1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  *
4  * Copyright (c) 2011, Microsoft Corporation.
5  *
6  * Authors:
7  *   Haiyang Zhang <haiyangz@microsoft.com>
8  *   Hank Janssen  <hjanssen@microsoft.com>
9  *   K. Y. Srinivasan <kys@microsoft.com>
10  */
11 
12 #ifndef _HYPERV_H
13 #define _HYPERV_H
14 
15 #include <uapi/linux/hyperv.h>
16 
17 #include <linux/mm.h>
18 #include <linux/types.h>
19 #include <linux/scatterlist.h>
20 #include <linux/list.h>
21 #include <linux/timer.h>
22 #include <linux/completion.h>
23 #include <linux/device.h>
24 #include <linux/mod_devicetable.h>
25 #include <linux/interrupt.h>
26 #include <linux/reciprocal_div.h>
27 #include <asm/hyperv-tlfs.h>
28 
29 #define MAX_PAGE_BUFFER_COUNT				32
30 #define MAX_MULTIPAGE_BUFFER_COUNT			32 /* 128K */
31 
32 #pragma pack(push, 1)
33 
34 /*
35  * Types for GPADL, decides is how GPADL header is created.
36  *
37  * It doesn't make much difference between BUFFER and RING if PAGE_SIZE is the
38  * same as HV_HYP_PAGE_SIZE.
39  *
40  * If PAGE_SIZE is bigger than HV_HYP_PAGE_SIZE, the headers of ring buffers
41  * will be of PAGE_SIZE, however, only the first HV_HYP_PAGE will be put
42  * into gpadl, therefore the number for HV_HYP_PAGE and the indexes of each
43  * HV_HYP_PAGE will be different between different types of GPADL, for example
44  * if PAGE_SIZE is 64K:
45  *
46  * BUFFER:
47  *
48  * gva:    |--       64k      --|--       64k      --| ... |
49  * gpa:    | 4k | 4k | ... | 4k | 4k | 4k | ... | 4k |
50  * index:  0    1    2     15   16   17   18 .. 31   32 ...
51  *         |    |    ...   |    |    |   ...    |   ...
52  *         v    V          V    V    V          V
53  * gpadl:  | 4k | 4k | ... | 4k | 4k | 4k | ... | 4k | ... |
54  * index:  0    1    2 ... 15   16   17   18 .. 31   32 ...
55  *
56  * RING:
57  *
58  *         | header  |           data           | header  |     data      |
59  * gva:    |-- 64k --|--       64k      --| ... |-- 64k --|-- 64k --| ... |
60  * gpa:    | 4k | .. | 4k | 4k | ... | 4k | ... | 4k | .. | 4k | .. | ... |
61  * index:  0    1    16   17   18    31   ...   n   n+1  n+16 ...         2n
62  *         |         /    /          /          |         /               /
63  *         |        /    /          /           |        /               /
64  *         |       /    /   ...    /    ...     |       /      ...      /
65  *         |      /    /          /             |      /               /
66  *         |     /    /          /              |     /               /
67  *         V    V    V          V               V    V               v
68  * gpadl:  | 4k | 4k |   ...    |    ...        | 4k | 4k |  ...     |
69  * index:  0    1    2   ...    16   ...       n-15 n-14 n-13  ...  2n-30
70  */
71 enum hv_gpadl_type {
72 	HV_GPADL_BUFFER,
73 	HV_GPADL_RING
74 };
75 
76 /* Single-page buffer */
77 struct hv_page_buffer {
78 	u32 len;
79 	u32 offset;
80 	u64 pfn;
81 };
82 
83 /* Multiple-page buffer */
84 struct hv_multipage_buffer {
85 	/* Length and Offset determines the # of pfns in the array */
86 	u32 len;
87 	u32 offset;
88 	u64 pfn_array[MAX_MULTIPAGE_BUFFER_COUNT];
89 };
90 
91 /*
92  * Multiple-page buffer array; the pfn array is variable size:
93  * The number of entries in the PFN array is determined by
94  * "len" and "offset".
95  */
96 struct hv_mpb_array {
97 	/* Length and Offset determines the # of pfns in the array */
98 	u32 len;
99 	u32 offset;
100 	u64 pfn_array[];
101 };
102 
103 /* 0x18 includes the proprietary packet header */
104 #define MAX_PAGE_BUFFER_PACKET		(0x18 +			\
105 					(sizeof(struct hv_page_buffer) * \
106 					 MAX_PAGE_BUFFER_COUNT))
107 #define MAX_MULTIPAGE_BUFFER_PACKET	(0x18 +			\
108 					 sizeof(struct hv_multipage_buffer))
109 
110 
111 #pragma pack(pop)
112 
113 struct hv_ring_buffer {
114 	/* Offset in bytes from the start of ring data below */
115 	u32 write_index;
116 
117 	/* Offset in bytes from the start of ring data below */
118 	u32 read_index;
119 
120 	u32 interrupt_mask;
121 
122 	/*
123 	 * WS2012/Win8 and later versions of Hyper-V implement interrupt
124 	 * driven flow management. The feature bit feat_pending_send_sz
125 	 * is set by the host on the host->guest ring buffer, and by the
126 	 * guest on the guest->host ring buffer.
127 	 *
128 	 * The meaning of the feature bit is a bit complex in that it has
129 	 * semantics that apply to both ring buffers.  If the guest sets
130 	 * the feature bit in the guest->host ring buffer, the guest is
131 	 * telling the host that:
132 	 * 1) It will set the pending_send_sz field in the guest->host ring
133 	 *    buffer when it is waiting for space to become available, and
134 	 * 2) It will read the pending_send_sz field in the host->guest
135 	 *    ring buffer and interrupt the host when it frees enough space
136 	 *
137 	 * Similarly, if the host sets the feature bit in the host->guest
138 	 * ring buffer, the host is telling the guest that:
139 	 * 1) It will set the pending_send_sz field in the host->guest ring
140 	 *    buffer when it is waiting for space to become available, and
141 	 * 2) It will read the pending_send_sz field in the guest->host
142 	 *    ring buffer and interrupt the guest when it frees enough space
143 	 *
144 	 * If either the guest or host does not set the feature bit that it
145 	 * owns, that guest or host must do polling if it encounters a full
146 	 * ring buffer, and not signal the other end with an interrupt.
147 	 */
148 	u32 pending_send_sz;
149 	u32 reserved1[12];
150 	union {
151 		struct {
152 			u32 feat_pending_send_sz:1;
153 		};
154 		u32 value;
155 	} feature_bits;
156 
157 	/* Pad it to PAGE_SIZE so that data starts on page boundary */
158 	u8	reserved2[PAGE_SIZE - 68];
159 
160 	/*
161 	 * Ring data starts here + RingDataStartOffset
162 	 * !!! DO NOT place any fields below this !!!
163 	 */
164 	u8 buffer[];
165 } __packed;
166 
167 /* Calculate the proper size of a ringbuffer, it must be page-aligned */
168 #define VMBUS_RING_SIZE(payload_sz) PAGE_ALIGN(sizeof(struct hv_ring_buffer) + \
169 					       (payload_sz))
170 
171 struct hv_ring_buffer_info {
172 	struct hv_ring_buffer *ring_buffer;
173 	u32 ring_size;			/* Include the shared header */
174 	struct reciprocal_value ring_size_div10_reciprocal;
175 	spinlock_t ring_lock;
176 
177 	u32 ring_datasize;		/* < ring_size */
178 	u32 priv_read_index;
179 	/*
180 	 * The ring buffer mutex lock. This lock prevents the ring buffer from
181 	 * being freed while the ring buffer is being accessed.
182 	 */
183 	struct mutex ring_buffer_mutex;
184 
185 	/* Buffer that holds a copy of an incoming host packet */
186 	void *pkt_buffer;
187 	u32 pkt_buffer_size;
188 };
189 
190 
hv_get_bytes_to_read(const struct hv_ring_buffer_info * rbi)191 static inline u32 hv_get_bytes_to_read(const struct hv_ring_buffer_info *rbi)
192 {
193 	u32 read_loc, write_loc, dsize, read;
194 
195 	dsize = rbi->ring_datasize;
196 	read_loc = rbi->ring_buffer->read_index;
197 	write_loc = READ_ONCE(rbi->ring_buffer->write_index);
198 
199 	read = write_loc >= read_loc ? (write_loc - read_loc) :
200 		(dsize - read_loc) + write_loc;
201 
202 	return read;
203 }
204 
hv_get_bytes_to_write(const struct hv_ring_buffer_info * rbi)205 static inline u32 hv_get_bytes_to_write(const struct hv_ring_buffer_info *rbi)
206 {
207 	u32 read_loc, write_loc, dsize, write;
208 
209 	dsize = rbi->ring_datasize;
210 	read_loc = READ_ONCE(rbi->ring_buffer->read_index);
211 	write_loc = rbi->ring_buffer->write_index;
212 
213 	write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
214 		read_loc - write_loc;
215 	return write;
216 }
217 
hv_get_avail_to_write_percent(const struct hv_ring_buffer_info * rbi)218 static inline u32 hv_get_avail_to_write_percent(
219 		const struct hv_ring_buffer_info *rbi)
220 {
221 	u32 avail_write = hv_get_bytes_to_write(rbi);
222 
223 	return reciprocal_divide(
224 			(avail_write  << 3) + (avail_write << 1),
225 			rbi->ring_size_div10_reciprocal);
226 }
227 
228 /*
229  * VMBUS version is 32 bit entity broken up into
230  * two 16 bit quantities: major_number. minor_number.
231  *
232  * 0 . 13 (Windows Server 2008)
233  * 1 . 1  (Windows 7, WS2008 R2)
234  * 2 . 4  (Windows 8, WS2012)
235  * 3 . 0  (Windows 8.1, WS2012 R2)
236  * 4 . 0  (Windows 10)
237  * 4 . 1  (Windows 10 RS3)
238  * 5 . 0  (Newer Windows 10)
239  * 5 . 1  (Windows 10 RS4)
240  * 5 . 2  (Windows Server 2019, RS5)
241  * 5 . 3  (Windows Server 2022)
242  *
243  * The WS2008 and WIN7 versions are listed here for
244  * completeness but are no longer supported in the
245  * Linux kernel.
246  */
247 
248 #define VERSION_WS2008  ((0 << 16) | (13))
249 #define VERSION_WIN7    ((1 << 16) | (1))
250 #define VERSION_WIN8    ((2 << 16) | (4))
251 #define VERSION_WIN8_1    ((3 << 16) | (0))
252 #define VERSION_WIN10 ((4 << 16) | (0))
253 #define VERSION_WIN10_V4_1 ((4 << 16) | (1))
254 #define VERSION_WIN10_V5 ((5 << 16) | (0))
255 #define VERSION_WIN10_V5_1 ((5 << 16) | (1))
256 #define VERSION_WIN10_V5_2 ((5 << 16) | (2))
257 #define VERSION_WIN10_V5_3 ((5 << 16) | (3))
258 
259 /* Make maximum size of pipe payload of 16K */
260 #define MAX_PIPE_DATA_PAYLOAD		(sizeof(u8) * 16384)
261 
262 /* Define PipeMode values. */
263 #define VMBUS_PIPE_TYPE_BYTE		0x00000000
264 #define VMBUS_PIPE_TYPE_MESSAGE		0x00000004
265 
266 /* The size of the user defined data buffer for non-pipe offers. */
267 #define MAX_USER_DEFINED_BYTES		120
268 
269 /* The size of the user defined data buffer for pipe offers. */
270 #define MAX_PIPE_USER_DEFINED_BYTES	116
271 
272 /*
273  * At the center of the Channel Management library is the Channel Offer. This
274  * struct contains the fundamental information about an offer.
275  */
276 struct vmbus_channel_offer {
277 	guid_t if_type;
278 	guid_t if_instance;
279 
280 	/*
281 	 * These two fields are not currently used.
282 	 */
283 	u64 reserved1;
284 	u64 reserved2;
285 
286 	u16 chn_flags;
287 	u16 mmio_megabytes;		/* in bytes * 1024 * 1024 */
288 
289 	union {
290 		/* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
291 		struct {
292 			unsigned char user_def[MAX_USER_DEFINED_BYTES];
293 		} std;
294 
295 		/*
296 		 * Pipes:
297 		 * The following structure is an integrated pipe protocol, which
298 		 * is implemented on top of standard user-defined data. Pipe
299 		 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
300 		 * use.
301 		 */
302 		struct {
303 			u32  pipe_mode;
304 			unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
305 		} pipe;
306 	} u;
307 	/*
308 	 * The sub_channel_index is defined in Win8: a value of zero means a
309 	 * primary channel and a value of non-zero means a sub-channel.
310 	 *
311 	 * Before Win8, the field is reserved, meaning it's always zero.
312 	 */
313 	u16 sub_channel_index;
314 	u16 reserved3;
315 } __packed;
316 
317 /* Server Flags */
318 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE	1
319 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES	2
320 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS		4
321 #define VMBUS_CHANNEL_NAMED_PIPE_MODE			0x10
322 #define VMBUS_CHANNEL_LOOPBACK_OFFER			0x100
323 #define VMBUS_CHANNEL_PARENT_OFFER			0x200
324 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION	0x400
325 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER		0x2000
326 
327 struct vmpacket_descriptor {
328 	u16 type;
329 	u16 offset8;
330 	u16 len8;
331 	u16 flags;
332 	u64 trans_id;
333 } __packed;
334 
335 struct vmpacket_header {
336 	u32 prev_pkt_start_offset;
337 	struct vmpacket_descriptor descriptor;
338 } __packed;
339 
340 struct vmtransfer_page_range {
341 	u32 byte_count;
342 	u32 byte_offset;
343 } __packed;
344 
345 struct vmtransfer_page_packet_header {
346 	struct vmpacket_descriptor d;
347 	u16 xfer_pageset_id;
348 	u8  sender_owns_set;
349 	u8 reserved;
350 	u32 range_cnt;
351 	struct vmtransfer_page_range ranges[];
352 } __packed;
353 
354 struct vmgpadl_packet_header {
355 	struct vmpacket_descriptor d;
356 	u32 gpadl;
357 	u32 reserved;
358 } __packed;
359 
360 struct vmadd_remove_transfer_page_set {
361 	struct vmpacket_descriptor d;
362 	u32 gpadl;
363 	u16 xfer_pageset_id;
364 	u16 reserved;
365 } __packed;
366 
367 /*
368  * This structure defines a range in guest physical space that can be made to
369  * look virtually contiguous.
370  */
371 struct gpa_range {
372 	u32 byte_count;
373 	u32 byte_offset;
374 	u64 pfn_array[];
375 };
376 
377 /*
378  * This is the format for an Establish Gpadl packet, which contains a handle by
379  * which this GPADL will be known and a set of GPA ranges associated with it.
380  * This can be converted to a MDL by the guest OS.  If there are multiple GPA
381  * ranges, then the resulting MDL will be "chained," representing multiple VA
382  * ranges.
383  */
384 struct vmestablish_gpadl {
385 	struct vmpacket_descriptor d;
386 	u32 gpadl;
387 	u32 range_cnt;
388 	struct gpa_range range[1];
389 } __packed;
390 
391 /*
392  * This is the format for a Teardown Gpadl packet, which indicates that the
393  * GPADL handle in the Establish Gpadl packet will never be referenced again.
394  */
395 struct vmteardown_gpadl {
396 	struct vmpacket_descriptor d;
397 	u32 gpadl;
398 	u32 reserved;	/* for alignment to a 8-byte boundary */
399 } __packed;
400 
401 /*
402  * This is the format for a GPA-Direct packet, which contains a set of GPA
403  * ranges, in addition to commands and/or data.
404  */
405 struct vmdata_gpa_direct {
406 	struct vmpacket_descriptor d;
407 	u32 reserved;
408 	u32 range_cnt;
409 	struct gpa_range range[1];
410 } __packed;
411 
412 /* This is the format for a Additional Data Packet. */
413 struct vmadditional_data {
414 	struct vmpacket_descriptor d;
415 	u64 total_bytes;
416 	u32 offset;
417 	u32 byte_cnt;
418 	unsigned char data[1];
419 } __packed;
420 
421 union vmpacket_largest_possible_header {
422 	struct vmpacket_descriptor simple_hdr;
423 	struct vmtransfer_page_packet_header xfer_page_hdr;
424 	struct vmgpadl_packet_header gpadl_hdr;
425 	struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr;
426 	struct vmestablish_gpadl establish_gpadl_hdr;
427 	struct vmteardown_gpadl teardown_gpadl_hdr;
428 	struct vmdata_gpa_direct data_gpa_direct_hdr;
429 };
430 
431 #define VMPACKET_DATA_START_ADDRESS(__packet)	\
432 	(void *)(((unsigned char *)__packet) +	\
433 	 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
434 
435 #define VMPACKET_DATA_LENGTH(__packet)		\
436 	((((struct vmpacket_descriptor)__packet)->len8 -	\
437 	  ((struct vmpacket_descriptor)__packet)->offset8) * 8)
438 
439 #define VMPACKET_TRANSFER_MODE(__packet)	\
440 	(((struct IMPACT)__packet)->type)
441 
442 enum vmbus_packet_type {
443 	VM_PKT_INVALID				= 0x0,
444 	VM_PKT_SYNCH				= 0x1,
445 	VM_PKT_ADD_XFER_PAGESET			= 0x2,
446 	VM_PKT_RM_XFER_PAGESET			= 0x3,
447 	VM_PKT_ESTABLISH_GPADL			= 0x4,
448 	VM_PKT_TEARDOWN_GPADL			= 0x5,
449 	VM_PKT_DATA_INBAND			= 0x6,
450 	VM_PKT_DATA_USING_XFER_PAGES		= 0x7,
451 	VM_PKT_DATA_USING_GPADL			= 0x8,
452 	VM_PKT_DATA_USING_GPA_DIRECT		= 0x9,
453 	VM_PKT_CANCEL_REQUEST			= 0xa,
454 	VM_PKT_COMP				= 0xb,
455 	VM_PKT_DATA_USING_ADDITIONAL_PKT	= 0xc,
456 	VM_PKT_ADDITIONAL_DATA			= 0xd
457 };
458 
459 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED	1
460 
461 
462 /* Version 1 messages */
463 enum vmbus_channel_message_type {
464 	CHANNELMSG_INVALID			=  0,
465 	CHANNELMSG_OFFERCHANNEL		=  1,
466 	CHANNELMSG_RESCIND_CHANNELOFFER	=  2,
467 	CHANNELMSG_REQUESTOFFERS		=  3,
468 	CHANNELMSG_ALLOFFERS_DELIVERED	=  4,
469 	CHANNELMSG_OPENCHANNEL		=  5,
470 	CHANNELMSG_OPENCHANNEL_RESULT		=  6,
471 	CHANNELMSG_CLOSECHANNEL		=  7,
472 	CHANNELMSG_GPADL_HEADER		=  8,
473 	CHANNELMSG_GPADL_BODY			=  9,
474 	CHANNELMSG_GPADL_CREATED		= 10,
475 	CHANNELMSG_GPADL_TEARDOWN		= 11,
476 	CHANNELMSG_GPADL_TORNDOWN		= 12,
477 	CHANNELMSG_RELID_RELEASED		= 13,
478 	CHANNELMSG_INITIATE_CONTACT		= 14,
479 	CHANNELMSG_VERSION_RESPONSE		= 15,
480 	CHANNELMSG_UNLOAD			= 16,
481 	CHANNELMSG_UNLOAD_RESPONSE		= 17,
482 	CHANNELMSG_18				= 18,
483 	CHANNELMSG_19				= 19,
484 	CHANNELMSG_20				= 20,
485 	CHANNELMSG_TL_CONNECT_REQUEST		= 21,
486 	CHANNELMSG_MODIFYCHANNEL		= 22,
487 	CHANNELMSG_TL_CONNECT_RESULT		= 23,
488 	CHANNELMSG_MODIFYCHANNEL_RESPONSE	= 24,
489 	CHANNELMSG_COUNT
490 };
491 
492 /* Hyper-V supports about 2048 channels, and the RELIDs start with 1. */
493 #define INVALID_RELID	U32_MAX
494 
495 struct vmbus_channel_message_header {
496 	enum vmbus_channel_message_type msgtype;
497 	u32 padding;
498 } __packed;
499 
500 /* Query VMBus Version parameters */
501 struct vmbus_channel_query_vmbus_version {
502 	struct vmbus_channel_message_header header;
503 	u32 version;
504 } __packed;
505 
506 /* VMBus Version Supported parameters */
507 struct vmbus_channel_version_supported {
508 	struct vmbus_channel_message_header header;
509 	u8 version_supported;
510 } __packed;
511 
512 /* Offer Channel parameters */
513 struct vmbus_channel_offer_channel {
514 	struct vmbus_channel_message_header header;
515 	struct vmbus_channel_offer offer;
516 	u32 child_relid;
517 	u8 monitorid;
518 	/*
519 	 * win7 and beyond splits this field into a bit field.
520 	 */
521 	u8 monitor_allocated:1;
522 	u8 reserved:7;
523 	/*
524 	 * These are new fields added in win7 and later.
525 	 * Do not access these fields without checking the
526 	 * negotiated protocol.
527 	 *
528 	 * If "is_dedicated_interrupt" is set, we must not set the
529 	 * associated bit in the channel bitmap while sending the
530 	 * interrupt to the host.
531 	 *
532 	 * connection_id is to be used in signaling the host.
533 	 */
534 	u16 is_dedicated_interrupt:1;
535 	u16 reserved1:15;
536 	u32 connection_id;
537 } __packed;
538 
539 /* Rescind Offer parameters */
540 struct vmbus_channel_rescind_offer {
541 	struct vmbus_channel_message_header header;
542 	u32 child_relid;
543 } __packed;
544 
545 /*
546  * Request Offer -- no parameters, SynIC message contains the partition ID
547  * Set Snoop -- no parameters, SynIC message contains the partition ID
548  * Clear Snoop -- no parameters, SynIC message contains the partition ID
549  * All Offers Delivered -- no parameters, SynIC message contains the partition
550  *		           ID
551  * Flush Client -- no parameters, SynIC message contains the partition ID
552  */
553 
554 /* Open Channel parameters */
555 struct vmbus_channel_open_channel {
556 	struct vmbus_channel_message_header header;
557 
558 	/* Identifies the specific VMBus channel that is being opened. */
559 	u32 child_relid;
560 
561 	/* ID making a particular open request at a channel offer unique. */
562 	u32 openid;
563 
564 	/* GPADL for the channel's ring buffer. */
565 	u32 ringbuffer_gpadlhandle;
566 
567 	/*
568 	 * Starting with win8, this field will be used to specify
569 	 * the target virtual processor on which to deliver the interrupt for
570 	 * the host to guest communication.
571 	 * Prior to win8, incoming channel interrupts would only
572 	 * be delivered on cpu 0. Setting this value to 0 would
573 	 * preserve the earlier behavior.
574 	 */
575 	u32 target_vp;
576 
577 	/*
578 	 * The upstream ring buffer begins at offset zero in the memory
579 	 * described by RingBufferGpadlHandle. The downstream ring buffer
580 	 * follows it at this offset (in pages).
581 	 */
582 	u32 downstream_ringbuffer_pageoffset;
583 
584 	/* User-specific data to be passed along to the server endpoint. */
585 	unsigned char userdata[MAX_USER_DEFINED_BYTES];
586 } __packed;
587 
588 /* Open Channel Result parameters */
589 struct vmbus_channel_open_result {
590 	struct vmbus_channel_message_header header;
591 	u32 child_relid;
592 	u32 openid;
593 	u32 status;
594 } __packed;
595 
596 /* Modify Channel Result parameters */
597 struct vmbus_channel_modifychannel_response {
598 	struct vmbus_channel_message_header header;
599 	u32 child_relid;
600 	u32 status;
601 } __packed;
602 
603 /* Close channel parameters; */
604 struct vmbus_channel_close_channel {
605 	struct vmbus_channel_message_header header;
606 	u32 child_relid;
607 } __packed;
608 
609 /* Channel Message GPADL */
610 #define GPADL_TYPE_RING_BUFFER		1
611 #define GPADL_TYPE_SERVER_SAVE_AREA	2
612 #define GPADL_TYPE_TRANSACTION		8
613 
614 /*
615  * The number of PFNs in a GPADL message is defined by the number of
616  * pages that would be spanned by ByteCount and ByteOffset.  If the
617  * implied number of PFNs won't fit in this packet, there will be a
618  * follow-up packet that contains more.
619  */
620 struct vmbus_channel_gpadl_header {
621 	struct vmbus_channel_message_header header;
622 	u32 child_relid;
623 	u32 gpadl;
624 	u16 range_buflen;
625 	u16 rangecount;
626 	struct gpa_range range[];
627 } __packed;
628 
629 /* This is the followup packet that contains more PFNs. */
630 struct vmbus_channel_gpadl_body {
631 	struct vmbus_channel_message_header header;
632 	u32 msgnumber;
633 	u32 gpadl;
634 	u64 pfn[];
635 } __packed;
636 
637 struct vmbus_channel_gpadl_created {
638 	struct vmbus_channel_message_header header;
639 	u32 child_relid;
640 	u32 gpadl;
641 	u32 creation_status;
642 } __packed;
643 
644 struct vmbus_channel_gpadl_teardown {
645 	struct vmbus_channel_message_header header;
646 	u32 child_relid;
647 	u32 gpadl;
648 } __packed;
649 
650 struct vmbus_channel_gpadl_torndown {
651 	struct vmbus_channel_message_header header;
652 	u32 gpadl;
653 } __packed;
654 
655 struct vmbus_channel_relid_released {
656 	struct vmbus_channel_message_header header;
657 	u32 child_relid;
658 } __packed;
659 
660 struct vmbus_channel_initiate_contact {
661 	struct vmbus_channel_message_header header;
662 	u32 vmbus_version_requested;
663 	u32 target_vcpu; /* The VCPU the host should respond to */
664 	union {
665 		u64 interrupt_page;
666 		struct {
667 			u8	msg_sint;
668 			u8	msg_vtl;
669 			u8	reserved[6];
670 		};
671 	};
672 	u64 monitor_page1;
673 	u64 monitor_page2;
674 } __packed;
675 
676 /* Hyper-V socket: guest's connect()-ing to host */
677 struct vmbus_channel_tl_connect_request {
678 	struct vmbus_channel_message_header header;
679 	guid_t guest_endpoint_id;
680 	guid_t host_service_id;
681 } __packed;
682 
683 /* Modify Channel parameters, cf. vmbus_send_modifychannel() */
684 struct vmbus_channel_modifychannel {
685 	struct vmbus_channel_message_header header;
686 	u32 child_relid;
687 	u32 target_vp;
688 } __packed;
689 
690 struct vmbus_channel_version_response {
691 	struct vmbus_channel_message_header header;
692 	u8 version_supported;
693 
694 	u8 connection_state;
695 	u16 padding;
696 
697 	/*
698 	 * On new hosts that support VMBus protocol 5.0, we must use
699 	 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
700 	 * and for subsequent messages, we must use the Message Connection ID
701 	 * field in the host-returned Version Response Message.
702 	 *
703 	 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
704 	 */
705 	u32 msg_conn_id;
706 } __packed;
707 
708 enum vmbus_channel_state {
709 	CHANNEL_OFFER_STATE,
710 	CHANNEL_OPENING_STATE,
711 	CHANNEL_OPEN_STATE,
712 	CHANNEL_OPENED_STATE,
713 };
714 
715 /*
716  * Represents each channel msg on the vmbus connection This is a
717  * variable-size data structure depending on the msg type itself
718  */
719 struct vmbus_channel_msginfo {
720 	/* Bookkeeping stuff */
721 	struct list_head msglistentry;
722 
723 	/* So far, this is only used to handle gpadl body message */
724 	struct list_head submsglist;
725 
726 	/* Synchronize the request/response if needed */
727 	struct completion  waitevent;
728 	struct vmbus_channel *waiting_channel;
729 	union {
730 		struct vmbus_channel_version_supported version_supported;
731 		struct vmbus_channel_open_result open_result;
732 		struct vmbus_channel_gpadl_torndown gpadl_torndown;
733 		struct vmbus_channel_gpadl_created gpadl_created;
734 		struct vmbus_channel_version_response version_response;
735 		struct vmbus_channel_modifychannel_response modify_response;
736 	} response;
737 
738 	u32 msgsize;
739 	/*
740 	 * The channel message that goes out on the "wire".
741 	 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
742 	 */
743 	unsigned char msg[];
744 };
745 
746 struct vmbus_close_msg {
747 	struct vmbus_channel_msginfo info;
748 	struct vmbus_channel_close_channel msg;
749 };
750 
751 /* Define connection identifier type. */
752 union hv_connection_id {
753 	u32 asu32;
754 	struct {
755 		u32 id:24;
756 		u32 reserved:8;
757 	} u;
758 };
759 
760 enum vmbus_device_type {
761 	HV_IDE = 0,
762 	HV_SCSI,
763 	HV_FC,
764 	HV_NIC,
765 	HV_ND,
766 	HV_PCIE,
767 	HV_FB,
768 	HV_KBD,
769 	HV_MOUSE,
770 	HV_KVP,
771 	HV_TS,
772 	HV_HB,
773 	HV_SHUTDOWN,
774 	HV_FCOPY,
775 	HV_BACKUP,
776 	HV_DM,
777 	HV_UNKNOWN,
778 };
779 
780 /*
781  * Provides request ids for VMBus. Encapsulates guest memory
782  * addresses and stores the next available slot in req_arr
783  * to generate new ids in constant time.
784  */
785 struct vmbus_requestor {
786 	u64 *req_arr;
787 	unsigned long *req_bitmap; /* is a given slot available? */
788 	u32 size;
789 	u64 next_request_id;
790 	spinlock_t req_lock; /* provides atomicity */
791 };
792 
793 #define VMBUS_NO_RQSTOR U64_MAX
794 #define VMBUS_RQST_ERROR (U64_MAX - 1)
795 #define VMBUS_RQST_ADDR_ANY U64_MAX
796 /* NetVSC-specific */
797 #define VMBUS_RQST_ID_NO_RESPONSE (U64_MAX - 2)
798 /* StorVSC-specific */
799 #define VMBUS_RQST_INIT (U64_MAX - 2)
800 #define VMBUS_RQST_RESET (U64_MAX - 3)
801 
802 struct vmbus_device {
803 	u16  dev_type;
804 	guid_t guid;
805 	bool perf_device;
806 	bool allowed_in_isolated;
807 };
808 
809 #define VMBUS_DEFAULT_MAX_PKT_SIZE 4096
810 
811 struct vmbus_gpadl {
812 	u32 gpadl_handle;
813 	u32 size;
814 	void *buffer;
815 };
816 
817 struct vmbus_channel {
818 	struct list_head listentry;
819 
820 	struct hv_device *device_obj;
821 
822 	enum vmbus_channel_state state;
823 
824 	struct vmbus_channel_offer_channel offermsg;
825 	/*
826 	 * These are based on the OfferMsg.MonitorId.
827 	 * Save it here for easy access.
828 	 */
829 	u8 monitor_grp;
830 	u8 monitor_bit;
831 
832 	bool rescind; /* got rescind msg */
833 	bool rescind_ref; /* got rescind msg, got channel reference */
834 	struct completion rescind_event;
835 
836 	struct vmbus_gpadl ringbuffer_gpadlhandle;
837 
838 	/* Allocated memory for ring buffer */
839 	struct page *ringbuffer_page;
840 	u32 ringbuffer_pagecount;
841 	u32 ringbuffer_send_offset;
842 	struct hv_ring_buffer_info outbound;	/* send to parent */
843 	struct hv_ring_buffer_info inbound;	/* receive from parent */
844 
845 	struct vmbus_close_msg close_msg;
846 
847 	/* Statistics */
848 	u64	interrupts;	/* Host to Guest interrupts */
849 	u64	sig_events;	/* Guest to Host events */
850 
851 	/*
852 	 * Guest to host interrupts caused by the outbound ring buffer changing
853 	 * from empty to not empty.
854 	 */
855 	u64 intr_out_empty;
856 
857 	/*
858 	 * Indicates that a full outbound ring buffer was encountered. The flag
859 	 * is set to true when a full outbound ring buffer is encountered and
860 	 * set to false when a write to the outbound ring buffer is completed.
861 	 */
862 	bool out_full_flag;
863 
864 	/* Channel callback's invoked in softirq context */
865 	struct tasklet_struct callback_event;
866 	void (*onchannel_callback)(void *context);
867 	void *channel_callback_context;
868 
869 	void (*change_target_cpu_callback)(struct vmbus_channel *channel,
870 			u32 old, u32 new);
871 
872 	/*
873 	 * Synchronize channel scheduling and channel removal; see the inline
874 	 * comments in vmbus_chan_sched() and vmbus_reset_channel_cb().
875 	 */
876 	spinlock_t sched_lock;
877 
878 	/*
879 	 * A channel can be marked for one of three modes of reading:
880 	 *   BATCHED - callback called from taslket and should read
881 	 *            channel until empty. Interrupts from the host
882 	 *            are masked while read is in process (default).
883 	 *   DIRECT - callback called from tasklet (softirq).
884 	 *   ISR - callback called in interrupt context and must
885 	 *         invoke its own deferred processing.
886 	 *         Host interrupts are disabled and must be re-enabled
887 	 *         when ring is empty.
888 	 */
889 	enum hv_callback_mode {
890 		HV_CALL_BATCHED,
891 		HV_CALL_DIRECT,
892 		HV_CALL_ISR
893 	} callback_mode;
894 
895 	bool is_dedicated_interrupt;
896 	u64 sig_event;
897 
898 	/*
899 	 * Starting with win8, this field will be used to specify the
900 	 * target CPU on which to deliver the interrupt for the host
901 	 * to guest communication.
902 	 *
903 	 * Prior to win8, incoming channel interrupts would only be
904 	 * delivered on CPU 0. Setting this value to 0 would preserve
905 	 * the earlier behavior.
906 	 */
907 	u32 target_cpu;
908 	/*
909 	 * Support for sub-channels. For high performance devices,
910 	 * it will be useful to have multiple sub-channels to support
911 	 * a scalable communication infrastructure with the host.
912 	 * The support for sub-channels is implemented as an extension
913 	 * to the current infrastructure.
914 	 * The initial offer is considered the primary channel and this
915 	 * offer message will indicate if the host supports sub-channels.
916 	 * The guest is free to ask for sub-channels to be offered and can
917 	 * open these sub-channels as a normal "primary" channel. However,
918 	 * all sub-channels will have the same type and instance guids as the
919 	 * primary channel. Requests sent on a given channel will result in a
920 	 * response on the same channel.
921 	 */
922 
923 	/*
924 	 * Sub-channel creation callback. This callback will be called in
925 	 * process context when a sub-channel offer is received from the host.
926 	 * The guest can open the sub-channel in the context of this callback.
927 	 */
928 	void (*sc_creation_callback)(struct vmbus_channel *new_sc);
929 
930 	/*
931 	 * Channel rescind callback. Some channels (the hvsock ones), need to
932 	 * register a callback which is invoked in vmbus_onoffer_rescind().
933 	 */
934 	void (*chn_rescind_callback)(struct vmbus_channel *channel);
935 
936 	/*
937 	 * All Sub-channels of a primary channel are linked here.
938 	 */
939 	struct list_head sc_list;
940 	/*
941 	 * The primary channel this sub-channel belongs to.
942 	 * This will be NULL for the primary channel.
943 	 */
944 	struct vmbus_channel *primary_channel;
945 	/*
946 	 * Support per-channel state for use by vmbus drivers.
947 	 */
948 	void *per_channel_state;
949 
950 	/*
951 	 * Defer freeing channel until after all cpu's have
952 	 * gone through grace period.
953 	 */
954 	struct rcu_head rcu;
955 
956 	/*
957 	 * For sysfs per-channel properties.
958 	 */
959 	struct kobject			kobj;
960 
961 	/*
962 	 * For performance critical channels (storage, networking
963 	 * etc,), Hyper-V has a mechanism to enhance the throughput
964 	 * at the expense of latency:
965 	 * When the host is to be signaled, we just set a bit in a shared page
966 	 * and this bit will be inspected by the hypervisor within a certain
967 	 * window and if the bit is set, the host will be signaled. The window
968 	 * of time is the monitor latency - currently around 100 usecs. This
969 	 * mechanism improves throughput by:
970 	 *
971 	 * A) Making the host more efficient - each time it wakes up,
972 	 *    potentially it will process more number of packets. The
973 	 *    monitor latency allows a batch to build up.
974 	 * B) By deferring the hypercall to signal, we will also minimize
975 	 *    the interrupts.
976 	 *
977 	 * Clearly, these optimizations improve throughput at the expense of
978 	 * latency. Furthermore, since the channel is shared for both
979 	 * control and data messages, control messages currently suffer
980 	 * unnecessary latency adversely impacting performance and boot
981 	 * time. To fix this issue, permit tagging the channel as being
982 	 * in "low latency" mode. In this mode, we will bypass the monitor
983 	 * mechanism.
984 	 */
985 	bool low_latency;
986 
987 	bool probe_done;
988 
989 	/*
990 	 * Cache the device ID here for easy access; this is useful, in
991 	 * particular, in situations where the channel's device_obj has
992 	 * not been allocated/initialized yet.
993 	 */
994 	u16 device_id;
995 
996 	/*
997 	 * We must offload the handling of the primary/sub channels
998 	 * from the single-threaded vmbus_connection.work_queue to
999 	 * two different workqueue, otherwise we can block
1000 	 * vmbus_connection.work_queue and hang: see vmbus_process_offer().
1001 	 */
1002 	struct work_struct add_channel_work;
1003 
1004 	/*
1005 	 * Guest to host interrupts caused by the inbound ring buffer changing
1006 	 * from full to not full while a packet is waiting.
1007 	 */
1008 	u64 intr_in_full;
1009 
1010 	/*
1011 	 * The total number of write operations that encountered a full
1012 	 * outbound ring buffer.
1013 	 */
1014 	u64 out_full_total;
1015 
1016 	/*
1017 	 * The number of write operations that were the first to encounter a
1018 	 * full outbound ring buffer.
1019 	 */
1020 	u64 out_full_first;
1021 
1022 	/* enabling/disabling fuzz testing on the channel (default is false)*/
1023 	bool fuzz_testing_state;
1024 
1025 	/*
1026 	 * Interrupt delay will delay the guest from emptying the ring buffer
1027 	 * for a specific amount of time. The delay is in microseconds and will
1028 	 * be between 1 to a maximum of 1000, its default is 0 (no delay).
1029 	 * The  Message delay will delay guest reading on a per message basis
1030 	 * in microseconds between 1 to 1000 with the default being 0
1031 	 * (no delay).
1032 	 */
1033 	u32 fuzz_testing_interrupt_delay;
1034 	u32 fuzz_testing_message_delay;
1035 
1036 	/* callback to generate a request ID from a request address */
1037 	u64 (*next_request_id_callback)(struct vmbus_channel *channel, u64 rqst_addr);
1038 	/* callback to retrieve a request address from a request ID */
1039 	u64 (*request_addr_callback)(struct vmbus_channel *channel, u64 rqst_id);
1040 
1041 	/* request/transaction ids for VMBus */
1042 	struct vmbus_requestor requestor;
1043 	u32 rqstor_size;
1044 
1045 	/* The max size of a packet on this channel */
1046 	u32 max_pkt_size;
1047 };
1048 
1049 #define lock_requestor(channel, flags)					\
1050 do {									\
1051 	struct vmbus_requestor *rqstor = &(channel)->requestor;		\
1052 									\
1053 	spin_lock_irqsave(&rqstor->req_lock, flags);			\
1054 } while (0)
1055 
unlock_requestor(struct vmbus_channel * channel,unsigned long flags)1056 static __always_inline void unlock_requestor(struct vmbus_channel *channel,
1057 					     unsigned long flags)
1058 {
1059 	struct vmbus_requestor *rqstor = &channel->requestor;
1060 
1061 	spin_unlock_irqrestore(&rqstor->req_lock, flags);
1062 }
1063 
1064 u64 vmbus_next_request_id(struct vmbus_channel *channel, u64 rqst_addr);
1065 u64 __vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1066 			       u64 rqst_addr);
1067 u64 vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1068 			     u64 rqst_addr);
1069 u64 vmbus_request_addr(struct vmbus_channel *channel, u64 trans_id);
1070 
is_hvsock_offer(const struct vmbus_channel_offer_channel * o)1071 static inline bool is_hvsock_offer(const struct vmbus_channel_offer_channel *o)
1072 {
1073 	return !!(o->offer.chn_flags & VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER);
1074 }
1075 
is_hvsock_channel(const struct vmbus_channel * c)1076 static inline bool is_hvsock_channel(const struct vmbus_channel *c)
1077 {
1078 	return is_hvsock_offer(&c->offermsg);
1079 }
1080 
is_sub_channel(const struct vmbus_channel * c)1081 static inline bool is_sub_channel(const struct vmbus_channel *c)
1082 {
1083 	return c->offermsg.offer.sub_channel_index != 0;
1084 }
1085 
set_channel_read_mode(struct vmbus_channel * c,enum hv_callback_mode mode)1086 static inline void set_channel_read_mode(struct vmbus_channel *c,
1087 					enum hv_callback_mode mode)
1088 {
1089 	c->callback_mode = mode;
1090 }
1091 
set_per_channel_state(struct vmbus_channel * c,void * s)1092 static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
1093 {
1094 	c->per_channel_state = s;
1095 }
1096 
get_per_channel_state(struct vmbus_channel * c)1097 static inline void *get_per_channel_state(struct vmbus_channel *c)
1098 {
1099 	return c->per_channel_state;
1100 }
1101 
set_channel_pending_send_size(struct vmbus_channel * c,u32 size)1102 static inline void set_channel_pending_send_size(struct vmbus_channel *c,
1103 						 u32 size)
1104 {
1105 	unsigned long flags;
1106 
1107 	if (size) {
1108 		spin_lock_irqsave(&c->outbound.ring_lock, flags);
1109 		++c->out_full_total;
1110 
1111 		if (!c->out_full_flag) {
1112 			++c->out_full_first;
1113 			c->out_full_flag = true;
1114 		}
1115 		spin_unlock_irqrestore(&c->outbound.ring_lock, flags);
1116 	} else {
1117 		c->out_full_flag = false;
1118 	}
1119 
1120 	c->outbound.ring_buffer->pending_send_sz = size;
1121 }
1122 
1123 void vmbus_onmessage(struct vmbus_channel_message_header *hdr);
1124 
1125 int vmbus_request_offers(void);
1126 
1127 /*
1128  * APIs for managing sub-channels.
1129  */
1130 
1131 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1132 			void (*sc_cr_cb)(struct vmbus_channel *new_sc));
1133 
1134 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1135 		void (*chn_rescind_cb)(struct vmbus_channel *));
1136 
1137 /* The format must be the same as struct vmdata_gpa_direct */
1138 struct vmbus_channel_packet_page_buffer {
1139 	u16 type;
1140 	u16 dataoffset8;
1141 	u16 length8;
1142 	u16 flags;
1143 	u64 transactionid;
1144 	u32 reserved;
1145 	u32 rangecount;
1146 	struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
1147 } __packed;
1148 
1149 /* The format must be the same as struct vmdata_gpa_direct */
1150 struct vmbus_channel_packet_multipage_buffer {
1151 	u16 type;
1152 	u16 dataoffset8;
1153 	u16 length8;
1154 	u16 flags;
1155 	u64 transactionid;
1156 	u32 reserved;
1157 	u32 rangecount;		/* Always 1 in this case */
1158 	struct hv_multipage_buffer range;
1159 } __packed;
1160 
1161 /* The format must be the same as struct vmdata_gpa_direct */
1162 struct vmbus_packet_mpb_array {
1163 	u16 type;
1164 	u16 dataoffset8;
1165 	u16 length8;
1166 	u16 flags;
1167 	u64 transactionid;
1168 	u32 reserved;
1169 	u32 rangecount;         /* Always 1 in this case */
1170 	struct hv_mpb_array range;
1171 } __packed;
1172 
1173 int vmbus_alloc_ring(struct vmbus_channel *channel,
1174 		     u32 send_size, u32 recv_size);
1175 void vmbus_free_ring(struct vmbus_channel *channel);
1176 
1177 int vmbus_connect_ring(struct vmbus_channel *channel,
1178 		       void (*onchannel_callback)(void *context),
1179 		       void *context);
1180 int vmbus_disconnect_ring(struct vmbus_channel *channel);
1181 
1182 extern int vmbus_open(struct vmbus_channel *channel,
1183 			    u32 send_ringbuffersize,
1184 			    u32 recv_ringbuffersize,
1185 			    void *userdata,
1186 			    u32 userdatalen,
1187 			    void (*onchannel_callback)(void *context),
1188 			    void *context);
1189 
1190 extern void vmbus_close(struct vmbus_channel *channel);
1191 
1192 extern int vmbus_sendpacket_getid(struct vmbus_channel *channel,
1193 				  void *buffer,
1194 				  u32 bufferLen,
1195 				  u64 requestid,
1196 				  u64 *trans_id,
1197 				  enum vmbus_packet_type type,
1198 				  u32 flags);
1199 extern int vmbus_sendpacket(struct vmbus_channel *channel,
1200 				  void *buffer,
1201 				  u32 bufferLen,
1202 				  u64 requestid,
1203 				  enum vmbus_packet_type type,
1204 				  u32 flags);
1205 
1206 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1207 					    struct hv_page_buffer pagebuffers[],
1208 					    u32 pagecount,
1209 					    void *buffer,
1210 					    u32 bufferlen,
1211 					    u64 requestid);
1212 
1213 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1214 				     struct vmbus_packet_mpb_array *mpb,
1215 				     u32 desc_size,
1216 				     void *buffer,
1217 				     u32 bufferlen,
1218 				     u64 requestid);
1219 
1220 extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
1221 				      void *kbuffer,
1222 				      u32 size,
1223 				      struct vmbus_gpadl *gpadl);
1224 
1225 extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
1226 				     struct vmbus_gpadl *gpadl);
1227 
1228 void vmbus_reset_channel_cb(struct vmbus_channel *channel);
1229 
1230 extern int vmbus_recvpacket(struct vmbus_channel *channel,
1231 				  void *buffer,
1232 				  u32 bufferlen,
1233 				  u32 *buffer_actual_len,
1234 				  u64 *requestid);
1235 
1236 extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
1237 				     void *buffer,
1238 				     u32 bufferlen,
1239 				     u32 *buffer_actual_len,
1240 				     u64 *requestid);
1241 
1242 /* Base driver object */
1243 struct hv_driver {
1244 	const char *name;
1245 
1246 	/*
1247 	 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1248 	 * channel flag, actually doesn't mean a synthetic device because the
1249 	 * offer's if_type/if_instance can change for every new hvsock
1250 	 * connection.
1251 	 *
1252 	 * However, to facilitate the notification of new-offer/rescind-offer
1253 	 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1254 	 * a special vmbus device, and hence we need the below flag to
1255 	 * indicate if the driver is the hvsock driver or not: we need to
1256 	 * specially treat the hvosck offer & driver in vmbus_match().
1257 	 */
1258 	bool hvsock;
1259 
1260 	/* the device type supported by this driver */
1261 	guid_t dev_type;
1262 	const struct hv_vmbus_device_id *id_table;
1263 
1264 	struct device_driver driver;
1265 
1266 	/* dynamic device GUID's */
1267 	struct  {
1268 		spinlock_t lock;
1269 		struct list_head list;
1270 	} dynids;
1271 
1272 	int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
1273 	void (*remove)(struct hv_device *dev);
1274 	void (*shutdown)(struct hv_device *);
1275 
1276 	int (*suspend)(struct hv_device *);
1277 	int (*resume)(struct hv_device *);
1278 
1279 };
1280 
1281 /* Base device object */
1282 struct hv_device {
1283 	/* the device type id of this device */
1284 	guid_t dev_type;
1285 
1286 	/* the device instance id of this device */
1287 	guid_t dev_instance;
1288 	u16 vendor_id;
1289 	u16 device_id;
1290 
1291 	struct device device;
1292 	/*
1293 	 * Driver name to force a match.  Do not set directly, because core
1294 	 * frees it.  Use driver_set_override() to set or clear it.
1295 	 */
1296 	const char *driver_override;
1297 
1298 	struct vmbus_channel *channel;
1299 	struct kset	     *channels_kset;
1300 	struct device_dma_parameters dma_parms;
1301 	u64 dma_mask;
1302 
1303 	/* place holder to keep track of the dir for hv device in debugfs */
1304 	struct dentry *debug_dir;
1305 
1306 };
1307 
1308 
1309 #define device_to_hv_device(d)	container_of_const(d, struct hv_device, device)
1310 
drv_to_hv_drv(struct device_driver * d)1311 static inline struct hv_driver *drv_to_hv_drv(struct device_driver *d)
1312 {
1313 	return container_of(d, struct hv_driver, driver);
1314 }
1315 
hv_set_drvdata(struct hv_device * dev,void * data)1316 static inline void hv_set_drvdata(struct hv_device *dev, void *data)
1317 {
1318 	dev_set_drvdata(&dev->device, data);
1319 }
1320 
hv_get_drvdata(struct hv_device * dev)1321 static inline void *hv_get_drvdata(struct hv_device *dev)
1322 {
1323 	return dev_get_drvdata(&dev->device);
1324 }
1325 
1326 struct hv_ring_buffer_debug_info {
1327 	u32 current_interrupt_mask;
1328 	u32 current_read_index;
1329 	u32 current_write_index;
1330 	u32 bytes_avail_toread;
1331 	u32 bytes_avail_towrite;
1332 };
1333 
1334 
1335 int hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
1336 				struct hv_ring_buffer_debug_info *debug_info);
1337 
1338 bool hv_ringbuffer_spinlock_busy(struct vmbus_channel *channel);
1339 
1340 /* Vmbus interface */
1341 #define vmbus_driver_register(driver)	\
1342 	__vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1343 int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
1344 					 struct module *owner,
1345 					 const char *mod_name);
1346 void vmbus_driver_unregister(struct hv_driver *hv_driver);
1347 
1348 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel);
1349 
1350 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1351 			resource_size_t min, resource_size_t max,
1352 			resource_size_t size, resource_size_t align,
1353 			bool fb_overlap_ok);
1354 void vmbus_free_mmio(resource_size_t start, resource_size_t size);
1355 
1356 /*
1357  * GUID definitions of various offer types - services offered to the guest.
1358  */
1359 
1360 /*
1361  * Network GUID
1362  * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1363  */
1364 #define HV_NIC_GUID \
1365 	.guid = GUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1366 			  0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1367 
1368 /*
1369  * IDE GUID
1370  * {32412632-86cb-44a2-9b5c-50d1417354f5}
1371  */
1372 #define HV_IDE_GUID \
1373 	.guid = GUID_INIT(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1374 			  0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1375 
1376 /*
1377  * SCSI GUID
1378  * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1379  */
1380 #define HV_SCSI_GUID \
1381 	.guid = GUID_INIT(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1382 			  0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1383 
1384 /*
1385  * Shutdown GUID
1386  * {0e0b6031-5213-4934-818b-38d90ced39db}
1387  */
1388 #define HV_SHUTDOWN_GUID \
1389 	.guid = GUID_INIT(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1390 			  0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1391 
1392 /*
1393  * Time Synch GUID
1394  * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1395  */
1396 #define HV_TS_GUID \
1397 	.guid = GUID_INIT(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1398 			  0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1399 
1400 /*
1401  * Heartbeat GUID
1402  * {57164f39-9115-4e78-ab55-382f3bd5422d}
1403  */
1404 #define HV_HEART_BEAT_GUID \
1405 	.guid = GUID_INIT(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1406 			  0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1407 
1408 /*
1409  * KVP GUID
1410  * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1411  */
1412 #define HV_KVP_GUID \
1413 	.guid = GUID_INIT(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1414 			  0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1415 
1416 /*
1417  * Dynamic memory GUID
1418  * {525074dc-8985-46e2-8057-a307dc18a502}
1419  */
1420 #define HV_DM_GUID \
1421 	.guid = GUID_INIT(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1422 			  0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1423 
1424 /*
1425  * Mouse GUID
1426  * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1427  */
1428 #define HV_MOUSE_GUID \
1429 	.guid = GUID_INIT(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1430 			  0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1431 
1432 /*
1433  * Keyboard GUID
1434  * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1435  */
1436 #define HV_KBD_GUID \
1437 	.guid = GUID_INIT(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1438 			  0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1439 
1440 /*
1441  * VSS (Backup/Restore) GUID
1442  */
1443 #define HV_VSS_GUID \
1444 	.guid = GUID_INIT(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1445 			  0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1446 /*
1447  * Synthetic Video GUID
1448  * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1449  */
1450 #define HV_SYNTHVID_GUID \
1451 	.guid = GUID_INIT(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1452 			  0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1453 
1454 /*
1455  * Synthetic FC GUID
1456  * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1457  */
1458 #define HV_SYNTHFC_GUID \
1459 	.guid = GUID_INIT(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1460 			  0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1461 
1462 /*
1463  * Guest File Copy Service
1464  * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1465  */
1466 
1467 #define HV_FCOPY_GUID \
1468 	.guid = GUID_INIT(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1469 			  0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1470 
1471 /*
1472  * NetworkDirect. This is the guest RDMA service.
1473  * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1474  */
1475 #define HV_ND_GUID \
1476 	.guid = GUID_INIT(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1477 			  0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1478 
1479 /*
1480  * PCI Express Pass Through
1481  * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1482  */
1483 
1484 #define HV_PCIE_GUID \
1485 	.guid = GUID_INIT(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1486 			  0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1487 
1488 /*
1489  * Linux doesn't support these 4 devices: the first two are for
1490  * Automatic Virtual Machine Activation, the third is for
1491  * Remote Desktop Virtualization, and the fourth is Initial
1492  * Machine Configuration (IMC) used only by Windows guests.
1493  * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1494  * {3375baf4-9e15-4b30-b765-67acb10d607b}
1495  * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1496  * {c376c1c3-d276-48d2-90a9-c04748072c60}
1497  */
1498 
1499 #define HV_AVMA1_GUID \
1500 	.guid = GUID_INIT(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1501 			  0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1502 
1503 #define HV_AVMA2_GUID \
1504 	.guid = GUID_INIT(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1505 			  0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1506 
1507 #define HV_RDV_GUID \
1508 	.guid = GUID_INIT(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1509 			  0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1510 
1511 #define HV_IMC_GUID \
1512 	.guid = GUID_INIT(0xc376c1c3, 0xd276, 0x48d2, 0x90, 0xa9, \
1513 			  0xc0, 0x47, 0x48, 0x07, 0x2c, 0x60)
1514 
1515 /*
1516  * Common header for Hyper-V ICs
1517  */
1518 
1519 #define ICMSGTYPE_NEGOTIATE		0
1520 #define ICMSGTYPE_HEARTBEAT		1
1521 #define ICMSGTYPE_KVPEXCHANGE		2
1522 #define ICMSGTYPE_SHUTDOWN		3
1523 #define ICMSGTYPE_TIMESYNC		4
1524 #define ICMSGTYPE_VSS			5
1525 #define ICMSGTYPE_FCOPY			7
1526 
1527 #define ICMSGHDRFLAG_TRANSACTION	1
1528 #define ICMSGHDRFLAG_REQUEST		2
1529 #define ICMSGHDRFLAG_RESPONSE		4
1530 
1531 
1532 /*
1533  * While we want to handle util services as regular devices,
1534  * there is only one instance of each of these services; so
1535  * we statically allocate the service specific state.
1536  */
1537 
1538 struct hv_util_service {
1539 	u8 *recv_buffer;
1540 	void *channel;
1541 	void (*util_cb)(void *);
1542 	int (*util_init)(struct hv_util_service *);
1543 	void (*util_deinit)(void);
1544 	int (*util_pre_suspend)(void);
1545 	int (*util_pre_resume)(void);
1546 };
1547 
1548 struct vmbuspipe_hdr {
1549 	u32 flags;
1550 	u32 msgsize;
1551 } __packed;
1552 
1553 struct ic_version {
1554 	u16 major;
1555 	u16 minor;
1556 } __packed;
1557 
1558 struct icmsg_hdr {
1559 	struct ic_version icverframe;
1560 	u16 icmsgtype;
1561 	struct ic_version icvermsg;
1562 	u16 icmsgsize;
1563 	u32 status;
1564 	u8 ictransaction_id;
1565 	u8 icflags;
1566 	u8 reserved[2];
1567 } __packed;
1568 
1569 #define IC_VERSION_NEGOTIATION_MAX_VER_COUNT 100
1570 #define ICMSG_HDR (sizeof(struct vmbuspipe_hdr) + sizeof(struct icmsg_hdr))
1571 #define ICMSG_NEGOTIATE_PKT_SIZE(icframe_vercnt, icmsg_vercnt) \
1572 	(ICMSG_HDR + sizeof(struct icmsg_negotiate) + \
1573 	 (((icframe_vercnt) + (icmsg_vercnt)) * sizeof(struct ic_version)))
1574 
1575 struct icmsg_negotiate {
1576 	u16 icframe_vercnt;
1577 	u16 icmsg_vercnt;
1578 	u32 reserved;
1579 	struct ic_version icversion_data[]; /* any size array */
1580 } __packed;
1581 
1582 struct shutdown_msg_data {
1583 	u32 reason_code;
1584 	u32 timeout_seconds;
1585 	u32 flags;
1586 	u8  display_message[2048];
1587 } __packed;
1588 
1589 struct heartbeat_msg_data {
1590 	u64 seq_num;
1591 	u32 reserved[8];
1592 } __packed;
1593 
1594 /* Time Sync IC defs */
1595 #define ICTIMESYNCFLAG_PROBE	0
1596 #define ICTIMESYNCFLAG_SYNC	1
1597 #define ICTIMESYNCFLAG_SAMPLE	2
1598 
1599 #ifdef __x86_64__
1600 #define WLTIMEDELTA	116444736000000000L	/* in 100ns unit */
1601 #else
1602 #define WLTIMEDELTA	116444736000000000LL
1603 #endif
1604 
1605 struct ictimesync_data {
1606 	u64 parenttime;
1607 	u64 childtime;
1608 	u64 roundtriptime;
1609 	u8 flags;
1610 } __packed;
1611 
1612 struct ictimesync_ref_data {
1613 	u64 parenttime;
1614 	u64 vmreferencetime;
1615 	u8 flags;
1616 	char leapflags;
1617 	char stratum;
1618 	u8 reserved[3];
1619 } __packed;
1620 
1621 struct hyperv_service_callback {
1622 	u8 msg_type;
1623 	char *log_msg;
1624 	guid_t data;
1625 	struct vmbus_channel *channel;
1626 	void (*callback)(void *context);
1627 };
1628 
1629 struct hv_dma_range {
1630 	dma_addr_t dma;
1631 	u32 mapping_size;
1632 };
1633 
1634 #define MAX_SRV_VER	0x7ffffff
1635 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf, u32 buflen,
1636 				const int *fw_version, int fw_vercnt,
1637 				const int *srv_version, int srv_vercnt,
1638 				int *nego_fw_version, int *nego_srv_version);
1639 
1640 void hv_process_channel_removal(struct vmbus_channel *channel);
1641 
1642 void vmbus_setevent(struct vmbus_channel *channel);
1643 /*
1644  * Negotiated version with the Host.
1645  */
1646 
1647 extern __u32 vmbus_proto_version;
1648 
1649 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
1650 				  const guid_t *shv_host_servie_id);
1651 int vmbus_send_modifychannel(struct vmbus_channel *channel, u32 target_vp);
1652 void vmbus_set_event(struct vmbus_channel *channel);
1653 
1654 /* Get the start of the ring buffer. */
1655 static inline void *
hv_get_ring_buffer(const struct hv_ring_buffer_info * ring_info)1656 hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info)
1657 {
1658 	return ring_info->ring_buffer->buffer;
1659 }
1660 
1661 /*
1662  * Mask off host interrupt callback notifications
1663  */
hv_begin_read(struct hv_ring_buffer_info * rbi)1664 static inline void hv_begin_read(struct hv_ring_buffer_info *rbi)
1665 {
1666 	rbi->ring_buffer->interrupt_mask = 1;
1667 
1668 	/* make sure mask update is not reordered */
1669 	virt_mb();
1670 }
1671 
1672 /*
1673  * Re-enable host callback and return number of outstanding bytes
1674  */
hv_end_read(struct hv_ring_buffer_info * rbi)1675 static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi)
1676 {
1677 
1678 	rbi->ring_buffer->interrupt_mask = 0;
1679 
1680 	/* make sure mask update is not reordered */
1681 	virt_mb();
1682 
1683 	/*
1684 	 * Now check to see if the ring buffer is still empty.
1685 	 * If it is not, we raced and we need to process new
1686 	 * incoming messages.
1687 	 */
1688 	return hv_get_bytes_to_read(rbi);
1689 }
1690 
1691 /*
1692  * An API to support in-place processing of incoming VMBUS packets.
1693  */
1694 
1695 /* Get data payload associated with descriptor */
hv_pkt_data(const struct vmpacket_descriptor * desc)1696 static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc)
1697 {
1698 	return (void *)((unsigned long)desc + (desc->offset8 << 3));
1699 }
1700 
1701 /* Get data size associated with descriptor */
hv_pkt_datalen(const struct vmpacket_descriptor * desc)1702 static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc)
1703 {
1704 	return (desc->len8 << 3) - (desc->offset8 << 3);
1705 }
1706 
1707 /* Get packet length associated with descriptor */
hv_pkt_len(const struct vmpacket_descriptor * desc)1708 static inline u32 hv_pkt_len(const struct vmpacket_descriptor *desc)
1709 {
1710 	return desc->len8 << 3;
1711 }
1712 
1713 struct vmpacket_descriptor *
1714 hv_pkt_iter_first(struct vmbus_channel *channel);
1715 
1716 struct vmpacket_descriptor *
1717 __hv_pkt_iter_next(struct vmbus_channel *channel,
1718 		   const struct vmpacket_descriptor *pkt);
1719 
1720 void hv_pkt_iter_close(struct vmbus_channel *channel);
1721 
1722 static inline struct vmpacket_descriptor *
hv_pkt_iter_next(struct vmbus_channel * channel,const struct vmpacket_descriptor * pkt)1723 hv_pkt_iter_next(struct vmbus_channel *channel,
1724 		 const struct vmpacket_descriptor *pkt)
1725 {
1726 	struct vmpacket_descriptor *nxt;
1727 
1728 	nxt = __hv_pkt_iter_next(channel, pkt);
1729 	if (!nxt)
1730 		hv_pkt_iter_close(channel);
1731 
1732 	return nxt;
1733 }
1734 
1735 #define foreach_vmbus_pkt(pkt, channel) \
1736 	for (pkt = hv_pkt_iter_first(channel); pkt; \
1737 	    pkt = hv_pkt_iter_next(channel, pkt))
1738 
1739 /*
1740  * Interface for passing data between SR-IOV PF and VF drivers. The VF driver
1741  * sends requests to read and write blocks. Each block must be 128 bytes or
1742  * smaller. Optionally, the VF driver can register a callback function which
1743  * will be invoked when the host says that one or more of the first 64 block
1744  * IDs is "invalid" which means that the VF driver should reread them.
1745  */
1746 #define HV_CONFIG_BLOCK_SIZE_MAX 128
1747 
1748 int hyperv_read_cfg_blk(struct pci_dev *dev, void *buf, unsigned int buf_len,
1749 			unsigned int block_id, unsigned int *bytes_returned);
1750 int hyperv_write_cfg_blk(struct pci_dev *dev, void *buf, unsigned int len,
1751 			 unsigned int block_id);
1752 int hyperv_reg_block_invalidate(struct pci_dev *dev, void *context,
1753 				void (*block_invalidate)(void *context,
1754 							 u64 block_mask));
1755 
1756 struct hyperv_pci_block_ops {
1757 	int (*read_block)(struct pci_dev *dev, void *buf, unsigned int buf_len,
1758 			  unsigned int block_id, unsigned int *bytes_returned);
1759 	int (*write_block)(struct pci_dev *dev, void *buf, unsigned int len,
1760 			   unsigned int block_id);
1761 	int (*reg_blk_invalidate)(struct pci_dev *dev, void *context,
1762 				  void (*block_invalidate)(void *context,
1763 							   u64 block_mask));
1764 };
1765 
1766 extern struct hyperv_pci_block_ops hvpci_block_ops;
1767 
virt_to_hvpfn(void * addr)1768 static inline unsigned long virt_to_hvpfn(void *addr)
1769 {
1770 	phys_addr_t paddr;
1771 
1772 	if (is_vmalloc_addr(addr))
1773 		paddr = page_to_phys(vmalloc_to_page(addr)) +
1774 				     offset_in_page(addr);
1775 	else
1776 		paddr = __pa(addr);
1777 
1778 	return  paddr >> HV_HYP_PAGE_SHIFT;
1779 }
1780 
1781 #define NR_HV_HYP_PAGES_IN_PAGE	(PAGE_SIZE / HV_HYP_PAGE_SIZE)
1782 #define offset_in_hvpage(ptr)	((unsigned long)(ptr) & ~HV_HYP_PAGE_MASK)
1783 #define HVPFN_UP(x)	(((x) + HV_HYP_PAGE_SIZE-1) >> HV_HYP_PAGE_SHIFT)
1784 #define HVPFN_DOWN(x)	((x) >> HV_HYP_PAGE_SHIFT)
1785 #define page_to_hvpfn(page)	(page_to_pfn(page) * NR_HV_HYP_PAGES_IN_PAGE)
1786 
1787 #endif /* _HYPERV_H */
1788