root/include/linux/hyperv.h

INCLUDED FROM

DEFINITIONS

1. hv_get_bytes_to_read
2. hv_get_bytes_to_write
3. hv_get_avail_to_write_percent
4. hv_ringbuffer_pending_size
5. is_hvsock_channel
6. is_sub_channel
7. set_channel_affinity_state
8. set_channel_read_mode
9. set_per_channel_state
10. get_per_channel_state
11. set_channel_pending_send_size
12. set_low_latency_mode
13. clear_low_latency_mode
14. device_to_hv_device
15. drv_to_hv_drv
16. hv_set_drvdata
17. hv_get_drvdata
18. hv_get_ring_buffer
19. hv_begin_read
20. hv_end_read
21. hv_pkt_data
22. hv_pkt_datalen
23. hv_pkt_iter_next

   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/types.h>
  18 #include <linux/scatterlist.h>
  19 #include <linux/list.h>
  20 #include <linux/timer.h>
  21 #include <linux/completion.h>
  22 #include <linux/device.h>
  23 #include <linux/mod_devicetable.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/reciprocal_div.h>
  26 
  27 #define MAX_PAGE_BUFFER_COUNT                           32
  28 #define MAX_MULTIPAGE_BUFFER_COUNT                      32 /* 128K */
  29 
  30 #pragma pack(push, 1)
  31 
  32 /* Single-page buffer */
  33 struct hv_page_buffer {
  34         u32 len;
  35         u32 offset;
  36         u64 pfn;
  37 };
  38 
  39 /* Multiple-page buffer */
  40 struct hv_multipage_buffer {
  41         /* Length and Offset determines the # of pfns in the array */
  42         u32 len;
  43         u32 offset;
  44         u64 pfn_array[MAX_MULTIPAGE_BUFFER_COUNT];
  45 };
  46 
  47 /*
  48  * Multiple-page buffer array; the pfn array is variable size:
  49  * The number of entries in the PFN array is determined by
  50  * "len" and "offset".
  51  */
  52 struct hv_mpb_array {
  53         /* Length and Offset determines the # of pfns in the array */
  54         u32 len;
  55         u32 offset;
  56         u64 pfn_array[];
  57 };
  58 
  59 /* 0x18 includes the proprietary packet header */
  60 #define MAX_PAGE_BUFFER_PACKET          (0x18 +                 \
  61                                         (sizeof(struct hv_page_buffer) * \
  62                                          MAX_PAGE_BUFFER_COUNT))
  63 #define MAX_MULTIPAGE_BUFFER_PACKET     (0x18 +                 \
  64                                          sizeof(struct hv_multipage_buffer))
  65 
  66 
  67 #pragma pack(pop)
  68 
  69 struct hv_ring_buffer {
  70         /* Offset in bytes from the start of ring data below */
  71         u32 write_index;
  72 
  73         /* Offset in bytes from the start of ring data below */
  74         u32 read_index;
  75 
  76         u32 interrupt_mask;
  77 
  78         /*
  79          * WS2012/Win8 and later versions of Hyper-V implement interrupt
  80          * driven flow management. The feature bit feat_pending_send_sz
  81          * is set by the host on the host->guest ring buffer, and by the
  82          * guest on the guest->host ring buffer.
  83          *
  84          * The meaning of the feature bit is a bit complex in that it has
  85          * semantics that apply to both ring buffers.  If the guest sets
  86          * the feature bit in the guest->host ring buffer, the guest is
  87          * telling the host that:
  88          * 1) It will set the pending_send_sz field in the guest->host ring
  89          *    buffer when it is waiting for space to become available, and
  90          * 2) It will read the pending_send_sz field in the host->guest
  91          *    ring buffer and interrupt the host when it frees enough space
  92          *
  93          * Similarly, if the host sets the feature bit in the host->guest
  94          * ring buffer, the host is telling the guest that:
  95          * 1) It will set the pending_send_sz field in the host->guest ring
  96          *    buffer when it is waiting for space to become available, and
  97          * 2) It will read the pending_send_sz field in the guest->host
  98          *    ring buffer and interrupt the guest when it frees enough space
  99          *
 100          * If either the guest or host does not set the feature bit that it
 101          * owns, that guest or host must do polling if it encounters a full
 102          * ring buffer, and not signal the other end with an interrupt.
 103          */
 104         u32 pending_send_sz;
 105         u32 reserved1[12];
 106         union {
 107                 struct {
 108                         u32 feat_pending_send_sz:1;
 109                 };
 110                 u32 value;
 111         } feature_bits;
 112 
 113         /* Pad it to PAGE_SIZE so that data starts on page boundary */
 114         u8      reserved2[4028];
 115 
 116         /*
 117          * Ring data starts here + RingDataStartOffset
 118          * !!! DO NOT place any fields below this !!!
 119          */
 120         u8 buffer[0];
 121 } __packed;
 122 
 123 struct hv_ring_buffer_info {
 124         struct hv_ring_buffer *ring_buffer;
 125         u32 ring_size;                  /* Include the shared header */
 126         struct reciprocal_value ring_size_div10_reciprocal;
 127         spinlock_t ring_lock;
 128 
 129         u32 ring_datasize;              /* < ring_size */
 130         u32 priv_read_index;
 131         /*
 132          * The ring buffer mutex lock. This lock prevents the ring buffer from
 133          * being freed while the ring buffer is being accessed.
 134          */
 135         struct mutex ring_buffer_mutex;
 136 };
 137 
 138 
 139 static inline u32 hv_get_bytes_to_read(const struct hv_ring_buffer_info *rbi)
 140 {
 141         u32 read_loc, write_loc, dsize, read;
 142 
 143         dsize = rbi->ring_datasize;
 144         read_loc = rbi->ring_buffer->read_index;
 145         write_loc = READ_ONCE(rbi->ring_buffer->write_index);
 146 
 147         read = write_loc >= read_loc ? (write_loc - read_loc) :
 148                 (dsize - read_loc) + write_loc;
 149 
 150         return read;
 151 }
 152 
 153 static inline u32 hv_get_bytes_to_write(const struct hv_ring_buffer_info *rbi)
 154 {
 155         u32 read_loc, write_loc, dsize, write;
 156 
 157         dsize = rbi->ring_datasize;
 158         read_loc = READ_ONCE(rbi->ring_buffer->read_index);
 159         write_loc = rbi->ring_buffer->write_index;
 160 
 161         write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
 162                 read_loc - write_loc;
 163         return write;
 164 }
 165 
 166 static inline u32 hv_get_avail_to_write_percent(
 167                 const struct hv_ring_buffer_info *rbi)
 168 {
 169         u32 avail_write = hv_get_bytes_to_write(rbi);
 170 
 171         return reciprocal_divide(
 172                         (avail_write  << 3) + (avail_write << 1),
 173                         rbi->ring_size_div10_reciprocal);
 174 }
 175 
 176 /*
 177  * VMBUS version is 32 bit entity broken up into
 178  * two 16 bit quantities: major_number. minor_number.
 179  *
 180  * 0 . 13 (Windows Server 2008)
 181  * 1 . 1  (Windows 7)
 182  * 2 . 4  (Windows 8)
 183  * 3 . 0  (Windows 8 R2)
 184  * 4 . 0  (Windows 10)
 185  * 5 . 0  (Newer Windows 10)
 186  */
 187 
 188 #define VERSION_WS2008  ((0 << 16) | (13))
 189 #define VERSION_WIN7    ((1 << 16) | (1))
 190 #define VERSION_WIN8    ((2 << 16) | (4))
 191 #define VERSION_WIN8_1    ((3 << 16) | (0))
 192 #define VERSION_WIN10   ((4 << 16) | (0))
 193 #define VERSION_WIN10_V5 ((5 << 16) | (0))
 194 
 195 #define VERSION_INVAL -1
 196 
 197 #define VERSION_CURRENT VERSION_WIN10_V5
 198 
 199 /* Make maximum size of pipe payload of 16K */
 200 #define MAX_PIPE_DATA_PAYLOAD           (sizeof(u8) * 16384)
 201 
 202 /* Define PipeMode values. */
 203 #define VMBUS_PIPE_TYPE_BYTE            0x00000000
 204 #define VMBUS_PIPE_TYPE_MESSAGE         0x00000004
 205 
 206 /* The size of the user defined data buffer for non-pipe offers. */
 207 #define MAX_USER_DEFINED_BYTES          120
 208 
 209 /* The size of the user defined data buffer for pipe offers. */
 210 #define MAX_PIPE_USER_DEFINED_BYTES     116
 211 
 212 /*
 213  * At the center of the Channel Management library is the Channel Offer. This
 214  * struct contains the fundamental information about an offer.
 215  */
 216 struct vmbus_channel_offer {
 217         guid_t if_type;
 218         guid_t if_instance;
 219 
 220         /*
 221          * These two fields are not currently used.
 222          */
 223         u64 reserved1;
 224         u64 reserved2;
 225 
 226         u16 chn_flags;
 227         u16 mmio_megabytes;             /* in bytes * 1024 * 1024 */
 228 
 229         union {
 230                 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
 231                 struct {
 232                         unsigned char user_def[MAX_USER_DEFINED_BYTES];
 233                 } std;
 234 
 235                 /*
 236                  * Pipes:
 237                  * The following sructure is an integrated pipe protocol, which
 238                  * is implemented on top of standard user-defined data. Pipe
 239                  * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
 240                  * use.
 241                  */
 242                 struct {
 243                         u32  pipe_mode;
 244                         unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
 245                 } pipe;
 246         } u;
 247         /*
 248          * The sub_channel_index is defined in Win8: a value of zero means a
 249          * primary channel and a value of non-zero means a sub-channel.
 250          *
 251          * Before Win8, the field is reserved, meaning it's always zero.
 252          */
 253         u16 sub_channel_index;
 254         u16 reserved3;
 255 } __packed;
 256 
 257 /* Server Flags */
 258 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE        1
 259 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES    2
 260 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS            4
 261 #define VMBUS_CHANNEL_NAMED_PIPE_MODE                   0x10
 262 #define VMBUS_CHANNEL_LOOPBACK_OFFER                    0x100
 263 #define VMBUS_CHANNEL_PARENT_OFFER                      0x200
 264 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION    0x400
 265 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER              0x2000
 266 
 267 struct vmpacket_descriptor {
 268         u16 type;
 269         u16 offset8;
 270         u16 len8;
 271         u16 flags;
 272         u64 trans_id;
 273 } __packed;
 274 
 275 struct vmpacket_header {
 276         u32 prev_pkt_start_offset;
 277         struct vmpacket_descriptor descriptor;
 278 } __packed;
 279 
 280 struct vmtransfer_page_range {
 281         u32 byte_count;
 282         u32 byte_offset;
 283 } __packed;
 284 
 285 struct vmtransfer_page_packet_header {
 286         struct vmpacket_descriptor d;
 287         u16 xfer_pageset_id;
 288         u8  sender_owns_set;
 289         u8 reserved;
 290         u32 range_cnt;
 291         struct vmtransfer_page_range ranges[1];
 292 } __packed;
 293 
 294 struct vmgpadl_packet_header {
 295         struct vmpacket_descriptor d;
 296         u32 gpadl;
 297         u32 reserved;
 298 } __packed;
 299 
 300 struct vmadd_remove_transfer_page_set {
 301         struct vmpacket_descriptor d;
 302         u32 gpadl;
 303         u16 xfer_pageset_id;
 304         u16 reserved;
 305 } __packed;
 306 
 307 /*
 308  * This structure defines a range in guest physical space that can be made to
 309  * look virtually contiguous.
 310  */
 311 struct gpa_range {
 312         u32 byte_count;
 313         u32 byte_offset;
 314         u64 pfn_array[0];
 315 };
 316 
 317 /*
 318  * This is the format for an Establish Gpadl packet, which contains a handle by
 319  * which this GPADL will be known and a set of GPA ranges associated with it.
 320  * This can be converted to a MDL by the guest OS.  If there are multiple GPA
 321  * ranges, then the resulting MDL will be "chained," representing multiple VA
 322  * ranges.
 323  */
 324 struct vmestablish_gpadl {
 325         struct vmpacket_descriptor d;
 326         u32 gpadl;
 327         u32 range_cnt;
 328         struct gpa_range range[1];
 329 } __packed;
 330 
 331 /*
 332  * This is the format for a Teardown Gpadl packet, which indicates that the
 333  * GPADL handle in the Establish Gpadl packet will never be referenced again.
 334  */
 335 struct vmteardown_gpadl {
 336         struct vmpacket_descriptor d;
 337         u32 gpadl;
 338         u32 reserved;   /* for alignment to a 8-byte boundary */
 339 } __packed;
 340 
 341 /*
 342  * This is the format for a GPA-Direct packet, which contains a set of GPA
 343  * ranges, in addition to commands and/or data.
 344  */
 345 struct vmdata_gpa_direct {
 346         struct vmpacket_descriptor d;
 347         u32 reserved;
 348         u32 range_cnt;
 349         struct gpa_range range[1];
 350 } __packed;
 351 
 352 /* This is the format for a Additional Data Packet. */
 353 struct vmadditional_data {
 354         struct vmpacket_descriptor d;
 355         u64 total_bytes;
 356         u32 offset;
 357         u32 byte_cnt;
 358         unsigned char data[1];
 359 } __packed;
 360 
 361 union vmpacket_largest_possible_header {
 362         struct vmpacket_descriptor simple_hdr;
 363         struct vmtransfer_page_packet_header xfer_page_hdr;
 364         struct vmgpadl_packet_header gpadl_hdr;
 365         struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr;
 366         struct vmestablish_gpadl establish_gpadl_hdr;
 367         struct vmteardown_gpadl teardown_gpadl_hdr;
 368         struct vmdata_gpa_direct data_gpa_direct_hdr;
 369 };
 370 
 371 #define VMPACKET_DATA_START_ADDRESS(__packet)   \
 372         (void *)(((unsigned char *)__packet) +  \
 373          ((struct vmpacket_descriptor)__packet)->offset8 * 8)
 374 
 375 #define VMPACKET_DATA_LENGTH(__packet)          \
 376         ((((struct vmpacket_descriptor)__packet)->len8 -        \
 377           ((struct vmpacket_descriptor)__packet)->offset8) * 8)
 378 
 379 #define VMPACKET_TRANSFER_MODE(__packet)        \
 380         (((struct IMPACT)__packet)->type)
 381 
 382 enum vmbus_packet_type {
 383         VM_PKT_INVALID                          = 0x0,
 384         VM_PKT_SYNCH                            = 0x1,
 385         VM_PKT_ADD_XFER_PAGESET                 = 0x2,
 386         VM_PKT_RM_XFER_PAGESET                  = 0x3,
 387         VM_PKT_ESTABLISH_GPADL                  = 0x4,
 388         VM_PKT_TEARDOWN_GPADL                   = 0x5,
 389         VM_PKT_DATA_INBAND                      = 0x6,
 390         VM_PKT_DATA_USING_XFER_PAGES            = 0x7,
 391         VM_PKT_DATA_USING_GPADL                 = 0x8,
 392         VM_PKT_DATA_USING_GPA_DIRECT            = 0x9,
 393         VM_PKT_CANCEL_REQUEST                   = 0xa,
 394         VM_PKT_COMP                             = 0xb,
 395         VM_PKT_DATA_USING_ADDITIONAL_PKT        = 0xc,
 396         VM_PKT_ADDITIONAL_DATA                  = 0xd
 397 };
 398 
 399 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED     1
 400 
 401 
 402 /* Version 1 messages */
 403 enum vmbus_channel_message_type {
 404         CHANNELMSG_INVALID                      =  0,
 405         CHANNELMSG_OFFERCHANNEL         =  1,
 406         CHANNELMSG_RESCIND_CHANNELOFFER =  2,
 407         CHANNELMSG_REQUESTOFFERS                =  3,
 408         CHANNELMSG_ALLOFFERS_DELIVERED  =  4,
 409         CHANNELMSG_OPENCHANNEL          =  5,
 410         CHANNELMSG_OPENCHANNEL_RESULT           =  6,
 411         CHANNELMSG_CLOSECHANNEL         =  7,
 412         CHANNELMSG_GPADL_HEADER         =  8,
 413         CHANNELMSG_GPADL_BODY                   =  9,
 414         CHANNELMSG_GPADL_CREATED                = 10,
 415         CHANNELMSG_GPADL_TEARDOWN               = 11,
 416         CHANNELMSG_GPADL_TORNDOWN               = 12,
 417         CHANNELMSG_RELID_RELEASED               = 13,
 418         CHANNELMSG_INITIATE_CONTACT             = 14,
 419         CHANNELMSG_VERSION_RESPONSE             = 15,
 420         CHANNELMSG_UNLOAD                       = 16,
 421         CHANNELMSG_UNLOAD_RESPONSE              = 17,
 422         CHANNELMSG_18                           = 18,
 423         CHANNELMSG_19                           = 19,
 424         CHANNELMSG_20                           = 20,
 425         CHANNELMSG_TL_CONNECT_REQUEST           = 21,
 426         CHANNELMSG_COUNT
 427 };
 428 
 429 /* Hyper-V supports about 2048 channels, and the RELIDs start with 1. */
 430 #define INVALID_RELID   U32_MAX
 431 
 432 struct vmbus_channel_message_header {
 433         enum vmbus_channel_message_type msgtype;
 434         u32 padding;
 435 } __packed;
 436 
 437 /* Query VMBus Version parameters */
 438 struct vmbus_channel_query_vmbus_version {
 439         struct vmbus_channel_message_header header;
 440         u32 version;
 441 } __packed;
 442 
 443 /* VMBus Version Supported parameters */
 444 struct vmbus_channel_version_supported {
 445         struct vmbus_channel_message_header header;
 446         u8 version_supported;
 447 } __packed;
 448 
 449 /* Offer Channel parameters */
 450 struct vmbus_channel_offer_channel {
 451         struct vmbus_channel_message_header header;
 452         struct vmbus_channel_offer offer;
 453         u32 child_relid;
 454         u8 monitorid;
 455         /*
 456          * win7 and beyond splits this field into a bit field.
 457          */
 458         u8 monitor_allocated:1;
 459         u8 reserved:7;
 460         /*
 461          * These are new fields added in win7 and later.
 462          * Do not access these fields without checking the
 463          * negotiated protocol.
 464          *
 465          * If "is_dedicated_interrupt" is set, we must not set the
 466          * associated bit in the channel bitmap while sending the
 467          * interrupt to the host.
 468          *
 469          * connection_id is to be used in signaling the host.
 470          */
 471         u16 is_dedicated_interrupt:1;
 472         u16 reserved1:15;
 473         u32 connection_id;
 474 } __packed;
 475 
 476 /* Rescind Offer parameters */
 477 struct vmbus_channel_rescind_offer {
 478         struct vmbus_channel_message_header header;
 479         u32 child_relid;
 480 } __packed;
 481 
 482 static inline u32
 483 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info *rbi)
 484 {
 485         return rbi->ring_buffer->pending_send_sz;
 486 }
 487 
 488 /*
 489  * Request Offer -- no parameters, SynIC message contains the partition ID
 490  * Set Snoop -- no parameters, SynIC message contains the partition ID
 491  * Clear Snoop -- no parameters, SynIC message contains the partition ID
 492  * All Offers Delivered -- no parameters, SynIC message contains the partition
 493  *                         ID
 494  * Flush Client -- no parameters, SynIC message contains the partition ID
 495  */
 496 
 497 /* Open Channel parameters */
 498 struct vmbus_channel_open_channel {
 499         struct vmbus_channel_message_header header;
 500 
 501         /* Identifies the specific VMBus channel that is being opened. */
 502         u32 child_relid;
 503 
 504         /* ID making a particular open request at a channel offer unique. */
 505         u32 openid;
 506 
 507         /* GPADL for the channel's ring buffer. */
 508         u32 ringbuffer_gpadlhandle;
 509 
 510         /*
 511          * Starting with win8, this field will be used to specify
 512          * the target virtual processor on which to deliver the interrupt for
 513          * the host to guest communication.
 514          * Prior to win8, incoming channel interrupts would only
 515          * be delivered on cpu 0. Setting this value to 0 would
 516          * preserve the earlier behavior.
 517          */
 518         u32 target_vp;
 519 
 520         /*
 521          * The upstream ring buffer begins at offset zero in the memory
 522          * described by RingBufferGpadlHandle. The downstream ring buffer
 523          * follows it at this offset (in pages).
 524          */
 525         u32 downstream_ringbuffer_pageoffset;
 526 
 527         /* User-specific data to be passed along to the server endpoint. */
 528         unsigned char userdata[MAX_USER_DEFINED_BYTES];
 529 } __packed;
 530 
 531 /* Open Channel Result parameters */
 532 struct vmbus_channel_open_result {
 533         struct vmbus_channel_message_header header;
 534         u32 child_relid;
 535         u32 openid;
 536         u32 status;
 537 } __packed;
 538 
 539 /* Close channel parameters; */
 540 struct vmbus_channel_close_channel {
 541         struct vmbus_channel_message_header header;
 542         u32 child_relid;
 543 } __packed;
 544 
 545 /* Channel Message GPADL */
 546 #define GPADL_TYPE_RING_BUFFER          1
 547 #define GPADL_TYPE_SERVER_SAVE_AREA     2
 548 #define GPADL_TYPE_TRANSACTION          8
 549 
 550 /*
 551  * The number of PFNs in a GPADL message is defined by the number of
 552  * pages that would be spanned by ByteCount and ByteOffset.  If the
 553  * implied number of PFNs won't fit in this packet, there will be a
 554  * follow-up packet that contains more.
 555  */
 556 struct vmbus_channel_gpadl_header {
 557         struct vmbus_channel_message_header header;
 558         u32 child_relid;
 559         u32 gpadl;
 560         u16 range_buflen;
 561         u16 rangecount;
 562         struct gpa_range range[0];
 563 } __packed;
 564 
 565 /* This is the followup packet that contains more PFNs. */
 566 struct vmbus_channel_gpadl_body {
 567         struct vmbus_channel_message_header header;
 568         u32 msgnumber;
 569         u32 gpadl;
 570         u64 pfn[0];
 571 } __packed;
 572 
 573 struct vmbus_channel_gpadl_created {
 574         struct vmbus_channel_message_header header;
 575         u32 child_relid;
 576         u32 gpadl;
 577         u32 creation_status;
 578 } __packed;
 579 
 580 struct vmbus_channel_gpadl_teardown {
 581         struct vmbus_channel_message_header header;
 582         u32 child_relid;
 583         u32 gpadl;
 584 } __packed;
 585 
 586 struct vmbus_channel_gpadl_torndown {
 587         struct vmbus_channel_message_header header;
 588         u32 gpadl;
 589 } __packed;
 590 
 591 struct vmbus_channel_relid_released {
 592         struct vmbus_channel_message_header header;
 593         u32 child_relid;
 594 } __packed;
 595 
 596 struct vmbus_channel_initiate_contact {
 597         struct vmbus_channel_message_header header;
 598         u32 vmbus_version_requested;
 599         u32 target_vcpu; /* The VCPU the host should respond to */
 600         union {
 601                 u64 interrupt_page;
 602                 struct {
 603                         u8      msg_sint;
 604                         u8      padding1[3];
 605                         u32     padding2;
 606                 };
 607         };
 608         u64 monitor_page1;
 609         u64 monitor_page2;
 610 } __packed;
 611 
 612 /* Hyper-V socket: guest's connect()-ing to host */
 613 struct vmbus_channel_tl_connect_request {
 614         struct vmbus_channel_message_header header;
 615         guid_t guest_endpoint_id;
 616         guid_t host_service_id;
 617 } __packed;
 618 
 619 struct vmbus_channel_version_response {
 620         struct vmbus_channel_message_header header;
 621         u8 version_supported;
 622 
 623         u8 connection_state;
 624         u16 padding;
 625 
 626         /*
 627          * On new hosts that support VMBus protocol 5.0, we must use
 628          * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
 629          * and for subsequent messages, we must use the Message Connection ID
 630          * field in the host-returned Version Response Message.
 631          *
 632          * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
 633          */
 634         u32 msg_conn_id;
 635 } __packed;
 636 
 637 enum vmbus_channel_state {
 638         CHANNEL_OFFER_STATE,
 639         CHANNEL_OPENING_STATE,
 640         CHANNEL_OPEN_STATE,
 641         CHANNEL_OPENED_STATE,
 642 };
 643 
 644 /*
 645  * Represents each channel msg on the vmbus connection This is a
 646  * variable-size data structure depending on the msg type itself
 647  */
 648 struct vmbus_channel_msginfo {
 649         /* Bookkeeping stuff */
 650         struct list_head msglistentry;
 651 
 652         /* So far, this is only used to handle gpadl body message */
 653         struct list_head submsglist;
 654 
 655         /* Synchronize the request/response if needed */
 656         struct completion  waitevent;
 657         struct vmbus_channel *waiting_channel;
 658         union {
 659                 struct vmbus_channel_version_supported version_supported;
 660                 struct vmbus_channel_open_result open_result;
 661                 struct vmbus_channel_gpadl_torndown gpadl_torndown;
 662                 struct vmbus_channel_gpadl_created gpadl_created;
 663                 struct vmbus_channel_version_response version_response;
 664         } response;
 665 
 666         u32 msgsize;
 667         /*
 668          * The channel message that goes out on the "wire".
 669          * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
 670          */
 671         unsigned char msg[0];
 672 };
 673 
 674 struct vmbus_close_msg {
 675         struct vmbus_channel_msginfo info;
 676         struct vmbus_channel_close_channel msg;
 677 };
 678 
 679 /* Define connection identifier type. */
 680 union hv_connection_id {
 681         u32 asu32;
 682         struct {
 683                 u32 id:24;
 684                 u32 reserved:8;
 685         } u;
 686 };
 687 
 688 enum hv_numa_policy {
 689         HV_BALANCED = 0,
 690         HV_LOCALIZED,
 691 };
 692 
 693 enum vmbus_device_type {
 694         HV_IDE = 0,
 695         HV_SCSI,
 696         HV_FC,
 697         HV_NIC,
 698         HV_ND,
 699         HV_PCIE,
 700         HV_FB,
 701         HV_KBD,
 702         HV_MOUSE,
 703         HV_KVP,
 704         HV_TS,
 705         HV_HB,
 706         HV_SHUTDOWN,
 707         HV_FCOPY,
 708         HV_BACKUP,
 709         HV_DM,
 710         HV_UNKNOWN,
 711 };
 712 
 713 struct vmbus_device {
 714         u16  dev_type;
 715         guid_t guid;
 716         bool perf_device;
 717 };
 718 
 719 struct vmbus_channel {
 720         struct list_head listentry;
 721 
 722         struct hv_device *device_obj;
 723 
 724         enum vmbus_channel_state state;
 725 
 726         struct vmbus_channel_offer_channel offermsg;
 727         /*
 728          * These are based on the OfferMsg.MonitorId.
 729          * Save it here for easy access.
 730          */
 731         u8 monitor_grp;
 732         u8 monitor_bit;
 733 
 734         bool rescind; /* got rescind msg */
 735         struct completion rescind_event;
 736 
 737         u32 ringbuffer_gpadlhandle;
 738 
 739         /* Allocated memory for ring buffer */
 740         struct page *ringbuffer_page;
 741         u32 ringbuffer_pagecount;
 742         u32 ringbuffer_send_offset;
 743         struct hv_ring_buffer_info outbound;    /* send to parent */
 744         struct hv_ring_buffer_info inbound;     /* receive from parent */
 745 
 746         struct vmbus_close_msg close_msg;
 747 
 748         /* Statistics */
 749         u64     interrupts;     /* Host to Guest interrupts */
 750         u64     sig_events;     /* Guest to Host events */
 751 
 752         /*
 753          * Guest to host interrupts caused by the outbound ring buffer changing
 754          * from empty to not empty.
 755          */
 756         u64 intr_out_empty;
 757 
 758         /*
 759          * Indicates that a full outbound ring buffer was encountered. The flag
 760          * is set to true when a full outbound ring buffer is encountered and
 761          * set to false when a write to the outbound ring buffer is completed.
 762          */
 763         bool out_full_flag;
 764 
 765         /* Channel callback's invoked in softirq context */
 766         struct tasklet_struct callback_event;
 767         void (*onchannel_callback)(void *context);
 768         void *channel_callback_context;
 769 
 770         /*
 771          * A channel can be marked for one of three modes of reading:
 772          *   BATCHED - callback called from taslket and should read
 773          *            channel until empty. Interrupts from the host
 774          *            are masked while read is in process (default).
 775          *   DIRECT - callback called from tasklet (softirq).
 776          *   ISR - callback called in interrupt context and must
 777          *         invoke its own deferred processing.
 778          *         Host interrupts are disabled and must be re-enabled
 779          *         when ring is empty.
 780          */
 781         enum hv_callback_mode {
 782                 HV_CALL_BATCHED,
 783                 HV_CALL_DIRECT,
 784                 HV_CALL_ISR
 785         } callback_mode;
 786 
 787         bool is_dedicated_interrupt;
 788         u64 sig_event;
 789 
 790         /*
 791          * Starting with win8, this field will be used to specify
 792          * the target virtual processor on which to deliver the interrupt for
 793          * the host to guest communication.
 794          * Prior to win8, incoming channel interrupts would only
 795          * be delivered on cpu 0. Setting this value to 0 would
 796          * preserve the earlier behavior.
 797          */
 798         u32 target_vp;
 799         /* The corresponding CPUID in the guest */
 800         u32 target_cpu;
 801         /*
 802          * State to manage the CPU affiliation of channels.
 803          */
 804         struct cpumask alloced_cpus_in_node;
 805         int numa_node;
 806         /*
 807          * Support for sub-channels. For high performance devices,
 808          * it will be useful to have multiple sub-channels to support
 809          * a scalable communication infrastructure with the host.
 810          * The support for sub-channels is implemented as an extention
 811          * to the current infrastructure.
 812          * The initial offer is considered the primary channel and this
 813          * offer message will indicate if the host supports sub-channels.
 814          * The guest is free to ask for sub-channels to be offerred and can
 815          * open these sub-channels as a normal "primary" channel. However,
 816          * all sub-channels will have the same type and instance guids as the
 817          * primary channel. Requests sent on a given channel will result in a
 818          * response on the same channel.
 819          */
 820 
 821         /*
 822          * Sub-channel creation callback. This callback will be called in
 823          * process context when a sub-channel offer is received from the host.
 824          * The guest can open the sub-channel in the context of this callback.
 825          */
 826         void (*sc_creation_callback)(struct vmbus_channel *new_sc);
 827 
 828         /*
 829          * Channel rescind callback. Some channels (the hvsock ones), need to
 830          * register a callback which is invoked in vmbus_onoffer_rescind().
 831          */
 832         void (*chn_rescind_callback)(struct vmbus_channel *channel);
 833 
 834         /*
 835          * The spinlock to protect the structure. It is being used to protect
 836          * test-and-set access to various attributes of the structure as well
 837          * as all sc_list operations.
 838          */
 839         spinlock_t lock;
 840         /*
 841          * All Sub-channels of a primary channel are linked here.
 842          */
 843         struct list_head sc_list;
 844         /*
 845          * The primary channel this sub-channel belongs to.
 846          * This will be NULL for the primary channel.
 847          */
 848         struct vmbus_channel *primary_channel;
 849         /*
 850          * Support per-channel state for use by vmbus drivers.
 851          */
 852         void *per_channel_state;
 853         /*
 854          * To support per-cpu lookup mapping of relid to channel,
 855          * link up channels based on their CPU affinity.
 856          */
 857         struct list_head percpu_list;
 858 
 859         /*
 860          * Defer freeing channel until after all cpu's have
 861          * gone through grace period.
 862          */
 863         struct rcu_head rcu;
 864 
 865         /*
 866          * For sysfs per-channel properties.
 867          */
 868         struct kobject                  kobj;
 869 
 870         /*
 871          * For performance critical channels (storage, networking
 872          * etc,), Hyper-V has a mechanism to enhance the throughput
 873          * at the expense of latency:
 874          * When the host is to be signaled, we just set a bit in a shared page
 875          * and this bit will be inspected by the hypervisor within a certain
 876          * window and if the bit is set, the host will be signaled. The window
 877          * of time is the monitor latency - currently around 100 usecs. This
 878          * mechanism improves throughput by:
 879          *
 880          * A) Making the host more efficient - each time it wakes up,
 881          *    potentially it will process morev number of packets. The
 882          *    monitor latency allows a batch to build up.
 883          * B) By deferring the hypercall to signal, we will also minimize
 884          *    the interrupts.
 885          *
 886          * Clearly, these optimizations improve throughput at the expense of
 887          * latency. Furthermore, since the channel is shared for both
 888          * control and data messages, control messages currently suffer
 889          * unnecessary latency adversley impacting performance and boot
 890          * time. To fix this issue, permit tagging the channel as being
 891          * in "low latency" mode. In this mode, we will bypass the monitor
 892          * mechanism.
 893          */
 894         bool low_latency;
 895 
 896         /*
 897          * NUMA distribution policy:
 898          * We support two policies:
 899          * 1) Balanced: Here all performance critical channels are
 900          *    distributed evenly amongst all the NUMA nodes.
 901          *    This policy will be the default policy.
 902          * 2) Localized: All channels of a given instance of a
 903          *    performance critical service will be assigned CPUs
 904          *    within a selected NUMA node.
 905          */
 906         enum hv_numa_policy affinity_policy;
 907 
 908         bool probe_done;
 909 
 910         /*
 911          * We must offload the handling of the primary/sub channels
 912          * from the single-threaded vmbus_connection.work_queue to
 913          * two different workqueue, otherwise we can block
 914          * vmbus_connection.work_queue and hang: see vmbus_process_offer().
 915          */
 916         struct work_struct add_channel_work;
 917 
 918         /*
 919          * Guest to host interrupts caused by the inbound ring buffer changing
 920          * from full to not full while a packet is waiting.
 921          */
 922         u64 intr_in_full;
 923 
 924         /*
 925          * The total number of write operations that encountered a full
 926          * outbound ring buffer.
 927          */
 928         u64 out_full_total;
 929 
 930         /*
 931          * The number of write operations that were the first to encounter a
 932          * full outbound ring buffer.
 933          */
 934         u64 out_full_first;
 935 };
 936 
 937 static inline bool is_hvsock_channel(const struct vmbus_channel *c)
 938 {
 939         return !!(c->offermsg.offer.chn_flags &
 940                   VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER);
 941 }
 942 
 943 static inline bool is_sub_channel(const struct vmbus_channel *c)
 944 {
 945         return c->offermsg.offer.sub_channel_index != 0;
 946 }
 947 
 948 static inline void set_channel_affinity_state(struct vmbus_channel *c,
 949                                               enum hv_numa_policy policy)
 950 {
 951         c->affinity_policy = policy;
 952 }
 953 
 954 static inline void set_channel_read_mode(struct vmbus_channel *c,
 955                                         enum hv_callback_mode mode)
 956 {
 957         c->callback_mode = mode;
 958 }
 959 
 960 static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
 961 {
 962         c->per_channel_state = s;
 963 }
 964 
 965 static inline void *get_per_channel_state(struct vmbus_channel *c)
 966 {
 967         return c->per_channel_state;
 968 }
 969 
 970 static inline void set_channel_pending_send_size(struct vmbus_channel *c,
 971                                                  u32 size)
 972 {
 973         unsigned long flags;
 974 
 975         if (size) {
 976                 spin_lock_irqsave(&c->outbound.ring_lock, flags);
 977                 ++c->out_full_total;
 978 
 979                 if (!c->out_full_flag) {
 980                         ++c->out_full_first;
 981                         c->out_full_flag = true;
 982                 }
 983                 spin_unlock_irqrestore(&c->outbound.ring_lock, flags);
 984         } else {
 985                 c->out_full_flag = false;
 986         }
 987 
 988         c->outbound.ring_buffer->pending_send_sz = size;
 989 }
 990 
 991 static inline void set_low_latency_mode(struct vmbus_channel *c)
 992 {
 993         c->low_latency = true;
 994 }
 995 
 996 static inline void clear_low_latency_mode(struct vmbus_channel *c)
 997 {
 998         c->low_latency = false;
 999 }
1000 
1001 void vmbus_onmessage(void *context);
1002 
1003 int vmbus_request_offers(void);
1004 
1005 /*
1006  * APIs for managing sub-channels.
1007  */
1008 
1009 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1010                         void (*sc_cr_cb)(struct vmbus_channel *new_sc));
1011 
1012 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1013                 void (*chn_rescind_cb)(struct vmbus_channel *));
1014 
1015 /*
1016  * Check if sub-channels have already been offerred. This API will be useful
1017  * when the driver is unloaded after establishing sub-channels. In this case,
1018  * when the driver is re-loaded, the driver would have to check if the
1019  * subchannels have already been established before attempting to request
1020  * the creation of sub-channels.
1021  * This function returns TRUE to indicate that subchannels have already been
1022  * created.
1023  * This function should be invoked after setting the callback function for
1024  * sub-channel creation.
1025  */
1026 bool vmbus_are_subchannels_present(struct vmbus_channel *primary);
1027 
1028 /* The format must be the same as struct vmdata_gpa_direct */
1029 struct vmbus_channel_packet_page_buffer {
1030         u16 type;
1031         u16 dataoffset8;
1032         u16 length8;
1033         u16 flags;
1034         u64 transactionid;
1035         u32 reserved;
1036         u32 rangecount;
1037         struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
1038 } __packed;
1039 
1040 /* The format must be the same as struct vmdata_gpa_direct */
1041 struct vmbus_channel_packet_multipage_buffer {
1042         u16 type;
1043         u16 dataoffset8;
1044         u16 length8;
1045         u16 flags;
1046         u64 transactionid;
1047         u32 reserved;
1048         u32 rangecount;         /* Always 1 in this case */
1049         struct hv_multipage_buffer range;
1050 } __packed;
1051 
1052 /* The format must be the same as struct vmdata_gpa_direct */
1053 struct vmbus_packet_mpb_array {
1054         u16 type;
1055         u16 dataoffset8;
1056         u16 length8;
1057         u16 flags;
1058         u64 transactionid;
1059         u32 reserved;
1060         u32 rangecount;         /* Always 1 in this case */
1061         struct hv_mpb_array range;
1062 } __packed;
1063 
1064 int vmbus_alloc_ring(struct vmbus_channel *channel,
1065                      u32 send_size, u32 recv_size);
1066 void vmbus_free_ring(struct vmbus_channel *channel);
1067 
1068 int vmbus_connect_ring(struct vmbus_channel *channel,
1069                        void (*onchannel_callback)(void *context),
1070                        void *context);
1071 int vmbus_disconnect_ring(struct vmbus_channel *channel);
1072 
1073 extern int vmbus_open(struct vmbus_channel *channel,
1074                             u32 send_ringbuffersize,
1075                             u32 recv_ringbuffersize,
1076                             void *userdata,
1077                             u32 userdatalen,
1078                             void (*onchannel_callback)(void *context),
1079                             void *context);
1080 
1081 extern void vmbus_close(struct vmbus_channel *channel);
1082 
1083 extern int vmbus_sendpacket(struct vmbus_channel *channel,
1084                                   void *buffer,
1085                                   u32 bufferLen,
1086                                   u64 requestid,
1087                                   enum vmbus_packet_type type,
1088                                   u32 flags);
1089 
1090 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1091                                             struct hv_page_buffer pagebuffers[],
1092                                             u32 pagecount,
1093                                             void *buffer,
1094                                             u32 bufferlen,
1095                                             u64 requestid);
1096 
1097 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1098                                      struct vmbus_packet_mpb_array *mpb,
1099                                      u32 desc_size,
1100                                      void *buffer,
1101                                      u32 bufferlen,
1102                                      u64 requestid);
1103 
1104 extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
1105                                       void *kbuffer,
1106                                       u32 size,
1107                                       u32 *gpadl_handle);
1108 
1109 extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
1110                                      u32 gpadl_handle);
1111 
1112 void vmbus_reset_channel_cb(struct vmbus_channel *channel);
1113 
1114 extern int vmbus_recvpacket(struct vmbus_channel *channel,
1115                                   void *buffer,
1116                                   u32 bufferlen,
1117                                   u32 *buffer_actual_len,
1118                                   u64 *requestid);
1119 
1120 extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
1121                                      void *buffer,
1122                                      u32 bufferlen,
1123                                      u32 *buffer_actual_len,
1124                                      u64 *requestid);
1125 
1126 
1127 extern void vmbus_ontimer(unsigned long data);
1128 
1129 /* Base driver object */
1130 struct hv_driver {
1131         const char *name;
1132 
1133         /*
1134          * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1135          * channel flag, actually doesn't mean a synthetic device because the
1136          * offer's if_type/if_instance can change for every new hvsock
1137          * connection.
1138          *
1139          * However, to facilitate the notification of new-offer/rescind-offer
1140          * from vmbus driver to hvsock driver, we can handle hvsock offer as
1141          * a special vmbus device, and hence we need the below flag to
1142          * indicate if the driver is the hvsock driver or not: we need to
1143          * specially treat the hvosck offer & driver in vmbus_match().
1144          */
1145         bool hvsock;
1146 
1147         /* the device type supported by this driver */
1148         guid_t dev_type;
1149         const struct hv_vmbus_device_id *id_table;
1150 
1151         struct device_driver driver;
1152 
1153         /* dynamic device GUID's */
1154         struct  {
1155                 spinlock_t lock;
1156                 struct list_head list;
1157         } dynids;
1158 
1159         int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
1160         int (*remove)(struct hv_device *);
1161         void (*shutdown)(struct hv_device *);
1162 
1163         int (*suspend)(struct hv_device *);
1164         int (*resume)(struct hv_device *);
1165 
1166 };
1167 
1168 /* Base device object */
1169 struct hv_device {
1170         /* the device type id of this device */
1171         guid_t dev_type;
1172 
1173         /* the device instance id of this device */
1174         guid_t dev_instance;
1175         u16 vendor_id;
1176         u16 device_id;
1177 
1178         struct device device;
1179         char *driver_override; /* Driver name to force a match */
1180 
1181         struct vmbus_channel *channel;
1182         struct kset          *channels_kset;
1183 };
1184 
1185 
1186 static inline struct hv_device *device_to_hv_device(struct device *d)
1187 {
1188         return container_of(d, struct hv_device, device);
1189 }
1190 
1191 static inline struct hv_driver *drv_to_hv_drv(struct device_driver *d)
1192 {
1193         return container_of(d, struct hv_driver, driver);
1194 }
1195 
1196 static inline void hv_set_drvdata(struct hv_device *dev, void *data)
1197 {
1198         dev_set_drvdata(&dev->device, data);
1199 }
1200 
1201 static inline void *hv_get_drvdata(struct hv_device *dev)
1202 {
1203         return dev_get_drvdata(&dev->device);
1204 }
1205 
1206 struct hv_ring_buffer_debug_info {
1207         u32 current_interrupt_mask;
1208         u32 current_read_index;
1209         u32 current_write_index;
1210         u32 bytes_avail_toread;
1211         u32 bytes_avail_towrite;
1212 };
1213 
1214 
1215 int hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
1216                                 struct hv_ring_buffer_debug_info *debug_info);
1217 
1218 /* Vmbus interface */
1219 #define vmbus_driver_register(driver)   \
1220         __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1221 int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
1222                                          struct module *owner,
1223                                          const char *mod_name);
1224 void vmbus_driver_unregister(struct hv_driver *hv_driver);
1225 
1226 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel);
1227 
1228 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1229                         resource_size_t min, resource_size_t max,
1230                         resource_size_t size, resource_size_t align,
1231                         bool fb_overlap_ok);
1232 void vmbus_free_mmio(resource_size_t start, resource_size_t size);
1233 
1234 /*
1235  * GUID definitions of various offer types - services offered to the guest.
1236  */
1237 
1238 /*
1239  * Network GUID
1240  * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1241  */
1242 #define HV_NIC_GUID \
1243         .guid = GUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1244                           0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1245 
1246 /*
1247  * IDE GUID
1248  * {32412632-86cb-44a2-9b5c-50d1417354f5}
1249  */
1250 #define HV_IDE_GUID \
1251         .guid = GUID_INIT(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1252                           0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1253 
1254 /*
1255  * SCSI GUID
1256  * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1257  */
1258 #define HV_SCSI_GUID \
1259         .guid = GUID_INIT(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1260                           0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1261 
1262 /*
1263  * Shutdown GUID
1264  * {0e0b6031-5213-4934-818b-38d90ced39db}
1265  */
1266 #define HV_SHUTDOWN_GUID \
1267         .guid = GUID_INIT(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1268                           0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1269 
1270 /*
1271  * Time Synch GUID
1272  * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1273  */
1274 #define HV_TS_GUID \
1275         .guid = GUID_INIT(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1276                           0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1277 
1278 /*
1279  * Heartbeat GUID
1280  * {57164f39-9115-4e78-ab55-382f3bd5422d}
1281  */
1282 #define HV_HEART_BEAT_GUID \
1283         .guid = GUID_INIT(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1284                           0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1285 
1286 /*
1287  * KVP GUID
1288  * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1289  */
1290 #define HV_KVP_GUID \
1291         .guid = GUID_INIT(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1292                           0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1293 
1294 /*
1295  * Dynamic memory GUID
1296  * {525074dc-8985-46e2-8057-a307dc18a502}
1297  */
1298 #define HV_DM_GUID \
1299         .guid = GUID_INIT(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1300                           0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1301 
1302 /*
1303  * Mouse GUID
1304  * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1305  */
1306 #define HV_MOUSE_GUID \
1307         .guid = GUID_INIT(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1308                           0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1309 
1310 /*
1311  * Keyboard GUID
1312  * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1313  */
1314 #define HV_KBD_GUID \
1315         .guid = GUID_INIT(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1316                           0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1317 
1318 /*
1319  * VSS (Backup/Restore) GUID
1320  */
1321 #define HV_VSS_GUID \
1322         .guid = GUID_INIT(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1323                           0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1324 /*
1325  * Synthetic Video GUID
1326  * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1327  */
1328 #define HV_SYNTHVID_GUID \
1329         .guid = GUID_INIT(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1330                           0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1331 
1332 /*
1333  * Synthetic FC GUID
1334  * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1335  */
1336 #define HV_SYNTHFC_GUID \
1337         .guid = GUID_INIT(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1338                           0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1339 
1340 /*
1341  * Guest File Copy Service
1342  * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1343  */
1344 
1345 #define HV_FCOPY_GUID \
1346         .guid = GUID_INIT(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1347                           0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1348 
1349 /*
1350  * NetworkDirect. This is the guest RDMA service.
1351  * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1352  */
1353 #define HV_ND_GUID \
1354         .guid = GUID_INIT(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1355                           0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1356 
1357 /*
1358  * PCI Express Pass Through
1359  * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1360  */
1361 
1362 #define HV_PCIE_GUID \
1363         .guid = GUID_INIT(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1364                           0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1365 
1366 /*
1367  * Linux doesn't support the 3 devices: the first two are for
1368  * Automatic Virtual Machine Activation, and the third is for
1369  * Remote Desktop Virtualization.
1370  * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1371  * {3375baf4-9e15-4b30-b765-67acb10d607b}
1372  * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1373  */
1374 
1375 #define HV_AVMA1_GUID \
1376         .guid = GUID_INIT(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1377                           0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1378 
1379 #define HV_AVMA2_GUID \
1380         .guid = GUID_INIT(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1381                           0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1382 
1383 #define HV_RDV_GUID \
1384         .guid = GUID_INIT(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1385                           0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1386 
1387 /*
1388  * Common header for Hyper-V ICs
1389  */
1390 
1391 #define ICMSGTYPE_NEGOTIATE             0
1392 #define ICMSGTYPE_HEARTBEAT             1
1393 #define ICMSGTYPE_KVPEXCHANGE           2
1394 #define ICMSGTYPE_SHUTDOWN              3
1395 #define ICMSGTYPE_TIMESYNC              4
1396 #define ICMSGTYPE_VSS                   5
1397 
1398 #define ICMSGHDRFLAG_TRANSACTION        1
1399 #define ICMSGHDRFLAG_REQUEST            2
1400 #define ICMSGHDRFLAG_RESPONSE           4
1401 
1402 
1403 /*
1404  * While we want to handle util services as regular devices,
1405  * there is only one instance of each of these services; so
1406  * we statically allocate the service specific state.
1407  */
1408 
1409 struct hv_util_service {
1410         u8 *recv_buffer;
1411         void *channel;
1412         void (*util_cb)(void *);
1413         int (*util_init)(struct hv_util_service *);
1414         void (*util_deinit)(void);
1415 };
1416 
1417 struct vmbuspipe_hdr {
1418         u32 flags;
1419         u32 msgsize;
1420 } __packed;
1421 
1422 struct ic_version {
1423         u16 major;
1424         u16 minor;
1425 } __packed;
1426 
1427 struct icmsg_hdr {
1428         struct ic_version icverframe;
1429         u16 icmsgtype;
1430         struct ic_version icvermsg;
1431         u16 icmsgsize;
1432         u32 status;
1433         u8 ictransaction_id;
1434         u8 icflags;
1435         u8 reserved[2];
1436 } __packed;
1437 
1438 struct icmsg_negotiate {
1439         u16 icframe_vercnt;
1440         u16 icmsg_vercnt;
1441         u32 reserved;
1442         struct ic_version icversion_data[1]; /* any size array */
1443 } __packed;
1444 
1445 struct shutdown_msg_data {
1446         u32 reason_code;
1447         u32 timeout_seconds;
1448         u32 flags;
1449         u8  display_message[2048];
1450 } __packed;
1451 
1452 struct heartbeat_msg_data {
1453         u64 seq_num;
1454         u32 reserved[8];
1455 } __packed;
1456 
1457 /* Time Sync IC defs */
1458 #define ICTIMESYNCFLAG_PROBE    0
1459 #define ICTIMESYNCFLAG_SYNC     1
1460 #define ICTIMESYNCFLAG_SAMPLE   2
1461 
1462 #ifdef __x86_64__
1463 #define WLTIMEDELTA     116444736000000000L     /* in 100ns unit */
1464 #else
1465 #define WLTIMEDELTA     116444736000000000LL
1466 #endif
1467 
1468 struct ictimesync_data {
1469         u64 parenttime;
1470         u64 childtime;
1471         u64 roundtriptime;
1472         u8 flags;
1473 } __packed;
1474 
1475 struct ictimesync_ref_data {
1476         u64 parenttime;
1477         u64 vmreferencetime;
1478         u8 flags;
1479         char leapflags;
1480         char stratum;
1481         u8 reserved[3];
1482 } __packed;
1483 
1484 struct hyperv_service_callback {
1485         u8 msg_type;
1486         char *log_msg;
1487         guid_t data;
1488         struct vmbus_channel *channel;
1489         void (*callback)(void *context);
1490 };
1491 
1492 #define MAX_SRV_VER     0x7ffffff
1493 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf,
1494                                 const int *fw_version, int fw_vercnt,
1495                                 const int *srv_version, int srv_vercnt,
1496                                 int *nego_fw_version, int *nego_srv_version);
1497 
1498 void hv_process_channel_removal(struct vmbus_channel *channel);
1499 
1500 void vmbus_setevent(struct vmbus_channel *channel);
1501 /*
1502  * Negotiated version with the Host.
1503  */
1504 
1505 extern __u32 vmbus_proto_version;
1506 
1507 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
1508                                   const guid_t *shv_host_servie_id);
1509 void vmbus_set_event(struct vmbus_channel *channel);
1510 
1511 /* Get the start of the ring buffer. */
1512 static inline void *
1513 hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info)
1514 {
1515         return ring_info->ring_buffer->buffer;
1516 }
1517 
1518 /*
1519  * Mask off host interrupt callback notifications
1520  */
1521 static inline void hv_begin_read(struct hv_ring_buffer_info *rbi)
1522 {
1523         rbi->ring_buffer->interrupt_mask = 1;
1524 
1525         /* make sure mask update is not reordered */
1526         virt_mb();
1527 }
1528 
1529 /*
1530  * Re-enable host callback and return number of outstanding bytes
1531  */
1532 static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi)
1533 {
1534 
1535         rbi->ring_buffer->interrupt_mask = 0;
1536 
1537         /* make sure mask update is not reordered */
1538         virt_mb();
1539 
1540         /*
1541          * Now check to see if the ring buffer is still empty.
1542          * If it is not, we raced and we need to process new
1543          * incoming messages.
1544          */
1545         return hv_get_bytes_to_read(rbi);
1546 }
1547 
1548 /*
1549  * An API to support in-place processing of incoming VMBUS packets.
1550  */
1551 
1552 /* Get data payload associated with descriptor */
1553 static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc)
1554 {
1555         return (void *)((unsigned long)desc + (desc->offset8 << 3));
1556 }
1557 
1558 /* Get data size associated with descriptor */
1559 static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc)
1560 {
1561         return (desc->len8 << 3) - (desc->offset8 << 3);
1562 }
1563 
1564 
1565 struct vmpacket_descriptor *
1566 hv_pkt_iter_first(struct vmbus_channel *channel);
1567 
1568 struct vmpacket_descriptor *
1569 __hv_pkt_iter_next(struct vmbus_channel *channel,
1570                    const struct vmpacket_descriptor *pkt);
1571 
1572 void hv_pkt_iter_close(struct vmbus_channel *channel);
1573 
1574 /*
1575  * Get next packet descriptor from iterator
1576  * If at end of list, return NULL and update host.
1577  */
1578 static inline struct vmpacket_descriptor *
1579 hv_pkt_iter_next(struct vmbus_channel *channel,
1580                  const struct vmpacket_descriptor *pkt)
1581 {
1582         struct vmpacket_descriptor *nxt;
1583 
1584         nxt = __hv_pkt_iter_next(channel, pkt);
1585         if (!nxt)
1586                 hv_pkt_iter_close(channel);
1587 
1588         return nxt;
1589 }
1590 
1591 #define foreach_vmbus_pkt(pkt, channel) \
1592         for (pkt = hv_pkt_iter_first(channel); pkt; \
1593             pkt = hv_pkt_iter_next(channel, pkt))
1594 
1595 /*
1596  * Interface for passing data between SR-IOV PF and VF drivers. The VF driver
1597  * sends requests to read and write blocks. Each block must be 128 bytes or
1598  * smaller. Optionally, the VF driver can register a callback function which
1599  * will be invoked when the host says that one or more of the first 64 block
1600  * IDs is "invalid" which means that the VF driver should reread them.
1601  */
1602 #define HV_CONFIG_BLOCK_SIZE_MAX 128
1603 
1604 int hyperv_read_cfg_blk(struct pci_dev *dev, void *buf, unsigned int buf_len,
1605                         unsigned int block_id, unsigned int *bytes_returned);
1606 int hyperv_write_cfg_blk(struct pci_dev *dev, void *buf, unsigned int len,
1607                          unsigned int block_id);
1608 int hyperv_reg_block_invalidate(struct pci_dev *dev, void *context,
1609                                 void (*block_invalidate)(void *context,
1610                                                          u64 block_mask));
1611 
1612 struct hyperv_pci_block_ops {
1613         int (*read_block)(struct pci_dev *dev, void *buf, unsigned int buf_len,
1614                           unsigned int block_id, unsigned int *bytes_returned);
1615         int (*write_block)(struct pci_dev *dev, void *buf, unsigned int len,
1616                            unsigned int block_id);
1617         int (*reg_blk_invalidate)(struct pci_dev *dev, void *context,
1618                                   void (*block_invalidate)(void *context,
1619                                                            u64 block_mask));
1620 };
1621 
1622 extern struct hyperv_pci_block_ops hvpci_block_ops;
1623 
1624 #endif /* _HYPERV_H */