1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* 3 * 4 * Copyright (c) 2011, Microsoft Corporation. 5 * 6 * Authors: 7 * Haiyang Zhang <haiyangz@microsoft.com> 8 * Hank Janssen <hjanssen@microsoft.com> 9 * K. Y. Srinivasan <kys@microsoft.com> 10 */ 11 12 #ifndef _HYPERV_H 13 #define _HYPERV_H 14 15 #include <uapi/linux/hyperv.h> 16 17 #include <linux/mm.h> 18 #include <linux/types.h> 19 #include <linux/scatterlist.h> 20 #include <linux/list.h> 21 #include <linux/timer.h> 22 #include <linux/completion.h> 23 #include <linux/device.h> 24 #include <linux/mod_devicetable.h> 25 #include <linux/interrupt.h> 26 #include <linux/reciprocal_div.h> 27 #include <asm/hyperv-tlfs.h> 28 29 #define MAX_PAGE_BUFFER_COUNT 32 30 #define MAX_MULTIPAGE_BUFFER_COUNT 32 /* 128K */ 31 32 #pragma pack(push, 1) 33 34 /* 35 * Types for GPADL, decides is how GPADL header is created. 36 * 37 * It doesn't make much difference between BUFFER and RING if PAGE_SIZE is the 38 * same as HV_HYP_PAGE_SIZE. 39 * 40 * If PAGE_SIZE is bigger than HV_HYP_PAGE_SIZE, the headers of ring buffers 41 * will be of PAGE_SIZE, however, only the first HV_HYP_PAGE will be put 42 * into gpadl, therefore the number for HV_HYP_PAGE and the indexes of each 43 * HV_HYP_PAGE will be different between different types of GPADL, for example 44 * if PAGE_SIZE is 64K: 45 * 46 * BUFFER: 47 * 48 * gva: |-- 64k --|-- 64k --| ... | 49 * gpa: | 4k | 4k | ... | 4k | 4k | 4k | ... | 4k | 50 * index: 0 1 2 15 16 17 18 .. 31 32 ... 51 * | | ... | | | ... | ... 52 * v V V V V V 53 * gpadl: | 4k | 4k | ... | 4k | 4k | 4k | ... | 4k | ... | 54 * index: 0 1 2 ... 15 16 17 18 .. 31 32 ... 55 * 56 * RING: 57 * 58 * | header | data | header | data | 59 * gva: |-- 64k --|-- 64k --| ... |-- 64k --|-- 64k --| ... | 60 * gpa: | 4k | .. | 4k | 4k | ... | 4k | ... | 4k | .. | 4k | .. | ... | 61 * index: 0 1 16 17 18 31 ... n n+1 n+16 ... 2n 62 * | / / / | / / 63 * | / / / | / / 64 * | / / ... / ... | / ... / 65 * | / / / | / / 66 * | / / / | / / 67 * V V V V V V v 68 * gpadl: | 4k | 4k | ... | ... | 4k | 4k | ... | 69 * index: 0 1 2 ... 16 ... n-15 n-14 n-13 ... 2n-30 70 */ 71 enum hv_gpadl_type { 72 HV_GPADL_BUFFER, 73 HV_GPADL_RING 74 }; 75 76 /* Single-page buffer */ 77 struct hv_page_buffer { 78 u32 len; 79 u32 offset; 80 u64 pfn; 81 }; 82 83 /* Multiple-page buffer */ 84 struct hv_multipage_buffer { 85 /* Length and Offset determines the # of pfns in the array */ 86 u32 len; 87 u32 offset; 88 u64 pfn_array[MAX_MULTIPAGE_BUFFER_COUNT]; 89 }; 90 91 /* 92 * Multiple-page buffer array; the pfn array is variable size: 93 * The number of entries in the PFN array is determined by 94 * "len" and "offset". 95 */ 96 struct hv_mpb_array { 97 /* Length and Offset determines the # of pfns in the array */ 98 u32 len; 99 u32 offset; 100 u64 pfn_array[]; 101 }; 102 103 /* 0x18 includes the proprietary packet header */ 104 #define MAX_PAGE_BUFFER_PACKET (0x18 + \ 105 (sizeof(struct hv_page_buffer) * \ 106 MAX_PAGE_BUFFER_COUNT)) 107 #define MAX_MULTIPAGE_BUFFER_PACKET (0x18 + \ 108 sizeof(struct hv_multipage_buffer)) 109 110 111 #pragma pack(pop) 112 113 struct hv_ring_buffer { 114 /* Offset in bytes from the start of ring data below */ 115 u32 write_index; 116 117 /* Offset in bytes from the start of ring data below */ 118 u32 read_index; 119 120 u32 interrupt_mask; 121 122 /* 123 * WS2012/Win8 and later versions of Hyper-V implement interrupt 124 * driven flow management. The feature bit feat_pending_send_sz 125 * is set by the host on the host->guest ring buffer, and by the 126 * guest on the guest->host ring buffer. 127 * 128 * The meaning of the feature bit is a bit complex in that it has 129 * semantics that apply to both ring buffers. If the guest sets 130 * the feature bit in the guest->host ring buffer, the guest is 131 * telling the host that: 132 * 1) It will set the pending_send_sz field in the guest->host ring 133 * buffer when it is waiting for space to become available, and 134 * 2) It will read the pending_send_sz field in the host->guest 135 * ring buffer and interrupt the host when it frees enough space 136 * 137 * Similarly, if the host sets the feature bit in the host->guest 138 * ring buffer, the host is telling the guest that: 139 * 1) It will set the pending_send_sz field in the host->guest ring 140 * buffer when it is waiting for space to become available, and 141 * 2) It will read the pending_send_sz field in the guest->host 142 * ring buffer and interrupt the guest when it frees enough space 143 * 144 * If either the guest or host does not set the feature bit that it 145 * owns, that guest or host must do polling if it encounters a full 146 * ring buffer, and not signal the other end with an interrupt. 147 */ 148 u32 pending_send_sz; 149 u32 reserved1[12]; 150 union { 151 struct { 152 u32 feat_pending_send_sz:1; 153 }; 154 u32 value; 155 } feature_bits; 156 157 /* Pad it to PAGE_SIZE so that data starts on page boundary */ 158 u8 reserved2[PAGE_SIZE - 68]; 159 160 /* 161 * Ring data starts here + RingDataStartOffset 162 * !!! DO NOT place any fields below this !!! 163 */ 164 u8 buffer[]; 165 } __packed; 166 167 /* Calculate the proper size of a ringbuffer, it must be page-aligned */ 168 #define VMBUS_RING_SIZE(payload_sz) PAGE_ALIGN(sizeof(struct hv_ring_buffer) + \ 169 (payload_sz)) 170 171 struct hv_ring_buffer_info { 172 struct hv_ring_buffer *ring_buffer; 173 u32 ring_size; /* Include the shared header */ 174 struct reciprocal_value ring_size_div10_reciprocal; 175 spinlock_t ring_lock; 176 177 u32 ring_datasize; /* < ring_size */ 178 u32 priv_read_index; 179 /* 180 * The ring buffer mutex lock. This lock prevents the ring buffer from 181 * being freed while the ring buffer is being accessed. 182 */ 183 struct mutex ring_buffer_mutex; 184 }; 185 186 187 static inline u32 hv_get_bytes_to_read(const struct hv_ring_buffer_info *rbi) 188 { 189 u32 read_loc, write_loc, dsize, read; 190 191 dsize = rbi->ring_datasize; 192 read_loc = rbi->ring_buffer->read_index; 193 write_loc = READ_ONCE(rbi->ring_buffer->write_index); 194 195 read = write_loc >= read_loc ? (write_loc - read_loc) : 196 (dsize - read_loc) + write_loc; 197 198 return read; 199 } 200 201 static inline u32 hv_get_bytes_to_write(const struct hv_ring_buffer_info *rbi) 202 { 203 u32 read_loc, write_loc, dsize, write; 204 205 dsize = rbi->ring_datasize; 206 read_loc = READ_ONCE(rbi->ring_buffer->read_index); 207 write_loc = rbi->ring_buffer->write_index; 208 209 write = write_loc >= read_loc ? dsize - (write_loc - read_loc) : 210 read_loc - write_loc; 211 return write; 212 } 213 214 static inline u32 hv_get_avail_to_write_percent( 215 const struct hv_ring_buffer_info *rbi) 216 { 217 u32 avail_write = hv_get_bytes_to_write(rbi); 218 219 return reciprocal_divide( 220 (avail_write << 3) + (avail_write << 1), 221 rbi->ring_size_div10_reciprocal); 222 } 223 224 /* 225 * VMBUS version is 32 bit entity broken up into 226 * two 16 bit quantities: major_number. minor_number. 227 * 228 * 0 . 13 (Windows Server 2008) 229 * 1 . 1 (Windows 7) 230 * 2 . 4 (Windows 8) 231 * 3 . 0 (Windows 8 R2) 232 * 4 . 0 (Windows 10) 233 * 4 . 1 (Windows 10 RS3) 234 * 5 . 0 (Newer Windows 10) 235 * 5 . 1 (Windows 10 RS4) 236 * 5 . 2 (Windows Server 2019, RS5) 237 */ 238 239 #define VERSION_WS2008 ((0 << 16) | (13)) 240 #define VERSION_WIN7 ((1 << 16) | (1)) 241 #define VERSION_WIN8 ((2 << 16) | (4)) 242 #define VERSION_WIN8_1 ((3 << 16) | (0)) 243 #define VERSION_WIN10 ((4 << 16) | (0)) 244 #define VERSION_WIN10_V4_1 ((4 << 16) | (1)) 245 #define VERSION_WIN10_V5 ((5 << 16) | (0)) 246 #define VERSION_WIN10_V5_1 ((5 << 16) | (1)) 247 #define VERSION_WIN10_V5_2 ((5 << 16) | (2)) 248 249 /* Make maximum size of pipe payload of 16K */ 250 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384) 251 252 /* Define PipeMode values. */ 253 #define VMBUS_PIPE_TYPE_BYTE 0x00000000 254 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004 255 256 /* The size of the user defined data buffer for non-pipe offers. */ 257 #define MAX_USER_DEFINED_BYTES 120 258 259 /* The size of the user defined data buffer for pipe offers. */ 260 #define MAX_PIPE_USER_DEFINED_BYTES 116 261 262 /* 263 * At the center of the Channel Management library is the Channel Offer. This 264 * struct contains the fundamental information about an offer. 265 */ 266 struct vmbus_channel_offer { 267 guid_t if_type; 268 guid_t if_instance; 269 270 /* 271 * These two fields are not currently used. 272 */ 273 u64 reserved1; 274 u64 reserved2; 275 276 u16 chn_flags; 277 u16 mmio_megabytes; /* in bytes * 1024 * 1024 */ 278 279 union { 280 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */ 281 struct { 282 unsigned char user_def[MAX_USER_DEFINED_BYTES]; 283 } std; 284 285 /* 286 * Pipes: 287 * The following sructure is an integrated pipe protocol, which 288 * is implemented on top of standard user-defined data. Pipe 289 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own 290 * use. 291 */ 292 struct { 293 u32 pipe_mode; 294 unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES]; 295 } pipe; 296 } u; 297 /* 298 * The sub_channel_index is defined in Win8: a value of zero means a 299 * primary channel and a value of non-zero means a sub-channel. 300 * 301 * Before Win8, the field is reserved, meaning it's always zero. 302 */ 303 u16 sub_channel_index; 304 u16 reserved3; 305 } __packed; 306 307 /* Server Flags */ 308 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1 309 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2 310 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4 311 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10 312 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100 313 #define VMBUS_CHANNEL_PARENT_OFFER 0x200 314 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400 315 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000 316 317 struct vmpacket_descriptor { 318 u16 type; 319 u16 offset8; 320 u16 len8; 321 u16 flags; 322 u64 trans_id; 323 } __packed; 324 325 struct vmpacket_header { 326 u32 prev_pkt_start_offset; 327 struct vmpacket_descriptor descriptor; 328 } __packed; 329 330 struct vmtransfer_page_range { 331 u32 byte_count; 332 u32 byte_offset; 333 } __packed; 334 335 struct vmtransfer_page_packet_header { 336 struct vmpacket_descriptor d; 337 u16 xfer_pageset_id; 338 u8 sender_owns_set; 339 u8 reserved; 340 u32 range_cnt; 341 struct vmtransfer_page_range ranges[1]; 342 } __packed; 343 344 struct vmgpadl_packet_header { 345 struct vmpacket_descriptor d; 346 u32 gpadl; 347 u32 reserved; 348 } __packed; 349 350 struct vmadd_remove_transfer_page_set { 351 struct vmpacket_descriptor d; 352 u32 gpadl; 353 u16 xfer_pageset_id; 354 u16 reserved; 355 } __packed; 356 357 /* 358 * This structure defines a range in guest physical space that can be made to 359 * look virtually contiguous. 360 */ 361 struct gpa_range { 362 u32 byte_count; 363 u32 byte_offset; 364 u64 pfn_array[]; 365 }; 366 367 /* 368 * This is the format for an Establish Gpadl packet, which contains a handle by 369 * which this GPADL will be known and a set of GPA ranges associated with it. 370 * This can be converted to a MDL by the guest OS. If there are multiple GPA 371 * ranges, then the resulting MDL will be "chained," representing multiple VA 372 * ranges. 373 */ 374 struct vmestablish_gpadl { 375 struct vmpacket_descriptor d; 376 u32 gpadl; 377 u32 range_cnt; 378 struct gpa_range range[1]; 379 } __packed; 380 381 /* 382 * This is the format for a Teardown Gpadl packet, which indicates that the 383 * GPADL handle in the Establish Gpadl packet will never be referenced again. 384 */ 385 struct vmteardown_gpadl { 386 struct vmpacket_descriptor d; 387 u32 gpadl; 388 u32 reserved; /* for alignment to a 8-byte boundary */ 389 } __packed; 390 391 /* 392 * This is the format for a GPA-Direct packet, which contains a set of GPA 393 * ranges, in addition to commands and/or data. 394 */ 395 struct vmdata_gpa_direct { 396 struct vmpacket_descriptor d; 397 u32 reserved; 398 u32 range_cnt; 399 struct gpa_range range[1]; 400 } __packed; 401 402 /* This is the format for a Additional Data Packet. */ 403 struct vmadditional_data { 404 struct vmpacket_descriptor d; 405 u64 total_bytes; 406 u32 offset; 407 u32 byte_cnt; 408 unsigned char data[1]; 409 } __packed; 410 411 union vmpacket_largest_possible_header { 412 struct vmpacket_descriptor simple_hdr; 413 struct vmtransfer_page_packet_header xfer_page_hdr; 414 struct vmgpadl_packet_header gpadl_hdr; 415 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr; 416 struct vmestablish_gpadl establish_gpadl_hdr; 417 struct vmteardown_gpadl teardown_gpadl_hdr; 418 struct vmdata_gpa_direct data_gpa_direct_hdr; 419 }; 420 421 #define VMPACKET_DATA_START_ADDRESS(__packet) \ 422 (void *)(((unsigned char *)__packet) + \ 423 ((struct vmpacket_descriptor)__packet)->offset8 * 8) 424 425 #define VMPACKET_DATA_LENGTH(__packet) \ 426 ((((struct vmpacket_descriptor)__packet)->len8 - \ 427 ((struct vmpacket_descriptor)__packet)->offset8) * 8) 428 429 #define VMPACKET_TRANSFER_MODE(__packet) \ 430 (((struct IMPACT)__packet)->type) 431 432 enum vmbus_packet_type { 433 VM_PKT_INVALID = 0x0, 434 VM_PKT_SYNCH = 0x1, 435 VM_PKT_ADD_XFER_PAGESET = 0x2, 436 VM_PKT_RM_XFER_PAGESET = 0x3, 437 VM_PKT_ESTABLISH_GPADL = 0x4, 438 VM_PKT_TEARDOWN_GPADL = 0x5, 439 VM_PKT_DATA_INBAND = 0x6, 440 VM_PKT_DATA_USING_XFER_PAGES = 0x7, 441 VM_PKT_DATA_USING_GPADL = 0x8, 442 VM_PKT_DATA_USING_GPA_DIRECT = 0x9, 443 VM_PKT_CANCEL_REQUEST = 0xa, 444 VM_PKT_COMP = 0xb, 445 VM_PKT_DATA_USING_ADDITIONAL_PKT = 0xc, 446 VM_PKT_ADDITIONAL_DATA = 0xd 447 }; 448 449 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1 450 451 452 /* Version 1 messages */ 453 enum vmbus_channel_message_type { 454 CHANNELMSG_INVALID = 0, 455 CHANNELMSG_OFFERCHANNEL = 1, 456 CHANNELMSG_RESCIND_CHANNELOFFER = 2, 457 CHANNELMSG_REQUESTOFFERS = 3, 458 CHANNELMSG_ALLOFFERS_DELIVERED = 4, 459 CHANNELMSG_OPENCHANNEL = 5, 460 CHANNELMSG_OPENCHANNEL_RESULT = 6, 461 CHANNELMSG_CLOSECHANNEL = 7, 462 CHANNELMSG_GPADL_HEADER = 8, 463 CHANNELMSG_GPADL_BODY = 9, 464 CHANNELMSG_GPADL_CREATED = 10, 465 CHANNELMSG_GPADL_TEARDOWN = 11, 466 CHANNELMSG_GPADL_TORNDOWN = 12, 467 CHANNELMSG_RELID_RELEASED = 13, 468 CHANNELMSG_INITIATE_CONTACT = 14, 469 CHANNELMSG_VERSION_RESPONSE = 15, 470 CHANNELMSG_UNLOAD = 16, 471 CHANNELMSG_UNLOAD_RESPONSE = 17, 472 CHANNELMSG_18 = 18, 473 CHANNELMSG_19 = 19, 474 CHANNELMSG_20 = 20, 475 CHANNELMSG_TL_CONNECT_REQUEST = 21, 476 CHANNELMSG_MODIFYCHANNEL = 22, 477 CHANNELMSG_TL_CONNECT_RESULT = 23, 478 CHANNELMSG_COUNT 479 }; 480 481 /* Hyper-V supports about 2048 channels, and the RELIDs start with 1. */ 482 #define INVALID_RELID U32_MAX 483 484 struct vmbus_channel_message_header { 485 enum vmbus_channel_message_type msgtype; 486 u32 padding; 487 } __packed; 488 489 /* Query VMBus Version parameters */ 490 struct vmbus_channel_query_vmbus_version { 491 struct vmbus_channel_message_header header; 492 u32 version; 493 } __packed; 494 495 /* VMBus Version Supported parameters */ 496 struct vmbus_channel_version_supported { 497 struct vmbus_channel_message_header header; 498 u8 version_supported; 499 } __packed; 500 501 /* Offer Channel parameters */ 502 struct vmbus_channel_offer_channel { 503 struct vmbus_channel_message_header header; 504 struct vmbus_channel_offer offer; 505 u32 child_relid; 506 u8 monitorid; 507 /* 508 * win7 and beyond splits this field into a bit field. 509 */ 510 u8 monitor_allocated:1; 511 u8 reserved:7; 512 /* 513 * These are new fields added in win7 and later. 514 * Do not access these fields without checking the 515 * negotiated protocol. 516 * 517 * If "is_dedicated_interrupt" is set, we must not set the 518 * associated bit in the channel bitmap while sending the 519 * interrupt to the host. 520 * 521 * connection_id is to be used in signaling the host. 522 */ 523 u16 is_dedicated_interrupt:1; 524 u16 reserved1:15; 525 u32 connection_id; 526 } __packed; 527 528 /* Rescind Offer parameters */ 529 struct vmbus_channel_rescind_offer { 530 struct vmbus_channel_message_header header; 531 u32 child_relid; 532 } __packed; 533 534 static inline u32 535 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info *rbi) 536 { 537 return rbi->ring_buffer->pending_send_sz; 538 } 539 540 /* 541 * Request Offer -- no parameters, SynIC message contains the partition ID 542 * Set Snoop -- no parameters, SynIC message contains the partition ID 543 * Clear Snoop -- no parameters, SynIC message contains the partition ID 544 * All Offers Delivered -- no parameters, SynIC message contains the partition 545 * ID 546 * Flush Client -- no parameters, SynIC message contains the partition ID 547 */ 548 549 /* Open Channel parameters */ 550 struct vmbus_channel_open_channel { 551 struct vmbus_channel_message_header header; 552 553 /* Identifies the specific VMBus channel that is being opened. */ 554 u32 child_relid; 555 556 /* ID making a particular open request at a channel offer unique. */ 557 u32 openid; 558 559 /* GPADL for the channel's ring buffer. */ 560 u32 ringbuffer_gpadlhandle; 561 562 /* 563 * Starting with win8, this field will be used to specify 564 * the target virtual processor on which to deliver the interrupt for 565 * the host to guest communication. 566 * Prior to win8, incoming channel interrupts would only 567 * be delivered on cpu 0. Setting this value to 0 would 568 * preserve the earlier behavior. 569 */ 570 u32 target_vp; 571 572 /* 573 * The upstream ring buffer begins at offset zero in the memory 574 * described by RingBufferGpadlHandle. The downstream ring buffer 575 * follows it at this offset (in pages). 576 */ 577 u32 downstream_ringbuffer_pageoffset; 578 579 /* User-specific data to be passed along to the server endpoint. */ 580 unsigned char userdata[MAX_USER_DEFINED_BYTES]; 581 } __packed; 582 583 /* Open Channel Result parameters */ 584 struct vmbus_channel_open_result { 585 struct vmbus_channel_message_header header; 586 u32 child_relid; 587 u32 openid; 588 u32 status; 589 } __packed; 590 591 /* Close channel parameters; */ 592 struct vmbus_channel_close_channel { 593 struct vmbus_channel_message_header header; 594 u32 child_relid; 595 } __packed; 596 597 /* Channel Message GPADL */ 598 #define GPADL_TYPE_RING_BUFFER 1 599 #define GPADL_TYPE_SERVER_SAVE_AREA 2 600 #define GPADL_TYPE_TRANSACTION 8 601 602 /* 603 * The number of PFNs in a GPADL message is defined by the number of 604 * pages that would be spanned by ByteCount and ByteOffset. If the 605 * implied number of PFNs won't fit in this packet, there will be a 606 * follow-up packet that contains more. 607 */ 608 struct vmbus_channel_gpadl_header { 609 struct vmbus_channel_message_header header; 610 u32 child_relid; 611 u32 gpadl; 612 u16 range_buflen; 613 u16 rangecount; 614 struct gpa_range range[]; 615 } __packed; 616 617 /* This is the followup packet that contains more PFNs. */ 618 struct vmbus_channel_gpadl_body { 619 struct vmbus_channel_message_header header; 620 u32 msgnumber; 621 u32 gpadl; 622 u64 pfn[]; 623 } __packed; 624 625 struct vmbus_channel_gpadl_created { 626 struct vmbus_channel_message_header header; 627 u32 child_relid; 628 u32 gpadl; 629 u32 creation_status; 630 } __packed; 631 632 struct vmbus_channel_gpadl_teardown { 633 struct vmbus_channel_message_header header; 634 u32 child_relid; 635 u32 gpadl; 636 } __packed; 637 638 struct vmbus_channel_gpadl_torndown { 639 struct vmbus_channel_message_header header; 640 u32 gpadl; 641 } __packed; 642 643 struct vmbus_channel_relid_released { 644 struct vmbus_channel_message_header header; 645 u32 child_relid; 646 } __packed; 647 648 struct vmbus_channel_initiate_contact { 649 struct vmbus_channel_message_header header; 650 u32 vmbus_version_requested; 651 u32 target_vcpu; /* The VCPU the host should respond to */ 652 union { 653 u64 interrupt_page; 654 struct { 655 u8 msg_sint; 656 u8 padding1[3]; 657 u32 padding2; 658 }; 659 }; 660 u64 monitor_page1; 661 u64 monitor_page2; 662 } __packed; 663 664 /* Hyper-V socket: guest's connect()-ing to host */ 665 struct vmbus_channel_tl_connect_request { 666 struct vmbus_channel_message_header header; 667 guid_t guest_endpoint_id; 668 guid_t host_service_id; 669 } __packed; 670 671 /* Modify Channel parameters, cf. vmbus_send_modifychannel() */ 672 struct vmbus_channel_modifychannel { 673 struct vmbus_channel_message_header header; 674 u32 child_relid; 675 u32 target_vp; 676 } __packed; 677 678 struct vmbus_channel_version_response { 679 struct vmbus_channel_message_header header; 680 u8 version_supported; 681 682 u8 connection_state; 683 u16 padding; 684 685 /* 686 * On new hosts that support VMBus protocol 5.0, we must use 687 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message, 688 * and for subsequent messages, we must use the Message Connection ID 689 * field in the host-returned Version Response Message. 690 * 691 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1). 692 */ 693 u32 msg_conn_id; 694 } __packed; 695 696 enum vmbus_channel_state { 697 CHANNEL_OFFER_STATE, 698 CHANNEL_OPENING_STATE, 699 CHANNEL_OPEN_STATE, 700 CHANNEL_OPENED_STATE, 701 }; 702 703 /* 704 * Represents each channel msg on the vmbus connection This is a 705 * variable-size data structure depending on the msg type itself 706 */ 707 struct vmbus_channel_msginfo { 708 /* Bookkeeping stuff */ 709 struct list_head msglistentry; 710 711 /* So far, this is only used to handle gpadl body message */ 712 struct list_head submsglist; 713 714 /* Synchronize the request/response if needed */ 715 struct completion waitevent; 716 struct vmbus_channel *waiting_channel; 717 union { 718 struct vmbus_channel_version_supported version_supported; 719 struct vmbus_channel_open_result open_result; 720 struct vmbus_channel_gpadl_torndown gpadl_torndown; 721 struct vmbus_channel_gpadl_created gpadl_created; 722 struct vmbus_channel_version_response version_response; 723 } response; 724 725 u32 msgsize; 726 /* 727 * The channel message that goes out on the "wire". 728 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header 729 */ 730 unsigned char msg[]; 731 }; 732 733 struct vmbus_close_msg { 734 struct vmbus_channel_msginfo info; 735 struct vmbus_channel_close_channel msg; 736 }; 737 738 /* Define connection identifier type. */ 739 union hv_connection_id { 740 u32 asu32; 741 struct { 742 u32 id:24; 743 u32 reserved:8; 744 } u; 745 }; 746 747 enum vmbus_device_type { 748 HV_IDE = 0, 749 HV_SCSI, 750 HV_FC, 751 HV_NIC, 752 HV_ND, 753 HV_PCIE, 754 HV_FB, 755 HV_KBD, 756 HV_MOUSE, 757 HV_KVP, 758 HV_TS, 759 HV_HB, 760 HV_SHUTDOWN, 761 HV_FCOPY, 762 HV_BACKUP, 763 HV_DM, 764 HV_UNKNOWN, 765 }; 766 767 /* 768 * Provides request ids for VMBus. Encapsulates guest memory 769 * addresses and stores the next available slot in req_arr 770 * to generate new ids in constant time. 771 */ 772 struct vmbus_requestor { 773 u64 *req_arr; 774 unsigned long *req_bitmap; /* is a given slot available? */ 775 u32 size; 776 u64 next_request_id; 777 spinlock_t req_lock; /* provides atomicity */ 778 }; 779 780 #define VMBUS_NO_RQSTOR U64_MAX 781 #define VMBUS_RQST_ERROR (U64_MAX - 1) 782 #define VMBUS_RQST_ID_NO_RESPONSE (U64_MAX - 2) 783 784 struct vmbus_device { 785 u16 dev_type; 786 guid_t guid; 787 bool perf_device; 788 bool allowed_in_isolated; 789 }; 790 791 struct vmbus_channel { 792 struct list_head listentry; 793 794 struct hv_device *device_obj; 795 796 enum vmbus_channel_state state; 797 798 struct vmbus_channel_offer_channel offermsg; 799 /* 800 * These are based on the OfferMsg.MonitorId. 801 * Save it here for easy access. 802 */ 803 u8 monitor_grp; 804 u8 monitor_bit; 805 806 bool rescind; /* got rescind msg */ 807 bool rescind_ref; /* got rescind msg, got channel reference */ 808 struct completion rescind_event; 809 810 u32 ringbuffer_gpadlhandle; 811 812 /* Allocated memory for ring buffer */ 813 struct page *ringbuffer_page; 814 u32 ringbuffer_pagecount; 815 u32 ringbuffer_send_offset; 816 struct hv_ring_buffer_info outbound; /* send to parent */ 817 struct hv_ring_buffer_info inbound; /* receive from parent */ 818 819 struct vmbus_close_msg close_msg; 820 821 /* Statistics */ 822 u64 interrupts; /* Host to Guest interrupts */ 823 u64 sig_events; /* Guest to Host events */ 824 825 /* 826 * Guest to host interrupts caused by the outbound ring buffer changing 827 * from empty to not empty. 828 */ 829 u64 intr_out_empty; 830 831 /* 832 * Indicates that a full outbound ring buffer was encountered. The flag 833 * is set to true when a full outbound ring buffer is encountered and 834 * set to false when a write to the outbound ring buffer is completed. 835 */ 836 bool out_full_flag; 837 838 /* Channel callback's invoked in softirq context */ 839 struct tasklet_struct callback_event; 840 void (*onchannel_callback)(void *context); 841 void *channel_callback_context; 842 843 void (*change_target_cpu_callback)(struct vmbus_channel *channel, 844 u32 old, u32 new); 845 846 /* 847 * Synchronize channel scheduling and channel removal; see the inline 848 * comments in vmbus_chan_sched() and vmbus_reset_channel_cb(). 849 */ 850 spinlock_t sched_lock; 851 852 /* 853 * A channel can be marked for one of three modes of reading: 854 * BATCHED - callback called from taslket and should read 855 * channel until empty. Interrupts from the host 856 * are masked while read is in process (default). 857 * DIRECT - callback called from tasklet (softirq). 858 * ISR - callback called in interrupt context and must 859 * invoke its own deferred processing. 860 * Host interrupts are disabled and must be re-enabled 861 * when ring is empty. 862 */ 863 enum hv_callback_mode { 864 HV_CALL_BATCHED, 865 HV_CALL_DIRECT, 866 HV_CALL_ISR 867 } callback_mode; 868 869 bool is_dedicated_interrupt; 870 u64 sig_event; 871 872 /* 873 * Starting with win8, this field will be used to specify the 874 * target CPU on which to deliver the interrupt for the host 875 * to guest communication. 876 * 877 * Prior to win8, incoming channel interrupts would only be 878 * delivered on CPU 0. Setting this value to 0 would preserve 879 * the earlier behavior. 880 */ 881 u32 target_cpu; 882 /* 883 * Support for sub-channels. For high performance devices, 884 * it will be useful to have multiple sub-channels to support 885 * a scalable communication infrastructure with the host. 886 * The support for sub-channels is implemented as an extention 887 * to the current infrastructure. 888 * The initial offer is considered the primary channel and this 889 * offer message will indicate if the host supports sub-channels. 890 * The guest is free to ask for sub-channels to be offerred and can 891 * open these sub-channels as a normal "primary" channel. However, 892 * all sub-channels will have the same type and instance guids as the 893 * primary channel. Requests sent on a given channel will result in a 894 * response on the same channel. 895 */ 896 897 /* 898 * Sub-channel creation callback. This callback will be called in 899 * process context when a sub-channel offer is received from the host. 900 * The guest can open the sub-channel in the context of this callback. 901 */ 902 void (*sc_creation_callback)(struct vmbus_channel *new_sc); 903 904 /* 905 * Channel rescind callback. Some channels (the hvsock ones), need to 906 * register a callback which is invoked in vmbus_onoffer_rescind(). 907 */ 908 void (*chn_rescind_callback)(struct vmbus_channel *channel); 909 910 /* 911 * All Sub-channels of a primary channel are linked here. 912 */ 913 struct list_head sc_list; 914 /* 915 * The primary channel this sub-channel belongs to. 916 * This will be NULL for the primary channel. 917 */ 918 struct vmbus_channel *primary_channel; 919 /* 920 * Support per-channel state for use by vmbus drivers. 921 */ 922 void *per_channel_state; 923 924 /* 925 * Defer freeing channel until after all cpu's have 926 * gone through grace period. 927 */ 928 struct rcu_head rcu; 929 930 /* 931 * For sysfs per-channel properties. 932 */ 933 struct kobject kobj; 934 935 /* 936 * For performance critical channels (storage, networking 937 * etc,), Hyper-V has a mechanism to enhance the throughput 938 * at the expense of latency: 939 * When the host is to be signaled, we just set a bit in a shared page 940 * and this bit will be inspected by the hypervisor within a certain 941 * window and if the bit is set, the host will be signaled. The window 942 * of time is the monitor latency - currently around 100 usecs. This 943 * mechanism improves throughput by: 944 * 945 * A) Making the host more efficient - each time it wakes up, 946 * potentially it will process morev number of packets. The 947 * monitor latency allows a batch to build up. 948 * B) By deferring the hypercall to signal, we will also minimize 949 * the interrupts. 950 * 951 * Clearly, these optimizations improve throughput at the expense of 952 * latency. Furthermore, since the channel is shared for both 953 * control and data messages, control messages currently suffer 954 * unnecessary latency adversley impacting performance and boot 955 * time. To fix this issue, permit tagging the channel as being 956 * in "low latency" mode. In this mode, we will bypass the monitor 957 * mechanism. 958 */ 959 bool low_latency; 960 961 bool probe_done; 962 963 /* 964 * Cache the device ID here for easy access; this is useful, in 965 * particular, in situations where the channel's device_obj has 966 * not been allocated/initialized yet. 967 */ 968 u16 device_id; 969 970 /* 971 * We must offload the handling of the primary/sub channels 972 * from the single-threaded vmbus_connection.work_queue to 973 * two different workqueue, otherwise we can block 974 * vmbus_connection.work_queue and hang: see vmbus_process_offer(). 975 */ 976 struct work_struct add_channel_work; 977 978 /* 979 * Guest to host interrupts caused by the inbound ring buffer changing 980 * from full to not full while a packet is waiting. 981 */ 982 u64 intr_in_full; 983 984 /* 985 * The total number of write operations that encountered a full 986 * outbound ring buffer. 987 */ 988 u64 out_full_total; 989 990 /* 991 * The number of write operations that were the first to encounter a 992 * full outbound ring buffer. 993 */ 994 u64 out_full_first; 995 996 /* enabling/disabling fuzz testing on the channel (default is false)*/ 997 bool fuzz_testing_state; 998 999 /* 1000 * Interrupt delay will delay the guest from emptying the ring buffer 1001 * for a specific amount of time. The delay is in microseconds and will 1002 * be between 1 to a maximum of 1000, its default is 0 (no delay). 1003 * The Message delay will delay guest reading on a per message basis 1004 * in microseconds between 1 to 1000 with the default being 0 1005 * (no delay). 1006 */ 1007 u32 fuzz_testing_interrupt_delay; 1008 u32 fuzz_testing_message_delay; 1009 1010 /* request/transaction ids for VMBus */ 1011 struct vmbus_requestor requestor; 1012 u32 rqstor_size; 1013 }; 1014 1015 u64 vmbus_next_request_id(struct vmbus_requestor *rqstor, u64 rqst_addr); 1016 u64 vmbus_request_addr(struct vmbus_requestor *rqstor, u64 trans_id); 1017 1018 static inline bool is_hvsock_channel(const struct vmbus_channel *c) 1019 { 1020 return !!(c->offermsg.offer.chn_flags & 1021 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER); 1022 } 1023 1024 static inline bool is_sub_channel(const struct vmbus_channel *c) 1025 { 1026 return c->offermsg.offer.sub_channel_index != 0; 1027 } 1028 1029 static inline void set_channel_read_mode(struct vmbus_channel *c, 1030 enum hv_callback_mode mode) 1031 { 1032 c->callback_mode = mode; 1033 } 1034 1035 static inline void set_per_channel_state(struct vmbus_channel *c, void *s) 1036 { 1037 c->per_channel_state = s; 1038 } 1039 1040 static inline void *get_per_channel_state(struct vmbus_channel *c) 1041 { 1042 return c->per_channel_state; 1043 } 1044 1045 static inline void set_channel_pending_send_size(struct vmbus_channel *c, 1046 u32 size) 1047 { 1048 unsigned long flags; 1049 1050 if (size) { 1051 spin_lock_irqsave(&c->outbound.ring_lock, flags); 1052 ++c->out_full_total; 1053 1054 if (!c->out_full_flag) { 1055 ++c->out_full_first; 1056 c->out_full_flag = true; 1057 } 1058 spin_unlock_irqrestore(&c->outbound.ring_lock, flags); 1059 } else { 1060 c->out_full_flag = false; 1061 } 1062 1063 c->outbound.ring_buffer->pending_send_sz = size; 1064 } 1065 1066 static inline void set_low_latency_mode(struct vmbus_channel *c) 1067 { 1068 c->low_latency = true; 1069 } 1070 1071 static inline void clear_low_latency_mode(struct vmbus_channel *c) 1072 { 1073 c->low_latency = false; 1074 } 1075 1076 void vmbus_onmessage(struct vmbus_channel_message_header *hdr); 1077 1078 int vmbus_request_offers(void); 1079 1080 /* 1081 * APIs for managing sub-channels. 1082 */ 1083 1084 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel, 1085 void (*sc_cr_cb)(struct vmbus_channel *new_sc)); 1086 1087 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel, 1088 void (*chn_rescind_cb)(struct vmbus_channel *)); 1089 1090 /* 1091 * Check if sub-channels have already been offerred. This API will be useful 1092 * when the driver is unloaded after establishing sub-channels. In this case, 1093 * when the driver is re-loaded, the driver would have to check if the 1094 * subchannels have already been established before attempting to request 1095 * the creation of sub-channels. 1096 * This function returns TRUE to indicate that subchannels have already been 1097 * created. 1098 * This function should be invoked after setting the callback function for 1099 * sub-channel creation. 1100 */ 1101 bool vmbus_are_subchannels_present(struct vmbus_channel *primary); 1102 1103 /* The format must be the same as struct vmdata_gpa_direct */ 1104 struct vmbus_channel_packet_page_buffer { 1105 u16 type; 1106 u16 dataoffset8; 1107 u16 length8; 1108 u16 flags; 1109 u64 transactionid; 1110 u32 reserved; 1111 u32 rangecount; 1112 struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT]; 1113 } __packed; 1114 1115 /* The format must be the same as struct vmdata_gpa_direct */ 1116 struct vmbus_channel_packet_multipage_buffer { 1117 u16 type; 1118 u16 dataoffset8; 1119 u16 length8; 1120 u16 flags; 1121 u64 transactionid; 1122 u32 reserved; 1123 u32 rangecount; /* Always 1 in this case */ 1124 struct hv_multipage_buffer range; 1125 } __packed; 1126 1127 /* The format must be the same as struct vmdata_gpa_direct */ 1128 struct vmbus_packet_mpb_array { 1129 u16 type; 1130 u16 dataoffset8; 1131 u16 length8; 1132 u16 flags; 1133 u64 transactionid; 1134 u32 reserved; 1135 u32 rangecount; /* Always 1 in this case */ 1136 struct hv_mpb_array range; 1137 } __packed; 1138 1139 int vmbus_alloc_ring(struct vmbus_channel *channel, 1140 u32 send_size, u32 recv_size); 1141 void vmbus_free_ring(struct vmbus_channel *channel); 1142 1143 int vmbus_connect_ring(struct vmbus_channel *channel, 1144 void (*onchannel_callback)(void *context), 1145 void *context); 1146 int vmbus_disconnect_ring(struct vmbus_channel *channel); 1147 1148 extern int vmbus_open(struct vmbus_channel *channel, 1149 u32 send_ringbuffersize, 1150 u32 recv_ringbuffersize, 1151 void *userdata, 1152 u32 userdatalen, 1153 void (*onchannel_callback)(void *context), 1154 void *context); 1155 1156 extern void vmbus_close(struct vmbus_channel *channel); 1157 1158 extern int vmbus_sendpacket(struct vmbus_channel *channel, 1159 void *buffer, 1160 u32 bufferLen, 1161 u64 requestid, 1162 enum vmbus_packet_type type, 1163 u32 flags); 1164 1165 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel, 1166 struct hv_page_buffer pagebuffers[], 1167 u32 pagecount, 1168 void *buffer, 1169 u32 bufferlen, 1170 u64 requestid); 1171 1172 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel, 1173 struct vmbus_packet_mpb_array *mpb, 1174 u32 desc_size, 1175 void *buffer, 1176 u32 bufferlen, 1177 u64 requestid); 1178 1179 extern int vmbus_establish_gpadl(struct vmbus_channel *channel, 1180 void *kbuffer, 1181 u32 size, 1182 u32 *gpadl_handle); 1183 1184 extern int vmbus_teardown_gpadl(struct vmbus_channel *channel, 1185 u32 gpadl_handle); 1186 1187 void vmbus_reset_channel_cb(struct vmbus_channel *channel); 1188 1189 extern int vmbus_recvpacket(struct vmbus_channel *channel, 1190 void *buffer, 1191 u32 bufferlen, 1192 u32 *buffer_actual_len, 1193 u64 *requestid); 1194 1195 extern int vmbus_recvpacket_raw(struct vmbus_channel *channel, 1196 void *buffer, 1197 u32 bufferlen, 1198 u32 *buffer_actual_len, 1199 u64 *requestid); 1200 1201 1202 extern void vmbus_ontimer(unsigned long data); 1203 1204 /* Base driver object */ 1205 struct hv_driver { 1206 const char *name; 1207 1208 /* 1209 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 1210 * channel flag, actually doesn't mean a synthetic device because the 1211 * offer's if_type/if_instance can change for every new hvsock 1212 * connection. 1213 * 1214 * However, to facilitate the notification of new-offer/rescind-offer 1215 * from vmbus driver to hvsock driver, we can handle hvsock offer as 1216 * a special vmbus device, and hence we need the below flag to 1217 * indicate if the driver is the hvsock driver or not: we need to 1218 * specially treat the hvosck offer & driver in vmbus_match(). 1219 */ 1220 bool hvsock; 1221 1222 /* the device type supported by this driver */ 1223 guid_t dev_type; 1224 const struct hv_vmbus_device_id *id_table; 1225 1226 struct device_driver driver; 1227 1228 /* dynamic device GUID's */ 1229 struct { 1230 spinlock_t lock; 1231 struct list_head list; 1232 } dynids; 1233 1234 int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *); 1235 int (*remove)(struct hv_device *); 1236 void (*shutdown)(struct hv_device *); 1237 1238 int (*suspend)(struct hv_device *); 1239 int (*resume)(struct hv_device *); 1240 1241 }; 1242 1243 /* Base device object */ 1244 struct hv_device { 1245 /* the device type id of this device */ 1246 guid_t dev_type; 1247 1248 /* the device instance id of this device */ 1249 guid_t dev_instance; 1250 u16 vendor_id; 1251 u16 device_id; 1252 1253 struct device device; 1254 char *driver_override; /* Driver name to force a match */ 1255 1256 struct vmbus_channel *channel; 1257 struct kset *channels_kset; 1258 1259 /* place holder to keep track of the dir for hv device in debugfs */ 1260 struct dentry *debug_dir; 1261 1262 }; 1263 1264 1265 static inline struct hv_device *device_to_hv_device(struct device *d) 1266 { 1267 return container_of(d, struct hv_device, device); 1268 } 1269 1270 static inline struct hv_driver *drv_to_hv_drv(struct device_driver *d) 1271 { 1272 return container_of(d, struct hv_driver, driver); 1273 } 1274 1275 static inline void hv_set_drvdata(struct hv_device *dev, void *data) 1276 { 1277 dev_set_drvdata(&dev->device, data); 1278 } 1279 1280 static inline void *hv_get_drvdata(struct hv_device *dev) 1281 { 1282 return dev_get_drvdata(&dev->device); 1283 } 1284 1285 struct hv_ring_buffer_debug_info { 1286 u32 current_interrupt_mask; 1287 u32 current_read_index; 1288 u32 current_write_index; 1289 u32 bytes_avail_toread; 1290 u32 bytes_avail_towrite; 1291 }; 1292 1293 1294 int hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info, 1295 struct hv_ring_buffer_debug_info *debug_info); 1296 1297 /* Vmbus interface */ 1298 #define vmbus_driver_register(driver) \ 1299 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME) 1300 int __must_check __vmbus_driver_register(struct hv_driver *hv_driver, 1301 struct module *owner, 1302 const char *mod_name); 1303 void vmbus_driver_unregister(struct hv_driver *hv_driver); 1304 1305 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel); 1306 1307 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj, 1308 resource_size_t min, resource_size_t max, 1309 resource_size_t size, resource_size_t align, 1310 bool fb_overlap_ok); 1311 void vmbus_free_mmio(resource_size_t start, resource_size_t size); 1312 1313 /* 1314 * GUID definitions of various offer types - services offered to the guest. 1315 */ 1316 1317 /* 1318 * Network GUID 1319 * {f8615163-df3e-46c5-913f-f2d2f965ed0e} 1320 */ 1321 #define HV_NIC_GUID \ 1322 .guid = GUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \ 1323 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e) 1324 1325 /* 1326 * IDE GUID 1327 * {32412632-86cb-44a2-9b5c-50d1417354f5} 1328 */ 1329 #define HV_IDE_GUID \ 1330 .guid = GUID_INIT(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \ 1331 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5) 1332 1333 /* 1334 * SCSI GUID 1335 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f} 1336 */ 1337 #define HV_SCSI_GUID \ 1338 .guid = GUID_INIT(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \ 1339 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f) 1340 1341 /* 1342 * Shutdown GUID 1343 * {0e0b6031-5213-4934-818b-38d90ced39db} 1344 */ 1345 #define HV_SHUTDOWN_GUID \ 1346 .guid = GUID_INIT(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \ 1347 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb) 1348 1349 /* 1350 * Time Synch GUID 1351 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF} 1352 */ 1353 #define HV_TS_GUID \ 1354 .guid = GUID_INIT(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \ 1355 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf) 1356 1357 /* 1358 * Heartbeat GUID 1359 * {57164f39-9115-4e78-ab55-382f3bd5422d} 1360 */ 1361 #define HV_HEART_BEAT_GUID \ 1362 .guid = GUID_INIT(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \ 1363 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d) 1364 1365 /* 1366 * KVP GUID 1367 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6} 1368 */ 1369 #define HV_KVP_GUID \ 1370 .guid = GUID_INIT(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \ 1371 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6) 1372 1373 /* 1374 * Dynamic memory GUID 1375 * {525074dc-8985-46e2-8057-a307dc18a502} 1376 */ 1377 #define HV_DM_GUID \ 1378 .guid = GUID_INIT(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \ 1379 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02) 1380 1381 /* 1382 * Mouse GUID 1383 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a} 1384 */ 1385 #define HV_MOUSE_GUID \ 1386 .guid = GUID_INIT(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \ 1387 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a) 1388 1389 /* 1390 * Keyboard GUID 1391 * {f912ad6d-2b17-48ea-bd65-f927a61c7684} 1392 */ 1393 #define HV_KBD_GUID \ 1394 .guid = GUID_INIT(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \ 1395 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84) 1396 1397 /* 1398 * VSS (Backup/Restore) GUID 1399 */ 1400 #define HV_VSS_GUID \ 1401 .guid = GUID_INIT(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \ 1402 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40) 1403 /* 1404 * Synthetic Video GUID 1405 * {DA0A7802-E377-4aac-8E77-0558EB1073F8} 1406 */ 1407 #define HV_SYNTHVID_GUID \ 1408 .guid = GUID_INIT(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \ 1409 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8) 1410 1411 /* 1412 * Synthetic FC GUID 1413 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda} 1414 */ 1415 #define HV_SYNTHFC_GUID \ 1416 .guid = GUID_INIT(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \ 1417 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda) 1418 1419 /* 1420 * Guest File Copy Service 1421 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192} 1422 */ 1423 1424 #define HV_FCOPY_GUID \ 1425 .guid = GUID_INIT(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \ 1426 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92) 1427 1428 /* 1429 * NetworkDirect. This is the guest RDMA service. 1430 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501} 1431 */ 1432 #define HV_ND_GUID \ 1433 .guid = GUID_INIT(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \ 1434 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01) 1435 1436 /* 1437 * PCI Express Pass Through 1438 * {44C4F61D-4444-4400-9D52-802E27EDE19F} 1439 */ 1440 1441 #define HV_PCIE_GUID \ 1442 .guid = GUID_INIT(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \ 1443 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f) 1444 1445 /* 1446 * Linux doesn't support the 3 devices: the first two are for 1447 * Automatic Virtual Machine Activation, and the third is for 1448 * Remote Desktop Virtualization. 1449 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5} 1450 * {3375baf4-9e15-4b30-b765-67acb10d607b} 1451 * {276aacf4-ac15-426c-98dd-7521ad3f01fe} 1452 */ 1453 1454 #define HV_AVMA1_GUID \ 1455 .guid = GUID_INIT(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \ 1456 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5) 1457 1458 #define HV_AVMA2_GUID \ 1459 .guid = GUID_INIT(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \ 1460 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b) 1461 1462 #define HV_RDV_GUID \ 1463 .guid = GUID_INIT(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \ 1464 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe) 1465 1466 /* 1467 * Common header for Hyper-V ICs 1468 */ 1469 1470 #define ICMSGTYPE_NEGOTIATE 0 1471 #define ICMSGTYPE_HEARTBEAT 1 1472 #define ICMSGTYPE_KVPEXCHANGE 2 1473 #define ICMSGTYPE_SHUTDOWN 3 1474 #define ICMSGTYPE_TIMESYNC 4 1475 #define ICMSGTYPE_VSS 5 1476 #define ICMSGTYPE_FCOPY 7 1477 1478 #define ICMSGHDRFLAG_TRANSACTION 1 1479 #define ICMSGHDRFLAG_REQUEST 2 1480 #define ICMSGHDRFLAG_RESPONSE 4 1481 1482 1483 /* 1484 * While we want to handle util services as regular devices, 1485 * there is only one instance of each of these services; so 1486 * we statically allocate the service specific state. 1487 */ 1488 1489 struct hv_util_service { 1490 u8 *recv_buffer; 1491 void *channel; 1492 void (*util_cb)(void *); 1493 int (*util_init)(struct hv_util_service *); 1494 void (*util_deinit)(void); 1495 int (*util_pre_suspend)(void); 1496 int (*util_pre_resume)(void); 1497 }; 1498 1499 struct vmbuspipe_hdr { 1500 u32 flags; 1501 u32 msgsize; 1502 } __packed; 1503 1504 struct ic_version { 1505 u16 major; 1506 u16 minor; 1507 } __packed; 1508 1509 struct icmsg_hdr { 1510 struct ic_version icverframe; 1511 u16 icmsgtype; 1512 struct ic_version icvermsg; 1513 u16 icmsgsize; 1514 u32 status; 1515 u8 ictransaction_id; 1516 u8 icflags; 1517 u8 reserved[2]; 1518 } __packed; 1519 1520 #define IC_VERSION_NEGOTIATION_MAX_VER_COUNT 100 1521 #define ICMSG_HDR (sizeof(struct vmbuspipe_hdr) + sizeof(struct icmsg_hdr)) 1522 #define ICMSG_NEGOTIATE_PKT_SIZE(icframe_vercnt, icmsg_vercnt) \ 1523 (ICMSG_HDR + sizeof(struct icmsg_negotiate) + \ 1524 (((icframe_vercnt) + (icmsg_vercnt)) * sizeof(struct ic_version))) 1525 1526 struct icmsg_negotiate { 1527 u16 icframe_vercnt; 1528 u16 icmsg_vercnt; 1529 u32 reserved; 1530 struct ic_version icversion_data[]; /* any size array */ 1531 } __packed; 1532 1533 struct shutdown_msg_data { 1534 u32 reason_code; 1535 u32 timeout_seconds; 1536 u32 flags; 1537 u8 display_message[2048]; 1538 } __packed; 1539 1540 struct heartbeat_msg_data { 1541 u64 seq_num; 1542 u32 reserved[8]; 1543 } __packed; 1544 1545 /* Time Sync IC defs */ 1546 #define ICTIMESYNCFLAG_PROBE 0 1547 #define ICTIMESYNCFLAG_SYNC 1 1548 #define ICTIMESYNCFLAG_SAMPLE 2 1549 1550 #ifdef __x86_64__ 1551 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */ 1552 #else 1553 #define WLTIMEDELTA 116444736000000000LL 1554 #endif 1555 1556 struct ictimesync_data { 1557 u64 parenttime; 1558 u64 childtime; 1559 u64 roundtriptime; 1560 u8 flags; 1561 } __packed; 1562 1563 struct ictimesync_ref_data { 1564 u64 parenttime; 1565 u64 vmreferencetime; 1566 u8 flags; 1567 char leapflags; 1568 char stratum; 1569 u8 reserved[3]; 1570 } __packed; 1571 1572 struct hyperv_service_callback { 1573 u8 msg_type; 1574 char *log_msg; 1575 guid_t data; 1576 struct vmbus_channel *channel; 1577 void (*callback)(void *context); 1578 }; 1579 1580 #define MAX_SRV_VER 0x7ffffff 1581 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf, u32 buflen, 1582 const int *fw_version, int fw_vercnt, 1583 const int *srv_version, int srv_vercnt, 1584 int *nego_fw_version, int *nego_srv_version); 1585 1586 void hv_process_channel_removal(struct vmbus_channel *channel); 1587 1588 void vmbus_setevent(struct vmbus_channel *channel); 1589 /* 1590 * Negotiated version with the Host. 1591 */ 1592 1593 extern __u32 vmbus_proto_version; 1594 1595 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id, 1596 const guid_t *shv_host_servie_id); 1597 int vmbus_send_modifychannel(u32 child_relid, u32 target_vp); 1598 void vmbus_set_event(struct vmbus_channel *channel); 1599 1600 /* Get the start of the ring buffer. */ 1601 static inline void * 1602 hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info) 1603 { 1604 return ring_info->ring_buffer->buffer; 1605 } 1606 1607 /* 1608 * Mask off host interrupt callback notifications 1609 */ 1610 static inline void hv_begin_read(struct hv_ring_buffer_info *rbi) 1611 { 1612 rbi->ring_buffer->interrupt_mask = 1; 1613 1614 /* make sure mask update is not reordered */ 1615 virt_mb(); 1616 } 1617 1618 /* 1619 * Re-enable host callback and return number of outstanding bytes 1620 */ 1621 static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi) 1622 { 1623 1624 rbi->ring_buffer->interrupt_mask = 0; 1625 1626 /* make sure mask update is not reordered */ 1627 virt_mb(); 1628 1629 /* 1630 * Now check to see if the ring buffer is still empty. 1631 * If it is not, we raced and we need to process new 1632 * incoming messages. 1633 */ 1634 return hv_get_bytes_to_read(rbi); 1635 } 1636 1637 /* 1638 * An API to support in-place processing of incoming VMBUS packets. 1639 */ 1640 1641 /* Get data payload associated with descriptor */ 1642 static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc) 1643 { 1644 return (void *)((unsigned long)desc + (desc->offset8 << 3)); 1645 } 1646 1647 /* Get data size associated with descriptor */ 1648 static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc) 1649 { 1650 return (desc->len8 << 3) - (desc->offset8 << 3); 1651 } 1652 1653 1654 struct vmpacket_descriptor * 1655 hv_pkt_iter_first(struct vmbus_channel *channel); 1656 1657 struct vmpacket_descriptor * 1658 __hv_pkt_iter_next(struct vmbus_channel *channel, 1659 const struct vmpacket_descriptor *pkt); 1660 1661 void hv_pkt_iter_close(struct vmbus_channel *channel); 1662 1663 /* 1664 * Get next packet descriptor from iterator 1665 * If at end of list, return NULL and update host. 1666 */ 1667 static inline struct vmpacket_descriptor * 1668 hv_pkt_iter_next(struct vmbus_channel *channel, 1669 const struct vmpacket_descriptor *pkt) 1670 { 1671 struct vmpacket_descriptor *nxt; 1672 1673 nxt = __hv_pkt_iter_next(channel, pkt); 1674 if (!nxt) 1675 hv_pkt_iter_close(channel); 1676 1677 return nxt; 1678 } 1679 1680 #define foreach_vmbus_pkt(pkt, channel) \ 1681 for (pkt = hv_pkt_iter_first(channel); pkt; \ 1682 pkt = hv_pkt_iter_next(channel, pkt)) 1683 1684 /* 1685 * Interface for passing data between SR-IOV PF and VF drivers. The VF driver 1686 * sends requests to read and write blocks. Each block must be 128 bytes or 1687 * smaller. Optionally, the VF driver can register a callback function which 1688 * will be invoked when the host says that one or more of the first 64 block 1689 * IDs is "invalid" which means that the VF driver should reread them. 1690 */ 1691 #define HV_CONFIG_BLOCK_SIZE_MAX 128 1692 1693 int hyperv_read_cfg_blk(struct pci_dev *dev, void *buf, unsigned int buf_len, 1694 unsigned int block_id, unsigned int *bytes_returned); 1695 int hyperv_write_cfg_blk(struct pci_dev *dev, void *buf, unsigned int len, 1696 unsigned int block_id); 1697 int hyperv_reg_block_invalidate(struct pci_dev *dev, void *context, 1698 void (*block_invalidate)(void *context, 1699 u64 block_mask)); 1700 1701 struct hyperv_pci_block_ops { 1702 int (*read_block)(struct pci_dev *dev, void *buf, unsigned int buf_len, 1703 unsigned int block_id, unsigned int *bytes_returned); 1704 int (*write_block)(struct pci_dev *dev, void *buf, unsigned int len, 1705 unsigned int block_id); 1706 int (*reg_blk_invalidate)(struct pci_dev *dev, void *context, 1707 void (*block_invalidate)(void *context, 1708 u64 block_mask)); 1709 }; 1710 1711 extern struct hyperv_pci_block_ops hvpci_block_ops; 1712 1713 static inline unsigned long virt_to_hvpfn(void *addr) 1714 { 1715 phys_addr_t paddr; 1716 1717 if (is_vmalloc_addr(addr)) 1718 paddr = page_to_phys(vmalloc_to_page(addr)) + 1719 offset_in_page(addr); 1720 else 1721 paddr = __pa(addr); 1722 1723 return paddr >> HV_HYP_PAGE_SHIFT; 1724 } 1725 1726 #define NR_HV_HYP_PAGES_IN_PAGE (PAGE_SIZE / HV_HYP_PAGE_SIZE) 1727 #define offset_in_hvpage(ptr) ((unsigned long)(ptr) & ~HV_HYP_PAGE_MASK) 1728 #define HVPFN_UP(x) (((x) + HV_HYP_PAGE_SIZE-1) >> HV_HYP_PAGE_SHIFT) 1729 #define page_to_hvpfn(page) (page_to_pfn(page) * NR_HV_HYP_PAGES_IN_PAGE) 1730 1731 #endif /* _HYPERV_H */ 1732