hyperv.h

   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 <hyperv/hvhdk.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 */
  71enum hv_gpadl_type {
  72	HV_GPADL_BUFFER,
  73	HV_GPADL_RING
  74};
  75
  76/* Single-page buffer */
  77struct hv_page_buffer {
  78	u32 len;
  79	u32 offset;
  80	u64 pfn;
  81};
  82
  83/* Multiple-page buffer */
  84struct 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 */
  96struct 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
 113struct 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
 168/*
 169 * If the requested ring buffer size is at least 8 times the size of the
 170 * header, steal space from the ring buffer for the header. Otherwise, add
 171 * space for the header so that is doesn't take too much of the ring buffer
 172 * space.
 173 *
 174 * The factor of 8 is somewhat arbitrary. The goal is to prevent adding a
 175 * relatively small header (4 Kbytes on x86) to a large-ish power-of-2 ring
 176 * buffer size (such as 128 Kbytes) and so end up making a nearly twice as
 177 * large allocation that will be almost half wasted. As a contrasting example,
 178 * on ARM64 with 64 Kbyte page size, we don't want to take 64 Kbytes for the
 179 * header from a 128 Kbyte allocation, leaving only 64 Kbytes for the ring.
 180 * In this latter case, we must add 64 Kbytes for the header and not worry
 181 * about what's wasted.
 182 */
 183#define VMBUS_HEADER_ADJ(payload_sz) \
 184	((payload_sz) >=  8 * sizeof(struct hv_ring_buffer) ? \
 185	0 : sizeof(struct hv_ring_buffer))
 186
 187/* Calculate the proper size of a ringbuffer, it must be page-aligned */
 188#define VMBUS_RING_SIZE(payload_sz) PAGE_ALIGN(VMBUS_HEADER_ADJ(payload_sz) + \
 189					       (payload_sz))
 190
 191struct hv_ring_buffer_info {
 192	struct hv_ring_buffer *ring_buffer;
 193	u32 ring_size;			/* Include the shared header */
 194	struct reciprocal_value ring_size_div10_reciprocal;
 195	spinlock_t ring_lock;
 196
 197	u32 ring_datasize;		/* < ring_size */
 198	u32 priv_read_index;
 199	/*
 200	 * The ring buffer mutex lock. This lock prevents the ring buffer from
 201	 * being freed while the ring buffer is being accessed.
 202	 */
 203	struct mutex ring_buffer_mutex;
 204
 205	/* Buffer that holds a copy of an incoming host packet */
 206	void *pkt_buffer;
 207	u32 pkt_buffer_size;
 208};
 209
 210
 211static inline u32 hv_get_bytes_to_read(const struct hv_ring_buffer_info *rbi)
 212{
 213	u32 read_loc, write_loc, dsize, read;
 214
 215	dsize = rbi->ring_datasize;
 216	read_loc = rbi->ring_buffer->read_index;
 217	write_loc = READ_ONCE(rbi->ring_buffer->write_index);
 218
 219	read = write_loc >= read_loc ? (write_loc - read_loc) :
 220		(dsize - read_loc) + write_loc;
 221
 222	return read;
 223}
 224
 225static inline u32 hv_get_bytes_to_write(const struct hv_ring_buffer_info *rbi)
 226{
 227	u32 read_loc, write_loc, dsize, write;
 228
 229	dsize = rbi->ring_datasize;
 230	read_loc = READ_ONCE(rbi->ring_buffer->read_index);
 231	write_loc = rbi->ring_buffer->write_index;
 232
 233	write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
 234		read_loc - write_loc;
 235	return write;
 236}
 237
 238static inline u32 hv_get_avail_to_write_percent(
 239		const struct hv_ring_buffer_info *rbi)
 240{
 241	u32 avail_write = hv_get_bytes_to_write(rbi);
 242
 243	return reciprocal_divide(
 244			(avail_write  << 3) + (avail_write << 1),
 245			rbi->ring_size_div10_reciprocal);
 246}
 247
 248/*
 249 * VMBUS version is 32 bit entity broken up into
 250 * two 16 bit quantities: major_number. minor_number.
 251 *
 252 * 0 . 13 (Windows Server 2008)
 253 * 1 . 1  (Windows 7, WS2008 R2)
 254 * 2 . 4  (Windows 8, WS2012)
 255 * 3 . 0  (Windows 8.1, WS2012 R2)
 256 * 4 . 0  (Windows 10)
 257 * 4 . 1  (Windows 10 RS3)
 258 * 5 . 0  (Newer Windows 10)
 259 * 5 . 1  (Windows 10 RS4)
 260 * 5 . 2  (Windows Server 2019, RS5)
 261 * 5 . 3  (Windows Server 2022)
 262 *
 263 * The WS2008 and WIN7 versions are listed here for
 264 * completeness but are no longer supported in the
 265 * Linux kernel.
 266 */
 267
 268#define VMBUS_MAKE_VERSION(MAJ, MIN)	((((u32)MAJ) << 16) | (MIN))
 269#define VERSION_WS2008					VMBUS_MAKE_VERSION(0, 13)
 270#define VERSION_WIN7					VMBUS_MAKE_VERSION(1, 1)
 271#define VERSION_WIN8					VMBUS_MAKE_VERSION(2, 4)
 272#define VERSION_WIN8_1					VMBUS_MAKE_VERSION(3, 0)
 273#define VERSION_WIN10					VMBUS_MAKE_VERSION(4, 0)
 274#define VERSION_WIN10_V4_1				VMBUS_MAKE_VERSION(4, 1)
 275#define VERSION_WIN10_V5				VMBUS_MAKE_VERSION(5, 0)
 276#define VERSION_WIN10_V5_1				VMBUS_MAKE_VERSION(5, 1)
 277#define VERSION_WIN10_V5_2				VMBUS_MAKE_VERSION(5, 2)
 278#define VERSION_WIN10_V5_3				VMBUS_MAKE_VERSION(5, 3)
 279#define VERSION_WIN10_V6_0				VMBUS_MAKE_VERSION(6, 0)
 280
 281/* Make maximum size of pipe payload of 16K */
 282#define MAX_PIPE_DATA_PAYLOAD		(sizeof(u8) * 16384)
 283
 284/* Define PipeMode values. */
 285#define VMBUS_PIPE_TYPE_BYTE		0x00000000
 286#define VMBUS_PIPE_TYPE_MESSAGE		0x00000004
 287
 288/* The size of the user defined data buffer for non-pipe offers. */
 289#define MAX_USER_DEFINED_BYTES		120
 290
 291/* The size of the user defined data buffer for pipe offers. */
 292#define MAX_PIPE_USER_DEFINED_BYTES	116
 293
 294/*
 295 * At the center of the Channel Management library is the Channel Offer. This
 296 * struct contains the fundamental information about an offer.
 297 */
 298struct vmbus_channel_offer {
 299	guid_t if_type;
 300	guid_t if_instance;
 301
 302	/*
 303	 * These two fields are not currently used.
 304	 */
 305	u64 reserved1;
 306	u64 reserved2;
 307
 308	u16 chn_flags;
 309	u16 mmio_megabytes;		/* in bytes * 1024 * 1024 */
 310
 311	union {
 312		/* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
 313		struct {
 314			unsigned char user_def[MAX_USER_DEFINED_BYTES];
 315		} std;
 316
 317		/*
 318		 * Pipes:
 319		 * The following structure is an integrated pipe protocol, which
 320		 * is implemented on top of standard user-defined data. Pipe
 321		 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
 322		 * use.
 323		 */
 324		struct {
 325			u32  pipe_mode;
 326			unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
 327		} pipe;
 328	} u;
 329	/*
 330	 * The sub_channel_index is defined in Win8: a value of zero means a
 331	 * primary channel and a value of non-zero means a sub-channel.
 332	 *
 333	 * Before Win8, the field is reserved, meaning it's always zero.
 334	 */
 335	u16 sub_channel_index;
 336	u16 reserved3;
 337} __packed;
 338
 339/* Server Flags */
 340#define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE		0x0001
 341/*
 342 * This flag indicates that the channel is offered by the paravisor, and must
 343 * use encrypted memory for the channel ring buffer.
 344 */
 345#define VMBUS_CHANNEL_CONFIDENTIAL_RING_BUFFER			0x0002
 346/*
 347 * This flag indicates that the channel is offered by the paravisor, and must
 348 * use encrypted memory for GPA direct packets and additional GPADLs.
 349 */
 350#define VMBUS_CHANNEL_CONFIDENTIAL_EXTERNAL_MEMORY		0x0004
 351#define VMBUS_CHANNEL_NAMED_PIPE_MODE					0x0010
 352#define VMBUS_CHANNEL_LOOPBACK_OFFER					0x0100
 353#define VMBUS_CHANNEL_PARENT_OFFER						0x0200
 354#define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION	0x0400
 355#define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER				0x2000
 356
 357struct vmpacket_descriptor {
 358	u16 type;
 359	u16 offset8;
 360	u16 len8;
 361	u16 flags;
 362	u64 trans_id;
 363} __packed;
 364
 365struct vmpacket_header {
 366	u32 prev_pkt_start_offset;
 367	struct vmpacket_descriptor descriptor;
 368} __packed;
 369
 370struct vmtransfer_page_range {
 371	u32 byte_count;
 372	u32 byte_offset;
 373} __packed;
 374
 375struct vmtransfer_page_packet_header {
 376	struct vmpacket_descriptor d;
 377	u16 xfer_pageset_id;
 378	u8  sender_owns_set;
 379	u8 reserved;
 380	u32 range_cnt;
 381	struct vmtransfer_page_range ranges[];
 382} __packed;
 383
 384/*
 385 * This structure defines a range in guest physical space that can be made to
 386 * look virtually contiguous.
 387 */
 388struct gpa_range {
 389	u32 byte_count;
 390	u32 byte_offset;
 391	u64 pfn_array[];
 392};
 393
 394/*
 395 * This is the format for a GPA-Direct packet, which contains a set of GPA
 396 * ranges, in addition to commands and/or data.
 397 */
 398struct vmdata_gpa_direct {
 399	struct vmpacket_descriptor d;
 400	u32 reserved;
 401	u32 range_cnt;
 402	struct gpa_range range[1];
 403} __packed;
 404
 405#define VMPACKET_DATA_START_ADDRESS(__packet)	\
 406	(void *)(((unsigned char *)__packet) +	\
 407	 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
 408
 409#define VMPACKET_DATA_LENGTH(__packet)		\
 410	((((struct vmpacket_descriptor)__packet)->len8 -	\
 411	  ((struct vmpacket_descriptor)__packet)->offset8) * 8)
 412
 413#define VMPACKET_TRANSFER_MODE(__packet)	\
 414	(((struct IMPACT)__packet)->type)
 415
 416enum vmbus_packet_type {
 417	VM_PKT_INVALID				= 0x0,
 418	VM_PKT_SYNCH				= 0x1,
 419	VM_PKT_ADD_XFER_PAGESET			= 0x2,
 420	VM_PKT_RM_XFER_PAGESET			= 0x3,
 421	VM_PKT_ESTABLISH_GPADL			= 0x4,
 422	VM_PKT_TEARDOWN_GPADL			= 0x5,
 423	VM_PKT_DATA_INBAND			= 0x6,
 424	VM_PKT_DATA_USING_XFER_PAGES		= 0x7,
 425	VM_PKT_DATA_USING_GPADL			= 0x8,
 426	VM_PKT_DATA_USING_GPA_DIRECT		= 0x9,
 427	VM_PKT_CANCEL_REQUEST			= 0xa,
 428	VM_PKT_COMP				= 0xb,
 429	VM_PKT_DATA_USING_ADDITIONAL_PKT	= 0xc,
 430	VM_PKT_ADDITIONAL_DATA			= 0xd
 431};
 432
 433#define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED	1
 434
 435
 436/* Version 1 messages */
 437enum vmbus_channel_message_type {
 438	CHANNELMSG_INVALID			=  0,
 439	CHANNELMSG_OFFERCHANNEL		=  1,
 440	CHANNELMSG_RESCIND_CHANNELOFFER	=  2,
 441	CHANNELMSG_REQUESTOFFERS		=  3,
 442	CHANNELMSG_ALLOFFERS_DELIVERED	=  4,
 443	CHANNELMSG_OPENCHANNEL		=  5,
 444	CHANNELMSG_OPENCHANNEL_RESULT		=  6,
 445	CHANNELMSG_CLOSECHANNEL		=  7,
 446	CHANNELMSG_GPADL_HEADER		=  8,
 447	CHANNELMSG_GPADL_BODY			=  9,
 448	CHANNELMSG_GPADL_CREATED		= 10,
 449	CHANNELMSG_GPADL_TEARDOWN		= 11,
 450	CHANNELMSG_GPADL_TORNDOWN		= 12,
 451	CHANNELMSG_RELID_RELEASED		= 13,
 452	CHANNELMSG_INITIATE_CONTACT		= 14,
 453	CHANNELMSG_VERSION_RESPONSE		= 15,
 454	CHANNELMSG_UNLOAD			= 16,
 455	CHANNELMSG_UNLOAD_RESPONSE		= 17,
 456	CHANNELMSG_18				= 18,
 457	CHANNELMSG_19				= 19,
 458	CHANNELMSG_20				= 20,
 459	CHANNELMSG_TL_CONNECT_REQUEST		= 21,
 460	CHANNELMSG_MODIFYCHANNEL		= 22,
 461	CHANNELMSG_TL_CONNECT_RESULT		= 23,
 462	CHANNELMSG_MODIFYCHANNEL_RESPONSE	= 24,
 463	CHANNELMSG_COUNT
 464};
 465
 466/* Hyper-V supports about 2048 channels, and the RELIDs start with 1. */
 467#define INVALID_RELID	U32_MAX
 468
 469struct vmbus_channel_message_header {
 470	enum vmbus_channel_message_type msgtype;
 471	u32 padding;
 472} __packed;
 473
 474/* Query VMBus Version parameters */
 475struct vmbus_channel_query_vmbus_version {
 476	struct vmbus_channel_message_header header;
 477	u32 version;
 478} __packed;
 479
 480/* VMBus Version Supported parameters */
 481struct vmbus_channel_version_supported {
 482	struct vmbus_channel_message_header header;
 483	u8 version_supported;
 484} __packed;
 485
 486/* Offer Channel parameters */
 487struct vmbus_channel_offer_channel {
 488	struct vmbus_channel_message_header header;
 489	struct vmbus_channel_offer offer;
 490	u32 child_relid;
 491	u8 monitorid;
 492	/*
 493	 * win7 and beyond splits this field into a bit field.
 494	 */
 495	u8 monitor_allocated:1;
 496	u8 reserved:7;
 497	/*
 498	 * These are new fields added in win7 and later.
 499	 * Do not access these fields without checking the
 500	 * negotiated protocol.
 501	 *
 502	 * If "is_dedicated_interrupt" is set, we must not set the
 503	 * associated bit in the channel bitmap while sending the
 504	 * interrupt to the host.
 505	 *
 506	 * connection_id is to be used in signaling the host.
 507	 */
 508	u16 is_dedicated_interrupt:1;
 509	u16 reserved1:15;
 510	u32 connection_id;
 511} __packed;
 512
 513/* Rescind Offer parameters */
 514struct vmbus_channel_rescind_offer {
 515	struct vmbus_channel_message_header header;
 516	u32 child_relid;
 517} __packed;
 518
 519/*
 520 * Request Offer -- no parameters, SynIC message contains the partition ID
 521 * Set Snoop -- no parameters, SynIC message contains the partition ID
 522 * Clear Snoop -- no parameters, SynIC message contains the partition ID
 523 * All Offers Delivered -- no parameters, SynIC message contains the partition
 524 *		           ID
 525 * Flush Client -- no parameters, SynIC message contains the partition ID
 526 */
 527
 528/* Open Channel parameters */
 529struct vmbus_channel_open_channel {
 530	struct vmbus_channel_message_header header;
 531
 532	/* Identifies the specific VMBus channel that is being opened. */
 533	u32 child_relid;
 534
 535	/* ID making a particular open request at a channel offer unique. */
 536	u32 openid;
 537
 538	/* GPADL for the channel's ring buffer. */
 539	u32 ringbuffer_gpadlhandle;
 540
 541	/*
 542	 * Starting with win8, this field will be used to specify
 543	 * the target virtual processor on which to deliver the interrupt for
 544	 * the host to guest communication.
 545	 * Prior to win8, incoming channel interrupts would only
 546	 * be delivered on cpu 0. Setting this value to 0 would
 547	 * preserve the earlier behavior.
 548	 */
 549	u32 target_vp;
 550
 551	/*
 552	 * The upstream ring buffer begins at offset zero in the memory
 553	 * described by RingBufferGpadlHandle. The downstream ring buffer
 554	 * follows it at this offset (in pages).
 555	 */
 556	u32 downstream_ringbuffer_pageoffset;
 557
 558	/* User-specific data to be passed along to the server endpoint. */
 559	unsigned char userdata[MAX_USER_DEFINED_BYTES];
 560} __packed;
 561
 562/* Open Channel Result parameters */
 563struct vmbus_channel_open_result {
 564	struct vmbus_channel_message_header header;
 565	u32 child_relid;
 566	u32 openid;
 567	u32 status;
 568} __packed;
 569
 570/* Modify Channel Result parameters */
 571struct vmbus_channel_modifychannel_response {
 572	struct vmbus_channel_message_header header;
 573	u32 child_relid;
 574	u32 status;
 575} __packed;
 576
 577/* Close channel parameters; */
 578struct vmbus_channel_close_channel {
 579	struct vmbus_channel_message_header header;
 580	u32 child_relid;
 581} __packed;
 582
 583/* Channel Message GPADL */
 584#define GPADL_TYPE_RING_BUFFER		1
 585#define GPADL_TYPE_SERVER_SAVE_AREA	2
 586#define GPADL_TYPE_TRANSACTION		8
 587
 588/*
 589 * The number of PFNs in a GPADL message is defined by the number of
 590 * pages that would be spanned by ByteCount and ByteOffset.  If the
 591 * implied number of PFNs won't fit in this packet, there will be a
 592 * follow-up packet that contains more.
 593 */
 594struct vmbus_channel_gpadl_header {
 595	struct vmbus_channel_message_header header;
 596	u32 child_relid;
 597	u32 gpadl;
 598	u16 range_buflen;
 599	u16 rangecount;
 600	struct gpa_range range[];
 601} __packed;
 602
 603/* This is the followup packet that contains more PFNs. */
 604struct vmbus_channel_gpadl_body {
 605	struct vmbus_channel_message_header header;
 606	u32 msgnumber;
 607	u32 gpadl;
 608	u64 pfn[];
 609} __packed;
 610
 611struct vmbus_channel_gpadl_created {
 612	struct vmbus_channel_message_header header;
 613	u32 child_relid;
 614	u32 gpadl;
 615	u32 creation_status;
 616} __packed;
 617
 618struct vmbus_channel_gpadl_teardown {
 619	struct vmbus_channel_message_header header;
 620	u32 child_relid;
 621	u32 gpadl;
 622} __packed;
 623
 624struct vmbus_channel_gpadl_torndown {
 625	struct vmbus_channel_message_header header;
 626	u32 gpadl;
 627} __packed;
 628
 629struct vmbus_channel_relid_released {
 630	struct vmbus_channel_message_header header;
 631	u32 child_relid;
 632} __packed;
 633
 634/*
 635 * Used by the paravisor only, means that the encrypted ring buffers and
 636 * the encrypted external memory are supported
 637 */
 638#define VMBUS_FEATURE_FLAG_CONFIDENTIAL_CHANNELS	0x10
 639
 640struct vmbus_channel_initiate_contact {
 641	struct vmbus_channel_message_header header;
 642	u32 vmbus_version_requested;
 643	u32 target_vcpu; /* The VCPU the host should respond to */
 644	union {
 645		u64 interrupt_page;
 646		struct {
 647			u8	msg_sint;
 648			u8	msg_vtl;
 649			u8	reserved[2];
 650			u32 feature_flags; /* VMBus version 6.0 */
 651		};
 652	};
 653	u64 monitor_page1;
 654	u64 monitor_page2;
 655} __packed;
 656
 657/* Hyper-V socket: guest's connect()-ing to host */
 658struct vmbus_channel_tl_connect_request {
 659	struct vmbus_channel_message_header header;
 660	guid_t guest_endpoint_id;
 661	guid_t host_service_id;
 662} __packed;
 663
 664/* Modify Channel parameters, cf. vmbus_send_modifychannel() */
 665struct vmbus_channel_modifychannel {
 666	struct vmbus_channel_message_header header;
 667	u32 child_relid;
 668	u32 target_vp;
 669} __packed;
 670
 671struct vmbus_channel_version_response {
 672	struct vmbus_channel_message_header header;
 673	u8 version_supported;
 674
 675	u8 connection_state;
 676	u16 padding;
 677
 678	/*
 679	 * On new hosts that support VMBus protocol 5.0, we must use
 680	 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
 681	 * and for subsequent messages, we must use the Message Connection ID
 682	 * field in the host-returned Version Response Message.
 683	 *
 684	 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
 685	 */
 686	u32 msg_conn_id;
 687} __packed;
 688
 689enum vmbus_channel_state {
 690	CHANNEL_OFFER_STATE,
 691	CHANNEL_OPENING_STATE,
 692	CHANNEL_OPEN_STATE,
 693	CHANNEL_OPENED_STATE,
 694};
 695
 696/*
 697 * Represents each channel msg on the vmbus connection This is a
 698 * variable-size data structure depending on the msg type itself
 699 */
 700struct vmbus_channel_msginfo {
 701	/* Bookkeeping stuff */
 702	struct list_head msglistentry;
 703
 704	/* So far, this is only used to handle gpadl body message */
 705	struct list_head submsglist;
 706
 707	/* Synchronize the request/response if needed */
 708	struct completion  waitevent;
 709	struct vmbus_channel *waiting_channel;
 710	union {
 711		struct vmbus_channel_version_supported version_supported;
 712		struct vmbus_channel_open_result open_result;
 713		struct vmbus_channel_gpadl_torndown gpadl_torndown;
 714		struct vmbus_channel_gpadl_created gpadl_created;
 715		struct vmbus_channel_version_response version_response;
 716		struct vmbus_channel_modifychannel_response modify_response;
 717	} response;
 718
 719	u32 msgsize;
 720	/*
 721	 * The channel message that goes out on the "wire".
 722	 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
 723	 */
 724	unsigned char msg[];
 725};
 726
 727enum vmbus_device_type {
 728	HV_IDE = 0,
 729	HV_SCSI,
 730	HV_FC,
 731	HV_NIC,
 732	HV_ND,
 733	HV_PCIE,
 734	HV_FB,
 735	HV_KBD,
 736	HV_MOUSE,
 737	HV_KVP,
 738	HV_TS,
 739	HV_HB,
 740	HV_SHUTDOWN,
 741	HV_FCOPY,
 742	HV_BACKUP,
 743	HV_DM,
 744	HV_UNKNOWN,
 745};
 746
 747/*
 748 * Provides request ids for VMBus. Encapsulates guest memory
 749 * addresses and stores the next available slot in req_arr
 750 * to generate new ids in constant time.
 751 */
 752struct vmbus_requestor {
 753	u64 *req_arr;
 754	unsigned long *req_bitmap; /* is a given slot available? */
 755	u32 size;
 756	u64 next_request_id;
 757	spinlock_t req_lock; /* provides atomicity */
 758};
 759
 760#define VMBUS_NO_RQSTOR U64_MAX
 761#define VMBUS_RQST_ERROR (U64_MAX - 1)
 762#define VMBUS_RQST_ADDR_ANY U64_MAX
 763/* NetVSC-specific */
 764#define VMBUS_RQST_ID_NO_RESPONSE (U64_MAX - 2)
 765/* StorVSC-specific */
 766#define VMBUS_RQST_INIT (U64_MAX - 2)
 767#define VMBUS_RQST_RESET (U64_MAX - 3)
 768
 769struct vmbus_device {
 770	/* preferred ring buffer size in KB, 0 means no preferred size for this device */
 771	size_t pref_ring_size;
 772	u16  dev_type;
 773	guid_t guid;
 774	bool perf_device;
 775	bool allowed_in_isolated;
 776};
 777
 778#define VMBUS_DEFAULT_MAX_PKT_SIZE 4096
 779
 780struct vmbus_gpadl {
 781	u32 gpadl_handle;
 782	u32 size;
 783	void *buffer;
 784	bool decrypted;
 785};
 786
 787struct vmbus_channel {
 788	struct list_head listentry;
 789
 790	struct hv_device *device_obj;
 791
 792	enum vmbus_channel_state state;
 793
 794	struct vmbus_channel_offer_channel offermsg;
 795	/*
 796	 * These are based on the OfferMsg.MonitorId.
 797	 * Save it here for easy access.
 798	 */
 799	u8 monitor_grp;
 800	u8 monitor_bit;
 801
 802	bool rescind; /* got rescind msg */
 803	bool rescind_ref; /* got rescind msg, got channel reference */
 804	struct completion rescind_event;
 805
 806	struct vmbus_gpadl ringbuffer_gpadlhandle;
 807
 808	/* Allocated memory for ring buffer */
 809	struct page *ringbuffer_page;
 810	u32 ringbuffer_pagecount;
 811	u32 ringbuffer_send_offset;
 812	struct hv_ring_buffer_info outbound;	/* send to parent */
 813	struct hv_ring_buffer_info inbound;	/* receive from parent */
 814
 815	struct vmbus_channel_close_channel close_msg;
 816
 817	/* Statistics */
 818	u64	interrupts;	/* Host to Guest interrupts */
 819	u64	sig_events;	/* Guest to Host events */
 820
 821	/*
 822	 * Guest to host interrupts caused by the outbound ring buffer changing
 823	 * from empty to not empty.
 824	 */
 825	u64 intr_out_empty;
 826
 827	/*
 828	 * Indicates that a full outbound ring buffer was encountered. The flag
 829	 * is set to true when a full outbound ring buffer is encountered and
 830	 * set to false when a write to the outbound ring buffer is completed.
 831	 */
 832	bool out_full_flag;
 833
 834	/* Channel callback's invoked in softirq context */
 835	struct tasklet_struct callback_event;
 836	void (*onchannel_callback)(void *context);
 837	void *channel_callback_context;
 838
 839	void (*change_target_cpu_callback)(struct vmbus_channel *channel,
 840			u32 old, u32 new);
 841
 842	/*
 843	 * Synchronize channel scheduling and channel removal; see the inline
 844	 * comments in vmbus_chan_sched() and vmbus_reset_channel_cb().
 845	 */
 846	spinlock_t sched_lock;
 847
 848	/*
 849	 * A channel can be marked for one of three modes of reading:
 850	 *   BATCHED - callback called from taslket and should read
 851	 *            channel until empty. Interrupts from the host
 852	 *            are masked while read is in process (default).
 853	 *   DIRECT - callback called from tasklet (softirq).
 854	 *   ISR - callback called in interrupt context and must
 855	 *         invoke its own deferred processing.
 856	 *         Host interrupts are disabled and must be re-enabled
 857	 *         when ring is empty.
 858	 */
 859	enum hv_callback_mode {
 860		HV_CALL_BATCHED,
 861		HV_CALL_DIRECT,
 862		HV_CALL_ISR
 863	} callback_mode;
 864
 865	bool is_dedicated_interrupt;
 866	u64 sig_event;
 867
 868	/*
 869	 * Starting with win8, this field will be used to specify the
 870	 * target CPU on which to deliver the interrupt for the host
 871	 * to guest communication.
 872	 *
 873	 * Prior to win8, incoming channel interrupts would only be
 874	 * delivered on CPU 0. Setting this value to 0 would preserve
 875	 * the earlier behavior.
 876	 */
 877	u32 target_cpu;
 878	/*
 879	 * Support for sub-channels. For high performance devices,
 880	 * it will be useful to have multiple sub-channels to support
 881	 * a scalable communication infrastructure with the host.
 882	 * The support for sub-channels is implemented as an extension
 883	 * to the current infrastructure.
 884	 * The initial offer is considered the primary channel and this
 885	 * offer message will indicate if the host supports sub-channels.
 886	 * The guest is free to ask for sub-channels to be offered and can
 887	 * open these sub-channels as a normal "primary" channel. However,
 888	 * all sub-channels will have the same type and instance guids as the
 889	 * primary channel. Requests sent on a given channel will result in a
 890	 * response on the same channel.
 891	 */
 892
 893	/*
 894	 * Sub-channel creation callback. This callback will be called in
 895	 * process context when a sub-channel offer is received from the host.
 896	 * The guest can open the sub-channel in the context of this callback.
 897	 */
 898	void (*sc_creation_callback)(struct vmbus_channel *new_sc);
 899
 900	/*
 901	 * Channel rescind callback. Some channels (the hvsock ones), need to
 902	 * register a callback which is invoked in vmbus_onoffer_rescind().
 903	 */
 904	void (*chn_rescind_callback)(struct vmbus_channel *channel);
 905
 906	/*
 907	 * All Sub-channels of a primary channel are linked here.
 908	 */
 909	struct list_head sc_list;
 910	/*
 911	 * The primary channel this sub-channel belongs to.
 912	 * This will be NULL for the primary channel.
 913	 */
 914	struct vmbus_channel *primary_channel;
 915	/*
 916	 * Support per-channel state for use by vmbus drivers.
 917	 */
 918	void *per_channel_state;
 919
 920	/*
 921	 * Defer freeing channel until after all cpu's have
 922	 * gone through grace period.
 923	 */
 924	struct rcu_head rcu;
 925
 926	/*
 927	 * For sysfs per-channel properties.
 928	 */
 929	struct kobject			kobj;
 930
 931	/*
 932	 * For performance critical channels (storage, networking
 933	 * etc,), Hyper-V has a mechanism to enhance the throughput
 934	 * at the expense of latency:
 935	 * When the host is to be signaled, we just set a bit in a shared page
 936	 * and this bit will be inspected by the hypervisor within a certain
 937	 * window and if the bit is set, the host will be signaled. The window
 938	 * of time is the monitor latency - currently around 100 usecs. This
 939	 * mechanism improves throughput by:
 940	 *
 941	 * A) Making the host more efficient - each time it wakes up,
 942	 *    potentially it will process more number of packets. The
 943	 *    monitor latency allows a batch to build up.
 944	 * B) By deferring the hypercall to signal, we will also minimize
 945	 *    the interrupts.
 946	 *
 947	 * Clearly, these optimizations improve throughput at the expense of
 948	 * latency. Furthermore, since the channel is shared for both
 949	 * control and data messages, control messages currently suffer
 950	 * unnecessary latency adversely impacting performance and boot
 951	 * time. To fix this issue, permit tagging the channel as being
 952	 * in "low latency" mode. In this mode, we will bypass the monitor
 953	 * mechanism.
 954	 */
 955	bool low_latency;
 956
 957	bool probe_done;
 958
 959	/*
 960	 * Cache the device ID here for easy access; this is useful, in
 961	 * particular, in situations where the channel's device_obj has
 962	 * not been allocated/initialized yet.
 963	 */
 964	u16 device_id;
 965
 966	/*
 967	 * We must offload the handling of the primary/sub channels
 968	 * from the single-threaded vmbus_connection.work_queue to
 969	 * two different workqueue, otherwise we can block
 970	 * vmbus_connection.work_queue and hang: see vmbus_process_offer().
 971	 */
 972	struct work_struct add_channel_work;
 973
 974	/*
 975	 * Guest to host interrupts caused by the inbound ring buffer changing
 976	 * from full to not full while a packet is waiting.
 977	 */
 978	u64 intr_in_full;
 979
 980	/*
 981	 * The total number of write operations that encountered a full
 982	 * outbound ring buffer.
 983	 */
 984	u64 out_full_total;
 985
 986	/*
 987	 * The number of write operations that were the first to encounter a
 988	 * full outbound ring buffer.
 989	 */
 990	u64 out_full_first;
 991
 992	/* enabling/disabling fuzz testing on the channel (default is false)*/
 993	bool fuzz_testing_state;
 994
 995	/*
 996	 * Interrupt delay will delay the guest from emptying the ring buffer
 997	 * for a specific amount of time. The delay is in microseconds and will
 998	 * be between 1 to a maximum of 1000, its default is 0 (no delay).
 999	 * The  Message delay will delay guest reading on a per message basis
1000	 * in microseconds between 1 to 1000 with the default being 0
1001	 * (no delay).
1002	 */
1003	u32 fuzz_testing_interrupt_delay;
1004	u32 fuzz_testing_message_delay;
1005
1006	/* callback to generate a request ID from a request address */
1007	u64 (*next_request_id_callback)(struct vmbus_channel *channel, u64 rqst_addr);
1008	/* callback to retrieve a request address from a request ID */
1009	u64 (*request_addr_callback)(struct vmbus_channel *channel, u64 rqst_id);
1010
1011	/* request/transaction ids for VMBus */
1012	struct vmbus_requestor requestor;
1013	u32 rqstor_size;
1014
1015	/* The max size of a packet on this channel */
1016	u32 max_pkt_size;
1017
1018	/* function to mmap ring buffer memory to the channel's sysfs ring attribute */
1019	int (*mmap_ring_buffer)(struct vmbus_channel *channel, struct vm_area_struct *vma);
1020
1021	/* boolean to control visibility of sysfs for ring buffer */
1022	bool ring_sysfs_visible;
1023	/* The ring buffer is encrypted */
1024	bool co_ring_buffer;
1025	/* The external memory is encrypted */
1026	bool co_external_memory;
1027};
1028
1029#define lock_requestor(channel, flags)					\
1030do {									\
1031	struct vmbus_requestor *rqstor = &(channel)->requestor;		\
1032									\
1033	spin_lock_irqsave(&rqstor->req_lock, flags);			\
1034} while (0)
1035
1036static __always_inline void unlock_requestor(struct vmbus_channel *channel,
1037					     unsigned long flags)
1038{
1039	struct vmbus_requestor *rqstor = &channel->requestor;
1040
1041	spin_unlock_irqrestore(&rqstor->req_lock, flags);
1042}
1043
1044u64 vmbus_next_request_id(struct vmbus_channel *channel, u64 rqst_addr);
1045u64 __vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1046			       u64 rqst_addr);
1047u64 vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1048			     u64 rqst_addr);
1049u64 vmbus_request_addr(struct vmbus_channel *channel, u64 trans_id);
1050
1051static inline bool is_co_ring_buffer(const struct vmbus_channel_offer_channel *o)
1052{
1053	return !!(o->offer.chn_flags & VMBUS_CHANNEL_CONFIDENTIAL_RING_BUFFER);
1054}
1055
1056static inline bool is_co_external_memory(const struct vmbus_channel_offer_channel *o)
1057{
1058	return !!(o->offer.chn_flags & VMBUS_CHANNEL_CONFIDENTIAL_EXTERNAL_MEMORY);
1059}
1060
1061static inline bool is_hvsock_offer(const struct vmbus_channel_offer_channel *o)
1062{
1063	return !!(o->offer.chn_flags & VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER);
1064}
1065
1066static inline bool is_hvsock_channel(const struct vmbus_channel *c)
1067{
1068	return is_hvsock_offer(&c->offermsg);
1069}
1070
1071static inline bool is_sub_channel(const struct vmbus_channel *c)
1072{
1073	return c->offermsg.offer.sub_channel_index != 0;
1074}
1075
1076static inline void set_channel_read_mode(struct vmbus_channel *c,
1077					enum hv_callback_mode mode)
1078{
1079	c->callback_mode = mode;
1080}
1081
1082static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
1083{
1084	c->per_channel_state = s;
1085}
1086
1087static inline void *get_per_channel_state(struct vmbus_channel *c)
1088{
1089	return c->per_channel_state;
1090}
1091
1092static inline void set_channel_pending_send_size(struct vmbus_channel *c,
1093						 u32 size)
1094{
1095	unsigned long flags;
1096
1097	if (size) {
1098		spin_lock_irqsave(&c->outbound.ring_lock, flags);
1099		++c->out_full_total;
1100
1101		if (!c->out_full_flag) {
1102			++c->out_full_first;
1103			c->out_full_flag = true;
1104		}
1105		spin_unlock_irqrestore(&c->outbound.ring_lock, flags);
1106	} else {
1107		c->out_full_flag = false;
1108	}
1109
1110	c->outbound.ring_buffer->pending_send_sz = size;
1111}
1112
1113void vmbus_onmessage(struct vmbus_channel_message_header *hdr);
1114
1115int vmbus_request_offers(void);
1116
1117/*
1118 * APIs for managing sub-channels.
1119 */
1120
1121void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1122			void (*sc_cr_cb)(struct vmbus_channel *new_sc));
1123
1124void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1125		void (*chn_rescind_cb)(struct vmbus_channel *));
1126
1127/* The format must be the same as struct vmdata_gpa_direct */
1128struct vmbus_channel_packet_page_buffer {
1129	u16 type;
1130	u16 dataoffset8;
1131	u16 length8;
1132	u16 flags;
1133	u64 transactionid;
1134	u32 reserved;
1135	u32 rangecount;
1136	struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
1137} __packed;
1138
1139/* The format must be the same as struct vmdata_gpa_direct */
1140struct vmbus_channel_packet_multipage_buffer {
1141	u16 type;
1142	u16 dataoffset8;
1143	u16 length8;
1144	u16 flags;
1145	u64 transactionid;
1146	u32 reserved;
1147	u32 rangecount;		/* Always 1 in this case */
1148	struct hv_multipage_buffer range;
1149} __packed;
1150
1151/* The format must be the same as struct vmdata_gpa_direct */
1152struct vmbus_packet_mpb_array {
1153	u16 type;
1154	u16 dataoffset8;
1155	u16 length8;
1156	u16 flags;
1157	u64 transactionid;
1158	u32 reserved;
1159	u32 rangecount;         /* Always 1 in this case */
1160	struct hv_mpb_array range;
1161} __packed;
1162
1163int vmbus_alloc_ring(struct vmbus_channel *channel,
1164		     u32 send_size, u32 recv_size);
1165void vmbus_free_ring(struct vmbus_channel *channel);
1166
1167int vmbus_connect_ring(struct vmbus_channel *channel,
1168		       void (*onchannel_callback)(void *context),
1169		       void *context);
1170int vmbus_disconnect_ring(struct vmbus_channel *channel);
1171
1172extern int vmbus_open(struct vmbus_channel *channel,
1173			    u32 send_ringbuffersize,
1174			    u32 recv_ringbuffersize,
1175			    void *userdata,
1176			    u32 userdatalen,
1177			    void (*onchannel_callback)(void *context),
1178			    void *context);
1179
1180extern void vmbus_close(struct vmbus_channel *channel);
1181
1182extern int vmbus_sendpacket_getid(struct vmbus_channel *channel,
1183				  void *buffer,
1184				  u32 bufferLen,
1185				  u64 requestid,
1186				  u64 *trans_id,
1187				  enum vmbus_packet_type type,
1188				  u32 flags);
1189extern int vmbus_sendpacket(struct vmbus_channel *channel,
1190				  void *buffer,
1191				  u32 bufferLen,
1192				  u64 requestid,
1193				  enum vmbus_packet_type type,
1194				  u32 flags);
1195
1196extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1197				     struct vmbus_packet_mpb_array *mpb,
1198				     u32 desc_size,
1199				     void *buffer,
1200				     u32 bufferlen,
1201				     u64 requestid);
1202
1203extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
1204				      void *kbuffer,
1205				      u32 size,
1206				      struct vmbus_gpadl *gpadl);
1207
1208extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
1209				     struct vmbus_gpadl *gpadl);
1210
1211void vmbus_reset_channel_cb(struct vmbus_channel *channel);
1212
1213extern int vmbus_recvpacket(struct vmbus_channel *channel,
1214				  void *buffer,
1215				  u32 bufferlen,
1216				  u32 *buffer_actual_len,
1217				  u64 *requestid);
1218
1219extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
1220				     void *buffer,
1221				     u32 bufferlen,
1222				     u32 *buffer_actual_len,
1223				     u64 *requestid);
1224
1225/* Base driver object */
1226struct hv_driver {
1227	const char *name;
1228
1229	/*
1230	 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1231	 * channel flag, actually doesn't mean a synthetic device because the
1232	 * offer's if_type/if_instance can change for every new hvsock
1233	 * connection.
1234	 *
1235	 * However, to facilitate the notification of new-offer/rescind-offer
1236	 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1237	 * a special vmbus device, and hence we need the below flag to
1238	 * indicate if the driver is the hvsock driver or not: we need to
1239	 * specially treat the hvosck offer & driver in vmbus_match().
1240	 */
1241	bool hvsock;
1242
1243	/* the device type supported by this driver */
1244	guid_t dev_type;
1245	const struct hv_vmbus_device_id *id_table;
1246
1247	struct device_driver driver;
1248
1249	/* dynamic device GUID's */
1250	struct  {
1251		spinlock_t lock;
1252		struct list_head list;
1253	} dynids;
1254
1255	int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
1256	void (*remove)(struct hv_device *dev);
1257	void (*shutdown)(struct hv_device *);
1258
1259	int (*suspend)(struct hv_device *);
1260	int (*resume)(struct hv_device *);
1261
1262};
1263
1264/* Base device object */
1265struct hv_device {
1266	/* the device type id of this device */
1267	guid_t dev_type;
1268
1269	/* the device instance id of this device */
1270	guid_t dev_instance;
1271	u16 vendor_id;
1272	u16 device_id;
1273
1274	struct device device;
1275	/*
1276	 * Driver name to force a match.  Do not set directly, because core
1277	 * frees it.  Use driver_set_override() to set or clear it.
1278	 */
1279	const char *driver_override;
1280
1281	struct vmbus_channel *channel;
1282	struct kset	     *channels_kset;
1283	struct device_dma_parameters dma_parms;
1284	u64 dma_mask;
1285
1286	/* place holder to keep track of the dir for hv device in debugfs */
1287	struct dentry *debug_dir;
1288
1289};
1290
1291
1292#define device_to_hv_device(d)	container_of_const(d, struct hv_device, device)
1293#define drv_to_hv_drv(d)	container_of_const(d, struct hv_driver, driver)
1294
1295static inline void hv_set_drvdata(struct hv_device *dev, void *data)
1296{
1297	dev_set_drvdata(&dev->device, data);
1298}
1299
1300static inline void *hv_get_drvdata(struct hv_device *dev)
1301{
1302	return dev_get_drvdata(&dev->device);
1303}
1304
1305struct device *hv_get_vmbus_root_device(void);
1306
1307struct hv_ring_buffer_debug_info {
1308	u32 current_interrupt_mask;
1309	u32 current_read_index;
1310	u32 current_write_index;
1311	u32 bytes_avail_toread;
1312	u32 bytes_avail_towrite;
1313};
1314
1315
1316int hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
1317				struct hv_ring_buffer_debug_info *debug_info);
1318
1319bool hv_ringbuffer_spinlock_busy(struct vmbus_channel *channel);
1320
1321/* Vmbus interface */
1322#define vmbus_driver_register(driver)	\
1323	__vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1324int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
1325					 struct module *owner,
1326					 const char *mod_name);
1327void vmbus_driver_unregister(struct hv_driver *hv_driver);
1328
1329void vmbus_hvsock_device_unregister(struct vmbus_channel *channel);
1330
1331int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1332			resource_size_t min, resource_size_t max,
1333			resource_size_t size, resource_size_t align,
1334			bool fb_overlap_ok);
1335void vmbus_free_mmio(resource_size_t start, resource_size_t size);
1336
1337/*
1338 * GUID definitions of various offer types - services offered to the guest.
1339 */
1340
1341/*
1342 * Network GUID
1343 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1344 */
1345#define HV_NIC_GUID \
1346	.guid = GUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1347			  0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1348
1349/*
1350 * IDE GUID
1351 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1352 */
1353#define HV_IDE_GUID \
1354	.guid = GUID_INIT(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1355			  0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1356
1357/*
1358 * SCSI GUID
1359 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1360 */
1361#define HV_SCSI_GUID \
1362	.guid = GUID_INIT(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1363			  0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1364
1365/*
1366 * Shutdown GUID
1367 * {0e0b6031-5213-4934-818b-38d90ced39db}
1368 */
1369#define HV_SHUTDOWN_GUID \
1370	.guid = GUID_INIT(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1371			  0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1372
1373/*
1374 * Time Synch GUID
1375 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1376 */
1377#define HV_TS_GUID \
1378	.guid = GUID_INIT(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1379			  0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1380
1381/*
1382 * Heartbeat GUID
1383 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1384 */
1385#define HV_HEART_BEAT_GUID \
1386	.guid = GUID_INIT(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1387			  0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1388
1389/*
1390 * KVP GUID
1391 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1392 */
1393#define HV_KVP_GUID \
1394	.guid = GUID_INIT(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1395			  0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1396
1397/*
1398 * Dynamic memory GUID
1399 * {525074dc-8985-46e2-8057-a307dc18a502}
1400 */
1401#define HV_DM_GUID \
1402	.guid = GUID_INIT(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1403			  0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1404
1405/*
1406 * Mouse GUID
1407 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1408 */
1409#define HV_MOUSE_GUID \
1410	.guid = GUID_INIT(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1411			  0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1412
1413/*
1414 * Keyboard GUID
1415 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1416 */
1417#define HV_KBD_GUID \
1418	.guid = GUID_INIT(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1419			  0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1420
1421/*
1422 * VSS (Backup/Restore) GUID
1423 */
1424#define HV_VSS_GUID \
1425	.guid = GUID_INIT(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1426			  0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1427/*
1428 * Synthetic Video GUID
1429 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1430 */
1431#define HV_SYNTHVID_GUID \
1432	.guid = GUID_INIT(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1433			  0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1434
1435/*
1436 * Synthetic FC GUID
1437 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1438 */
1439#define HV_SYNTHFC_GUID \
1440	.guid = GUID_INIT(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1441			  0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1442
1443/*
1444 * Guest File Copy Service
1445 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1446 */
1447
1448#define HV_FCOPY_GUID \
1449	.guid = GUID_INIT(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1450			  0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1451
1452/*
1453 * NetworkDirect. This is the guest RDMA service.
1454 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1455 */
1456#define HV_ND_GUID \
1457	.guid = GUID_INIT(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1458			  0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1459
1460/*
1461 * PCI Express Pass Through
1462 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1463 */
1464
1465#define HV_PCIE_GUID \
1466	.guid = GUID_INIT(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1467			  0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1468
1469/*
1470 * Linux doesn't support these 4 devices: the first two are for
1471 * Automatic Virtual Machine Activation, the third is for
1472 * Remote Desktop Virtualization, and the fourth is Initial
1473 * Machine Configuration (IMC) used only by Windows guests.
1474 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1475 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1476 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1477 * {c376c1c3-d276-48d2-90a9-c04748072c60}
1478 */
1479
1480#define HV_AVMA1_GUID \
1481	.guid = GUID_INIT(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1482			  0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1483
1484#define HV_AVMA2_GUID \
1485	.guid = GUID_INIT(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1486			  0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1487
1488#define HV_RDV_GUID \
1489	.guid = GUID_INIT(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1490			  0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1491
1492#define HV_IMC_GUID \
1493	.guid = GUID_INIT(0xc376c1c3, 0xd276, 0x48d2, 0x90, 0xa9, \
1494			  0xc0, 0x47, 0x48, 0x07, 0x2c, 0x60)
1495
1496/*
1497 * Common header for Hyper-V ICs
1498 */
1499
1500#define ICMSGTYPE_NEGOTIATE		0
1501#define ICMSGTYPE_HEARTBEAT		1
1502#define ICMSGTYPE_KVPEXCHANGE		2
1503#define ICMSGTYPE_SHUTDOWN		3
1504#define ICMSGTYPE_TIMESYNC		4
1505#define ICMSGTYPE_VSS			5
1506#define ICMSGTYPE_FCOPY			7
1507
1508#define ICMSGHDRFLAG_TRANSACTION	1
1509#define ICMSGHDRFLAG_REQUEST		2
1510#define ICMSGHDRFLAG_RESPONSE		4
1511
1512
1513/*
1514 * While we want to handle util services as regular devices,
1515 * there is only one instance of each of these services; so
1516 * we statically allocate the service specific state.
1517 */
1518
1519struct hv_util_service {
1520	u8 *recv_buffer;
1521	void *channel;
1522	void (*util_cb)(void *);
1523	int (*util_init)(struct hv_util_service *);
1524	int (*util_init_transport)(void);
1525	void (*util_deinit)(void);
1526	int (*util_pre_suspend)(void);
1527	int (*util_pre_resume)(void);
1528};
1529
1530struct vmbuspipe_hdr {
1531	u32 flags;
1532	u32 msgsize;
1533} __packed;
1534
1535struct ic_version {
1536	u16 major;
1537	u16 minor;
1538} __packed;
1539
1540struct icmsg_hdr {
1541	struct ic_version icverframe;
1542	u16 icmsgtype;
1543	struct ic_version icvermsg;
1544	u16 icmsgsize;
1545	u32 status;
1546	u8 ictransaction_id;
1547	u8 icflags;
1548	u8 reserved[2];
1549} __packed;
1550
1551#define IC_VERSION_NEGOTIATION_MAX_VER_COUNT 100
1552#define ICMSG_HDR (sizeof(struct vmbuspipe_hdr) + sizeof(struct icmsg_hdr))
1553#define ICMSG_NEGOTIATE_PKT_SIZE(icframe_vercnt, icmsg_vercnt) \
1554	(ICMSG_HDR + sizeof(struct icmsg_negotiate) + \
1555	 (((icframe_vercnt) + (icmsg_vercnt)) * sizeof(struct ic_version)))
1556
1557struct icmsg_negotiate {
1558	u16 icframe_vercnt;
1559	u16 icmsg_vercnt;
1560	u32 reserved;
1561	struct ic_version icversion_data[]; /* any size array */
1562} __packed;
1563
1564struct shutdown_msg_data {
1565	u32 reason_code;
1566	u32 timeout_seconds;
1567	u32 flags;
1568	u8  display_message[2048];
1569} __packed;
1570
1571struct heartbeat_msg_data {
1572	u64 seq_num;
1573	u32 reserved[8];
1574} __packed;
1575
1576/* Time Sync IC defs */
1577#define ICTIMESYNCFLAG_PROBE	0
1578#define ICTIMESYNCFLAG_SYNC	1
1579#define ICTIMESYNCFLAG_SAMPLE	2
1580
1581#ifdef __x86_64__
1582#define WLTIMEDELTA	116444736000000000L	/* in 100ns unit */
1583#else
1584#define WLTIMEDELTA	116444736000000000LL
1585#endif
1586
1587struct ictimesync_data {
1588	u64 parenttime;
1589	u64 childtime;
1590	u64 roundtriptime;
1591	u8 flags;
1592} __packed;
1593
1594struct ictimesync_ref_data {
1595	u64 parenttime;
1596	u64 vmreferencetime;
1597	u8 flags;
1598	char leapflags;
1599	char stratum;
1600	u8 reserved[3];
1601} __packed;
1602
1603struct hyperv_service_callback {
1604	u8 msg_type;
1605	char *log_msg;
1606	guid_t data;
1607	struct vmbus_channel *channel;
1608	void (*callback)(void *context);
1609};
1610
1611struct hv_dma_range {
1612	dma_addr_t dma;
1613	u32 mapping_size;
1614};
1615
1616#define MAX_SRV_VER	0x7ffffff
1617extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf, u32 buflen,
1618				const int *fw_version, int fw_vercnt,
1619				const int *srv_version, int srv_vercnt,
1620				int *nego_fw_version, int *nego_srv_version);
1621
1622void hv_process_channel_removal(struct vmbus_channel *channel);
1623
1624void vmbus_setevent(struct vmbus_channel *channel);
1625/*
1626 * Negotiated version with the Host.
1627 */
1628
1629extern __u32 vmbus_proto_version;
1630
1631int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
1632				  const guid_t *shv_host_servie_id);
1633int vmbus_send_modifychannel(struct vmbus_channel *channel, u32 target_vp);
1634void vmbus_set_event(struct vmbus_channel *channel);
1635int vmbus_channel_set_cpu(struct vmbus_channel *channel, u32 target_cpu);
1636
1637/* Get the start of the ring buffer. */
1638static inline void *
1639hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info)
1640{
1641	return ring_info->ring_buffer->buffer;
1642}
1643
1644/*
1645 * Mask off host interrupt callback notifications
1646 */
1647static inline void hv_begin_read(struct hv_ring_buffer_info *rbi)
1648{
1649	rbi->ring_buffer->interrupt_mask = 1;
1650
1651	/* make sure mask update is not reordered */
1652	virt_mb();
1653}
1654
1655/*
1656 * Re-enable host callback and return number of outstanding bytes
1657 */
1658static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi)
1659{
1660
1661	rbi->ring_buffer->interrupt_mask = 0;
1662
1663	/* make sure mask update is not reordered */
1664	virt_mb();
1665
1666	/*
1667	 * Now check to see if the ring buffer is still empty.
1668	 * If it is not, we raced and we need to process new
1669	 * incoming messages.
1670	 */
1671	return hv_get_bytes_to_read(rbi);
1672}
1673
1674/*
1675 * An API to support in-place processing of incoming VMBUS packets.
1676 */
1677
1678/* Get data payload associated with descriptor */
1679static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc)
1680{
1681	return (void *)((unsigned long)desc + (desc->offset8 << 3));
1682}
1683
1684/* Get data size associated with descriptor */
1685static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc)
1686{
1687	return (desc->len8 << 3) - (desc->offset8 << 3);
1688}
1689
1690/* Get packet length associated with descriptor */
1691static inline u32 hv_pkt_len(const struct vmpacket_descriptor *desc)
1692{
1693	return desc->len8 << 3;
1694}
1695
1696struct vmpacket_descriptor *
1697hv_pkt_iter_first(struct vmbus_channel *channel);
1698
1699struct vmpacket_descriptor *
1700__hv_pkt_iter_next(struct vmbus_channel *channel,
1701		   const struct vmpacket_descriptor *pkt);
1702
1703void hv_pkt_iter_close(struct vmbus_channel *channel);
1704
1705static inline struct vmpacket_descriptor *
1706hv_pkt_iter_next(struct vmbus_channel *channel,
1707		 const struct vmpacket_descriptor *pkt)
1708{
1709	struct vmpacket_descriptor *nxt;
1710
1711	nxt = __hv_pkt_iter_next(channel, pkt);
1712	if (!nxt)
1713		hv_pkt_iter_close(channel);
1714
1715	return nxt;
1716}
1717
1718#define foreach_vmbus_pkt(pkt, channel) \
1719	for (pkt = hv_pkt_iter_first(channel); pkt; \
1720	    pkt = hv_pkt_iter_next(channel, pkt))
1721
1722/*
1723 * Interface for passing data between SR-IOV PF and VF drivers. The VF driver
1724 * sends requests to read and write blocks. Each block must be 128 bytes or
1725 * smaller. Optionally, the VF driver can register a callback function which
1726 * will be invoked when the host says that one or more of the first 64 block
1727 * IDs is "invalid" which means that the VF driver should reread them.
1728 */
1729#define HV_CONFIG_BLOCK_SIZE_MAX 128
1730
1731int hyperv_read_cfg_blk(struct pci_dev *dev, void *buf, unsigned int buf_len,
1732			unsigned int block_id, unsigned int *bytes_returned);
1733int hyperv_write_cfg_blk(struct pci_dev *dev, void *buf, unsigned int len,
1734			 unsigned int block_id);
1735int hyperv_reg_block_invalidate(struct pci_dev *dev, void *context,
1736				void (*block_invalidate)(void *context,
1737							 u64 block_mask));
1738
1739struct hyperv_pci_block_ops {
1740	int (*read_block)(struct pci_dev *dev, void *buf, unsigned int buf_len,
1741			  unsigned int block_id, unsigned int *bytes_returned);
1742	int (*write_block)(struct pci_dev *dev, void *buf, unsigned int len,
1743			   unsigned int block_id);
1744	int (*reg_blk_invalidate)(struct pci_dev *dev, void *context,
1745				  void (*block_invalidate)(void *context,
1746							   u64 block_mask));
1747};
1748
1749extern struct hyperv_pci_block_ops hvpci_block_ops;
1750
1751static inline unsigned long virt_to_hvpfn(void *addr)
1752{
1753	phys_addr_t paddr;
1754
1755	if (is_vmalloc_addr(addr))
1756		paddr = page_to_phys(vmalloc_to_page(addr)) +
1757				     offset_in_page(addr);
1758	else
1759		paddr = __pa(addr);
1760
1761	return  paddr >> HV_HYP_PAGE_SHIFT;
1762}
1763
1764#define NR_HV_HYP_PAGES_IN_PAGE	(PAGE_SIZE / HV_HYP_PAGE_SIZE)
1765#define offset_in_hvpage(ptr)	((unsigned long)(ptr) & ~HV_HYP_PAGE_MASK)
1766#define HVPFN_UP(x)	(((x) + HV_HYP_PAGE_SIZE-1) >> HV_HYP_PAGE_SHIFT)
1767#define HVPFN_DOWN(x)	((x) >> HV_HYP_PAGE_SHIFT)
1768#define page_to_hvpfn(page)	(page_to_pfn(page) * NR_HV_HYP_PAGES_IN_PAGE)
1769
1770#endif /* _HYPERV_H */