xref: /linux-6.15/include/linux/hyperv.h (revision f2baaff2)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  *
4  * Copyright (c) 2011, Microsoft Corporation.
5  *
6  * Authors:
7  *   Haiyang Zhang <[email protected]>
8  *   Hank Janssen  <[email protected]>
9  *   K. Y. Srinivasan <[email protected]>
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[];
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  * 4 . 1  (Windows 10 RS3)
186  * 5 . 0  (Newer Windows 10)
187  * 5 . 1  (Windows 10 RS4)
188  * 5 . 2  (Windows Server 2019, RS5)
189  */
190 
191 #define VERSION_WS2008  ((0 << 16) | (13))
192 #define VERSION_WIN7    ((1 << 16) | (1))
193 #define VERSION_WIN8    ((2 << 16) | (4))
194 #define VERSION_WIN8_1    ((3 << 16) | (0))
195 #define VERSION_WIN10 ((4 << 16) | (0))
196 #define VERSION_WIN10_V4_1 ((4 << 16) | (1))
197 #define VERSION_WIN10_V5 ((5 << 16) | (0))
198 #define VERSION_WIN10_V5_1 ((5 << 16) | (1))
199 #define VERSION_WIN10_V5_2 ((5 << 16) | (2))
200 
201 /* Make maximum size of pipe payload of 16K */
202 #define MAX_PIPE_DATA_PAYLOAD		(sizeof(u8) * 16384)
203 
204 /* Define PipeMode values. */
205 #define VMBUS_PIPE_TYPE_BYTE		0x00000000
206 #define VMBUS_PIPE_TYPE_MESSAGE		0x00000004
207 
208 /* The size of the user defined data buffer for non-pipe offers. */
209 #define MAX_USER_DEFINED_BYTES		120
210 
211 /* The size of the user defined data buffer for pipe offers. */
212 #define MAX_PIPE_USER_DEFINED_BYTES	116
213 
214 /*
215  * At the center of the Channel Management library is the Channel Offer. This
216  * struct contains the fundamental information about an offer.
217  */
218 struct vmbus_channel_offer {
219 	guid_t if_type;
220 	guid_t if_instance;
221 
222 	/*
223 	 * These two fields are not currently used.
224 	 */
225 	u64 reserved1;
226 	u64 reserved2;
227 
228 	u16 chn_flags;
229 	u16 mmio_megabytes;		/* in bytes * 1024 * 1024 */
230 
231 	union {
232 		/* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
233 		struct {
234 			unsigned char user_def[MAX_USER_DEFINED_BYTES];
235 		} std;
236 
237 		/*
238 		 * Pipes:
239 		 * The following sructure is an integrated pipe protocol, which
240 		 * is implemented on top of standard user-defined data. Pipe
241 		 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
242 		 * use.
243 		 */
244 		struct {
245 			u32  pipe_mode;
246 			unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
247 		} pipe;
248 	} u;
249 	/*
250 	 * The sub_channel_index is defined in Win8: a value of zero means a
251 	 * primary channel and a value of non-zero means a sub-channel.
252 	 *
253 	 * Before Win8, the field is reserved, meaning it's always zero.
254 	 */
255 	u16 sub_channel_index;
256 	u16 reserved3;
257 } __packed;
258 
259 /* Server Flags */
260 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE	1
261 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES	2
262 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS		4
263 #define VMBUS_CHANNEL_NAMED_PIPE_MODE			0x10
264 #define VMBUS_CHANNEL_LOOPBACK_OFFER			0x100
265 #define VMBUS_CHANNEL_PARENT_OFFER			0x200
266 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION	0x400
267 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER		0x2000
268 
269 struct vmpacket_descriptor {
270 	u16 type;
271 	u16 offset8;
272 	u16 len8;
273 	u16 flags;
274 	u64 trans_id;
275 } __packed;
276 
277 struct vmpacket_header {
278 	u32 prev_pkt_start_offset;
279 	struct vmpacket_descriptor descriptor;
280 } __packed;
281 
282 struct vmtransfer_page_range {
283 	u32 byte_count;
284 	u32 byte_offset;
285 } __packed;
286 
287 struct vmtransfer_page_packet_header {
288 	struct vmpacket_descriptor d;
289 	u16 xfer_pageset_id;
290 	u8  sender_owns_set;
291 	u8 reserved;
292 	u32 range_cnt;
293 	struct vmtransfer_page_range ranges[1];
294 } __packed;
295 
296 struct vmgpadl_packet_header {
297 	struct vmpacket_descriptor d;
298 	u32 gpadl;
299 	u32 reserved;
300 } __packed;
301 
302 struct vmadd_remove_transfer_page_set {
303 	struct vmpacket_descriptor d;
304 	u32 gpadl;
305 	u16 xfer_pageset_id;
306 	u16 reserved;
307 } __packed;
308 
309 /*
310  * This structure defines a range in guest physical space that can be made to
311  * look virtually contiguous.
312  */
313 struct gpa_range {
314 	u32 byte_count;
315 	u32 byte_offset;
316 	u64 pfn_array[];
317 };
318 
319 /*
320  * This is the format for an Establish Gpadl packet, which contains a handle by
321  * which this GPADL will be known and a set of GPA ranges associated with it.
322  * This can be converted to a MDL by the guest OS.  If there are multiple GPA
323  * ranges, then the resulting MDL will be "chained," representing multiple VA
324  * ranges.
325  */
326 struct vmestablish_gpadl {
327 	struct vmpacket_descriptor d;
328 	u32 gpadl;
329 	u32 range_cnt;
330 	struct gpa_range range[1];
331 } __packed;
332 
333 /*
334  * This is the format for a Teardown Gpadl packet, which indicates that the
335  * GPADL handle in the Establish Gpadl packet will never be referenced again.
336  */
337 struct vmteardown_gpadl {
338 	struct vmpacket_descriptor d;
339 	u32 gpadl;
340 	u32 reserved;	/* for alignment to a 8-byte boundary */
341 } __packed;
342 
343 /*
344  * This is the format for a GPA-Direct packet, which contains a set of GPA
345  * ranges, in addition to commands and/or data.
346  */
347 struct vmdata_gpa_direct {
348 	struct vmpacket_descriptor d;
349 	u32 reserved;
350 	u32 range_cnt;
351 	struct gpa_range range[1];
352 } __packed;
353 
354 /* This is the format for a Additional Data Packet. */
355 struct vmadditional_data {
356 	struct vmpacket_descriptor d;
357 	u64 total_bytes;
358 	u32 offset;
359 	u32 byte_cnt;
360 	unsigned char data[1];
361 } __packed;
362 
363 union vmpacket_largest_possible_header {
364 	struct vmpacket_descriptor simple_hdr;
365 	struct vmtransfer_page_packet_header xfer_page_hdr;
366 	struct vmgpadl_packet_header gpadl_hdr;
367 	struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr;
368 	struct vmestablish_gpadl establish_gpadl_hdr;
369 	struct vmteardown_gpadl teardown_gpadl_hdr;
370 	struct vmdata_gpa_direct data_gpa_direct_hdr;
371 };
372 
373 #define VMPACKET_DATA_START_ADDRESS(__packet)	\
374 	(void *)(((unsigned char *)__packet) +	\
375 	 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
376 
377 #define VMPACKET_DATA_LENGTH(__packet)		\
378 	((((struct vmpacket_descriptor)__packet)->len8 -	\
379 	  ((struct vmpacket_descriptor)__packet)->offset8) * 8)
380 
381 #define VMPACKET_TRANSFER_MODE(__packet)	\
382 	(((struct IMPACT)__packet)->type)
383 
384 enum vmbus_packet_type {
385 	VM_PKT_INVALID				= 0x0,
386 	VM_PKT_SYNCH				= 0x1,
387 	VM_PKT_ADD_XFER_PAGESET			= 0x2,
388 	VM_PKT_RM_XFER_PAGESET			= 0x3,
389 	VM_PKT_ESTABLISH_GPADL			= 0x4,
390 	VM_PKT_TEARDOWN_GPADL			= 0x5,
391 	VM_PKT_DATA_INBAND			= 0x6,
392 	VM_PKT_DATA_USING_XFER_PAGES		= 0x7,
393 	VM_PKT_DATA_USING_GPADL			= 0x8,
394 	VM_PKT_DATA_USING_GPA_DIRECT		= 0x9,
395 	VM_PKT_CANCEL_REQUEST			= 0xa,
396 	VM_PKT_COMP				= 0xb,
397 	VM_PKT_DATA_USING_ADDITIONAL_PKT	= 0xc,
398 	VM_PKT_ADDITIONAL_DATA			= 0xd
399 };
400 
401 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED	1
402 
403 
404 /* Version 1 messages */
405 enum vmbus_channel_message_type {
406 	CHANNELMSG_INVALID			=  0,
407 	CHANNELMSG_OFFERCHANNEL		=  1,
408 	CHANNELMSG_RESCIND_CHANNELOFFER	=  2,
409 	CHANNELMSG_REQUESTOFFERS		=  3,
410 	CHANNELMSG_ALLOFFERS_DELIVERED	=  4,
411 	CHANNELMSG_OPENCHANNEL		=  5,
412 	CHANNELMSG_OPENCHANNEL_RESULT		=  6,
413 	CHANNELMSG_CLOSECHANNEL		=  7,
414 	CHANNELMSG_GPADL_HEADER		=  8,
415 	CHANNELMSG_GPADL_BODY			=  9,
416 	CHANNELMSG_GPADL_CREATED		= 10,
417 	CHANNELMSG_GPADL_TEARDOWN		= 11,
418 	CHANNELMSG_GPADL_TORNDOWN		= 12,
419 	CHANNELMSG_RELID_RELEASED		= 13,
420 	CHANNELMSG_INITIATE_CONTACT		= 14,
421 	CHANNELMSG_VERSION_RESPONSE		= 15,
422 	CHANNELMSG_UNLOAD			= 16,
423 	CHANNELMSG_UNLOAD_RESPONSE		= 17,
424 	CHANNELMSG_18				= 18,
425 	CHANNELMSG_19				= 19,
426 	CHANNELMSG_20				= 20,
427 	CHANNELMSG_TL_CONNECT_REQUEST		= 21,
428 	CHANNELMSG_MODIFYCHANNEL		= 22,
429 	CHANNELMSG_TL_CONNECT_RESULT		= 23,
430 	CHANNELMSG_COUNT
431 };
432 
433 /* Hyper-V supports about 2048 channels, and the RELIDs start with 1. */
434 #define INVALID_RELID	U32_MAX
435 
436 struct vmbus_channel_message_header {
437 	enum vmbus_channel_message_type msgtype;
438 	u32 padding;
439 } __packed;
440 
441 /* Query VMBus Version parameters */
442 struct vmbus_channel_query_vmbus_version {
443 	struct vmbus_channel_message_header header;
444 	u32 version;
445 } __packed;
446 
447 /* VMBus Version Supported parameters */
448 struct vmbus_channel_version_supported {
449 	struct vmbus_channel_message_header header;
450 	u8 version_supported;
451 } __packed;
452 
453 /* Offer Channel parameters */
454 struct vmbus_channel_offer_channel {
455 	struct vmbus_channel_message_header header;
456 	struct vmbus_channel_offer offer;
457 	u32 child_relid;
458 	u8 monitorid;
459 	/*
460 	 * win7 and beyond splits this field into a bit field.
461 	 */
462 	u8 monitor_allocated:1;
463 	u8 reserved:7;
464 	/*
465 	 * These are new fields added in win7 and later.
466 	 * Do not access these fields without checking the
467 	 * negotiated protocol.
468 	 *
469 	 * If "is_dedicated_interrupt" is set, we must not set the
470 	 * associated bit in the channel bitmap while sending the
471 	 * interrupt to the host.
472 	 *
473 	 * connection_id is to be used in signaling the host.
474 	 */
475 	u16 is_dedicated_interrupt:1;
476 	u16 reserved1:15;
477 	u32 connection_id;
478 } __packed;
479 
480 /* Rescind Offer parameters */
481 struct vmbus_channel_rescind_offer {
482 	struct vmbus_channel_message_header header;
483 	u32 child_relid;
484 } __packed;
485 
486 static inline u32
487 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info *rbi)
488 {
489 	return rbi->ring_buffer->pending_send_sz;
490 }
491 
492 /*
493  * Request Offer -- no parameters, SynIC message contains the partition ID
494  * Set Snoop -- no parameters, SynIC message contains the partition ID
495  * Clear Snoop -- no parameters, SynIC message contains the partition ID
496  * All Offers Delivered -- no parameters, SynIC message contains the partition
497  *		           ID
498  * Flush Client -- no parameters, SynIC message contains the partition ID
499  */
500 
501 /* Open Channel parameters */
502 struct vmbus_channel_open_channel {
503 	struct vmbus_channel_message_header header;
504 
505 	/* Identifies the specific VMBus channel that is being opened. */
506 	u32 child_relid;
507 
508 	/* ID making a particular open request at a channel offer unique. */
509 	u32 openid;
510 
511 	/* GPADL for the channel's ring buffer. */
512 	u32 ringbuffer_gpadlhandle;
513 
514 	/*
515 	 * Starting with win8, this field will be used to specify
516 	 * the target virtual processor on which to deliver the interrupt for
517 	 * the host to guest communication.
518 	 * Prior to win8, incoming channel interrupts would only
519 	 * be delivered on cpu 0. Setting this value to 0 would
520 	 * preserve the earlier behavior.
521 	 */
522 	u32 target_vp;
523 
524 	/*
525 	 * The upstream ring buffer begins at offset zero in the memory
526 	 * described by RingBufferGpadlHandle. The downstream ring buffer
527 	 * follows it at this offset (in pages).
528 	 */
529 	u32 downstream_ringbuffer_pageoffset;
530 
531 	/* User-specific data to be passed along to the server endpoint. */
532 	unsigned char userdata[MAX_USER_DEFINED_BYTES];
533 } __packed;
534 
535 /* Open Channel Result parameters */
536 struct vmbus_channel_open_result {
537 	struct vmbus_channel_message_header header;
538 	u32 child_relid;
539 	u32 openid;
540 	u32 status;
541 } __packed;
542 
543 /* Close channel parameters; */
544 struct vmbus_channel_close_channel {
545 	struct vmbus_channel_message_header header;
546 	u32 child_relid;
547 } __packed;
548 
549 /* Channel Message GPADL */
550 #define GPADL_TYPE_RING_BUFFER		1
551 #define GPADL_TYPE_SERVER_SAVE_AREA	2
552 #define GPADL_TYPE_TRANSACTION		8
553 
554 /*
555  * The number of PFNs in a GPADL message is defined by the number of
556  * pages that would be spanned by ByteCount and ByteOffset.  If the
557  * implied number of PFNs won't fit in this packet, there will be a
558  * follow-up packet that contains more.
559  */
560 struct vmbus_channel_gpadl_header {
561 	struct vmbus_channel_message_header header;
562 	u32 child_relid;
563 	u32 gpadl;
564 	u16 range_buflen;
565 	u16 rangecount;
566 	struct gpa_range range[];
567 } __packed;
568 
569 /* This is the followup packet that contains more PFNs. */
570 struct vmbus_channel_gpadl_body {
571 	struct vmbus_channel_message_header header;
572 	u32 msgnumber;
573 	u32 gpadl;
574 	u64 pfn[];
575 } __packed;
576 
577 struct vmbus_channel_gpadl_created {
578 	struct vmbus_channel_message_header header;
579 	u32 child_relid;
580 	u32 gpadl;
581 	u32 creation_status;
582 } __packed;
583 
584 struct vmbus_channel_gpadl_teardown {
585 	struct vmbus_channel_message_header header;
586 	u32 child_relid;
587 	u32 gpadl;
588 } __packed;
589 
590 struct vmbus_channel_gpadl_torndown {
591 	struct vmbus_channel_message_header header;
592 	u32 gpadl;
593 } __packed;
594 
595 struct vmbus_channel_relid_released {
596 	struct vmbus_channel_message_header header;
597 	u32 child_relid;
598 } __packed;
599 
600 struct vmbus_channel_initiate_contact {
601 	struct vmbus_channel_message_header header;
602 	u32 vmbus_version_requested;
603 	u32 target_vcpu; /* The VCPU the host should respond to */
604 	union {
605 		u64 interrupt_page;
606 		struct {
607 			u8	msg_sint;
608 			u8	padding1[3];
609 			u32	padding2;
610 		};
611 	};
612 	u64 monitor_page1;
613 	u64 monitor_page2;
614 } __packed;
615 
616 /* Hyper-V socket: guest's connect()-ing to host */
617 struct vmbus_channel_tl_connect_request {
618 	struct vmbus_channel_message_header header;
619 	guid_t guest_endpoint_id;
620 	guid_t host_service_id;
621 } __packed;
622 
623 /* Modify Channel parameters, cf. vmbus_send_modifychannel() */
624 struct vmbus_channel_modifychannel {
625 	struct vmbus_channel_message_header header;
626 	u32 child_relid;
627 	u32 target_vp;
628 } __packed;
629 
630 struct vmbus_channel_version_response {
631 	struct vmbus_channel_message_header header;
632 	u8 version_supported;
633 
634 	u8 connection_state;
635 	u16 padding;
636 
637 	/*
638 	 * On new hosts that support VMBus protocol 5.0, we must use
639 	 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
640 	 * and for subsequent messages, we must use the Message Connection ID
641 	 * field in the host-returned Version Response Message.
642 	 *
643 	 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
644 	 */
645 	u32 msg_conn_id;
646 } __packed;
647 
648 enum vmbus_channel_state {
649 	CHANNEL_OFFER_STATE,
650 	CHANNEL_OPENING_STATE,
651 	CHANNEL_OPEN_STATE,
652 	CHANNEL_OPENED_STATE,
653 };
654 
655 /*
656  * Represents each channel msg on the vmbus connection This is a
657  * variable-size data structure depending on the msg type itself
658  */
659 struct vmbus_channel_msginfo {
660 	/* Bookkeeping stuff */
661 	struct list_head msglistentry;
662 
663 	/* So far, this is only used to handle gpadl body message */
664 	struct list_head submsglist;
665 
666 	/* Synchronize the request/response if needed */
667 	struct completion  waitevent;
668 	struct vmbus_channel *waiting_channel;
669 	union {
670 		struct vmbus_channel_version_supported version_supported;
671 		struct vmbus_channel_open_result open_result;
672 		struct vmbus_channel_gpadl_torndown gpadl_torndown;
673 		struct vmbus_channel_gpadl_created gpadl_created;
674 		struct vmbus_channel_version_response version_response;
675 	} response;
676 
677 	u32 msgsize;
678 	/*
679 	 * The channel message that goes out on the "wire".
680 	 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
681 	 */
682 	unsigned char msg[];
683 };
684 
685 struct vmbus_close_msg {
686 	struct vmbus_channel_msginfo info;
687 	struct vmbus_channel_close_channel msg;
688 };
689 
690 /* Define connection identifier type. */
691 union hv_connection_id {
692 	u32 asu32;
693 	struct {
694 		u32 id:24;
695 		u32 reserved:8;
696 	} u;
697 };
698 
699 enum vmbus_device_type {
700 	HV_IDE = 0,
701 	HV_SCSI,
702 	HV_FC,
703 	HV_NIC,
704 	HV_ND,
705 	HV_PCIE,
706 	HV_FB,
707 	HV_KBD,
708 	HV_MOUSE,
709 	HV_KVP,
710 	HV_TS,
711 	HV_HB,
712 	HV_SHUTDOWN,
713 	HV_FCOPY,
714 	HV_BACKUP,
715 	HV_DM,
716 	HV_UNKNOWN,
717 };
718 
719 struct vmbus_device {
720 	u16  dev_type;
721 	guid_t guid;
722 	bool perf_device;
723 };
724 
725 struct vmbus_channel {
726 	struct list_head listentry;
727 
728 	struct hv_device *device_obj;
729 
730 	enum vmbus_channel_state state;
731 
732 	struct vmbus_channel_offer_channel offermsg;
733 	/*
734 	 * These are based on the OfferMsg.MonitorId.
735 	 * Save it here for easy access.
736 	 */
737 	u8 monitor_grp;
738 	u8 monitor_bit;
739 
740 	bool rescind; /* got rescind msg */
741 	struct completion rescind_event;
742 
743 	u32 ringbuffer_gpadlhandle;
744 
745 	/* Allocated memory for ring buffer */
746 	struct page *ringbuffer_page;
747 	u32 ringbuffer_pagecount;
748 	u32 ringbuffer_send_offset;
749 	struct hv_ring_buffer_info outbound;	/* send to parent */
750 	struct hv_ring_buffer_info inbound;	/* receive from parent */
751 
752 	struct vmbus_close_msg close_msg;
753 
754 	/* Statistics */
755 	u64	interrupts;	/* Host to Guest interrupts */
756 	u64	sig_events;	/* Guest to Host events */
757 
758 	/*
759 	 * Guest to host interrupts caused by the outbound ring buffer changing
760 	 * from empty to not empty.
761 	 */
762 	u64 intr_out_empty;
763 
764 	/*
765 	 * Indicates that a full outbound ring buffer was encountered. The flag
766 	 * is set to true when a full outbound ring buffer is encountered and
767 	 * set to false when a write to the outbound ring buffer is completed.
768 	 */
769 	bool out_full_flag;
770 
771 	/* Channel callback's invoked in softirq context */
772 	struct tasklet_struct callback_event;
773 	void (*onchannel_callback)(void *context);
774 	void *channel_callback_context;
775 
776 	void (*change_target_cpu_callback)(struct vmbus_channel *channel,
777 			u32 old, u32 new);
778 
779 	/*
780 	 * Synchronize channel scheduling and channel removal; see the inline
781 	 * comments in vmbus_chan_sched() and vmbus_reset_channel_cb().
782 	 */
783 	spinlock_t sched_lock;
784 
785 	/*
786 	 * A channel can be marked for one of three modes of reading:
787 	 *   BATCHED - callback called from taslket and should read
788 	 *            channel until empty. Interrupts from the host
789 	 *            are masked while read is in process (default).
790 	 *   DIRECT - callback called from tasklet (softirq).
791 	 *   ISR - callback called in interrupt context and must
792 	 *         invoke its own deferred processing.
793 	 *         Host interrupts are disabled and must be re-enabled
794 	 *         when ring is empty.
795 	 */
796 	enum hv_callback_mode {
797 		HV_CALL_BATCHED,
798 		HV_CALL_DIRECT,
799 		HV_CALL_ISR
800 	} callback_mode;
801 
802 	bool is_dedicated_interrupt;
803 	u64 sig_event;
804 
805 	/*
806 	 * Starting with win8, this field will be used to specify
807 	 * the target virtual processor on which to deliver the interrupt for
808 	 * the host to guest communication.
809 	 * Prior to win8, incoming channel interrupts would only
810 	 * be delivered on cpu 0. Setting this value to 0 would
811 	 * preserve the earlier behavior.
812 	 */
813 	u32 target_vp;
814 	/* The corresponding CPUID in the guest */
815 	u32 target_cpu;
816 	int numa_node;
817 	/*
818 	 * Support for sub-channels. For high performance devices,
819 	 * it will be useful to have multiple sub-channels to support
820 	 * a scalable communication infrastructure with the host.
821 	 * The support for sub-channels is implemented as an extention
822 	 * to the current infrastructure.
823 	 * The initial offer is considered the primary channel and this
824 	 * offer message will indicate if the host supports sub-channels.
825 	 * The guest is free to ask for sub-channels to be offerred and can
826 	 * open these sub-channels as a normal "primary" channel. However,
827 	 * all sub-channels will have the same type and instance guids as the
828 	 * primary channel. Requests sent on a given channel will result in a
829 	 * response on the same channel.
830 	 */
831 
832 	/*
833 	 * Sub-channel creation callback. This callback will be called in
834 	 * process context when a sub-channel offer is received from the host.
835 	 * The guest can open the sub-channel in the context of this callback.
836 	 */
837 	void (*sc_creation_callback)(struct vmbus_channel *new_sc);
838 
839 	/*
840 	 * Channel rescind callback. Some channels (the hvsock ones), need to
841 	 * register a callback which is invoked in vmbus_onoffer_rescind().
842 	 */
843 	void (*chn_rescind_callback)(struct vmbus_channel *channel);
844 
845 	/*
846 	 * The spinlock to protect the structure. It is being used to protect
847 	 * test-and-set access to various attributes of the structure as well
848 	 * as all sc_list operations.
849 	 */
850 	spinlock_t lock;
851 	/*
852 	 * All Sub-channels of a primary channel are linked here.
853 	 */
854 	struct list_head sc_list;
855 	/*
856 	 * The primary channel this sub-channel belongs to.
857 	 * This will be NULL for the primary channel.
858 	 */
859 	struct vmbus_channel *primary_channel;
860 	/*
861 	 * Support per-channel state for use by vmbus drivers.
862 	 */
863 	void *per_channel_state;
864 
865 	/*
866 	 * Defer freeing channel until after all cpu's have
867 	 * gone through grace period.
868 	 */
869 	struct rcu_head rcu;
870 
871 	/*
872 	 * For sysfs per-channel properties.
873 	 */
874 	struct kobject			kobj;
875 
876 	/*
877 	 * For performance critical channels (storage, networking
878 	 * etc,), Hyper-V has a mechanism to enhance the throughput
879 	 * at the expense of latency:
880 	 * When the host is to be signaled, we just set a bit in a shared page
881 	 * and this bit will be inspected by the hypervisor within a certain
882 	 * window and if the bit is set, the host will be signaled. The window
883 	 * of time is the monitor latency - currently around 100 usecs. This
884 	 * mechanism improves throughput by:
885 	 *
886 	 * A) Making the host more efficient - each time it wakes up,
887 	 *    potentially it will process morev number of packets. The
888 	 *    monitor latency allows a batch to build up.
889 	 * B) By deferring the hypercall to signal, we will also minimize
890 	 *    the interrupts.
891 	 *
892 	 * Clearly, these optimizations improve throughput at the expense of
893 	 * latency. Furthermore, since the channel is shared for both
894 	 * control and data messages, control messages currently suffer
895 	 * unnecessary latency adversley impacting performance and boot
896 	 * time. To fix this issue, permit tagging the channel as being
897 	 * in "low latency" mode. In this mode, we will bypass the monitor
898 	 * mechanism.
899 	 */
900 	bool low_latency;
901 
902 	bool probe_done;
903 
904 	/*
905 	 * Cache the device ID here for easy access; this is useful, in
906 	 * particular, in situations where the channel's device_obj has
907 	 * not been allocated/initialized yet.
908 	 */
909 	u16 device_id;
910 
911 	/*
912 	 * We must offload the handling of the primary/sub channels
913 	 * from the single-threaded vmbus_connection.work_queue to
914 	 * two different workqueue, otherwise we can block
915 	 * vmbus_connection.work_queue and hang: see vmbus_process_offer().
916 	 */
917 	struct work_struct add_channel_work;
918 
919 	/*
920 	 * Guest to host interrupts caused by the inbound ring buffer changing
921 	 * from full to not full while a packet is waiting.
922 	 */
923 	u64 intr_in_full;
924 
925 	/*
926 	 * The total number of write operations that encountered a full
927 	 * outbound ring buffer.
928 	 */
929 	u64 out_full_total;
930 
931 	/*
932 	 * The number of write operations that were the first to encounter a
933 	 * full outbound ring buffer.
934 	 */
935 	u64 out_full_first;
936 
937 	/* enabling/disabling fuzz testing on the channel (default is false)*/
938 	bool fuzz_testing_state;
939 
940 	/*
941 	 * Interrupt delay will delay the guest from emptying the ring buffer
942 	 * for a specific amount of time. The delay is in microseconds and will
943 	 * be between 1 to a maximum of 1000, its default is 0 (no delay).
944 	 * The  Message delay will delay guest reading on a per message basis
945 	 * in microseconds between 1 to 1000 with the default being 0
946 	 * (no delay).
947 	 */
948 	u32 fuzz_testing_interrupt_delay;
949 	u32 fuzz_testing_message_delay;
950 
951 };
952 
953 static inline bool is_hvsock_channel(const struct vmbus_channel *c)
954 {
955 	return !!(c->offermsg.offer.chn_flags &
956 		  VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER);
957 }
958 
959 static inline bool is_sub_channel(const struct vmbus_channel *c)
960 {
961 	return c->offermsg.offer.sub_channel_index != 0;
962 }
963 
964 static inline void set_channel_read_mode(struct vmbus_channel *c,
965 					enum hv_callback_mode mode)
966 {
967 	c->callback_mode = mode;
968 }
969 
970 static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
971 {
972 	c->per_channel_state = s;
973 }
974 
975 static inline void *get_per_channel_state(struct vmbus_channel *c)
976 {
977 	return c->per_channel_state;
978 }
979 
980 static inline void set_channel_pending_send_size(struct vmbus_channel *c,
981 						 u32 size)
982 {
983 	unsigned long flags;
984 
985 	if (size) {
986 		spin_lock_irqsave(&c->outbound.ring_lock, flags);
987 		++c->out_full_total;
988 
989 		if (!c->out_full_flag) {
990 			++c->out_full_first;
991 			c->out_full_flag = true;
992 		}
993 		spin_unlock_irqrestore(&c->outbound.ring_lock, flags);
994 	} else {
995 		c->out_full_flag = false;
996 	}
997 
998 	c->outbound.ring_buffer->pending_send_sz = size;
999 }
1000 
1001 static inline void set_low_latency_mode(struct vmbus_channel *c)
1002 {
1003 	c->low_latency = true;
1004 }
1005 
1006 static inline void clear_low_latency_mode(struct vmbus_channel *c)
1007 {
1008 	c->low_latency = false;
1009 }
1010 
1011 void vmbus_onmessage(struct vmbus_channel_message_header *hdr);
1012 
1013 int vmbus_request_offers(void);
1014 
1015 /*
1016  * APIs for managing sub-channels.
1017  */
1018 
1019 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1020 			void (*sc_cr_cb)(struct vmbus_channel *new_sc));
1021 
1022 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1023 		void (*chn_rescind_cb)(struct vmbus_channel *));
1024 
1025 /*
1026  * Check if sub-channels have already been offerred. This API will be useful
1027  * when the driver is unloaded after establishing sub-channels. In this case,
1028  * when the driver is re-loaded, the driver would have to check if the
1029  * subchannels have already been established before attempting to request
1030  * the creation of sub-channels.
1031  * This function returns TRUE to indicate that subchannels have already been
1032  * created.
1033  * This function should be invoked after setting the callback function for
1034  * sub-channel creation.
1035  */
1036 bool vmbus_are_subchannels_present(struct vmbus_channel *primary);
1037 
1038 /* The format must be the same as struct vmdata_gpa_direct */
1039 struct vmbus_channel_packet_page_buffer {
1040 	u16 type;
1041 	u16 dataoffset8;
1042 	u16 length8;
1043 	u16 flags;
1044 	u64 transactionid;
1045 	u32 reserved;
1046 	u32 rangecount;
1047 	struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
1048 } __packed;
1049 
1050 /* The format must be the same as struct vmdata_gpa_direct */
1051 struct vmbus_channel_packet_multipage_buffer {
1052 	u16 type;
1053 	u16 dataoffset8;
1054 	u16 length8;
1055 	u16 flags;
1056 	u64 transactionid;
1057 	u32 reserved;
1058 	u32 rangecount;		/* Always 1 in this case */
1059 	struct hv_multipage_buffer range;
1060 } __packed;
1061 
1062 /* The format must be the same as struct vmdata_gpa_direct */
1063 struct vmbus_packet_mpb_array {
1064 	u16 type;
1065 	u16 dataoffset8;
1066 	u16 length8;
1067 	u16 flags;
1068 	u64 transactionid;
1069 	u32 reserved;
1070 	u32 rangecount;         /* Always 1 in this case */
1071 	struct hv_mpb_array range;
1072 } __packed;
1073 
1074 int vmbus_alloc_ring(struct vmbus_channel *channel,
1075 		     u32 send_size, u32 recv_size);
1076 void vmbus_free_ring(struct vmbus_channel *channel);
1077 
1078 int vmbus_connect_ring(struct vmbus_channel *channel,
1079 		       void (*onchannel_callback)(void *context),
1080 		       void *context);
1081 int vmbus_disconnect_ring(struct vmbus_channel *channel);
1082 
1083 extern int vmbus_open(struct vmbus_channel *channel,
1084 			    u32 send_ringbuffersize,
1085 			    u32 recv_ringbuffersize,
1086 			    void *userdata,
1087 			    u32 userdatalen,
1088 			    void (*onchannel_callback)(void *context),
1089 			    void *context);
1090 
1091 extern void vmbus_close(struct vmbus_channel *channel);
1092 
1093 extern int vmbus_sendpacket(struct vmbus_channel *channel,
1094 				  void *buffer,
1095 				  u32 bufferLen,
1096 				  u64 requestid,
1097 				  enum vmbus_packet_type type,
1098 				  u32 flags);
1099 
1100 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1101 					    struct hv_page_buffer pagebuffers[],
1102 					    u32 pagecount,
1103 					    void *buffer,
1104 					    u32 bufferlen,
1105 					    u64 requestid);
1106 
1107 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1108 				     struct vmbus_packet_mpb_array *mpb,
1109 				     u32 desc_size,
1110 				     void *buffer,
1111 				     u32 bufferlen,
1112 				     u64 requestid);
1113 
1114 extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
1115 				      void *kbuffer,
1116 				      u32 size,
1117 				      u32 *gpadl_handle);
1118 
1119 extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
1120 				     u32 gpadl_handle);
1121 
1122 void vmbus_reset_channel_cb(struct vmbus_channel *channel);
1123 
1124 extern int vmbus_recvpacket(struct vmbus_channel *channel,
1125 				  void *buffer,
1126 				  u32 bufferlen,
1127 				  u32 *buffer_actual_len,
1128 				  u64 *requestid);
1129 
1130 extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
1131 				     void *buffer,
1132 				     u32 bufferlen,
1133 				     u32 *buffer_actual_len,
1134 				     u64 *requestid);
1135 
1136 
1137 extern void vmbus_ontimer(unsigned long data);
1138 
1139 /* Base driver object */
1140 struct hv_driver {
1141 	const char *name;
1142 
1143 	/*
1144 	 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1145 	 * channel flag, actually doesn't mean a synthetic device because the
1146 	 * offer's if_type/if_instance can change for every new hvsock
1147 	 * connection.
1148 	 *
1149 	 * However, to facilitate the notification of new-offer/rescind-offer
1150 	 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1151 	 * a special vmbus device, and hence we need the below flag to
1152 	 * indicate if the driver is the hvsock driver or not: we need to
1153 	 * specially treat the hvosck offer & driver in vmbus_match().
1154 	 */
1155 	bool hvsock;
1156 
1157 	/* the device type supported by this driver */
1158 	guid_t dev_type;
1159 	const struct hv_vmbus_device_id *id_table;
1160 
1161 	struct device_driver driver;
1162 
1163 	/* dynamic device GUID's */
1164 	struct  {
1165 		spinlock_t lock;
1166 		struct list_head list;
1167 	} dynids;
1168 
1169 	int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
1170 	int (*remove)(struct hv_device *);
1171 	void (*shutdown)(struct hv_device *);
1172 
1173 	int (*suspend)(struct hv_device *);
1174 	int (*resume)(struct hv_device *);
1175 
1176 };
1177 
1178 /* Base device object */
1179 struct hv_device {
1180 	/* the device type id of this device */
1181 	guid_t dev_type;
1182 
1183 	/* the device instance id of this device */
1184 	guid_t dev_instance;
1185 	u16 vendor_id;
1186 	u16 device_id;
1187 
1188 	struct device device;
1189 	char *driver_override; /* Driver name to force a match */
1190 
1191 	struct vmbus_channel *channel;
1192 	struct kset	     *channels_kset;
1193 
1194 	/* place holder to keep track of the dir for hv device in debugfs */
1195 	struct dentry *debug_dir;
1196 
1197 };
1198 
1199 
1200 static inline struct hv_device *device_to_hv_device(struct device *d)
1201 {
1202 	return container_of(d, struct hv_device, device);
1203 }
1204 
1205 static inline struct hv_driver *drv_to_hv_drv(struct device_driver *d)
1206 {
1207 	return container_of(d, struct hv_driver, driver);
1208 }
1209 
1210 static inline void hv_set_drvdata(struct hv_device *dev, void *data)
1211 {
1212 	dev_set_drvdata(&dev->device, data);
1213 }
1214 
1215 static inline void *hv_get_drvdata(struct hv_device *dev)
1216 {
1217 	return dev_get_drvdata(&dev->device);
1218 }
1219 
1220 struct hv_ring_buffer_debug_info {
1221 	u32 current_interrupt_mask;
1222 	u32 current_read_index;
1223 	u32 current_write_index;
1224 	u32 bytes_avail_toread;
1225 	u32 bytes_avail_towrite;
1226 };
1227 
1228 
1229 int hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
1230 				struct hv_ring_buffer_debug_info *debug_info);
1231 
1232 /* Vmbus interface */
1233 #define vmbus_driver_register(driver)	\
1234 	__vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1235 int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
1236 					 struct module *owner,
1237 					 const char *mod_name);
1238 void vmbus_driver_unregister(struct hv_driver *hv_driver);
1239 
1240 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel);
1241 
1242 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1243 			resource_size_t min, resource_size_t max,
1244 			resource_size_t size, resource_size_t align,
1245 			bool fb_overlap_ok);
1246 void vmbus_free_mmio(resource_size_t start, resource_size_t size);
1247 
1248 /*
1249  * GUID definitions of various offer types - services offered to the guest.
1250  */
1251 
1252 /*
1253  * Network GUID
1254  * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1255  */
1256 #define HV_NIC_GUID \
1257 	.guid = GUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1258 			  0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1259 
1260 /*
1261  * IDE GUID
1262  * {32412632-86cb-44a2-9b5c-50d1417354f5}
1263  */
1264 #define HV_IDE_GUID \
1265 	.guid = GUID_INIT(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1266 			  0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1267 
1268 /*
1269  * SCSI GUID
1270  * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1271  */
1272 #define HV_SCSI_GUID \
1273 	.guid = GUID_INIT(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1274 			  0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1275 
1276 /*
1277  * Shutdown GUID
1278  * {0e0b6031-5213-4934-818b-38d90ced39db}
1279  */
1280 #define HV_SHUTDOWN_GUID \
1281 	.guid = GUID_INIT(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1282 			  0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1283 
1284 /*
1285  * Time Synch GUID
1286  * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1287  */
1288 #define HV_TS_GUID \
1289 	.guid = GUID_INIT(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1290 			  0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1291 
1292 /*
1293  * Heartbeat GUID
1294  * {57164f39-9115-4e78-ab55-382f3bd5422d}
1295  */
1296 #define HV_HEART_BEAT_GUID \
1297 	.guid = GUID_INIT(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1298 			  0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1299 
1300 /*
1301  * KVP GUID
1302  * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1303  */
1304 #define HV_KVP_GUID \
1305 	.guid = GUID_INIT(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1306 			  0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1307 
1308 /*
1309  * Dynamic memory GUID
1310  * {525074dc-8985-46e2-8057-a307dc18a502}
1311  */
1312 #define HV_DM_GUID \
1313 	.guid = GUID_INIT(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1314 			  0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1315 
1316 /*
1317  * Mouse GUID
1318  * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1319  */
1320 #define HV_MOUSE_GUID \
1321 	.guid = GUID_INIT(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1322 			  0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1323 
1324 /*
1325  * Keyboard GUID
1326  * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1327  */
1328 #define HV_KBD_GUID \
1329 	.guid = GUID_INIT(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1330 			  0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1331 
1332 /*
1333  * VSS (Backup/Restore) GUID
1334  */
1335 #define HV_VSS_GUID \
1336 	.guid = GUID_INIT(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1337 			  0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1338 /*
1339  * Synthetic Video GUID
1340  * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1341  */
1342 #define HV_SYNTHVID_GUID \
1343 	.guid = GUID_INIT(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1344 			  0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1345 
1346 /*
1347  * Synthetic FC GUID
1348  * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1349  */
1350 #define HV_SYNTHFC_GUID \
1351 	.guid = GUID_INIT(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1352 			  0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1353 
1354 /*
1355  * Guest File Copy Service
1356  * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1357  */
1358 
1359 #define HV_FCOPY_GUID \
1360 	.guid = GUID_INIT(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1361 			  0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1362 
1363 /*
1364  * NetworkDirect. This is the guest RDMA service.
1365  * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1366  */
1367 #define HV_ND_GUID \
1368 	.guid = GUID_INIT(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1369 			  0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1370 
1371 /*
1372  * PCI Express Pass Through
1373  * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1374  */
1375 
1376 #define HV_PCIE_GUID \
1377 	.guid = GUID_INIT(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1378 			  0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1379 
1380 /*
1381  * Linux doesn't support the 3 devices: the first two are for
1382  * Automatic Virtual Machine Activation, and the third is for
1383  * Remote Desktop Virtualization.
1384  * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1385  * {3375baf4-9e15-4b30-b765-67acb10d607b}
1386  * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1387  */
1388 
1389 #define HV_AVMA1_GUID \
1390 	.guid = GUID_INIT(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1391 			  0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1392 
1393 #define HV_AVMA2_GUID \
1394 	.guid = GUID_INIT(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1395 			  0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1396 
1397 #define HV_RDV_GUID \
1398 	.guid = GUID_INIT(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1399 			  0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1400 
1401 /*
1402  * Common header for Hyper-V ICs
1403  */
1404 
1405 #define ICMSGTYPE_NEGOTIATE		0
1406 #define ICMSGTYPE_HEARTBEAT		1
1407 #define ICMSGTYPE_KVPEXCHANGE		2
1408 #define ICMSGTYPE_SHUTDOWN		3
1409 #define ICMSGTYPE_TIMESYNC		4
1410 #define ICMSGTYPE_VSS			5
1411 
1412 #define ICMSGHDRFLAG_TRANSACTION	1
1413 #define ICMSGHDRFLAG_REQUEST		2
1414 #define ICMSGHDRFLAG_RESPONSE		4
1415 
1416 
1417 /*
1418  * While we want to handle util services as regular devices,
1419  * there is only one instance of each of these services; so
1420  * we statically allocate the service specific state.
1421  */
1422 
1423 struct hv_util_service {
1424 	u8 *recv_buffer;
1425 	void *channel;
1426 	void (*util_cb)(void *);
1427 	int (*util_init)(struct hv_util_service *);
1428 	void (*util_deinit)(void);
1429 	int (*util_pre_suspend)(void);
1430 	int (*util_pre_resume)(void);
1431 };
1432 
1433 struct vmbuspipe_hdr {
1434 	u32 flags;
1435 	u32 msgsize;
1436 } __packed;
1437 
1438 struct ic_version {
1439 	u16 major;
1440 	u16 minor;
1441 } __packed;
1442 
1443 struct icmsg_hdr {
1444 	struct ic_version icverframe;
1445 	u16 icmsgtype;
1446 	struct ic_version icvermsg;
1447 	u16 icmsgsize;
1448 	u32 status;
1449 	u8 ictransaction_id;
1450 	u8 icflags;
1451 	u8 reserved[2];
1452 } __packed;
1453 
1454 struct icmsg_negotiate {
1455 	u16 icframe_vercnt;
1456 	u16 icmsg_vercnt;
1457 	u32 reserved;
1458 	struct ic_version icversion_data[1]; /* any size array */
1459 } __packed;
1460 
1461 struct shutdown_msg_data {
1462 	u32 reason_code;
1463 	u32 timeout_seconds;
1464 	u32 flags;
1465 	u8  display_message[2048];
1466 } __packed;
1467 
1468 struct heartbeat_msg_data {
1469 	u64 seq_num;
1470 	u32 reserved[8];
1471 } __packed;
1472 
1473 /* Time Sync IC defs */
1474 #define ICTIMESYNCFLAG_PROBE	0
1475 #define ICTIMESYNCFLAG_SYNC	1
1476 #define ICTIMESYNCFLAG_SAMPLE	2
1477 
1478 #ifdef __x86_64__
1479 #define WLTIMEDELTA	116444736000000000L	/* in 100ns unit */
1480 #else
1481 #define WLTIMEDELTA	116444736000000000LL
1482 #endif
1483 
1484 struct ictimesync_data {
1485 	u64 parenttime;
1486 	u64 childtime;
1487 	u64 roundtriptime;
1488 	u8 flags;
1489 } __packed;
1490 
1491 struct ictimesync_ref_data {
1492 	u64 parenttime;
1493 	u64 vmreferencetime;
1494 	u8 flags;
1495 	char leapflags;
1496 	char stratum;
1497 	u8 reserved[3];
1498 } __packed;
1499 
1500 struct hyperv_service_callback {
1501 	u8 msg_type;
1502 	char *log_msg;
1503 	guid_t data;
1504 	struct vmbus_channel *channel;
1505 	void (*callback)(void *context);
1506 };
1507 
1508 #define MAX_SRV_VER	0x7ffffff
1509 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf,
1510 				const int *fw_version, int fw_vercnt,
1511 				const int *srv_version, int srv_vercnt,
1512 				int *nego_fw_version, int *nego_srv_version);
1513 
1514 void hv_process_channel_removal(struct vmbus_channel *channel);
1515 
1516 void vmbus_setevent(struct vmbus_channel *channel);
1517 /*
1518  * Negotiated version with the Host.
1519  */
1520 
1521 extern __u32 vmbus_proto_version;
1522 
1523 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
1524 				  const guid_t *shv_host_servie_id);
1525 int vmbus_send_modifychannel(u32 child_relid, u32 target_vp);
1526 void vmbus_set_event(struct vmbus_channel *channel);
1527 
1528 /* Get the start of the ring buffer. */
1529 static inline void *
1530 hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info)
1531 {
1532 	return ring_info->ring_buffer->buffer;
1533 }
1534 
1535 /*
1536  * Mask off host interrupt callback notifications
1537  */
1538 static inline void hv_begin_read(struct hv_ring_buffer_info *rbi)
1539 {
1540 	rbi->ring_buffer->interrupt_mask = 1;
1541 
1542 	/* make sure mask update is not reordered */
1543 	virt_mb();
1544 }
1545 
1546 /*
1547  * Re-enable host callback and return number of outstanding bytes
1548  */
1549 static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi)
1550 {
1551 
1552 	rbi->ring_buffer->interrupt_mask = 0;
1553 
1554 	/* make sure mask update is not reordered */
1555 	virt_mb();
1556 
1557 	/*
1558 	 * Now check to see if the ring buffer is still empty.
1559 	 * If it is not, we raced and we need to process new
1560 	 * incoming messages.
1561 	 */
1562 	return hv_get_bytes_to_read(rbi);
1563 }
1564 
1565 /*
1566  * An API to support in-place processing of incoming VMBUS packets.
1567  */
1568 
1569 /* Get data payload associated with descriptor */
1570 static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc)
1571 {
1572 	return (void *)((unsigned long)desc + (desc->offset8 << 3));
1573 }
1574 
1575 /* Get data size associated with descriptor */
1576 static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc)
1577 {
1578 	return (desc->len8 << 3) - (desc->offset8 << 3);
1579 }
1580 
1581 
1582 struct vmpacket_descriptor *
1583 hv_pkt_iter_first(struct vmbus_channel *channel);
1584 
1585 struct vmpacket_descriptor *
1586 __hv_pkt_iter_next(struct vmbus_channel *channel,
1587 		   const struct vmpacket_descriptor *pkt);
1588 
1589 void hv_pkt_iter_close(struct vmbus_channel *channel);
1590 
1591 /*
1592  * Get next packet descriptor from iterator
1593  * If at end of list, return NULL and update host.
1594  */
1595 static inline struct vmpacket_descriptor *
1596 hv_pkt_iter_next(struct vmbus_channel *channel,
1597 		 const struct vmpacket_descriptor *pkt)
1598 {
1599 	struct vmpacket_descriptor *nxt;
1600 
1601 	nxt = __hv_pkt_iter_next(channel, pkt);
1602 	if (!nxt)
1603 		hv_pkt_iter_close(channel);
1604 
1605 	return nxt;
1606 }
1607 
1608 #define foreach_vmbus_pkt(pkt, channel) \
1609 	for (pkt = hv_pkt_iter_first(channel); pkt; \
1610 	    pkt = hv_pkt_iter_next(channel, pkt))
1611 
1612 /*
1613  * Interface for passing data between SR-IOV PF and VF drivers. The VF driver
1614  * sends requests to read and write blocks. Each block must be 128 bytes or
1615  * smaller. Optionally, the VF driver can register a callback function which
1616  * will be invoked when the host says that one or more of the first 64 block
1617  * IDs is "invalid" which means that the VF driver should reread them.
1618  */
1619 #define HV_CONFIG_BLOCK_SIZE_MAX 128
1620 
1621 int hyperv_read_cfg_blk(struct pci_dev *dev, void *buf, unsigned int buf_len,
1622 			unsigned int block_id, unsigned int *bytes_returned);
1623 int hyperv_write_cfg_blk(struct pci_dev *dev, void *buf, unsigned int len,
1624 			 unsigned int block_id);
1625 int hyperv_reg_block_invalidate(struct pci_dev *dev, void *context,
1626 				void (*block_invalidate)(void *context,
1627 							 u64 block_mask));
1628 
1629 struct hyperv_pci_block_ops {
1630 	int (*read_block)(struct pci_dev *dev, void *buf, unsigned int buf_len,
1631 			  unsigned int block_id, unsigned int *bytes_returned);
1632 	int (*write_block)(struct pci_dev *dev, void *buf, unsigned int len,
1633 			   unsigned int block_id);
1634 	int (*reg_blk_invalidate)(struct pci_dev *dev, void *context,
1635 				  void (*block_invalidate)(void *context,
1636 							   u64 block_mask));
1637 };
1638 
1639 extern struct hyperv_pci_block_ops hvpci_block_ops;
1640 
1641 #endif /* _HYPERV_H */
1642