 900 901 vmscsi_size_delta = 902 vmstor_protocols[i].vmscsi_size_delta; 903 904 break; 905 } 906 } 907 908 if (vstor_packet->status != 0) 909 return -EINVAL; 910 911 912 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 913 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES; 914 ret = storvsc_execute_vstor_op(device, request, true); 915 if (ret != 0) 916 return ret; 917 918 /* 919 * Check to see if multi-channel support is there. 920 * Hosts that implement protocol version of 5.1 and above 921 * support multi-channel. 922 */ 923 max_chns = vstor_packet->storage_channel_properties.max_channel_cnt; 924 925 /* 926 * Allocate state to manage the sub-channels. 927 * We allocate an array based on the numbers of possible CPUs 928 * (Hyper-V does not support cpu online/offline). 929 * This Array will be sparseley populated with unique 930 * channels - primary + sub-channels. 931 * We will however populate all the slots to evenly distribute 932 * the load. 933 */ 934 stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *), 935 GFP_KERNEL); 936 if (stor_device->stor_chns == NULL) 937 return -ENOMEM; 938 939 device->channel->change_target_cpu_callback = storvsc_change_target_cpu; 940 941 stor_device->stor_chns[device->channel->target_cpu] = device->channel; 942 cpumask_set_cpu(device->channel->target_cpu, 943 &stor_device->alloced_cpus); 944 945 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN8) { 946 if (vstor_packet->storage_channel_properties.flags & 947 STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL) 948 process_sub_channels = true; 949 } 950 stor_device->max_transfer_bytes = 951 vstor_packet->storage_channel_properties.max_transfer_bytes; 952 953 if (!is_fc) 954 goto done; 955 956 /* 957 * For FC devices retrieve FC HBA data. 958 */ 959 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 960 vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA; 961 ret = storvsc_execute_vstor_op(device, request, true); 962 if (ret != 0) 963 return ret; 964 965 /* 966 * Cache the currently active port and node ww names. 967 */ 968 cache_wwn(stor_device, vstor_packet); 969 970 done: 971 972 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 973 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION; 974 ret = storvsc_execute_vstor_op(device, request, true); 975 if (ret != 0) 976 return ret; 977 978 if (process_sub_channels) 979 handle_multichannel_storage(device, max_chns); 980 981 return ret; 982 } 983 984 static void storvsc_handle_error(struct vmscsi_request *vm_srb, 985 struct scsi_cmnd *scmnd, 986 struct Scsi_Host *host, 987 u8 asc, u8 ascq) 988 { 989 struct storvsc_scan_work *wrk; 990 void (*process_err_fn)(struct work_struct *work); 991 struct hv_host_device *host_dev = shost_priv(host); 992 bool do_work = false; 993 994 switch (SRB_STATUS(vm_srb->srb_status)) { 995 case SRB_STATUS_ERROR: 996 /* 997 * Let upper layer deal with error when 998 * sense message is present. 999 */ 1000 1001 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID) 1002 break; 1003 /* 1004 * If there is an error; offline the device since all 1005 * error recovery strategies would have already been 1006 * deployed on the host side. However, if the command 1007 * were a pass-through command deal with it appropriately. 1008 */ 1009 switch (scmnd->cmnd[0]) { 1010 case ATA_16: 1011 case ATA_12: 1012 set_host_byte(scmnd, DID_PASSTHROUGH); 1013 break; 1014 /* 1015 * On Some Windows hosts TEST_UNIT_READY command can return 1016 * SRB_STATUS_ERROR, let the upper level code deal with it 1017 * based on the sense information. 1018 */ 1019 case TEST_UNIT_READY: 1020 break; 1021 default: 1022 set_host_byte(scmnd, DID_ERROR); 1023 } 1024 break; 1025 case SRB_STATUS_INVALID_LUN: 1026 set_host_byte(scmnd, DID_NO_CONNECT); 1027 do_work = true; 1028 process_err_fn = storvsc_remove_lun; 1029 break; 1030 case SRB_STATUS_ABORTED: 1031 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID && 1032 (asc == 0x2a) && (ascq == 0x9)) { 1033 do_work = true; 1034 process_err_fn = storvsc_device_scan; 1035 /* 1036 * Retry the I/O that trigerred this. 1037 */ 1038 set_host_byte(scmnd, DID_REQUEUE); 1039 } 1040 break; 1041 } 1042 1043 if (!do_work) 1044 return; 1045 1046 /* 1047 * We need to schedule work to process this error; schedule it. 1048 */ 1049 wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC); 1050 if (!wrk) { 1051 set_host_byte(scmnd, DID_TARGET_FAILURE); 1052 return; 1053 } 1054 1055 wrk->host = host; 1056 wrk->lun = vm_srb->lun; 1057 wrk->tgt_id = vm_srb->target_id; 1058 INIT_WORK(&wrk->work, process_err_fn); 1059 queue_work(host_dev->handle_error_wq, &wrk->work); 1060 } 1061 1062 1063 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request, 1064 struct storvsc_device *stor_dev) 1065 { 1066 struct scsi_cmnd *scmnd = cmd_request->cmd; 1067 struct scsi_sense_hdr sense_hdr; 1068 struct vmscsi_request *vm_srb; 1069 u32 data_transfer_length; 1070 struct Scsi_Host *host; 1071 u32 payload_sz = cmd_request->payload_sz; 1072 void *payload = cmd_request->payload; 1073 1074 host = stor_dev->host; 1075 1076 vm_srb = &cmd_request->vstor_packet.vm_srb; 1077 data_transfer_length = vm_srb->data_transfer_length; 1078 1079 scmnd->result = vm_srb->scsi_status; 1080 1081 if (scmnd->result) { 1082 if (scsi_normalize_sense(scmnd->sense_buffer, 1083 SCSI_SENSE_BUFFERSIZE, &sense_hdr) && 1084 !(sense_hdr.sense_key == NOT_READY && 1085 sense_hdr.asc == 0x03A) && 1086 do_logging(STORVSC_LOGGING_ERROR)) 1087 scsi_print_sense_hdr(scmnd->device, "storvsc", 1088 &sense_hdr); 1089 } 1090 1091 if (vm_srb->srb_status != SRB_STATUS_SUCCESS) { 1092 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc, 1093 sense_hdr.ascq); 1094 /* 1095 * The Windows driver set data_transfer_length on 1096 * SRB_STATUS_DATA_OVERRUN. On other errors, this value 1097 * is untouched. In these cases we set it to 0. 1098 */ 1099 if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN) 1100 data_transfer_length = 0; 1101 } 1102 1103 scsi_set_resid(scmnd, 1104 cmd_request->payload->range.len - data_transfer_length); 1105 1106 scmnd->scsi_done(scmnd); 1107 1108 if (payload_sz > 1109 sizeof(struct vmbus_channel_packet_multipage_buffer)) 1110 kfree(payload); 1111 } 1112 1113 static void storvsc_on_io_completion(struct storvsc_device *stor_device, 1114 struct vstor_packet *vstor_packet, 1115 struct storvsc_cmd_request *request) 1116 { 1117 struct vstor_packet *stor_pkt; 1118 struct hv_device *device = stor_device->device; 1119 1120 stor_pkt = &request->vstor_packet; 1121 1122 /* 1123 * The current SCSI handling on the host side does 1124 * not correctly handle: 1125 * INQUIRY command with page code parameter set to 0x80 1126 * MODE_SENSE command with cmd[2] == 0x1c 1127 * 1128 * Setup srb and scsi status so this won't be fatal. 1129 * We do this so we can distinguish truly fatal failues 1130 * (srb status == 0x4) and off-line the device in that case. 1131 */ 1132 1133 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) || 1134 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) { 1135 vstor_packet->vm_srb.scsi_status = 0; 1136 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS; 1137 } 1138 1139 1140 /* Copy over the status...etc */ 1141 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status; 1142 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status; 1143 stor_pkt->vm_srb.sense_info_length = 1144 vstor_packet->vm_srb.sense_info_length; 1145 1146 if (vstor_packet->vm_srb.scsi_status != 0 || 1147 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS) 1148 storvsc_log(device, STORVSC_LOGGING_WARN, 1149 "cmd 0x%x scsi status 0x%x srb status 0x%x\n", 1150 stor_pkt->vm_srb.cdb[0], 1151 vstor_packet->vm_srb.scsi_status, 1152 vstor_packet->vm_srb.srb_status); 1153 1154 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) { 1155 /* CHECK_CONDITION */ 1156 if (vstor_packet->vm_srb.srb_status & 1157 SRB_STATUS_AUTOSENSE_VALID) { 1158 /* autosense data available */ 1159 1160 storvsc_log(device, STORVSC_LOGGING_WARN, 1161 "stor pkt %p autosense data valid - len %d\n", 1162 request, vstor_packet->vm_srb.sense_info_length); 1163 1164 memcpy(request->cmd->sense_buffer, 1165 vstor_packet->vm_srb.sense_data, 1166 vstor_packet->vm_srb.sense_info_length); 1167 1168 } 1169 } 1170 1171 stor_pkt->vm_srb.data_transfer_length = 1172 vstor_packet->vm_srb.data_transfer_length; 1173 1174 storvsc_command_completion(request, stor_device); 1175 1176 if (atomic_dec_and_test(&stor_device->num_outstanding_req) && 1177 stor_device->drain_notify) 1178 wake_up(&stor_device->waiting_to_drain); 1179 1180 1181 } 1182 1183 static void storvsc_on_receive(struct storvsc_device *stor_device, 1184 struct vstor_packet *vstor_packet, 1185 struct storvsc_cmd_request *request) 1186 { 1187 struct hv_host_device *host_dev; 1188 switch (vstor_packet->operation) { 1189 case VSTOR_OPERATION_COMPLETE_IO: 1190 storvsc_on_io_completion(stor_device, vstor_packet, request); 1191 break; 1192 1193 case VSTOR_OPERATION_REMOVE_DEVICE: 1194 case VSTOR_OPERATION_ENUMERATE_BUS: 1195 host_dev = shost_priv(stor_device->host); 1196 queue_work( 1197 host_dev->handle_error_wq, &host_dev->host_scan_work); 1198 break; 1199 1200 case VSTOR_OPERATION_FCHBA_DATA: 1201 cache_wwn(stor_device, vstor_packet); 1202 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1203 fc_host_node_name(stor_device->host) = stor_device->node_name; 1204 fc_host_port_name(stor_device->host) = stor_device->port_name; 1205 #endif 1206 break; 1207 default: 1208 break; 1209 } 1210 } 1211 1212 static void storvsc_on_channel_callback(void *context) 1213 { 1214 struct vmbus_channel *channel = (struct vmbus_channel *)context; 1215 const struct vmpacket_descriptor *desc; 1216 struct hv_device *device; 1217 struct storvsc_device *stor_device; 1218 1219 if (channel->primary_channel != NULL) 1220 device = channel->primary_channel->device_obj; 1221 else 1222 device = channel->device_obj; 1223 1224 stor_device = get_in_stor_device(device); 1225 if (!stor_device) 1226 return; 1227 1228 foreach_vmbus_pkt(desc, channel) { 1229 void *packet = hv_pkt_data(desc); 1230 struct storvsc_cmd_request *request; 1231 1232 request = (struct storvsc_cmd_request *) 1233 ((unsigned long)desc->trans_id); 1234 1235 if (request == &stor_device->init_request || 1236 request == &stor_device->reset_request) { 1237 memcpy(&request->vstor_packet, packet, 1238 (sizeof(struct vstor_packet) - vmscsi_size_delta)); 1239 complete(&request->wait_event); 1240 } else { 1241 storvsc_on_receive(stor_device, packet, request); 1242 } 1243 } 1244 } 1245 1246 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size, 1247 bool is_fc) 1248 { 1249 struct vmstorage_channel_properties props; 1250 int ret; 1251 1252 memset(&props, 0, sizeof(struct vmstorage_channel_properties)); 1253 1254 ret = vmbus_open(device->channel, 1255 ring_size, 1256 ring_size, 1257 (void *)&props, 1258 sizeof(struct vmstorage_channel_properties), 1259 storvsc_on_channel_callback, device->channel); 1260 1261 if (ret != 0) 1262 return ret; 1263 1264 ret = storvsc_channel_init(device, is_fc); 1265 1266 return ret; 1267 } 1268 1269 static int storvsc_dev_remove(struct hv_device *device) 1270 { 1271 struct storvsc_device *stor_device; 1272 1273 stor_device = hv_get_drvdata(device); 1274 1275 stor_device->destroy = true; 1276 1277 /* Make sure flag is set before waiting */ 1278 wmb(); 1279 1280 /* 1281 * At this point, all outbound traffic should be disable. We 1282 * only allow inbound traffic (responses) to proceed so that 1283 * outstanding requests can be completed. 1284 */ 1285 1286 storvsc_wait_to_drain(stor_device); 1287 1288 /* 1289 * Since we have already drained, we don't need to busy wait 1290 * as was done in final_release_stor_device() 1291 * Note that we cannot set the ext pointer to NULL until 1292 * we have drained - to drain the outgoing packets, we need to 1293 * allow incoming packets. 1294 */ 1295 hv_set_drvdata(device, NULL); 1296 1297 /* Close the channel */ 1298 vmbus_close(device->channel); 1299 1300 kfree(stor_device->stor_chns); 1301 kfree(stor_device); 1302 return 0; 1303 } 1304 1305 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device, 1306 u16 q_num) 1307 { 1308 u16 slot = 0; 1309 u16 hash_qnum; 1310 const struct cpumask *node_mask; 1311 int num_channels, tgt_cpu; 1312 1313 if (stor_device->num_sc == 0) { 1314 stor_device->stor_chns[q_num] = stor_device->device->channel; 1315 return stor_device->device->channel; 1316 } 1317 1318 /* 1319 * Our channel array is sparsley populated and we 1320 * initiated I/O on a processor/hw-q that does not 1321 * currently have a designated channel. Fix this. 1322 * The strategy is simple: 1323 * I. Ensure NUMA locality 1324 * II. Distribute evenly (best effort) 1325 */ 1326 1327 node_mask = cpumask_of_node(cpu_to_node(q_num)); 1328 1329 num_channels = 0; 1330 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) { 1331 if (cpumask_test_cpu(tgt_cpu, node_mask)) 1332 num_channels++; 1333 } 1334 if (num_channels == 0) { 1335 stor_device->stor_chns[q_num] = stor_device->device->channel; 1336 return stor_device->device->channel; 1337 } 1338 1339 hash_qnum = q_num; 1340 while (hash_qnum >= num_channels) 1341 hash_qnum -= num_channels; 1342 1343 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) { 1344 if (!cpumask_test_cpu(tgt_cpu, node_mask)) 1345 continue; 1346 if (slot == hash_qnum) 1347 break; 1348 slot++; 1349 } 1350 1351 stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu]; 1352 1353 return stor_device->stor_chns[q_num]; 1354 } 1355 1356 1357 static int storvsc_do_io(struct hv_device *device, 1358 struct storvsc_cmd_request *request, u16 q_num) 1359 { 1360 struct storvsc_device *stor_device; 1361 struct vstor_packet *vstor_packet; 1362 struct vmbus_channel *outgoing_channel, *channel; 1363 unsigned long flags; 1364 int ret = 0; 1365 const struct cpumask *node_mask; 1366 int tgt_cpu; 1367 1368 vstor_packet = &request->vstor_packet; 1369 stor_device = get_out_stor_device(device); 1370 1371 if (!stor_device) 1372 return -ENODEV; 1373 1374 1375 request->device = device; 1376 /* 1377 * Select an appropriate channel to send the request out. 1378 */ 1379 /* See storvsc_change_target_cpu(). */ 1380 outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]); 1381 if (outgoing_channel != NULL) { 1382 if (outgoing_channel->target_cpu == q_num) { 1383 /* 1384 * Ideally, we want to pick a different channel if 1385 * available on the same NUMA node. 1386 */ 1387 node_mask = cpumask_of_node(cpu_to_node(q_num)); 1388 for_each_cpu_wrap(tgt_cpu, 1389 &stor_device->alloced_cpus, q_num + 1) { 1390 if (!cpumask_test_cpu(tgt_cpu, node_mask)) 1391 continue; 1392 if (tgt_cpu == q_num) 1393 continue; 1394 channel = READ_ONCE( 1395 stor_device->stor_chns[tgt_cpu]); 1396 if (channel == NULL) 1397 continue; 1398 if (hv_get_avail_to_write_percent( 1399 &channel->outbound) 1400 > ring_avail_percent_lowater) { 1401 outgoing_channel = channel; 1402 goto found_channel; 1403 } 1404 } 1405 1406 /* 1407 * All the other channels on the same NUMA node are 1408 * busy. Try to use the channel on the current CPU 1409 */ 1410 if (hv_get_avail_to_write_percent( 1411 &outgoing_channel->outbound) 1412 > ring_avail_percent_lowater) 1413 goto found_channel; 1414 1415 /* 1416 * If we reach here, all the channels on the current 1417 * NUMA node are busy. Try to find a channel in 1418 * other NUMA nodes 1419 */ 1420 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) { 1421 if (cpumask_test_cpu(tgt_cpu, node_mask)) 1422 continue; 1423 channel = READ_ONCE( 1424 stor_device->stor_chns[tgt_cpu]); 1425 if (channel == NULL) 1426 continue; 1427 if (hv_get_avail_to_write_percent( 1428 &channel->outbound) 1429 > ring_avail_percent_lowater) { 1430 outgoing_channel = channel; 1431 goto found_channel; 1432 } 1433 } 1434 } 1435 } else { 1436 spin_lock_irqsave(&device->channel->lock, flags); 1437 outgoing_channel = stor_device->stor_chns[q_num]; 1438 if (outgoing_channel != NULL) { 1439 spin_unlock_irqrestore(&device->channel->lock, flags); 1440 goto found_channel; 1441 } 1442 outgoing_channel = get_og_chn(stor_device, q_num); 1443 spin_unlock_irqrestore(&device->channel->lock, flags); 1444 } 1445 1446 found_channel: 1447 vstor_packet->flags |= REQUEST_COMPLETION_FLAG; 1448 1449 vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) - 1450 vmscsi_size_delta); 1451 1452 1453 vstor_packet->vm_srb.sense_info_length = sense_buffer_size; 1454 1455 1456 vstor_packet->vm_srb.data_transfer_length = 1457 request->payload->range.len; 1458 1459 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB; 1460 1461 if (request->payload->range.len) { 1462 1463 ret = vmbus_sendpacket_mpb_desc(outgoing_channel, 1464 request->payload, request->payload_sz, 1465 vstor_packet, 1466 (sizeof(struct vstor_packet) - 1467 vmscsi_size_delta), 1468 (unsigned long)request); 1469 } else { 1470 ret = vmbus_sendpacket(outgoing_channel, vstor_packet, 1471 (sizeof(struct vstor_packet) - 1472 vmscsi_size_delta), 1473 (unsigned long)request, 1474 VM_PKT_DATA_INBAND, 1475 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 1476 } 1477 1478 if (ret != 0) 1479 return ret; 1480 1481 atomic_inc(&stor_device->num_outstanding_req); 1482 1483 return ret; 1484 } 1485 1486 static int storvsc_device_alloc(struct scsi_device *sdevice) 1487 { 1488 /* 1489 * Set blist flag to permit the reading of the VPD pages even when 1490 * the target may claim SPC-2 compliance. MSFT targets currently 1491 * claim SPC-2 compliance while they implement post SPC-2 features. 1492 * With this flag we can correctly handle WRITE_SAME_16 issues. 1493 * 1494 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but 1495 * still supports REPORT LUN. 1496 */ 1497 sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES; 1498 1499 return 0; 1500 } 1501 1502 static int storvsc_device_configure(struct scsi_device *sdevice) 1503 { 1504 blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ)); 1505 1506 sdevice->no_write_same = 1; 1507 1508 /* 1509 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3 1510 * if the device is a MSFT virtual device. If the host is 1511 * WIN10 or newer, allow write_same. 1512 */ 1513 if (!strncmp(sdevice->vendor, "Msft", 4)) { 1514 switch (vmstor_proto_version) { 1515 case VMSTOR_PROTO_VERSION_WIN8: 1516 case VMSTOR_PROTO_VERSION_WIN8_1: 1517 sdevice->scsi_level = SCSI_SPC_3; 1518 break; 1519 } 1520 1521 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10) 1522 sdevice->no_write_same = 0; 1523 } 1524 1525 return 0; 1526 } 1527 1528 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev, 1529 sector_t capacity, int *info) 1530 { 1531 sector_t nsect = capacity; 1532 sector_t cylinders = nsect; 1533 int heads, sectors_pt; 1534 1535 /* 1536 * We are making up these values; let us keep it simple. 1537 */ 1538 heads = 0xff; 1539 sectors_pt = 0x3f; /* Sectors per track */ 1540 sector_div(cylinders, heads * sectors_pt); 1541 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect) 1542 cylinders = 0xffff; 1543 1544 info[0] = heads; 1545 info[1] = sectors_pt; 1546 info[2] = (int)cylinders; 1547 1548 return 0; 1549 } 1550 1551 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd) 1552 { 1553 struct hv_host_device *host_dev = shost_priv(scmnd->device->host); 1554 struct hv_device *device = host_dev->dev; 1555 1556 struct storvsc_device *stor_device; 1557 struct storvsc_cmd_request *request; 1558 struct vstor_packet *vstor_packet; 1559 int ret, t; 1560 1561 1562 stor_device = get_out_stor_device(device); 1563 if (!stor_device) 1564 return FAILED; 1565 1566 request = &stor_device->reset_request; 1567 vstor_packet = &request->vstor_packet; 1568 1569 init_completion(&request->wait_event); 1570 1571 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS; 1572 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 1573 vstor_packet->vm_srb.path_id = stor_device->path_id; 1574 1575 ret = vmbus_sendpacket(device->channel, vstor_packet, 1576 (sizeof(struct vstor_packet) - 1577 vmscsi_size_delta), 1578 (unsigned long)&stor_device->reset_request, 1579 VM_PKT_DATA_INBAND, 1580 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 1581 if (ret != 0) 1582 return FAILED; 1583 1584 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 1585 if (t == 0) 1586 return TIMEOUT_ERROR; 1587 1588 1589 /* 1590 * At this point, all outstanding requests in the adapter 1591 * should have been flushed out and return to us 1592 * There is a potential race here where the host may be in 1593 * the process of responding when we return from here. 1594 * Just wait for all in-transit packets to be accounted for 1595 * before we return from here. 1596 */ 1597 storvsc_wait_to_drain(stor_device); 1598 1599 return SUCCESS; 1600 } 1601 1602 /* 1603 * The host guarantees to respond to each command, although I/O latencies might 1604 * be unbounded on Azure. Reset the timer unconditionally to give the host a 1605 * chance to perform EH. 1606 */ 1607 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd) 1608 { 1609 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1610 if (scmnd->device->host->transportt == fc_transport_template) 1611 return fc_eh_timed_out(scmnd); 1612 #endif 1613 return BLK_EH_RESET_TIMER; 1614 } 1615 1616 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd) 1617 { 1618 bool allowed = true; 1619 u8 scsi_op = scmnd->cmnd[0]; 1620 1621 switch (scsi_op) { 1622 /* the host does not handle WRITE_SAME, log accident usage */ 1623 case WRITE_SAME: 1624 /* 1625 * smartd sends this command and the host does not handle 1626 * this. So, don't send it. 1627 */ 1628 case SET_WINDOW: 1629 scmnd->result = ILLEGAL_REQUEST << 16; 1630 allowed = false; 1631 break; 1632 default: 1633 break; 1634 } 1635 return allowed; 1636 } 1637 1638 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd) 1639 { 1640 int ret; 1641 struct hv_host_device *host_dev = shost_priv(host); 1642 struct hv_device *dev = host_dev->dev; 1643 struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd); 1644 int i; 1645 struct scatterlist *sgl; 1646 unsigned int sg_count = 0; 1647 struct vmscsi_request *vm_srb; 1648 struct scatterlist *cur_sgl; 1649 struct vmbus_packet_mpb_array *payload; 1650 u32 payload_sz; 1651 u32 length; 1652 1653 if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) { 1654 /* 1655 * On legacy hosts filter unimplemented commands. 1656 * Future hosts are expected to correctly handle 1657 * unsupported commands. Furthermore, it is 1658 * possible that some of the currently 1659 * unsupported commands maybe supported in 1660 * future versions of the host. 1661 */ 1662 if (!storvsc_scsi_cmd_ok(scmnd)) { 1663 scmnd->scsi_done(scmnd); 1664 return 0; 1665 } 1666 } 1667 1668 /* Setup the cmd request */ 1669 cmd_request->cmd = scmnd; 1670 1671 vm_srb = &cmd_request->vstor_packet.vm_srb; 1672 vm_srb->win8_extension.time_out_value = 60; 1673 1674 vm_srb->win8_extension.srb_flags |= 1675 SRB_FLAGS_DISABLE_SYNCH_TRANSFER; 1676 1677 if (scmnd->device->tagged_supported) { 1678 vm_srb->win8_extension.srb_flags |= 1679 (SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE); 1680 vm_srb->win8_extension.queue_tag = SP_UNTAGGED; 1681 vm_srb->win8_extension.queue_action = SRB_SIMPLE_TAG_REQUEST; 1682 } 1683 1684 /* Build the SRB */ 1685 switch (scmnd->sc_data_direction) { 1686 case DMA_TO_DEVICE: 1687 vm_srb->data_in = WRITE_TYPE; 1688 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT; 1689 break; 1690 case DMA_FROM_DEVICE: 1691 vm_srb->data_in = READ_TYPE; 1692 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN; 1693 break; 1694 case DMA_NONE: 1695 vm_srb->data_in = UNKNOWN_TYPE; 1696 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER; 1697 break; 1698 default: 1699 /* 1700 * This is DMA_BIDIRECTIONAL or something else we are never 1701 * supposed to see here. 1702 */ 1703 WARN(1, "Unexpected data direction: %d\n", 1704 scmnd->sc_data_direction); 1705 return -EINVAL; 1706 } 1707 1708 1709 vm_srb->port_number = host_dev->port; 1710 vm_srb->path_id = scmnd->device->channel; 1711 vm_srb->target_id = scmnd->device->id; 1712 vm_srb->lun = scmnd->device->lun; 1713 1714 vm_srb->cdb_length = scmnd->cmd_len; 1715 1716 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length); 1717 1718 sgl = (struct scatterlist *)scsi_sglist(scmnd); 1719 sg_count = scsi_sg_count(scmnd); 1720 1721 length = scsi_bufflen(scmnd); 1722 payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb; 1723 payload_sz = sizeof(cmd_request->mpb); 1724 1725 if (sg_count) { 1726 if (sg_count > MAX_PAGE_BUFFER_COUNT) { 1727 1728 payload_sz = (sg_count * sizeof(u64) + 1729 sizeof(struct vmbus_packet_mpb_array)); 1730 payload = kzalloc(payload_sz, GFP_ATOMIC); 1731 if (!payload) 1732 return SCSI_MLQUEUE_DEVICE_BUSY; 1733 } 1734 1735 payload->range.len = length; 1736 payload->range.offset = sgl[0].offset; 1737 1738 cur_sgl = sgl; 1739 for (i = 0; i < sg_count; i++) { 1740 payload->range.pfn_array[i] = 1741 page_to_pfn(sg_page((cur_sgl))); 1742 cur_sgl = sg_next(cur_sgl); 1743 } 1744 } 1745 1746 cmd_request->payload = payload; 1747 cmd_request->payload_sz = payload_sz; 1748 1749 /* Invokes the vsc to start an IO */ 1750 ret = storvsc_do_io(dev, cmd_request, get_cpu()); 1751 put_cpu(); 1752 1753 if (ret == -EAGAIN) { 1754 if (payload_sz > sizeof(cmd_request->mpb)) 1755 kfree(payload); 1756 /* no more space */ 1757 return SCSI_MLQUEUE_DEVICE_BUSY; 1758 } 1759 1760 return 0; 1761 } 1762 1763 static struct scsi_host_template scsi_driver = { 1764 .module = THIS_MODULE, 1765 .name = "storvsc_host_t", 1766 .cmd_size = sizeof(struct storvsc_cmd_request), 1767 .bios_param = storvsc_get_chs, 1768 .queuecommand = storvsc_queuecommand, 1769 .eh_host_reset_handler = storvsc_host_reset_handler, 1770 .proc_name = "storvsc_host", 1771 .eh_timed_out = storvsc_eh_timed_out, 1772 .slave_alloc = storvsc_device_alloc, 1773 .slave_configure = storvsc_device_configure, 1774 .cmd_per_lun = 2048, 1775 .this_id = -1, 1776 /* Make sure we dont get a sg segment crosses a page boundary */ 1777 .dma_boundary = PAGE_SIZE-1, 1778 /* Ensure there are no gaps in presented sgls */ 1779 .virt_boundary_mask = PAGE_SIZE-1, 1780 .no_write_same = 1, 1781 .track_queue_depth = 1, 1782 .change_queue_depth = storvsc_change_queue_depth, 1783 }; 1784 1785 enum { 1786 SCSI_GUID, 1787 IDE_GUID, 1788 SFC_GUID, 1789 }; 1790 1791 static const struct hv_vmbus_device_id id_table[] = { 1792 /* SCSI guid */ 1793 { HV_SCSI_GUID, 1794 .driver_data = SCSI_GUID 1795 }, 1796 /* IDE guid */ 1797 { HV_IDE_GUID, 1798 .driver_data = IDE_GUID 1799 }, 1800 /* Fibre Channel GUID */ 1801 { 1802 HV_SYNTHFC_GUID, 1803 .driver_data = SFC_GUID 1804 }, 1805 { }, 1806 }; 1807 1808 MODULE_DEVICE_TABLE(vmbus, id_table); 1809 1810 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID }; 1811 1812 static bool hv_dev_is_fc(struct hv_device *hv_dev) 1813 { 1814 return guid_equal(&fc_guid.guid, &hv_dev->dev_type); 1815 } 1816 1817 static int storvsc_probe(struct hv_device *device, 1818 const struct hv_vmbus_device_id *dev_id) 1819 { 1820 int ret; 1821 int num_cpus = num_online_cpus(); 1822 struct Scsi_Host *host; 1823 struct hv_host_device *host_dev; 1824 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false); 1825 bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false); 1826 int target = 0; 1827 struct storvsc_device *stor_device; 1828 int max_luns_per_target; 1829 int max_targets; 1830 int max_channels; 1831 int max_sub_channels = 0; 1832 1833 /* 1834 * Based on the windows host we are running on, 1835 * set state to properly communicate with the host. 1836 */ 1837 1838 if (vmbus_proto_version < VERSION_WIN8) { 1839 max_luns_per_target = STORVSC_IDE_MAX_LUNS_PER_TARGET; 1840 max_targets = STORVSC_IDE_MAX_TARGETS; 1841 max_channels = STORVSC_IDE_MAX_CHANNELS; 1842 } else { 1843 max_luns_per_target = STORVSC_MAX_LUNS_PER_TARGET; 1844 max_targets = STORVSC_MAX_TARGETS; 1845 max_channels = STORVSC_MAX_CHANNELS; 1846 /* 1847 * On Windows8 and above, we support sub-channels for storage 1848 * on SCSI and FC controllers. 1849 * The number of sub-channels offerred is based on the number of 1850 * VCPUs in the guest. 1851 */ 1852 if (!dev_is_ide) 1853 max_sub_channels = 1854 (num_cpus - 1) / storvsc_vcpus_per_sub_channel; 1855 } 1856 1857 scsi_driver.can_queue = max_outstanding_req_per_channel * 1858 (max_sub_channels + 1) * 1859 (100 - ring_avail_percent_lowater) / 100; 1860 1861 host = scsi_host_alloc(&scsi_driver, 1862 sizeof(struct hv_host_device)); 1863 if (!host) 1864 return -ENOMEM; 1865 1866 host_dev = shost_priv(host); 1867 memset(host_dev, 0, sizeof(struct hv_host_device)); 1868 1869 host_dev->port = host->host_no; 1870 host_dev->dev = device; 1871 host_dev->host = host; 1872 1873 1874 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL); 1875 if (!stor_device) { 1876 ret = -ENOMEM; 1877 goto err_out0; 1878 } 1879 1880 stor_device->destroy = false; 1881 init_waitqueue_head(&stor_device->waiting_to_drain); 1882 stor_device->device = device; 1883 stor_device->host = host; 1884 hv_set_drvdata(device, stor_device); 1885 1886 stor_device->port_number = host->host_no; 1887 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc); 1888 if (ret) 1889 goto err_out1; 1890 1891 host_dev->path = stor_device->path_id; 1892 host_dev->target = stor_device->target_id; 1893 1894 switch (dev_id->driver_data) { 1895 case SFC_GUID: 1896 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET; 1897 host->max_id = STORVSC_FC_MAX_TARGETS; 1898 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1; 1899 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1900 host->transportt = fc_transport_template; 1901 #endif 1902 break; 1903 1904 case SCSI_GUID: 1905 host->max_lun = max_luns_per_target; 1906 host->max_id = max_targets; 1907 host->max_channel = max_channels - 1; 1908 break; 1909 1910 default: 1911 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET; 1912 host->max_id = STORVSC_IDE_MAX_TARGETS; 1913 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1; 1914 break; 1915 } 1916 /* max cmd length */ 1917 host->max_cmd_len = STORVSC_MAX_CMD_LEN; 1918 1919 /* 1920 * set the table size based on the info we got 1921 * from the host. 1922 */ 1923 host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT); 1924 /* 1925 * For non-IDE disks, the host supports multiple channels. 1926 * Set the number of HW queues we are supporting. 1927 */ 1928 if (!dev_is_ide) 1929 host->nr_hw_queues = num_present_cpus(); 1930 1931 /* 1932 * Set the error handler work queue. 1933 */ 1934 host_dev->handle_error_wq = 1935 alloc_ordered_workqueue("storvsc_error_wq_%d", 1936 WQ_MEM_RECLAIM, 1937 host->host_no); 1938 if (!host_dev->handle_error_wq) 1939 goto err_out2; 1940 INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan); 1941 /* Register the HBA and start the scsi bus scan */ 1942 ret = scsi_add_host(host, &device->device); 1943 if (ret != 0) 1944 goto err_out3; 1945 1946 if (!dev_is_ide) { 1947 scsi_scan_host(host); 1948 } else { 1949 target = (device->dev_instance.b[5] << 8 | 1950 device->dev_instance.b[4]); 1951 ret = scsi_add_device(host, 0, target, 0); 1952 if (ret) 1953 goto err_out4; 1954 } 1955 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1956 if (host->transportt == fc_transport_template) { 1957 struct fc_rport_identifiers ids = { 1958 .roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR, 1959 }; 1960 1961 fc_host_node_name(host) = stor_device->node_name; 1962 fc_host_port_name(host) = stor_device->port_name; 1963 stor_device->rport = fc_remote_port_add(host, 0, &ids); 1964 if (!stor_device->rport) { 1965 ret = -ENOMEM; 1966 goto err_out4; 1967 } 1968 } 1969 #endif 1970 return 0; 1971 1972 err_out4: 1973 scsi_remove_host(host); 1974 1975 err_out3: 1976 destroy_workqueue(host_dev->handle_error_wq); 1977 1978 err_out2: 1979 /* 1980 * Once we have connected with the host, we would need to 1981 * to invoke storvsc_dev_remove() to rollback this state and 1982 * this call also frees up the stor_device; hence the jump around 1983 * err_out1 label. 1984 */ 1985 storvsc_dev_remove(device); 1986 goto err_out0; 1987 1988 err_out1: 1989 kfree(stor_device->stor_chns); 1990 kfree(stor_device); 1991 1992 err_out0: 1993 scsi_host_put(host); 1994 return ret; 1995 } 1996 1997 /* Change a scsi target's queue depth */ 1998 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth) 1999 { 2000 if (queue_depth > scsi_driver.can_queue) 2001 queue_depth = scsi_driver.can_queue; 2002 2003 return scsi_change_queue_depth(sdev, queue_depth); 2004 } 2005 2006 static int storvsc_remove(struct hv_device *dev) 2007 { 2008 struct storvsc_device *stor_device = hv_get_drvdata(dev); 2009 struct Scsi_Host *host = stor_device->host; 2010 struct hv_host_device *host_dev = shost_priv(host); 2011 2012 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 2013 if (host->transportt == fc_transport_template) { 2014 fc_remote_port_delete(stor_device->rport); 2015 fc_remove_host(host); 2016 } 2017 #endif 2018 destroy_workqueue(host_dev->handle_error_wq); 2019 scsi_remove_host(host); 2020 storvsc_dev_remove(dev); 2021 scsi_host_put(host); 2022 2023 return 0; 2024 } 2025 2026 static int storvsc_suspend(struct hv_device *hv_dev) 2027 { 2028 struct storvsc_device *stor_device = hv_get_drvdata(hv_dev); 2029 struct Scsi_Host *host = stor_device->host; 2030 struct hv_host_device *host_dev = shost_priv(host); 2031 2032 storvsc_wait_to_drain(stor_device); 2033 2034 drain_workqueue(host_dev->handle_error_wq); 2035 2036 vmbus_close(hv_dev->channel); 2037 2038 kfree(stor_device->stor_chns); 2039 stor_device->stor_chns = NULL; 2040 2041 cpumask_clear(&stor_device->alloced_cpus); 2042 2043 return 0; 2044 } 2045 2046 static int storvsc_resume(struct hv_device *hv_dev) 2047 { 2048 int ret; 2049 2050 ret = storvsc_connect_to_vsp(hv_dev, storvsc_ringbuffer_size, 2051 hv_dev_is_fc(hv_dev)); 2052 return ret; 2053 } 2054 2055 static struct hv_driver storvsc_drv = { 2056 .name = KBUILD_MODNAME, 2057 .id_table = id_table, 2058 .probe = storvsc_probe, 2059 .remove = storvsc_remove, 2060 .suspend = storvsc_suspend, 2061 .resume = storvsc_resume, 2062 .driver = { 2063 .probe_type = PROBE_PREFER_ASYNCHRONOUS, 2064 }, 2065 }; 2066 2067 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 2068 static struct fc_function_template fc_transport_functions = { 2069 .show_host_node_name = 1, 2070 .show_host_port_name = 1, 2071 }; 2072 #endif 2073 2074 static int __init storvsc_drv_init(void) 2075 { 2076 int ret; 2077 2078 /* 2079 * Divide the ring buffer data size (which is 1 page less 2080 * than the ring buffer size since that page is reserved for 2081 * the ring buffer indices) by the max request size (which is 2082 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64) 2083 */ 2084 max_outstanding_req_per_channel = 2085 ((storvsc_ringbuffer_size - PAGE_SIZE) / 2086 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET + 2087 sizeof(struct vstor_packet) + sizeof(u64) - 2088 vmscsi_size_delta, 2089 sizeof(u64))); 2090 2091 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 2092 fc_transport_template = fc_attach_transport(&fc_transport_functions); 2093 if (!fc_transport_template) 2094 return -ENODEV; 2095 #endif 2096 2097 ret = vmbus_driver_register(&storvsc_drv); 2098 2099 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 2100 if (ret) 2101 fc_release_transport(fc_transport_template); 2102 #endif 2103 2104 return ret; 2105 } 2106 2107 static void __exit storvsc_drv_exit(void) 2108 { 2109 vmbus_driver_unregister(&storvsc_drv); 2110 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 2111 fc_release_transport(fc_transport_template); 2112 #endif 2113 } 2114 2115 MODULE_LICENSE("GPL"); 2116 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver"); 2117 module_init(storvsc_drv_init); 2118 module_exit(storvsc_drv_exit); 2119