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 struct scsi_host_template scsi_driver; 403 static void storvsc_on_channel_callback(void *context); 404 405 #define STORVSC_MAX_LUNS_PER_TARGET 255 406 #define STORVSC_MAX_TARGETS 2 407 #define STORVSC_MAX_CHANNELS 8 408 409 #define STORVSC_FC_MAX_LUNS_PER_TARGET 255 410 #define STORVSC_FC_MAX_TARGETS 128 411 #define STORVSC_FC_MAX_CHANNELS 8 412 413 #define STORVSC_IDE_MAX_LUNS_PER_TARGET 64 414 #define STORVSC_IDE_MAX_TARGETS 1 415 #define STORVSC_IDE_MAX_CHANNELS 1 416 417 /* 418 * Upper bound on the size of a storvsc packet. vmscsi_size_delta is not 419 * included in the calculation because it is set after STORVSC_MAX_PKT_SIZE 420 * is used in storvsc_connect_to_vsp 421 */ 422 #define STORVSC_MAX_PKT_SIZE (sizeof(struct vmpacket_descriptor) +\ 423 sizeof(struct vstor_packet)) 424 425 struct storvsc_cmd_request { 426 struct scsi_cmnd *cmd; 427 428 struct hv_device *device; 429 430 /* Synchronize the request/response if needed */ 431 struct completion wait_event; 432 433 struct vmbus_channel_packet_multipage_buffer mpb; 434 struct vmbus_packet_mpb_array *payload; 435 u32 payload_sz; 436 437 struct vstor_packet vstor_packet; 438 }; 439 440 441 /* A storvsc device is a device object that contains a vmbus channel */ 442 struct storvsc_device { 443 struct hv_device *device; 444 445 bool destroy; 446 bool drain_notify; 447 atomic_t num_outstanding_req; 448 struct Scsi_Host *host; 449 450 wait_queue_head_t waiting_to_drain; 451 452 /* 453 * Each unique Port/Path/Target represents 1 channel ie scsi 454 * controller. In reality, the pathid, targetid is always 0 455 * and the port is set by us 456 */ 457 unsigned int port_number; 458 unsigned char path_id; 459 unsigned char target_id; 460 461 /* 462 * Max I/O, the device can support. 463 */ 464 u32 max_transfer_bytes; 465 /* 466 * Number of sub-channels we will open. 467 */ 468 u16 num_sc; 469 struct vmbus_channel **stor_chns; 470 /* 471 * Mask of CPUs bound to subchannels. 472 */ 473 struct cpumask alloced_cpus; 474 /* 475 * Serializes modifications of stor_chns[] from storvsc_do_io() 476 * and storvsc_change_target_cpu(). 477 */ 478 spinlock_t lock; 479 /* Used for vsc/vsp channel reset process */ 480 struct storvsc_cmd_request init_request; 481 struct storvsc_cmd_request reset_request; 482 /* 483 * Currently active port and node names for FC devices. 484 */ 485 u64 node_name; 486 u64 port_name; 487 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 488 struct fc_rport *rport; 489 #endif 490 }; 491 492 struct hv_host_device { 493 struct hv_device *dev; 494 unsigned int port; 495 unsigned char path; 496 unsigned char target; 497 struct workqueue_struct *handle_error_wq; 498 struct work_struct host_scan_work; 499 struct Scsi_Host *host; 500 }; 501 502 struct storvsc_scan_work { 503 struct work_struct work; 504 struct Scsi_Host *host; 505 u8 lun; 506 u8 tgt_id; 507 }; 508 509 static void storvsc_device_scan(struct work_struct *work) 510 { 511 struct storvsc_scan_work *wrk; 512 struct scsi_device *sdev; 513 514 wrk = container_of(work, struct storvsc_scan_work, work); 515 516 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun); 517 if (!sdev) 518 goto done; 519 scsi_rescan_device(&sdev->sdev_gendev); 520 scsi_device_put(sdev); 521 522 done: 523 kfree(wrk); 524 } 525 526 static void storvsc_host_scan(struct work_struct *work) 527 { 528 struct Scsi_Host *host; 529 struct scsi_device *sdev; 530 struct hv_host_device *host_device = 531 container_of(work, struct hv_host_device, host_scan_work); 532 533 host = host_device->host; 534 /* 535 * Before scanning the host, first check to see if any of the 536 * currrently known devices have been hot removed. We issue a 537 * "unit ready" command against all currently known devices. 538 * This I/O will result in an error for devices that have been 539 * removed. As part of handling the I/O error, we remove the device. 540 * 541 * When a LUN is added or removed, the host sends us a signal to 542 * scan the host. Thus we are forced to discover the LUNs that 543 * may have been removed this way. 544 */ 545 mutex_lock(&host->scan_mutex); 546 shost_for_each_device(sdev, host) 547 scsi_test_unit_ready(sdev, 1, 1, NULL); 548 mutex_unlock(&host->scan_mutex); 549 /* 550 * Now scan the host to discover LUNs that may have been added. 551 */ 552 scsi_scan_host(host); 553 } 554 555 static void storvsc_remove_lun(struct work_struct *work) 556 { 557 struct storvsc_scan_work *wrk; 558 struct scsi_device *sdev; 559 560 wrk = container_of(work, struct storvsc_scan_work, work); 561 if (!scsi_host_get(wrk->host)) 562 goto done; 563 564 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun); 565 566 if (sdev) { 567 scsi_remove_device(sdev); 568 scsi_device_put(sdev); 569 } 570 scsi_host_put(wrk->host); 571 572 done: 573 kfree(wrk); 574 } 575 576 577 /* 578 * We can get incoming messages from the host that are not in response to 579 * messages that we have sent out. An example of this would be messages 580 * received by the guest to notify dynamic addition/removal of LUNs. To 581 * deal with potential race conditions where the driver may be in the 582 * midst of being unloaded when we might receive an unsolicited message 583 * from the host, we have implemented a mechanism to gurantee sequential 584 * consistency: 585 * 586 * 1) Once the device is marked as being destroyed, we will fail all 587 * outgoing messages. 588 * 2) We permit incoming messages when the device is being destroyed, 589 * only to properly account for messages already sent out. 590 */ 591 592 static inline struct storvsc_device *get_out_stor_device( 593 struct hv_device *device) 594 { 595 struct storvsc_device *stor_device; 596 597 stor_device = hv_get_drvdata(device); 598 599 if (stor_device && stor_device->destroy) 600 stor_device = NULL; 601 602 return stor_device; 603 } 604 605 606 static inline void storvsc_wait_to_drain(struct storvsc_device *dev) 607 { 608 dev->drain_notify = true; 609 wait_event(dev->waiting_to_drain, 610 atomic_read(&dev->num_outstanding_req) == 0); 611 dev->drain_notify = false; 612 } 613 614 static inline struct storvsc_device *get_in_stor_device( 615 struct hv_device *device) 616 { 617 struct storvsc_device *stor_device; 618 619 stor_device = hv_get_drvdata(device); 620 621 if (!stor_device) 622 goto get_in_err; 623 624 /* 625 * If the device is being destroyed; allow incoming 626 * traffic only to cleanup outstanding requests. 627 */ 628 629 if (stor_device->destroy && 630 (atomic_read(&stor_device->num_outstanding_req) == 0)) 631 stor_device = NULL; 632 633 get_in_err: 634 return stor_device; 635 636 } 637 638 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old, 639 u32 new) 640 { 641 struct storvsc_device *stor_device; 642 struct vmbus_channel *cur_chn; 643 bool old_is_alloced = false; 644 struct hv_device *device; 645 unsigned long flags; 646 int cpu; 647 648 device = channel->primary_channel ? 649 channel->primary_channel->device_obj 650 : channel->device_obj; 651 stor_device = get_out_stor_device(device); 652 if (!stor_device) 653 return; 654 655 /* See storvsc_do_io() -> get_og_chn(). */ 656 spin_lock_irqsave(&stor_device->lock, flags); 657 658 /* 659 * Determines if the storvsc device has other channels assigned to 660 * the "old" CPU to update the alloced_cpus mask and the stor_chns 661 * array. 662 */ 663 if (device->channel != channel && device->channel->target_cpu == old) { 664 cur_chn = device->channel; 665 old_is_alloced = true; 666 goto old_is_alloced; 667 } 668 list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) { 669 if (cur_chn == channel) 670 continue; 671 if (cur_chn->target_cpu == old) { 672 old_is_alloced = true; 673 goto old_is_alloced; 674 } 675 } 676 677 old_is_alloced: 678 if (old_is_alloced) 679 WRITE_ONCE(stor_device->stor_chns[old], cur_chn); 680 else 681 cpumask_clear_cpu(old, &stor_device->alloced_cpus); 682 683 /* "Flush" the stor_chns array. */ 684 for_each_possible_cpu(cpu) { 685 if (stor_device->stor_chns[cpu] && !cpumask_test_cpu( 686 cpu, &stor_device->alloced_cpus)) 687 WRITE_ONCE(stor_device->stor_chns[cpu], NULL); 688 } 689 690 WRITE_ONCE(stor_device->stor_chns[new], channel); 691 cpumask_set_cpu(new, &stor_device->alloced_cpus); 692 693 spin_unlock_irqrestore(&stor_device->lock, flags); 694 } 695 696 static void handle_sc_creation(struct vmbus_channel *new_sc) 697 { 698 struct hv_device *device = new_sc->primary_channel->device_obj; 699 struct device *dev = &device->device; 700 struct storvsc_device *stor_device; 701 struct vmstorage_channel_properties props; 702 int ret; 703 704 stor_device = get_out_stor_device(device); 705 if (!stor_device) 706 return; 707 708 memset(&props, 0, sizeof(struct vmstorage_channel_properties)); 709 new_sc->max_pkt_size = STORVSC_MAX_PKT_SIZE; 710 711 /* 712 * The size of vmbus_requestor is an upper bound on the number of requests 713 * that can be in-progress at any one time across all channels. 714 */ 715 new_sc->rqstor_size = scsi_driver.can_queue; 716 717 ret = vmbus_open(new_sc, 718 storvsc_ringbuffer_size, 719 storvsc_ringbuffer_size, 720 (void *)&props, 721 sizeof(struct vmstorage_channel_properties), 722 storvsc_on_channel_callback, new_sc); 723 724 /* In case vmbus_open() fails, we don't use the sub-channel. */ 725 if (ret != 0) { 726 dev_err(dev, "Failed to open sub-channel: err=%d\n", ret); 727 return; 728 } 729 730 new_sc->change_target_cpu_callback = storvsc_change_target_cpu; 731 732 /* Add the sub-channel to the array of available channels. */ 733 stor_device->stor_chns[new_sc->target_cpu] = new_sc; 734 cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus); 735 } 736 737 static void handle_multichannel_storage(struct hv_device *device, int max_chns) 738 { 739 struct device *dev = &device->device; 740 struct storvsc_device *stor_device; 741 int num_sc; 742 struct storvsc_cmd_request *request; 743 struct vstor_packet *vstor_packet; 744 int ret, t; 745 746 /* 747 * If the number of CPUs is artificially restricted, such as 748 * with maxcpus=1 on the kernel boot line, Hyper-V could offer 749 * sub-channels >= the number of CPUs. These sub-channels 750 * should not be created. The primary channel is already created 751 * and assigned to one CPU, so check against # CPUs - 1. 752 */ 753 num_sc = min((int)(num_online_cpus() - 1), max_chns); 754 if (!num_sc) 755 return; 756 757 stor_device = get_out_stor_device(device); 758 if (!stor_device) 759 return; 760 761 stor_device->num_sc = num_sc; 762 request = &stor_device->init_request; 763 vstor_packet = &request->vstor_packet; 764 765 /* 766 * Establish a handler for dealing with subchannels. 767 */ 768 vmbus_set_sc_create_callback(device->channel, handle_sc_creation); 769 770 /* 771 * Request the host to create sub-channels. 772 */ 773 memset(request, 0, sizeof(struct storvsc_cmd_request)); 774 init_completion(&request->wait_event); 775 vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS; 776 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 777 vstor_packet->sub_channel_count = num_sc; 778 779 ret = vmbus_sendpacket(device->channel, vstor_packet, 780 (sizeof(struct vstor_packet) - 781 vmscsi_size_delta), 782 (unsigned long)request, 783 VM_PKT_DATA_INBAND, 784 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 785 786 if (ret != 0) { 787 dev_err(dev, "Failed to create sub-channel: err=%d\n", ret); 788 return; 789 } 790 791 t = wait_for_completion_timeout(&request->wait_event, 10*HZ); 792 if (t == 0) { 793 dev_err(dev, "Failed to create sub-channel: timed out\n"); 794 return; 795 } 796 797 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 798 vstor_packet->status != 0) { 799 dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n", 800 vstor_packet->operation, vstor_packet->status); 801 return; 802 } 803 804 /* 805 * We need to do nothing here, because vmbus_process_offer() 806 * invokes channel->sc_creation_callback, which will open and use 807 * the sub-channel(s). 808 */ 809 } 810 811 static void cache_wwn(struct storvsc_device *stor_device, 812 struct vstor_packet *vstor_packet) 813 { 814 /* 815 * Cache the currently active port and node ww names. 816 */ 817 if (vstor_packet->wwn_packet.primary_active) { 818 stor_device->node_name = 819 wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn); 820 stor_device->port_name = 821 wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn); 822 } else { 823 stor_device->node_name = 824 wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn); 825 stor_device->port_name = 826 wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn); 827 } 828 } 829 830 831 static int storvsc_execute_vstor_op(struct hv_device *device, 832 struct storvsc_cmd_request *request, 833 bool status_check) 834 { 835 struct vstor_packet *vstor_packet; 836 int ret, t; 837 838 vstor_packet = &request->vstor_packet; 839 840 init_completion(&request->wait_event); 841 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 842 843 ret = vmbus_sendpacket(device->channel, vstor_packet, 844 (sizeof(struct vstor_packet) - 845 vmscsi_size_delta), 846 (unsigned long)request, 847 VM_PKT_DATA_INBAND, 848 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 849 if (ret != 0) 850 return ret; 851 852 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 853 if (t == 0) 854 return -ETIMEDOUT; 855 856 if (!status_check) 857 return ret; 858 859 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 860 vstor_packet->status != 0) 861 return -EINVAL; 862 863 return ret; 864 } 865 866 static int storvsc_channel_init(struct hv_device *device, bool is_fc) 867 { 868 struct storvsc_device *stor_device; 869 struct storvsc_cmd_request *request; 870 struct vstor_packet *vstor_packet; 871 int ret, i; 872 int max_chns; 873 bool process_sub_channels = false; 874 875 stor_device = get_out_stor_device(device); 876 if (!stor_device) 877 return -ENODEV; 878 879 request = &stor_device->init_request; 880 vstor_packet = &request->vstor_packet; 881 882 /* 883 * Now, initiate the vsc/vsp initialization protocol on the open 884 * channel 885 */ 886 memset(request, 0, sizeof(struct storvsc_cmd_request)); 887 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION; 888 ret = storvsc_execute_vstor_op(device, request, true); 889 if (ret) 890 return ret; 891 /* 892 * Query host supported protocol version. 893 */ 894 895 for (i = 0; i < ARRAY_SIZE(vmstor_protocols); i++) { 896 /* reuse the packet for version range supported */ 897 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 898 vstor_packet->operation = 899 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION; 900 901 vstor_packet->version.major_minor = 902 vmstor_protocols[i].protocol_version; 903 904 /* 905 * The revision number is only used in Windows; set it to 0. 906 */ 907 vstor_packet->version.revision = 0; 908 ret = storvsc_execute_vstor_op(device, request, false); 909 if (ret != 0) 910 return ret; 911 912 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO) 913 return -EINVAL; 914 915 if (vstor_packet->status == 0) { 916 vmstor_proto_version = 917 vmstor_protocols[i].protocol_version; 918 919 sense_buffer_size = 920 vmstor_protocols[i].sense_buffer_size; 921 922 vmscsi_size_delta = 923 vmstor_protocols[i].vmscsi_size_delta; 924 925 break; 926 } 927 } 928 929 if (vstor_packet->status != 0) 930 return -EINVAL; 931 932 933 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 934 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES; 935 ret = storvsc_execute_vstor_op(device, request, true); 936 if (ret != 0) 937 return ret; 938 939 /* 940 * Check to see if multi-channel support is there. 941 * Hosts that implement protocol version of 5.1 and above 942 * support multi-channel. 943 */ 944 max_chns = vstor_packet->storage_channel_properties.max_channel_cnt; 945 946 /* 947 * Allocate state to manage the sub-channels. 948 * We allocate an array based on the numbers of possible CPUs 949 * (Hyper-V does not support cpu online/offline). 950 * This Array will be sparseley populated with unique 951 * channels - primary + sub-channels. 952 * We will however populate all the slots to evenly distribute 953 * the load. 954 */ 955 stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *), 956 GFP_KERNEL); 957 if (stor_device->stor_chns == NULL) 958 return -ENOMEM; 959 960 device->channel->change_target_cpu_callback = storvsc_change_target_cpu; 961 962 stor_device->stor_chns[device->channel->target_cpu] = device->channel; 963 cpumask_set_cpu(device->channel->target_cpu, 964 &stor_device->alloced_cpus); 965 966 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN8) { 967 if (vstor_packet->storage_channel_properties.flags & 968 STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL) 969 process_sub_channels = true; 970 } 971 stor_device->max_transfer_bytes = 972 vstor_packet->storage_channel_properties.max_transfer_bytes; 973 974 if (!is_fc) 975 goto done; 976 977 /* 978 * For FC devices retrieve FC HBA data. 979 */ 980 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 981 vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA; 982 ret = storvsc_execute_vstor_op(device, request, true); 983 if (ret != 0) 984 return ret; 985 986 /* 987 * Cache the currently active port and node ww names. 988 */ 989 cache_wwn(stor_device, vstor_packet); 990 991 done: 992 993 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 994 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION; 995 ret = storvsc_execute_vstor_op(device, request, true); 996 if (ret != 0) 997 return ret; 998 999 if (process_sub_channels) 1000 handle_multichannel_storage(device, max_chns); 1001 1002 return ret; 1003 } 1004 1005 static void storvsc_handle_error(struct vmscsi_request *vm_srb, 1006 struct scsi_cmnd *scmnd, 1007 struct Scsi_Host *host, 1008 u8 asc, u8 ascq) 1009 { 1010 struct storvsc_scan_work *wrk; 1011 void (*process_err_fn)(struct work_struct *work); 1012 struct hv_host_device *host_dev = shost_priv(host); 1013 bool do_work = false; 1014 1015 switch (SRB_STATUS(vm_srb->srb_status)) { 1016 case SRB_STATUS_ERROR: 1017 /* 1018 * Let upper layer deal with error when 1019 * sense message is present. 1020 */ 1021 1022 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID) 1023 break; 1024 /* 1025 * If there is an error; offline the device since all 1026 * error recovery strategies would have already been 1027 * deployed on the host side. However, if the command 1028 * were a pass-through command deal with it appropriately. 1029 */ 1030 switch (scmnd->cmnd[0]) { 1031 case ATA_16: 1032 case ATA_12: 1033 set_host_byte(scmnd, DID_PASSTHROUGH); 1034 break; 1035 /* 1036 * On Some Windows hosts TEST_UNIT_READY command can return 1037 * SRB_STATUS_ERROR, let the upper level code deal with it 1038 * based on the sense information. 1039 */ 1040 case TEST_UNIT_READY: 1041 break; 1042 default: 1043 set_host_byte(scmnd, DID_ERROR); 1044 } 1045 break; 1046 case SRB_STATUS_INVALID_LUN: 1047 set_host_byte(scmnd, DID_NO_CONNECT); 1048 do_work = true; 1049 process_err_fn = storvsc_remove_lun; 1050 break; 1051 case SRB_STATUS_ABORTED: 1052 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID && 1053 (asc == 0x2a) && (ascq == 0x9)) { 1054 do_work = true; 1055 process_err_fn = storvsc_device_scan; 1056 /* 1057 * Retry the I/O that triggered this. 1058 */ 1059 set_host_byte(scmnd, DID_REQUEUE); 1060 } 1061 break; 1062 } 1063 1064 if (!do_work) 1065 return; 1066 1067 /* 1068 * We need to schedule work to process this error; schedule it. 1069 */ 1070 wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC); 1071 if (!wrk) { 1072 set_host_byte(scmnd, DID_TARGET_FAILURE); 1073 return; 1074 } 1075 1076 wrk->host = host; 1077 wrk->lun = vm_srb->lun; 1078 wrk->tgt_id = vm_srb->target_id; 1079 INIT_WORK(&wrk->work, process_err_fn); 1080 queue_work(host_dev->handle_error_wq, &wrk->work); 1081 } 1082 1083 1084 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request, 1085 struct storvsc_device *stor_dev) 1086 { 1087 struct scsi_cmnd *scmnd = cmd_request->cmd; 1088 struct scsi_sense_hdr sense_hdr; 1089 struct vmscsi_request *vm_srb; 1090 u32 data_transfer_length; 1091 struct Scsi_Host *host; 1092 u32 payload_sz = cmd_request->payload_sz; 1093 void *payload = cmd_request->payload; 1094 1095 host = stor_dev->host; 1096 1097 vm_srb = &cmd_request->vstor_packet.vm_srb; 1098 data_transfer_length = vm_srb->data_transfer_length; 1099 1100 scmnd->result = vm_srb->scsi_status; 1101 1102 if (scmnd->result) { 1103 if (scsi_normalize_sense(scmnd->sense_buffer, 1104 SCSI_SENSE_BUFFERSIZE, &sense_hdr) && 1105 !(sense_hdr.sense_key == NOT_READY && 1106 sense_hdr.asc == 0x03A) && 1107 do_logging(STORVSC_LOGGING_ERROR)) 1108 scsi_print_sense_hdr(scmnd->device, "storvsc", 1109 &sense_hdr); 1110 } 1111 1112 if (vm_srb->srb_status != SRB_STATUS_SUCCESS) { 1113 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc, 1114 sense_hdr.ascq); 1115 /* 1116 * The Windows driver set data_transfer_length on 1117 * SRB_STATUS_DATA_OVERRUN. On other errors, this value 1118 * is untouched. In these cases we set it to 0. 1119 */ 1120 if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN) 1121 data_transfer_length = 0; 1122 } 1123 1124 /* Validate data_transfer_length (from Hyper-V) */ 1125 if (data_transfer_length > cmd_request->payload->range.len) 1126 data_transfer_length = cmd_request->payload->range.len; 1127 1128 scsi_set_resid(scmnd, 1129 cmd_request->payload->range.len - data_transfer_length); 1130 1131 scmnd->scsi_done(scmnd); 1132 1133 if (payload_sz > 1134 sizeof(struct vmbus_channel_packet_multipage_buffer)) 1135 kfree(payload); 1136 } 1137 1138 static void storvsc_on_io_completion(struct storvsc_device *stor_device, 1139 struct vstor_packet *vstor_packet, 1140 struct storvsc_cmd_request *request) 1141 { 1142 struct vstor_packet *stor_pkt; 1143 struct hv_device *device = stor_device->device; 1144 1145 stor_pkt = &request->vstor_packet; 1146 1147 /* 1148 * The current SCSI handling on the host side does 1149 * not correctly handle: 1150 * INQUIRY command with page code parameter set to 0x80 1151 * MODE_SENSE command with cmd[2] == 0x1c 1152 * 1153 * Setup srb and scsi status so this won't be fatal. 1154 * We do this so we can distinguish truly fatal failues 1155 * (srb status == 0x4) and off-line the device in that case. 1156 */ 1157 1158 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) || 1159 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) { 1160 vstor_packet->vm_srb.scsi_status = 0; 1161 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS; 1162 } 1163 1164 1165 /* Copy over the status...etc */ 1166 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status; 1167 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status; 1168 1169 /* Validate sense_info_length (from Hyper-V) */ 1170 if (vstor_packet->vm_srb.sense_info_length > sense_buffer_size) 1171 vstor_packet->vm_srb.sense_info_length = sense_buffer_size; 1172 1173 stor_pkt->vm_srb.sense_info_length = 1174 vstor_packet->vm_srb.sense_info_length; 1175 1176 if (vstor_packet->vm_srb.scsi_status != 0 || 1177 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS) 1178 storvsc_log(device, STORVSC_LOGGING_WARN, 1179 "cmd 0x%x scsi status 0x%x srb status 0x%x\n", 1180 stor_pkt->vm_srb.cdb[0], 1181 vstor_packet->vm_srb.scsi_status, 1182 vstor_packet->vm_srb.srb_status); 1183 1184 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) { 1185 /* CHECK_CONDITION */ 1186 if (vstor_packet->vm_srb.srb_status & 1187 SRB_STATUS_AUTOSENSE_VALID) { 1188 /* autosense data available */ 1189 1190 storvsc_log(device, STORVSC_LOGGING_WARN, 1191 "stor pkt %p autosense data valid - len %d\n", 1192 request, vstor_packet->vm_srb.sense_info_length); 1193 1194 memcpy(request->cmd->sense_buffer, 1195 vstor_packet->vm_srb.sense_data, 1196 vstor_packet->vm_srb.sense_info_length); 1197 1198 } 1199 } 1200 1201 stor_pkt->vm_srb.data_transfer_length = 1202 vstor_packet->vm_srb.data_transfer_length; 1203 1204 storvsc_command_completion(request, stor_device); 1205 1206 if (atomic_dec_and_test(&stor_device->num_outstanding_req) && 1207 stor_device->drain_notify) 1208 wake_up(&stor_device->waiting_to_drain); 1209 1210 1211 } 1212 1213 static void storvsc_on_receive(struct storvsc_device *stor_device, 1214 struct vstor_packet *vstor_packet, 1215 struct storvsc_cmd_request *request) 1216 { 1217 struct hv_host_device *host_dev; 1218 switch (vstor_packet->operation) { 1219 case VSTOR_OPERATION_COMPLETE_IO: 1220 storvsc_on_io_completion(stor_device, vstor_packet, request); 1221 break; 1222 1223 case VSTOR_OPERATION_REMOVE_DEVICE: 1224 case VSTOR_OPERATION_ENUMERATE_BUS: 1225 host_dev = shost_priv(stor_device->host); 1226 queue_work( 1227 host_dev->handle_error_wq, &host_dev->host_scan_work); 1228 break; 1229 1230 case VSTOR_OPERATION_FCHBA_DATA: 1231 cache_wwn(stor_device, vstor_packet); 1232 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1233 fc_host_node_name(stor_device->host) = stor_device->node_name; 1234 fc_host_port_name(stor_device->host) = stor_device->port_name; 1235 #endif 1236 break; 1237 default: 1238 break; 1239 } 1240 } 1241 1242 static void storvsc_on_channel_callback(void *context) 1243 { 1244 struct vmbus_channel *channel = (struct vmbus_channel *)context; 1245 const struct vmpacket_descriptor *desc; 1246 struct hv_device *device; 1247 struct storvsc_device *stor_device; 1248 1249 if (channel->primary_channel != NULL) 1250 device = channel->primary_channel->device_obj; 1251 else 1252 device = channel->device_obj; 1253 1254 stor_device = get_in_stor_device(device); 1255 if (!stor_device) 1256 return; 1257 1258 foreach_vmbus_pkt(desc, channel) { 1259 void *packet = hv_pkt_data(desc); 1260 struct storvsc_cmd_request *request; 1261 u64 cmd_rqst; 1262 1263 cmd_rqst = vmbus_request_addr(&channel->requestor, 1264 desc->trans_id); 1265 if (cmd_rqst == VMBUS_RQST_ERROR) { 1266 dev_err(&device->device, 1267 "Incorrect transaction id\n"); 1268 continue; 1269 } 1270 1271 request = (struct storvsc_cmd_request *)(unsigned long)cmd_rqst; 1272 1273 if (request == &stor_device->init_request || 1274 request == &stor_device->reset_request) { 1275 memcpy(&request->vstor_packet, packet, 1276 (sizeof(struct vstor_packet) - vmscsi_size_delta)); 1277 complete(&request->wait_event); 1278 } else { 1279 storvsc_on_receive(stor_device, packet, request); 1280 } 1281 } 1282 } 1283 1284 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size, 1285 bool is_fc) 1286 { 1287 struct vmstorage_channel_properties props; 1288 int ret; 1289 1290 memset(&props, 0, sizeof(struct vmstorage_channel_properties)); 1291 1292 device->channel->max_pkt_size = STORVSC_MAX_PKT_SIZE; 1293 /* 1294 * The size of vmbus_requestor is an upper bound on the number of requests 1295 * that can be in-progress at any one time across all channels. 1296 */ 1297 device->channel->rqstor_size = scsi_driver.can_queue; 1298 1299 ret = vmbus_open(device->channel, 1300 ring_size, 1301 ring_size, 1302 (void *)&props, 1303 sizeof(struct vmstorage_channel_properties), 1304 storvsc_on_channel_callback, device->channel); 1305 1306 if (ret != 0) 1307 return ret; 1308 1309 ret = storvsc_channel_init(device, is_fc); 1310 1311 return ret; 1312 } 1313 1314 static int storvsc_dev_remove(struct hv_device *device) 1315 { 1316 struct storvsc_device *stor_device; 1317 1318 stor_device = hv_get_drvdata(device); 1319 1320 stor_device->destroy = true; 1321 1322 /* Make sure flag is set before waiting */ 1323 wmb(); 1324 1325 /* 1326 * At this point, all outbound traffic should be disable. We 1327 * only allow inbound traffic (responses) to proceed so that 1328 * outstanding requests can be completed. 1329 */ 1330 1331 storvsc_wait_to_drain(stor_device); 1332 1333 /* 1334 * Since we have already drained, we don't need to busy wait 1335 * as was done in final_release_stor_device() 1336 * Note that we cannot set the ext pointer to NULL until 1337 * we have drained - to drain the outgoing packets, we need to 1338 * allow incoming packets. 1339 */ 1340 hv_set_drvdata(device, NULL); 1341 1342 /* Close the channel */ 1343 vmbus_close(device->channel); 1344 1345 kfree(stor_device->stor_chns); 1346 kfree(stor_device); 1347 return 0; 1348 } 1349 1350 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device, 1351 u16 q_num) 1352 { 1353 u16 slot = 0; 1354 u16 hash_qnum; 1355 const struct cpumask *node_mask; 1356 int num_channels, tgt_cpu; 1357 1358 if (stor_device->num_sc == 0) { 1359 stor_device->stor_chns[q_num] = stor_device->device->channel; 1360 return stor_device->device->channel; 1361 } 1362 1363 /* 1364 * Our channel array is sparsley populated and we 1365 * initiated I/O on a processor/hw-q that does not 1366 * currently have a designated channel. Fix this. 1367 * The strategy is simple: 1368 * I. Ensure NUMA locality 1369 * II. Distribute evenly (best effort) 1370 */ 1371 1372 node_mask = cpumask_of_node(cpu_to_node(q_num)); 1373 1374 num_channels = 0; 1375 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) { 1376 if (cpumask_test_cpu(tgt_cpu, node_mask)) 1377 num_channels++; 1378 } 1379 if (num_channels == 0) { 1380 stor_device->stor_chns[q_num] = stor_device->device->channel; 1381 return stor_device->device->channel; 1382 } 1383 1384 hash_qnum = q_num; 1385 while (hash_qnum >= num_channels) 1386 hash_qnum -= num_channels; 1387 1388 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) { 1389 if (!cpumask_test_cpu(tgt_cpu, node_mask)) 1390 continue; 1391 if (slot == hash_qnum) 1392 break; 1393 slot++; 1394 } 1395 1396 stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu]; 1397 1398 return stor_device->stor_chns[q_num]; 1399 } 1400 1401 1402 static int storvsc_do_io(struct hv_device *device, 1403 struct storvsc_cmd_request *request, u16 q_num) 1404 { 1405 struct storvsc_device *stor_device; 1406 struct vstor_packet *vstor_packet; 1407 struct vmbus_channel *outgoing_channel, *channel; 1408 unsigned long flags; 1409 int ret = 0; 1410 const struct cpumask *node_mask; 1411 int tgt_cpu; 1412 1413 vstor_packet = &request->vstor_packet; 1414 stor_device = get_out_stor_device(device); 1415 1416 if (!stor_device) 1417 return -ENODEV; 1418 1419 1420 request->device = device; 1421 /* 1422 * Select an appropriate channel to send the request out. 1423 */ 1424 /* See storvsc_change_target_cpu(). */ 1425 outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]); 1426 if (outgoing_channel != NULL) { 1427 if (outgoing_channel->target_cpu == q_num) { 1428 /* 1429 * Ideally, we want to pick a different channel if 1430 * available on the same NUMA node. 1431 */ 1432 node_mask = cpumask_of_node(cpu_to_node(q_num)); 1433 for_each_cpu_wrap(tgt_cpu, 1434 &stor_device->alloced_cpus, q_num + 1) { 1435 if (!cpumask_test_cpu(tgt_cpu, node_mask)) 1436 continue; 1437 if (tgt_cpu == q_num) 1438 continue; 1439 channel = READ_ONCE( 1440 stor_device->stor_chns[tgt_cpu]); 1441 if (channel == NULL) 1442 continue; 1443 if (hv_get_avail_to_write_percent( 1444 &channel->outbound) 1445 > ring_avail_percent_lowater) { 1446 outgoing_channel = channel; 1447 goto found_channel; 1448 } 1449 } 1450 1451 /* 1452 * All the other channels on the same NUMA node are 1453 * busy. Try to use the channel on the current CPU 1454 */ 1455 if (hv_get_avail_to_write_percent( 1456 &outgoing_channel->outbound) 1457 > ring_avail_percent_lowater) 1458 goto found_channel; 1459 1460 /* 1461 * If we reach here, all the channels on the current 1462 * NUMA node are busy. Try to find a channel in 1463 * other NUMA nodes 1464 */ 1465 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) { 1466 if (cpumask_test_cpu(tgt_cpu, node_mask)) 1467 continue; 1468 channel = READ_ONCE( 1469 stor_device->stor_chns[tgt_cpu]); 1470 if (channel == NULL) 1471 continue; 1472 if (hv_get_avail_to_write_percent( 1473 &channel->outbound) 1474 > ring_avail_percent_lowater) { 1475 outgoing_channel = channel; 1476 goto found_channel; 1477 } 1478 } 1479 } 1480 } else { 1481 spin_lock_irqsave(&stor_device->lock, flags); 1482 outgoing_channel = stor_device->stor_chns[q_num]; 1483 if (outgoing_channel != NULL) { 1484 spin_unlock_irqrestore(&stor_device->lock, flags); 1485 goto found_channel; 1486 } 1487 outgoing_channel = get_og_chn(stor_device, q_num); 1488 spin_unlock_irqrestore(&stor_device->lock, flags); 1489 } 1490 1491 found_channel: 1492 vstor_packet->flags |= REQUEST_COMPLETION_FLAG; 1493 1494 vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) - 1495 vmscsi_size_delta); 1496 1497 1498 vstor_packet->vm_srb.sense_info_length = sense_buffer_size; 1499 1500 1501 vstor_packet->vm_srb.data_transfer_length = 1502 request->payload->range.len; 1503 1504 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB; 1505 1506 if (request->payload->range.len) { 1507 1508 ret = vmbus_sendpacket_mpb_desc(outgoing_channel, 1509 request->payload, request->payload_sz, 1510 vstor_packet, 1511 (sizeof(struct vstor_packet) - 1512 vmscsi_size_delta), 1513 (unsigned long)request); 1514 } else { 1515 ret = vmbus_sendpacket(outgoing_channel, vstor_packet, 1516 (sizeof(struct vstor_packet) - 1517 vmscsi_size_delta), 1518 (unsigned long)request, 1519 VM_PKT_DATA_INBAND, 1520 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 1521 } 1522 1523 if (ret != 0) 1524 return ret; 1525 1526 atomic_inc(&stor_device->num_outstanding_req); 1527 1528 return ret; 1529 } 1530 1531 static int storvsc_device_alloc(struct scsi_device *sdevice) 1532 { 1533 /* 1534 * Set blist flag to permit the reading of the VPD pages even when 1535 * the target may claim SPC-2 compliance. MSFT targets currently 1536 * claim SPC-2 compliance while they implement post SPC-2 features. 1537 * With this flag we can correctly handle WRITE_SAME_16 issues. 1538 * 1539 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but 1540 * still supports REPORT LUN. 1541 */ 1542 sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES; 1543 1544 return 0; 1545 } 1546 1547 static int storvsc_device_configure(struct scsi_device *sdevice) 1548 { 1549 blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ)); 1550 1551 sdevice->no_write_same = 1; 1552 1553 /* 1554 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3 1555 * if the device is a MSFT virtual device. If the host is 1556 * WIN10 or newer, allow write_same. 1557 */ 1558 if (!strncmp(sdevice->vendor, "Msft", 4)) { 1559 switch (vmstor_proto_version) { 1560 case VMSTOR_PROTO_VERSION_WIN8: 1561 case VMSTOR_PROTO_VERSION_WIN8_1: 1562 sdevice->scsi_level = SCSI_SPC_3; 1563 break; 1564 } 1565 1566 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10) 1567 sdevice->no_write_same = 0; 1568 } 1569 1570 return 0; 1571 } 1572 1573 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev, 1574 sector_t capacity, int *info) 1575 { 1576 sector_t nsect = capacity; 1577 sector_t cylinders = nsect; 1578 int heads, sectors_pt; 1579 1580 /* 1581 * We are making up these values; let us keep it simple. 1582 */ 1583 heads = 0xff; 1584 sectors_pt = 0x3f; /* Sectors per track */ 1585 sector_div(cylinders, heads * sectors_pt); 1586 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect) 1587 cylinders = 0xffff; 1588 1589 info[0] = heads; 1590 info[1] = sectors_pt; 1591 info[2] = (int)cylinders; 1592 1593 return 0; 1594 } 1595 1596 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd) 1597 { 1598 struct hv_host_device *host_dev = shost_priv(scmnd->device->host); 1599 struct hv_device *device = host_dev->dev; 1600 1601 struct storvsc_device *stor_device; 1602 struct storvsc_cmd_request *request; 1603 struct vstor_packet *vstor_packet; 1604 int ret, t; 1605 1606 stor_device = get_out_stor_device(device); 1607 if (!stor_device) 1608 return FAILED; 1609 1610 request = &stor_device->reset_request; 1611 vstor_packet = &request->vstor_packet; 1612 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 1613 1614 init_completion(&request->wait_event); 1615 1616 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS; 1617 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 1618 vstor_packet->vm_srb.path_id = stor_device->path_id; 1619 1620 ret = vmbus_sendpacket(device->channel, vstor_packet, 1621 (sizeof(struct vstor_packet) - 1622 vmscsi_size_delta), 1623 (unsigned long)&stor_device->reset_request, 1624 VM_PKT_DATA_INBAND, 1625 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 1626 if (ret != 0) 1627 return FAILED; 1628 1629 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 1630 if (t == 0) 1631 return TIMEOUT_ERROR; 1632 1633 1634 /* 1635 * At this point, all outstanding requests in the adapter 1636 * should have been flushed out and return to us 1637 * There is a potential race here where the host may be in 1638 * the process of responding when we return from here. 1639 * Just wait for all in-transit packets to be accounted for 1640 * before we return from here. 1641 */ 1642 storvsc_wait_to_drain(stor_device); 1643 1644 return SUCCESS; 1645 } 1646 1647 /* 1648 * The host guarantees to respond to each command, although I/O latencies might 1649 * be unbounded on Azure. Reset the timer unconditionally to give the host a 1650 * chance to perform EH. 1651 */ 1652 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd) 1653 { 1654 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1655 if (scmnd->device->host->transportt == fc_transport_template) 1656 return fc_eh_timed_out(scmnd); 1657 #endif 1658 return BLK_EH_RESET_TIMER; 1659 } 1660 1661 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd) 1662 { 1663 bool allowed = true; 1664 u8 scsi_op = scmnd->cmnd[0]; 1665 1666 switch (scsi_op) { 1667 /* the host does not handle WRITE_SAME, log accident usage */ 1668 case WRITE_SAME: 1669 /* 1670 * smartd sends this command and the host does not handle 1671 * this. So, don't send it. 1672 */ 1673 case SET_WINDOW: 1674 scmnd->result = ILLEGAL_REQUEST << 16; 1675 allowed = false; 1676 break; 1677 default: 1678 break; 1679 } 1680 return allowed; 1681 } 1682 1683 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd) 1684 { 1685 int ret; 1686 struct hv_host_device *host_dev = shost_priv(host); 1687 struct hv_device *dev = host_dev->dev; 1688 struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd); 1689 int i; 1690 struct scatterlist *sgl; 1691 unsigned int sg_count = 0; 1692 struct vmscsi_request *vm_srb; 1693 struct scatterlist *cur_sgl; 1694 struct vmbus_packet_mpb_array *payload; 1695 u32 payload_sz; 1696 u32 length; 1697 1698 if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) { 1699 /* 1700 * On legacy hosts filter unimplemented commands. 1701 * Future hosts are expected to correctly handle 1702 * unsupported commands. Furthermore, it is 1703 * possible that some of the currently 1704 * unsupported commands maybe supported in 1705 * future versions of the host. 1706 */ 1707 if (!storvsc_scsi_cmd_ok(scmnd)) { 1708 scmnd->scsi_done(scmnd); 1709 return 0; 1710 } 1711 } 1712 1713 /* Setup the cmd request */ 1714 cmd_request->cmd = scmnd; 1715 1716 memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet)); 1717 vm_srb = &cmd_request->vstor_packet.vm_srb; 1718 vm_srb->win8_extension.time_out_value = 60; 1719 1720 vm_srb->win8_extension.srb_flags |= 1721 SRB_FLAGS_DISABLE_SYNCH_TRANSFER; 1722 1723 if (scmnd->device->tagged_supported) { 1724 vm_srb->win8_extension.srb_flags |= 1725 (SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE); 1726 vm_srb->win8_extension.queue_tag = SP_UNTAGGED; 1727 vm_srb->win8_extension.queue_action = SRB_SIMPLE_TAG_REQUEST; 1728 } 1729 1730 /* Build the SRB */ 1731 switch (scmnd->sc_data_direction) { 1732 case DMA_TO_DEVICE: 1733 vm_srb->data_in = WRITE_TYPE; 1734 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT; 1735 break; 1736 case DMA_FROM_DEVICE: 1737 vm_srb->data_in = READ_TYPE; 1738 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN; 1739 break; 1740 case DMA_NONE: 1741 vm_srb->data_in = UNKNOWN_TYPE; 1742 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER; 1743 break; 1744 default: 1745 /* 1746 * This is DMA_BIDIRECTIONAL or something else we are never 1747 * supposed to see here. 1748 */ 1749 WARN(1, "Unexpected data direction: %d\n", 1750 scmnd->sc_data_direction); 1751 return -EINVAL; 1752 } 1753 1754 1755 vm_srb->port_number = host_dev->port; 1756 vm_srb->path_id = scmnd->device->channel; 1757 vm_srb->target_id = scmnd->device->id; 1758 vm_srb->lun = scmnd->device->lun; 1759 1760 vm_srb->cdb_length = scmnd->cmd_len; 1761 1762 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length); 1763 1764 sgl = (struct scatterlist *)scsi_sglist(scmnd); 1765 sg_count = scsi_sg_count(scmnd); 1766 1767 length = scsi_bufflen(scmnd); 1768 payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb; 1769 payload_sz = sizeof(cmd_request->mpb); 1770 1771 if (sg_count) { 1772 unsigned int hvpgoff = 0; 1773 unsigned long offset_in_hvpg = sgl->offset & ~HV_HYP_PAGE_MASK; 1774 unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length); 1775 u64 hvpfn; 1776 1777 if (hvpg_count > MAX_PAGE_BUFFER_COUNT) { 1778 1779 payload_sz = (hvpg_count * sizeof(u64) + 1780 sizeof(struct vmbus_packet_mpb_array)); 1781 payload = kzalloc(payload_sz, GFP_ATOMIC); 1782 if (!payload) 1783 return SCSI_MLQUEUE_DEVICE_BUSY; 1784 } 1785 1786 /* 1787 * sgl is a list of PAGEs, and payload->range.pfn_array 1788 * expects the page number in the unit of HV_HYP_PAGE_SIZE (the 1789 * page size that Hyper-V uses, so here we need to divide PAGEs 1790 * into HV_HYP_PAGE in case that PAGE_SIZE > HV_HYP_PAGE_SIZE. 1791 * Besides, payload->range.offset should be the offset in one 1792 * HV_HYP_PAGE. 1793 */ 1794 payload->range.len = length; 1795 payload->range.offset = offset_in_hvpg; 1796 hvpgoff = sgl->offset >> HV_HYP_PAGE_SHIFT; 1797 1798 cur_sgl = sgl; 1799 for (i = 0; i < hvpg_count; i++) { 1800 /* 1801 * 'i' is the index of hv pages in the payload and 1802 * 'hvpgoff' is the offset (in hv pages) of the first 1803 * hv page in the the first page. The relationship 1804 * between the sum of 'i' and 'hvpgoff' and the offset 1805 * (in hv pages) in a payload page ('hvpgoff_in_page') 1806 * is as follow: 1807 * 1808 * |------------------ PAGE -------------------| 1809 * | NR_HV_HYP_PAGES_IN_PAGE hvpgs in total | 1810 * |hvpg|hvpg| ... |hvpg|... |hvpg| 1811 * ^ ^ ^ ^ 1812 * +-hvpgoff-+ +-hvpgoff_in_page-+ 1813 * ^ | 1814 * +--------------------- i ---------------------------+ 1815 */ 1816 unsigned int hvpgoff_in_page = 1817 (i + hvpgoff) % NR_HV_HYP_PAGES_IN_PAGE; 1818 1819 /* 1820 * Two cases that we need to fetch a page: 1821 * 1) i == 0, the first step or 1822 * 2) hvpgoff_in_page == 0, when we reach the boundary 1823 * of a page. 1824 */ 1825 if (hvpgoff_in_page == 0 || i == 0) { 1826 hvpfn = page_to_hvpfn(sg_page(cur_sgl)); 1827 cur_sgl = sg_next(cur_sgl); 1828 } 1829 1830 payload->range.pfn_array[i] = hvpfn + hvpgoff_in_page; 1831 } 1832 } 1833 1834 cmd_request->payload = payload; 1835 cmd_request->payload_sz = payload_sz; 1836 1837 /* Invokes the vsc to start an IO */ 1838 ret = storvsc_do_io(dev, cmd_request, get_cpu()); 1839 put_cpu(); 1840 1841 if (ret == -EAGAIN) { 1842 if (payload_sz > sizeof(cmd_request->mpb)) 1843 kfree(payload); 1844 /* no more space */ 1845 return SCSI_MLQUEUE_DEVICE_BUSY; 1846 } 1847 1848 return 0; 1849 } 1850 1851 static struct scsi_host_template scsi_driver = { 1852 .module = THIS_MODULE, 1853 .name = "storvsc_host_t", 1854 .cmd_size = sizeof(struct storvsc_cmd_request), 1855 .bios_param = storvsc_get_chs, 1856 .queuecommand = storvsc_queuecommand, 1857 .eh_host_reset_handler = storvsc_host_reset_handler, 1858 .proc_name = "storvsc_host", 1859 .eh_timed_out = storvsc_eh_timed_out, 1860 .slave_alloc = storvsc_device_alloc, 1861 .slave_configure = storvsc_device_configure, 1862 .cmd_per_lun = 2048, 1863 .this_id = -1, 1864 /* Make sure we dont get a sg segment crosses a page boundary */ 1865 .dma_boundary = PAGE_SIZE-1, 1866 /* Ensure there are no gaps in presented sgls */ 1867 .virt_boundary_mask = PAGE_SIZE-1, 1868 .no_write_same = 1, 1869 .track_queue_depth = 1, 1870 .change_queue_depth = storvsc_change_queue_depth, 1871 }; 1872 1873 enum { 1874 SCSI_GUID, 1875 IDE_GUID, 1876 SFC_GUID, 1877 }; 1878 1879 static const struct hv_vmbus_device_id id_table[] = { 1880 /* SCSI guid */ 1881 { HV_SCSI_GUID, 1882 .driver_data = SCSI_GUID 1883 }, 1884 /* IDE guid */ 1885 { HV_IDE_GUID, 1886 .driver_data = IDE_GUID 1887 }, 1888 /* Fibre Channel GUID */ 1889 { 1890 HV_SYNTHFC_GUID, 1891 .driver_data = SFC_GUID 1892 }, 1893 { }, 1894 }; 1895 1896 MODULE_DEVICE_TABLE(vmbus, id_table); 1897 1898 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID }; 1899 1900 static bool hv_dev_is_fc(struct hv_device *hv_dev) 1901 { 1902 return guid_equal(&fc_guid.guid, &hv_dev->dev_type); 1903 } 1904 1905 static int storvsc_probe(struct hv_device *device, 1906 const struct hv_vmbus_device_id *dev_id) 1907 { 1908 int ret; 1909 int num_cpus = num_online_cpus(); 1910 struct Scsi_Host *host; 1911 struct hv_host_device *host_dev; 1912 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false); 1913 bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false); 1914 int target = 0; 1915 struct storvsc_device *stor_device; 1916 int max_luns_per_target; 1917 int max_targets; 1918 int max_channels; 1919 int max_sub_channels = 0; 1920 1921 /* 1922 * Based on the windows host we are running on, 1923 * set state to properly communicate with the host. 1924 */ 1925 1926 if (vmbus_proto_version < VERSION_WIN8) { 1927 max_luns_per_target = STORVSC_IDE_MAX_LUNS_PER_TARGET; 1928 max_targets = STORVSC_IDE_MAX_TARGETS; 1929 max_channels = STORVSC_IDE_MAX_CHANNELS; 1930 } else { 1931 max_luns_per_target = STORVSC_MAX_LUNS_PER_TARGET; 1932 max_targets = STORVSC_MAX_TARGETS; 1933 max_channels = STORVSC_MAX_CHANNELS; 1934 /* 1935 * On Windows8 and above, we support sub-channels for storage 1936 * on SCSI and FC controllers. 1937 * The number of sub-channels offerred is based on the number of 1938 * VCPUs in the guest. 1939 */ 1940 if (!dev_is_ide) 1941 max_sub_channels = 1942 (num_cpus - 1) / storvsc_vcpus_per_sub_channel; 1943 } 1944 1945 scsi_driver.can_queue = max_outstanding_req_per_channel * 1946 (max_sub_channels + 1) * 1947 (100 - ring_avail_percent_lowater) / 100; 1948 1949 host = scsi_host_alloc(&scsi_driver, 1950 sizeof(struct hv_host_device)); 1951 if (!host) 1952 return -ENOMEM; 1953 1954 host_dev = shost_priv(host); 1955 memset(host_dev, 0, sizeof(struct hv_host_device)); 1956 1957 host_dev->port = host->host_no; 1958 host_dev->dev = device; 1959 host_dev->host = host; 1960 1961 1962 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL); 1963 if (!stor_device) { 1964 ret = -ENOMEM; 1965 goto err_out0; 1966 } 1967 1968 stor_device->destroy = false; 1969 init_waitqueue_head(&stor_device->waiting_to_drain); 1970 stor_device->device = device; 1971 stor_device->host = host; 1972 spin_lock_init(&stor_device->lock); 1973 hv_set_drvdata(device, stor_device); 1974 1975 stor_device->port_number = host->host_no; 1976 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc); 1977 if (ret) 1978 goto err_out1; 1979 1980 host_dev->path = stor_device->path_id; 1981 host_dev->target = stor_device->target_id; 1982 1983 switch (dev_id->driver_data) { 1984 case SFC_GUID: 1985 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET; 1986 host->max_id = STORVSC_FC_MAX_TARGETS; 1987 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1; 1988 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1989 host->transportt = fc_transport_template; 1990 #endif 1991 break; 1992 1993 case SCSI_GUID: 1994 host->max_lun = max_luns_per_target; 1995 host->max_id = max_targets; 1996 host->max_channel = max_channels - 1; 1997 break; 1998 1999 default: 2000 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET; 2001 host->max_id = STORVSC_IDE_MAX_TARGETS; 2002 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1; 2003 break; 2004 } 2005 /* max cmd length */ 2006 host->max_cmd_len = STORVSC_MAX_CMD_LEN; 2007 2008 /* 2009 * set the table size based on the info we got 2010 * from the host. 2011 */ 2012 host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT); 2013 /* 2014 * For non-IDE disks, the host supports multiple channels. 2015 * Set the number of HW queues we are supporting. 2016 */ 2017 if (!dev_is_ide) 2018 host->nr_hw_queues = num_present_cpus(); 2019 2020 /* 2021 * Set the error handler work queue. 2022 */ 2023 host_dev->handle_error_wq = 2024 alloc_ordered_workqueue("storvsc_error_wq_%d", 2025 WQ_MEM_RECLAIM, 2026 host->host_no); 2027 if (!host_dev->handle_error_wq) { 2028 ret = -ENOMEM; 2029 goto err_out2; 2030 } 2031 INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan); 2032 /* Register the HBA and start the scsi bus scan */ 2033 ret = scsi_add_host(host, &device->device); 2034 if (ret != 0) 2035 goto err_out3; 2036 2037 if (!dev_is_ide) { 2038 scsi_scan_host(host); 2039 } else { 2040 target = (device->dev_instance.b[5] << 8 | 2041 device->dev_instance.b[4]); 2042 ret = scsi_add_device(host, 0, target, 0); 2043 if (ret) 2044 goto err_out4; 2045 } 2046 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 2047 if (host->transportt == fc_transport_template) { 2048 struct fc_rport_identifiers ids = { 2049 .roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR, 2050 }; 2051 2052 fc_host_node_name(host) = stor_device->node_name; 2053 fc_host_port_name(host) = stor_device->port_name; 2054 stor_device->rport = fc_remote_port_add(host, 0, &ids); 2055 if (!stor_device->rport) { 2056 ret = -ENOMEM; 2057 goto err_out4; 2058 } 2059 } 2060 #endif 2061 return 0; 2062 2063 err_out4: 2064 scsi_remove_host(host); 2065 2066 err_out3: 2067 destroy_workqueue(host_dev->handle_error_wq); 2068 2069 err_out2: 2070 /* 2071 * Once we have connected with the host, we would need to 2072 * to invoke storvsc_dev_remove() to rollback this state and 2073 * this call also frees up the stor_device; hence the jump around 2074 * err_out1 label. 2075 */ 2076 storvsc_dev_remove(device); 2077 goto err_out0; 2078 2079 err_out1: 2080 kfree(stor_device->stor_chns); 2081 kfree(stor_device); 2082 2083 err_out0: 2084 scsi_host_put(host); 2085 return ret; 2086 } 2087 2088 /* Change a scsi target's queue depth */ 2089 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth) 2090 { 2091 if (queue_depth > scsi_driver.can_queue) 2092 queue_depth = scsi_driver.can_queue; 2093 2094 return scsi_change_queue_depth(sdev, queue_depth); 2095 } 2096 2097 static int storvsc_remove(struct hv_device *dev) 2098 { 2099 struct storvsc_device *stor_device = hv_get_drvdata(dev); 2100 struct Scsi_Host *host = stor_device->host; 2101 struct hv_host_device *host_dev = shost_priv(host); 2102 2103 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 2104 if (host->transportt == fc_transport_template) { 2105 fc_remote_port_delete(stor_device->rport); 2106 fc_remove_host(host); 2107 } 2108 #endif 2109 destroy_workqueue(host_dev->handle_error_wq); 2110 scsi_remove_host(host); 2111 storvsc_dev_remove(dev); 2112 scsi_host_put(host); 2113 2114 return 0; 2115 } 2116 2117 static int storvsc_suspend(struct hv_device *hv_dev) 2118 { 2119 struct storvsc_device *stor_device = hv_get_drvdata(hv_dev); 2120 struct Scsi_Host *host = stor_device->host; 2121 struct hv_host_device *host_dev = shost_priv(host); 2122 2123 storvsc_wait_to_drain(stor_device); 2124 2125 drain_workqueue(host_dev->handle_error_wq); 2126 2127 vmbus_close(hv_dev->channel); 2128 2129 kfree(stor_device->stor_chns); 2130 stor_device->stor_chns = NULL; 2131 2132 cpumask_clear(&stor_device->alloced_cpus); 2133 2134 return 0; 2135 } 2136 2137 static int storvsc_resume(struct hv_device *hv_dev) 2138 { 2139 int ret; 2140 2141 ret = storvsc_connect_to_vsp(hv_dev, storvsc_ringbuffer_size, 2142 hv_dev_is_fc(hv_dev)); 2143 return ret; 2144 } 2145 2146 static struct hv_driver storvsc_drv = { 2147 .name = KBUILD_MODNAME, 2148 .id_table = id_table, 2149 .probe = storvsc_probe, 2150 .remove = storvsc_remove, 2151 .suspend = storvsc_suspend, 2152 .resume = storvsc_resume, 2153 .driver = { 2154 .probe_type = PROBE_PREFER_ASYNCHRONOUS, 2155 }, 2156 }; 2157 2158 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 2159 static struct fc_function_template fc_transport_functions = { 2160 .show_host_node_name = 1, 2161 .show_host_port_name = 1, 2162 }; 2163 #endif 2164 2165 static int __init storvsc_drv_init(void) 2166 { 2167 int ret; 2168 2169 /* 2170 * Divide the ring buffer data size (which is 1 page less 2171 * than the ring buffer size since that page is reserved for 2172 * the ring buffer indices) by the max request size (which is 2173 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64) 2174 */ 2175 max_outstanding_req_per_channel = 2176 ((storvsc_ringbuffer_size - PAGE_SIZE) / 2177 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET + 2178 sizeof(struct vstor_packet) + sizeof(u64) - 2179 vmscsi_size_delta, 2180 sizeof(u64))); 2181 2182 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 2183 fc_transport_template = fc_attach_transport(&fc_transport_functions); 2184 if (!fc_transport_template) 2185 return -ENODEV; 2186 #endif 2187 2188 ret = vmbus_driver_register(&storvsc_drv); 2189 2190 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 2191 if (ret) 2192 fc_release_transport(fc_transport_template); 2193 #endif 2194 2195 return ret; 2196 } 2197 2198 static void __exit storvsc_drv_exit(void) 2199 { 2200 vmbus_driver_unregister(&storvsc_drv); 2201 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 2202 fc_release_transport(fc_transport_template); 2203 #endif 2204 } 2205 2206 MODULE_LICENSE("GPL"); 2207 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver"); 2208 module_init(storvsc_drv_init); 2209 module_exit(storvsc_drv_exit); 2210