xref: /linux/include/linux/hyperv.h (revision 37cb0c76ac6c3bf3d82d25a7c95950f2dac3ca41)
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 
168 /*
169  * If the requested ring buffer size is at least 8 times the size of the
170  * header, steal space from the ring buffer for the header. Otherwise, add
171  * space for the header so that is doesn't take too much of the ring buffer
172  * space.
173  *
174  * The factor of 8 is somewhat arbitrary. The goal is to prevent adding a
175  * relatively small header (4 Kbytes on x86) to a large-ish power-of-2 ring
176  * buffer size (such as 128 Kbytes) and so end up making a nearly twice as
177  * large allocation that will be almost half wasted. As a contrasting example,
178  * on ARM64 with 64 Kbyte page size, we don't want to take 64 Kbytes for the
179  * header from a 128 Kbyte allocation, leaving only 64 Kbytes for the ring.
180  * In this latter case, we must add 64 Kbytes for the header and not worry
181  * about what's wasted.
182  */
183 #define VMBUS_HEADER_ADJ(payload_sz) \
184 	((payload_sz) >=  8 * sizeof(struct hv_ring_buffer) ? \
185 	0 : sizeof(struct hv_ring_buffer))
186 
187 /* Calculate the proper size of a ringbuffer, it must be page-aligned */
188 #define VMBUS_RING_SIZE(payload_sz) PAGE_ALIGN(VMBUS_HEADER_ADJ(payload_sz) + \
189 					       (payload_sz))
190 
191 struct hv_ring_buffer_info {
192 	struct hv_ring_buffer *ring_buffer;
193 	u32 ring_size;			/* Include the shared header */
194 	struct reciprocal_value ring_size_div10_reciprocal;
195 	spinlock_t ring_lock;
196 
197 	u32 ring_datasize;		/* < ring_size */
198 	u32 priv_read_index;
199 	/*
200 	 * The ring buffer mutex lock. This lock prevents the ring buffer from
201 	 * being freed while the ring buffer is being accessed.
202 	 */
203 	struct mutex ring_buffer_mutex;
204 
205 	/* Buffer that holds a copy of an incoming host packet */
206 	void *pkt_buffer;
207 	u32 pkt_buffer_size;
208 };
209 
210 
hv_get_bytes_to_read(const struct hv_ring_buffer_info * rbi)211 static inline u32 hv_get_bytes_to_read(const struct hv_ring_buffer_info *rbi)
212 {
213 	u32 read_loc, write_loc, dsize, read;
214 
215 	dsize = rbi->ring_datasize;
216 	read_loc = rbi->ring_buffer->read_index;
217 	write_loc = READ_ONCE(rbi->ring_buffer->write_index);
218 
219 	read = write_loc >= read_loc ? (write_loc - read_loc) :
220 		(dsize - read_loc) + write_loc;
221 
222 	return read;
223 }
224 
hv_get_bytes_to_write(const struct hv_ring_buffer_info * rbi)225 static inline u32 hv_get_bytes_to_write(const struct hv_ring_buffer_info *rbi)
226 {
227 	u32 read_loc, write_loc, dsize, write;
228 
229 	dsize = rbi->ring_datasize;
230 	read_loc = READ_ONCE(rbi->ring_buffer->read_index);
231 	write_loc = rbi->ring_buffer->write_index;
232 
233 	write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
234 		read_loc - write_loc;
235 	return write;
236 }
237 
hv_get_avail_to_write_percent(const struct hv_ring_buffer_info * rbi)238 static inline u32 hv_get_avail_to_write_percent(
239 		const struct hv_ring_buffer_info *rbi)
240 {
241 	u32 avail_write = hv_get_bytes_to_write(rbi);
242 
243 	return reciprocal_divide(
244 			(avail_write  << 3) + (avail_write << 1),
245 			rbi->ring_size_div10_reciprocal);
246 }
247 
248 /*
249  * VMBUS version is 32 bit entity broken up into
250  * two 16 bit quantities: major_number. minor_number.
251  *
252  * 0 . 13 (Windows Server 2008)
253  * 1 . 1  (Windows 7, WS2008 R2)
254  * 2 . 4  (Windows 8, WS2012)
255  * 3 . 0  (Windows 8.1, WS2012 R2)
256  * 4 . 0  (Windows 10)
257  * 4 . 1  (Windows 10 RS3)
258  * 5 . 0  (Newer Windows 10)
259  * 5 . 1  (Windows 10 RS4)
260  * 5 . 2  (Windows Server 2019, RS5)
261  * 5 . 3  (Windows Server 2022)
262  *
263  * The WS2008 and WIN7 versions are listed here for
264  * completeness but are no longer supported in the
265  * Linux kernel.
266  */
267 
268 #define VERSION_WS2008  ((0 << 16) | (13))
269 #define VERSION_WIN7    ((1 << 16) | (1))
270 #define VERSION_WIN8    ((2 << 16) | (4))
271 #define VERSION_WIN8_1    ((3 << 16) | (0))
272 #define VERSION_WIN10 ((4 << 16) | (0))
273 #define VERSION_WIN10_V4_1 ((4 << 16) | (1))
274 #define VERSION_WIN10_V5 ((5 << 16) | (0))
275 #define VERSION_WIN10_V5_1 ((5 << 16) | (1))
276 #define VERSION_WIN10_V5_2 ((5 << 16) | (2))
277 #define VERSION_WIN10_V5_3 ((5 << 16) | (3))
278 
279 /* Make maximum size of pipe payload of 16K */
280 #define MAX_PIPE_DATA_PAYLOAD		(sizeof(u8) * 16384)
281 
282 /* Define PipeMode values. */
283 #define VMBUS_PIPE_TYPE_BYTE		0x00000000
284 #define VMBUS_PIPE_TYPE_MESSAGE		0x00000004
285 
286 /* The size of the user defined data buffer for non-pipe offers. */
287 #define MAX_USER_DEFINED_BYTES		120
288 
289 /* The size of the user defined data buffer for pipe offers. */
290 #define MAX_PIPE_USER_DEFINED_BYTES	116
291 
292 /*
293  * At the center of the Channel Management library is the Channel Offer. This
294  * struct contains the fundamental information about an offer.
295  */
296 struct vmbus_channel_offer {
297 	guid_t if_type;
298 	guid_t if_instance;
299 
300 	/*
301 	 * These two fields are not currently used.
302 	 */
303 	u64 reserved1;
304 	u64 reserved2;
305 
306 	u16 chn_flags;
307 	u16 mmio_megabytes;		/* in bytes * 1024 * 1024 */
308 
309 	union {
310 		/* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
311 		struct {
312 			unsigned char user_def[MAX_USER_DEFINED_BYTES];
313 		} std;
314 
315 		/*
316 		 * Pipes:
317 		 * The following structure is an integrated pipe protocol, which
318 		 * is implemented on top of standard user-defined data. Pipe
319 		 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
320 		 * use.
321 		 */
322 		struct {
323 			u32  pipe_mode;
324 			unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
325 		} pipe;
326 	} u;
327 	/*
328 	 * The sub_channel_index is defined in Win8: a value of zero means a
329 	 * primary channel and a value of non-zero means a sub-channel.
330 	 *
331 	 * Before Win8, the field is reserved, meaning it's always zero.
332 	 */
333 	u16 sub_channel_index;
334 	u16 reserved3;
335 } __packed;
336 
337 /* Server Flags */
338 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE	1
339 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES	2
340 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS		4
341 #define VMBUS_CHANNEL_NAMED_PIPE_MODE			0x10
342 #define VMBUS_CHANNEL_LOOPBACK_OFFER			0x100
343 #define VMBUS_CHANNEL_PARENT_OFFER			0x200
344 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION	0x400
345 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER		0x2000
346 
347 struct vmpacket_descriptor {
348 	u16 type;
349 	u16 offset8;
350 	u16 len8;
351 	u16 flags;
352 	u64 trans_id;
353 } __packed;
354 
355 struct vmpacket_header {
356 	u32 prev_pkt_start_offset;
357 	struct vmpacket_descriptor descriptor;
358 } __packed;
359 
360 struct vmtransfer_page_range {
361 	u32 byte_count;
362 	u32 byte_offset;
363 } __packed;
364 
365 struct vmtransfer_page_packet_header {
366 	struct vmpacket_descriptor d;
367 	u16 xfer_pageset_id;
368 	u8  sender_owns_set;
369 	u8 reserved;
370 	u32 range_cnt;
371 	struct vmtransfer_page_range ranges[];
372 } __packed;
373 
374 struct vmgpadl_packet_header {
375 	struct vmpacket_descriptor d;
376 	u32 gpadl;
377 	u32 reserved;
378 } __packed;
379 
380 struct vmadd_remove_transfer_page_set {
381 	struct vmpacket_descriptor d;
382 	u32 gpadl;
383 	u16 xfer_pageset_id;
384 	u16 reserved;
385 } __packed;
386 
387 /*
388  * This structure defines a range in guest physical space that can be made to
389  * look virtually contiguous.
390  */
391 struct gpa_range {
392 	u32 byte_count;
393 	u32 byte_offset;
394 	u64 pfn_array[];
395 };
396 
397 /*
398  * This is the format for an Establish Gpadl packet, which contains a handle by
399  * which this GPADL will be known and a set of GPA ranges associated with it.
400  * This can be converted to a MDL by the guest OS.  If there are multiple GPA
401  * ranges, then the resulting MDL will be "chained," representing multiple VA
402  * ranges.
403  */
404 struct vmestablish_gpadl {
405 	struct vmpacket_descriptor d;
406 	u32 gpadl;
407 	u32 range_cnt;
408 	struct gpa_range range[1];
409 } __packed;
410 
411 /*
412  * This is the format for a Teardown Gpadl packet, which indicates that the
413  * GPADL handle in the Establish Gpadl packet will never be referenced again.
414  */
415 struct vmteardown_gpadl {
416 	struct vmpacket_descriptor d;
417 	u32 gpadl;
418 	u32 reserved;	/* for alignment to a 8-byte boundary */
419 } __packed;
420 
421 /*
422  * This is the format for a GPA-Direct packet, which contains a set of GPA
423  * ranges, in addition to commands and/or data.
424  */
425 struct vmdata_gpa_direct {
426 	struct vmpacket_descriptor d;
427 	u32 reserved;
428 	u32 range_cnt;
429 	struct gpa_range range[1];
430 } __packed;
431 
432 /* This is the format for a Additional Data Packet. */
433 struct vmadditional_data {
434 	struct vmpacket_descriptor d;
435 	u64 total_bytes;
436 	u32 offset;
437 	u32 byte_cnt;
438 	unsigned char data[1];
439 } __packed;
440 
441 union vmpacket_largest_possible_header {
442 	struct vmpacket_descriptor simple_hdr;
443 	struct vmtransfer_page_packet_header xfer_page_hdr;
444 	struct vmgpadl_packet_header gpadl_hdr;
445 	struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr;
446 	struct vmestablish_gpadl establish_gpadl_hdr;
447 	struct vmteardown_gpadl teardown_gpadl_hdr;
448 	struct vmdata_gpa_direct data_gpa_direct_hdr;
449 };
450 
451 #define VMPACKET_DATA_START_ADDRESS(__packet)	\
452 	(void *)(((unsigned char *)__packet) +	\
453 	 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
454 
455 #define VMPACKET_DATA_LENGTH(__packet)		\
456 	((((struct vmpacket_descriptor)__packet)->len8 -	\
457 	  ((struct vmpacket_descriptor)__packet)->offset8) * 8)
458 
459 #define VMPACKET_TRANSFER_MODE(__packet)	\
460 	(((struct IMPACT)__packet)->type)
461 
462 enum vmbus_packet_type {
463 	VM_PKT_INVALID				= 0x0,
464 	VM_PKT_SYNCH				= 0x1,
465 	VM_PKT_ADD_XFER_PAGESET			= 0x2,
466 	VM_PKT_RM_XFER_PAGESET			= 0x3,
467 	VM_PKT_ESTABLISH_GPADL			= 0x4,
468 	VM_PKT_TEARDOWN_GPADL			= 0x5,
469 	VM_PKT_DATA_INBAND			= 0x6,
470 	VM_PKT_DATA_USING_XFER_PAGES		= 0x7,
471 	VM_PKT_DATA_USING_GPADL			= 0x8,
472 	VM_PKT_DATA_USING_GPA_DIRECT		= 0x9,
473 	VM_PKT_CANCEL_REQUEST			= 0xa,
474 	VM_PKT_COMP				= 0xb,
475 	VM_PKT_DATA_USING_ADDITIONAL_PKT	= 0xc,
476 	VM_PKT_ADDITIONAL_DATA			= 0xd
477 };
478 
479 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED	1
480 
481 
482 /* Version 1 messages */
483 enum vmbus_channel_message_type {
484 	CHANNELMSG_INVALID			=  0,
485 	CHANNELMSG_OFFERCHANNEL		=  1,
486 	CHANNELMSG_RESCIND_CHANNELOFFER	=  2,
487 	CHANNELMSG_REQUESTOFFERS		=  3,
488 	CHANNELMSG_ALLOFFERS_DELIVERED	=  4,
489 	CHANNELMSG_OPENCHANNEL		=  5,
490 	CHANNELMSG_OPENCHANNEL_RESULT		=  6,
491 	CHANNELMSG_CLOSECHANNEL		=  7,
492 	CHANNELMSG_GPADL_HEADER		=  8,
493 	CHANNELMSG_GPADL_BODY			=  9,
494 	CHANNELMSG_GPADL_CREATED		= 10,
495 	CHANNELMSG_GPADL_TEARDOWN		= 11,
496 	CHANNELMSG_GPADL_TORNDOWN		= 12,
497 	CHANNELMSG_RELID_RELEASED		= 13,
498 	CHANNELMSG_INITIATE_CONTACT		= 14,
499 	CHANNELMSG_VERSION_RESPONSE		= 15,
500 	CHANNELMSG_UNLOAD			= 16,
501 	CHANNELMSG_UNLOAD_RESPONSE		= 17,
502 	CHANNELMSG_18				= 18,
503 	CHANNELMSG_19				= 19,
504 	CHANNELMSG_20				= 20,
505 	CHANNELMSG_TL_CONNECT_REQUEST		= 21,
506 	CHANNELMSG_MODIFYCHANNEL		= 22,
507 	CHANNELMSG_TL_CONNECT_RESULT		= 23,
508 	CHANNELMSG_MODIFYCHANNEL_RESPONSE	= 24,
509 	CHANNELMSG_COUNT
510 };
511 
512 /* Hyper-V supports about 2048 channels, and the RELIDs start with 1. */
513 #define INVALID_RELID	U32_MAX
514 
515 struct vmbus_channel_message_header {
516 	enum vmbus_channel_message_type msgtype;
517 	u32 padding;
518 } __packed;
519 
520 /* Query VMBus Version parameters */
521 struct vmbus_channel_query_vmbus_version {
522 	struct vmbus_channel_message_header header;
523 	u32 version;
524 } __packed;
525 
526 /* VMBus Version Supported parameters */
527 struct vmbus_channel_version_supported {
528 	struct vmbus_channel_message_header header;
529 	u8 version_supported;
530 } __packed;
531 
532 /* Offer Channel parameters */
533 struct vmbus_channel_offer_channel {
534 	struct vmbus_channel_message_header header;
535 	struct vmbus_channel_offer offer;
536 	u32 child_relid;
537 	u8 monitorid;
538 	/*
539 	 * win7 and beyond splits this field into a bit field.
540 	 */
541 	u8 monitor_allocated:1;
542 	u8 reserved:7;
543 	/*
544 	 * These are new fields added in win7 and later.
545 	 * Do not access these fields without checking the
546 	 * negotiated protocol.
547 	 *
548 	 * If "is_dedicated_interrupt" is set, we must not set the
549 	 * associated bit in the channel bitmap while sending the
550 	 * interrupt to the host.
551 	 *
552 	 * connection_id is to be used in signaling the host.
553 	 */
554 	u16 is_dedicated_interrupt:1;
555 	u16 reserved1:15;
556 	u32 connection_id;
557 } __packed;
558 
559 /* Rescind Offer parameters */
560 struct vmbus_channel_rescind_offer {
561 	struct vmbus_channel_message_header header;
562 	u32 child_relid;
563 } __packed;
564 
565 /*
566  * Request Offer -- no parameters, SynIC message contains the partition ID
567  * Set Snoop -- no parameters, SynIC message contains the partition ID
568  * Clear Snoop -- no parameters, SynIC message contains the partition ID
569  * All Offers Delivered -- no parameters, SynIC message contains the partition
570  *		           ID
571  * Flush Client -- no parameters, SynIC message contains the partition ID
572  */
573 
574 /* Open Channel parameters */
575 struct vmbus_channel_open_channel {
576 	struct vmbus_channel_message_header header;
577 
578 	/* Identifies the specific VMBus channel that is being opened. */
579 	u32 child_relid;
580 
581 	/* ID making a particular open request at a channel offer unique. */
582 	u32 openid;
583 
584 	/* GPADL for the channel's ring buffer. */
585 	u32 ringbuffer_gpadlhandle;
586 
587 	/*
588 	 * Starting with win8, this field will be used to specify
589 	 * the target virtual processor on which to deliver the interrupt for
590 	 * the host to guest communication.
591 	 * Prior to win8, incoming channel interrupts would only
592 	 * be delivered on cpu 0. Setting this value to 0 would
593 	 * preserve the earlier behavior.
594 	 */
595 	u32 target_vp;
596 
597 	/*
598 	 * The upstream ring buffer begins at offset zero in the memory
599 	 * described by RingBufferGpadlHandle. The downstream ring buffer
600 	 * follows it at this offset (in pages).
601 	 */
602 	u32 downstream_ringbuffer_pageoffset;
603 
604 	/* User-specific data to be passed along to the server endpoint. */
605 	unsigned char userdata[MAX_USER_DEFINED_BYTES];
606 } __packed;
607 
608 /* Open Channel Result parameters */
609 struct vmbus_channel_open_result {
610 	struct vmbus_channel_message_header header;
611 	u32 child_relid;
612 	u32 openid;
613 	u32 status;
614 } __packed;
615 
616 /* Modify Channel Result parameters */
617 struct vmbus_channel_modifychannel_response {
618 	struct vmbus_channel_message_header header;
619 	u32 child_relid;
620 	u32 status;
621 } __packed;
622 
623 /* Close channel parameters; */
624 struct vmbus_channel_close_channel {
625 	struct vmbus_channel_message_header header;
626 	u32 child_relid;
627 } __packed;
628 
629 /* Channel Message GPADL */
630 #define GPADL_TYPE_RING_BUFFER		1
631 #define GPADL_TYPE_SERVER_SAVE_AREA	2
632 #define GPADL_TYPE_TRANSACTION		8
633 
634 /*
635  * The number of PFNs in a GPADL message is defined by the number of
636  * pages that would be spanned by ByteCount and ByteOffset.  If the
637  * implied number of PFNs won't fit in this packet, there will be a
638  * follow-up packet that contains more.
639  */
640 struct vmbus_channel_gpadl_header {
641 	struct vmbus_channel_message_header header;
642 	u32 child_relid;
643 	u32 gpadl;
644 	u16 range_buflen;
645 	u16 rangecount;
646 	struct gpa_range range[];
647 } __packed;
648 
649 /* This is the followup packet that contains more PFNs. */
650 struct vmbus_channel_gpadl_body {
651 	struct vmbus_channel_message_header header;
652 	u32 msgnumber;
653 	u32 gpadl;
654 	u64 pfn[];
655 } __packed;
656 
657 struct vmbus_channel_gpadl_created {
658 	struct vmbus_channel_message_header header;
659 	u32 child_relid;
660 	u32 gpadl;
661 	u32 creation_status;
662 } __packed;
663 
664 struct vmbus_channel_gpadl_teardown {
665 	struct vmbus_channel_message_header header;
666 	u32 child_relid;
667 	u32 gpadl;
668 } __packed;
669 
670 struct vmbus_channel_gpadl_torndown {
671 	struct vmbus_channel_message_header header;
672 	u32 gpadl;
673 } __packed;
674 
675 struct vmbus_channel_relid_released {
676 	struct vmbus_channel_message_header header;
677 	u32 child_relid;
678 } __packed;
679 
680 struct vmbus_channel_initiate_contact {
681 	struct vmbus_channel_message_header header;
682 	u32 vmbus_version_requested;
683 	u32 target_vcpu; /* The VCPU the host should respond to */
684 	union {
685 		u64 interrupt_page;
686 		struct {
687 			u8	msg_sint;
688 			u8	msg_vtl;
689 			u8	reserved[6];
690 		};
691 	};
692 	u64 monitor_page1;
693 	u64 monitor_page2;
694 } __packed;
695 
696 /* Hyper-V socket: guest's connect()-ing to host */
697 struct vmbus_channel_tl_connect_request {
698 	struct vmbus_channel_message_header header;
699 	guid_t guest_endpoint_id;
700 	guid_t host_service_id;
701 } __packed;
702 
703 /* Modify Channel parameters, cf. vmbus_send_modifychannel() */
704 struct vmbus_channel_modifychannel {
705 	struct vmbus_channel_message_header header;
706 	u32 child_relid;
707 	u32 target_vp;
708 } __packed;
709 
710 struct vmbus_channel_version_response {
711 	struct vmbus_channel_message_header header;
712 	u8 version_supported;
713 
714 	u8 connection_state;
715 	u16 padding;
716 
717 	/*
718 	 * On new hosts that support VMBus protocol 5.0, we must use
719 	 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
720 	 * and for subsequent messages, we must use the Message Connection ID
721 	 * field in the host-returned Version Response Message.
722 	 *
723 	 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
724 	 */
725 	u32 msg_conn_id;
726 } __packed;
727 
728 enum vmbus_channel_state {
729 	CHANNEL_OFFER_STATE,
730 	CHANNEL_OPENING_STATE,
731 	CHANNEL_OPEN_STATE,
732 	CHANNEL_OPENED_STATE,
733 };
734 
735 /*
736  * Represents each channel msg on the vmbus connection This is a
737  * variable-size data structure depending on the msg type itself
738  */
739 struct vmbus_channel_msginfo {
740 	/* Bookkeeping stuff */
741 	struct list_head msglistentry;
742 
743 	/* So far, this is only used to handle gpadl body message */
744 	struct list_head submsglist;
745 
746 	/* Synchronize the request/response if needed */
747 	struct completion  waitevent;
748 	struct vmbus_channel *waiting_channel;
749 	union {
750 		struct vmbus_channel_version_supported version_supported;
751 		struct vmbus_channel_open_result open_result;
752 		struct vmbus_channel_gpadl_torndown gpadl_torndown;
753 		struct vmbus_channel_gpadl_created gpadl_created;
754 		struct vmbus_channel_version_response version_response;
755 		struct vmbus_channel_modifychannel_response modify_response;
756 	} response;
757 
758 	u32 msgsize;
759 	/*
760 	 * The channel message that goes out on the "wire".
761 	 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
762 	 */
763 	unsigned char msg[];
764 };
765 
766 struct vmbus_close_msg {
767 	struct vmbus_channel_msginfo info;
768 	struct vmbus_channel_close_channel msg;
769 };
770 
771 /* Define connection identifier type. */
772 union hv_connection_id {
773 	u32 asu32;
774 	struct {
775 		u32 id:24;
776 		u32 reserved:8;
777 	} u;
778 };
779 
780 enum vmbus_device_type {
781 	HV_IDE = 0,
782 	HV_SCSI,
783 	HV_FC,
784 	HV_NIC,
785 	HV_ND,
786 	HV_PCIE,
787 	HV_FB,
788 	HV_KBD,
789 	HV_MOUSE,
790 	HV_KVP,
791 	HV_TS,
792 	HV_HB,
793 	HV_SHUTDOWN,
794 	HV_FCOPY,
795 	HV_BACKUP,
796 	HV_DM,
797 	HV_UNKNOWN,
798 };
799 
800 /*
801  * Provides request ids for VMBus. Encapsulates guest memory
802  * addresses and stores the next available slot in req_arr
803  * to generate new ids in constant time.
804  */
805 struct vmbus_requestor {
806 	u64 *req_arr;
807 	unsigned long *req_bitmap; /* is a given slot available? */
808 	u32 size;
809 	u64 next_request_id;
810 	spinlock_t req_lock; /* provides atomicity */
811 };
812 
813 #define VMBUS_NO_RQSTOR U64_MAX
814 #define VMBUS_RQST_ERROR (U64_MAX - 1)
815 #define VMBUS_RQST_ADDR_ANY U64_MAX
816 /* NetVSC-specific */
817 #define VMBUS_RQST_ID_NO_RESPONSE (U64_MAX - 2)
818 /* StorVSC-specific */
819 #define VMBUS_RQST_INIT (U64_MAX - 2)
820 #define VMBUS_RQST_RESET (U64_MAX - 3)
821 
822 struct vmbus_device {
823 	/* preferred ring buffer size in KB, 0 means no preferred size for this device */
824 	size_t pref_ring_size;
825 	u16  dev_type;
826 	guid_t guid;
827 	bool perf_device;
828 	bool allowed_in_isolated;
829 };
830 
831 #define VMBUS_DEFAULT_MAX_PKT_SIZE 4096
832 
833 struct vmbus_gpadl {
834 	u32 gpadl_handle;
835 	u32 size;
836 	void *buffer;
837 	bool decrypted;
838 };
839 
840 struct vmbus_channel {
841 	struct list_head listentry;
842 
843 	struct hv_device *device_obj;
844 
845 	enum vmbus_channel_state state;
846 
847 	struct vmbus_channel_offer_channel offermsg;
848 	/*
849 	 * These are based on the OfferMsg.MonitorId.
850 	 * Save it here for easy access.
851 	 */
852 	u8 monitor_grp;
853 	u8 monitor_bit;
854 
855 	bool rescind; /* got rescind msg */
856 	bool rescind_ref; /* got rescind msg, got channel reference */
857 	struct completion rescind_event;
858 
859 	struct vmbus_gpadl ringbuffer_gpadlhandle;
860 
861 	/* Allocated memory for ring buffer */
862 	struct page *ringbuffer_page;
863 	u32 ringbuffer_pagecount;
864 	u32 ringbuffer_send_offset;
865 	struct hv_ring_buffer_info outbound;	/* send to parent */
866 	struct hv_ring_buffer_info inbound;	/* receive from parent */
867 
868 	struct vmbus_close_msg close_msg;
869 
870 	/* Statistics */
871 	u64	interrupts;	/* Host to Guest interrupts */
872 	u64	sig_events;	/* Guest to Host events */
873 
874 	/*
875 	 * Guest to host interrupts caused by the outbound ring buffer changing
876 	 * from empty to not empty.
877 	 */
878 	u64 intr_out_empty;
879 
880 	/*
881 	 * Indicates that a full outbound ring buffer was encountered. The flag
882 	 * is set to true when a full outbound ring buffer is encountered and
883 	 * set to false when a write to the outbound ring buffer is completed.
884 	 */
885 	bool out_full_flag;
886 
887 	/* Channel callback's invoked in softirq context */
888 	struct tasklet_struct callback_event;
889 	void (*onchannel_callback)(void *context);
890 	void *channel_callback_context;
891 
892 	void (*change_target_cpu_callback)(struct vmbus_channel *channel,
893 			u32 old, u32 new);
894 
895 	/*
896 	 * Synchronize channel scheduling and channel removal; see the inline
897 	 * comments in vmbus_chan_sched() and vmbus_reset_channel_cb().
898 	 */
899 	spinlock_t sched_lock;
900 
901 	/*
902 	 * A channel can be marked for one of three modes of reading:
903 	 *   BATCHED - callback called from taslket and should read
904 	 *            channel until empty. Interrupts from the host
905 	 *            are masked while read is in process (default).
906 	 *   DIRECT - callback called from tasklet (softirq).
907 	 *   ISR - callback called in interrupt context and must
908 	 *         invoke its own deferred processing.
909 	 *         Host interrupts are disabled and must be re-enabled
910 	 *         when ring is empty.
911 	 */
912 	enum hv_callback_mode {
913 		HV_CALL_BATCHED,
914 		HV_CALL_DIRECT,
915 		HV_CALL_ISR
916 	} callback_mode;
917 
918 	bool is_dedicated_interrupt;
919 	u64 sig_event;
920 
921 	/*
922 	 * Starting with win8, this field will be used to specify the
923 	 * target CPU on which to deliver the interrupt for the host
924 	 * to guest communication.
925 	 *
926 	 * Prior to win8, incoming channel interrupts would only be
927 	 * delivered on CPU 0. Setting this value to 0 would preserve
928 	 * the earlier behavior.
929 	 */
930 	u32 target_cpu;
931 	/*
932 	 * Support for sub-channels. For high performance devices,
933 	 * it will be useful to have multiple sub-channels to support
934 	 * a scalable communication infrastructure with the host.
935 	 * The support for sub-channels is implemented as an extension
936 	 * to the current infrastructure.
937 	 * The initial offer is considered the primary channel and this
938 	 * offer message will indicate if the host supports sub-channels.
939 	 * The guest is free to ask for sub-channels to be offered and can
940 	 * open these sub-channels as a normal "primary" channel. However,
941 	 * all sub-channels will have the same type and instance guids as the
942 	 * primary channel. Requests sent on a given channel will result in a
943 	 * response on the same channel.
944 	 */
945 
946 	/*
947 	 * Sub-channel creation callback. This callback will be called in
948 	 * process context when a sub-channel offer is received from the host.
949 	 * The guest can open the sub-channel in the context of this callback.
950 	 */
951 	void (*sc_creation_callback)(struct vmbus_channel *new_sc);
952 
953 	/*
954 	 * Channel rescind callback. Some channels (the hvsock ones), need to
955 	 * register a callback which is invoked in vmbus_onoffer_rescind().
956 	 */
957 	void (*chn_rescind_callback)(struct vmbus_channel *channel);
958 
959 	/*
960 	 * All Sub-channels of a primary channel are linked here.
961 	 */
962 	struct list_head sc_list;
963 	/*
964 	 * The primary channel this sub-channel belongs to.
965 	 * This will be NULL for the primary channel.
966 	 */
967 	struct vmbus_channel *primary_channel;
968 	/*
969 	 * Support per-channel state for use by vmbus drivers.
970 	 */
971 	void *per_channel_state;
972 
973 	/*
974 	 * Defer freeing channel until after all cpu's have
975 	 * gone through grace period.
976 	 */
977 	struct rcu_head rcu;
978 
979 	/*
980 	 * For sysfs per-channel properties.
981 	 */
982 	struct kobject			kobj;
983 
984 	/*
985 	 * For performance critical channels (storage, networking
986 	 * etc,), Hyper-V has a mechanism to enhance the throughput
987 	 * at the expense of latency:
988 	 * When the host is to be signaled, we just set a bit in a shared page
989 	 * and this bit will be inspected by the hypervisor within a certain
990 	 * window and if the bit is set, the host will be signaled. The window
991 	 * of time is the monitor latency - currently around 100 usecs. This
992 	 * mechanism improves throughput by:
993 	 *
994 	 * A) Making the host more efficient - each time it wakes up,
995 	 *    potentially it will process more number of packets. The
996 	 *    monitor latency allows a batch to build up.
997 	 * B) By deferring the hypercall to signal, we will also minimize
998 	 *    the interrupts.
999 	 *
1000 	 * Clearly, these optimizations improve throughput at the expense of
1001 	 * latency. Furthermore, since the channel is shared for both
1002 	 * control and data messages, control messages currently suffer
1003 	 * unnecessary latency adversely impacting performance and boot
1004 	 * time. To fix this issue, permit tagging the channel as being
1005 	 * in "low latency" mode. In this mode, we will bypass the monitor
1006 	 * mechanism.
1007 	 */
1008 	bool low_latency;
1009 
1010 	bool probe_done;
1011 
1012 	/*
1013 	 * Cache the device ID here for easy access; this is useful, in
1014 	 * particular, in situations where the channel's device_obj has
1015 	 * not been allocated/initialized yet.
1016 	 */
1017 	u16 device_id;
1018 
1019 	/*
1020 	 * We must offload the handling of the primary/sub channels
1021 	 * from the single-threaded vmbus_connection.work_queue to
1022 	 * two different workqueue, otherwise we can block
1023 	 * vmbus_connection.work_queue and hang: see vmbus_process_offer().
1024 	 */
1025 	struct work_struct add_channel_work;
1026 
1027 	/*
1028 	 * Guest to host interrupts caused by the inbound ring buffer changing
1029 	 * from full to not full while a packet is waiting.
1030 	 */
1031 	u64 intr_in_full;
1032 
1033 	/*
1034 	 * The total number of write operations that encountered a full
1035 	 * outbound ring buffer.
1036 	 */
1037 	u64 out_full_total;
1038 
1039 	/*
1040 	 * The number of write operations that were the first to encounter a
1041 	 * full outbound ring buffer.
1042 	 */
1043 	u64 out_full_first;
1044 
1045 	/* enabling/disabling fuzz testing on the channel (default is false)*/
1046 	bool fuzz_testing_state;
1047 
1048 	/*
1049 	 * Interrupt delay will delay the guest from emptying the ring buffer
1050 	 * for a specific amount of time. The delay is in microseconds and will
1051 	 * be between 1 to a maximum of 1000, its default is 0 (no delay).
1052 	 * The  Message delay will delay guest reading on a per message basis
1053 	 * in microseconds between 1 to 1000 with the default being 0
1054 	 * (no delay).
1055 	 */
1056 	u32 fuzz_testing_interrupt_delay;
1057 	u32 fuzz_testing_message_delay;
1058 
1059 	/* callback to generate a request ID from a request address */
1060 	u64 (*next_request_id_callback)(struct vmbus_channel *channel, u64 rqst_addr);
1061 	/* callback to retrieve a request address from a request ID */
1062 	u64 (*request_addr_callback)(struct vmbus_channel *channel, u64 rqst_id);
1063 
1064 	/* request/transaction ids for VMBus */
1065 	struct vmbus_requestor requestor;
1066 	u32 rqstor_size;
1067 
1068 	/* The max size of a packet on this channel */
1069 	u32 max_pkt_size;
1070 };
1071 
1072 #define lock_requestor(channel, flags)					\
1073 do {									\
1074 	struct vmbus_requestor *rqstor = &(channel)->requestor;		\
1075 									\
1076 	spin_lock_irqsave(&rqstor->req_lock, flags);			\
1077 } while (0)
1078 
unlock_requestor(struct vmbus_channel * channel,unsigned long flags)1079 static __always_inline void unlock_requestor(struct vmbus_channel *channel,
1080 					     unsigned long flags)
1081 {
1082 	struct vmbus_requestor *rqstor = &channel->requestor;
1083 
1084 	spin_unlock_irqrestore(&rqstor->req_lock, flags);
1085 }
1086 
1087 u64 vmbus_next_request_id(struct vmbus_channel *channel, u64 rqst_addr);
1088 u64 __vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1089 			       u64 rqst_addr);
1090 u64 vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1091 			     u64 rqst_addr);
1092 u64 vmbus_request_addr(struct vmbus_channel *channel, u64 trans_id);
1093 
is_hvsock_offer(const struct vmbus_channel_offer_channel * o)1094 static inline bool is_hvsock_offer(const struct vmbus_channel_offer_channel *o)
1095 {
1096 	return !!(o->offer.chn_flags & VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER);
1097 }
1098 
is_hvsock_channel(const struct vmbus_channel * c)1099 static inline bool is_hvsock_channel(const struct vmbus_channel *c)
1100 {
1101 	return is_hvsock_offer(&c->offermsg);
1102 }
1103 
is_sub_channel(const struct vmbus_channel * c)1104 static inline bool is_sub_channel(const struct vmbus_channel *c)
1105 {
1106 	return c->offermsg.offer.sub_channel_index != 0;
1107 }
1108 
set_channel_read_mode(struct vmbus_channel * c,enum hv_callback_mode mode)1109 static inline void set_channel_read_mode(struct vmbus_channel *c,
1110 					enum hv_callback_mode mode)
1111 {
1112 	c->callback_mode = mode;
1113 }
1114 
set_per_channel_state(struct vmbus_channel * c,void * s)1115 static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
1116 {
1117 	c->per_channel_state = s;
1118 }
1119 
get_per_channel_state(struct vmbus_channel * c)1120 static inline void *get_per_channel_state(struct vmbus_channel *c)
1121 {
1122 	return c->per_channel_state;
1123 }
1124 
set_channel_pending_send_size(struct vmbus_channel * c,u32 size)1125 static inline void set_channel_pending_send_size(struct vmbus_channel *c,
1126 						 u32 size)
1127 {
1128 	unsigned long flags;
1129 
1130 	if (size) {
1131 		spin_lock_irqsave(&c->outbound.ring_lock, flags);
1132 		++c->out_full_total;
1133 
1134 		if (!c->out_full_flag) {
1135 			++c->out_full_first;
1136 			c->out_full_flag = true;
1137 		}
1138 		spin_unlock_irqrestore(&c->outbound.ring_lock, flags);
1139 	} else {
1140 		c->out_full_flag = false;
1141 	}
1142 
1143 	c->outbound.ring_buffer->pending_send_sz = size;
1144 }
1145 
1146 void vmbus_onmessage(struct vmbus_channel_message_header *hdr);
1147 
1148 int vmbus_request_offers(void);
1149 
1150 /*
1151  * APIs for managing sub-channels.
1152  */
1153 
1154 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1155 			void (*sc_cr_cb)(struct vmbus_channel *new_sc));
1156 
1157 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1158 		void (*chn_rescind_cb)(struct vmbus_channel *));
1159 
1160 /* The format must be the same as struct vmdata_gpa_direct */
1161 struct vmbus_channel_packet_page_buffer {
1162 	u16 type;
1163 	u16 dataoffset8;
1164 	u16 length8;
1165 	u16 flags;
1166 	u64 transactionid;
1167 	u32 reserved;
1168 	u32 rangecount;
1169 	struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
1170 } __packed;
1171 
1172 /* The format must be the same as struct vmdata_gpa_direct */
1173 struct vmbus_channel_packet_multipage_buffer {
1174 	u16 type;
1175 	u16 dataoffset8;
1176 	u16 length8;
1177 	u16 flags;
1178 	u64 transactionid;
1179 	u32 reserved;
1180 	u32 rangecount;		/* Always 1 in this case */
1181 	struct hv_multipage_buffer range;
1182 } __packed;
1183 
1184 /* The format must be the same as struct vmdata_gpa_direct */
1185 struct vmbus_packet_mpb_array {
1186 	u16 type;
1187 	u16 dataoffset8;
1188 	u16 length8;
1189 	u16 flags;
1190 	u64 transactionid;
1191 	u32 reserved;
1192 	u32 rangecount;         /* Always 1 in this case */
1193 	struct hv_mpb_array range;
1194 } __packed;
1195 
1196 int vmbus_alloc_ring(struct vmbus_channel *channel,
1197 		     u32 send_size, u32 recv_size);
1198 void vmbus_free_ring(struct vmbus_channel *channel);
1199 
1200 int vmbus_connect_ring(struct vmbus_channel *channel,
1201 		       void (*onchannel_callback)(void *context),
1202 		       void *context);
1203 int vmbus_disconnect_ring(struct vmbus_channel *channel);
1204 
1205 extern int vmbus_open(struct vmbus_channel *channel,
1206 			    u32 send_ringbuffersize,
1207 			    u32 recv_ringbuffersize,
1208 			    void *userdata,
1209 			    u32 userdatalen,
1210 			    void (*onchannel_callback)(void *context),
1211 			    void *context);
1212 
1213 extern void vmbus_close(struct vmbus_channel *channel);
1214 
1215 extern int vmbus_sendpacket_getid(struct vmbus_channel *channel,
1216 				  void *buffer,
1217 				  u32 bufferLen,
1218 				  u64 requestid,
1219 				  u64 *trans_id,
1220 				  enum vmbus_packet_type type,
1221 				  u32 flags);
1222 extern int vmbus_sendpacket(struct vmbus_channel *channel,
1223 				  void *buffer,
1224 				  u32 bufferLen,
1225 				  u64 requestid,
1226 				  enum vmbus_packet_type type,
1227 				  u32 flags);
1228 
1229 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1230 					    struct hv_page_buffer pagebuffers[],
1231 					    u32 pagecount,
1232 					    void *buffer,
1233 					    u32 bufferlen,
1234 					    u64 requestid);
1235 
1236 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1237 				     struct vmbus_packet_mpb_array *mpb,
1238 				     u32 desc_size,
1239 				     void *buffer,
1240 				     u32 bufferlen,
1241 				     u64 requestid);
1242 
1243 extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
1244 				      void *kbuffer,
1245 				      u32 size,
1246 				      struct vmbus_gpadl *gpadl);
1247 
1248 extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
1249 				     struct vmbus_gpadl *gpadl);
1250 
1251 void vmbus_reset_channel_cb(struct vmbus_channel *channel);
1252 
1253 extern int vmbus_recvpacket(struct vmbus_channel *channel,
1254 				  void *buffer,
1255 				  u32 bufferlen,
1256 				  u32 *buffer_actual_len,
1257 				  u64 *requestid);
1258 
1259 extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
1260 				     void *buffer,
1261 				     u32 bufferlen,
1262 				     u32 *buffer_actual_len,
1263 				     u64 *requestid);
1264 
1265 /* Base driver object */
1266 struct hv_driver {
1267 	const char *name;
1268 
1269 	/*
1270 	 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1271 	 * channel flag, actually doesn't mean a synthetic device because the
1272 	 * offer's if_type/if_instance can change for every new hvsock
1273 	 * connection.
1274 	 *
1275 	 * However, to facilitate the notification of new-offer/rescind-offer
1276 	 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1277 	 * a special vmbus device, and hence we need the below flag to
1278 	 * indicate if the driver is the hvsock driver or not: we need to
1279 	 * specially treat the hvosck offer & driver in vmbus_match().
1280 	 */
1281 	bool hvsock;
1282 
1283 	/* the device type supported by this driver */
1284 	guid_t dev_type;
1285 	const struct hv_vmbus_device_id *id_table;
1286 
1287 	struct device_driver driver;
1288 
1289 	/* dynamic device GUID's */
1290 	struct  {
1291 		spinlock_t lock;
1292 		struct list_head list;
1293 	} dynids;
1294 
1295 	int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
1296 	void (*remove)(struct hv_device *dev);
1297 	void (*shutdown)(struct hv_device *);
1298 
1299 	int (*suspend)(struct hv_device *);
1300 	int (*resume)(struct hv_device *);
1301 
1302 };
1303 
1304 /* Base device object */
1305 struct hv_device {
1306 	/* the device type id of this device */
1307 	guid_t dev_type;
1308 
1309 	/* the device instance id of this device */
1310 	guid_t dev_instance;
1311 	u16 vendor_id;
1312 	u16 device_id;
1313 
1314 	struct device device;
1315 	/*
1316 	 * Driver name to force a match.  Do not set directly, because core
1317 	 * frees it.  Use driver_set_override() to set or clear it.
1318 	 */
1319 	const char *driver_override;
1320 
1321 	struct vmbus_channel *channel;
1322 	struct kset	     *channels_kset;
1323 	struct device_dma_parameters dma_parms;
1324 	u64 dma_mask;
1325 
1326 	/* place holder to keep track of the dir for hv device in debugfs */
1327 	struct dentry *debug_dir;
1328 
1329 };
1330 
1331 
1332 #define device_to_hv_device(d)	container_of_const(d, struct hv_device, device)
1333 #define drv_to_hv_drv(d)	container_of_const(d, struct hv_driver, driver)
1334 
hv_set_drvdata(struct hv_device * dev,void * data)1335 static inline void hv_set_drvdata(struct hv_device *dev, void *data)
1336 {
1337 	dev_set_drvdata(&dev->device, data);
1338 }
1339 
hv_get_drvdata(struct hv_device * dev)1340 static inline void *hv_get_drvdata(struct hv_device *dev)
1341 {
1342 	return dev_get_drvdata(&dev->device);
1343 }
1344 
1345 struct hv_ring_buffer_debug_info {
1346 	u32 current_interrupt_mask;
1347 	u32 current_read_index;
1348 	u32 current_write_index;
1349 	u32 bytes_avail_toread;
1350 	u32 bytes_avail_towrite;
1351 };
1352 
1353 
1354 int hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
1355 				struct hv_ring_buffer_debug_info *debug_info);
1356 
1357 bool hv_ringbuffer_spinlock_busy(struct vmbus_channel *channel);
1358 
1359 /* Vmbus interface */
1360 #define vmbus_driver_register(driver)	\
1361 	__vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1362 int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
1363 					 struct module *owner,
1364 					 const char *mod_name);
1365 void vmbus_driver_unregister(struct hv_driver *hv_driver);
1366 
1367 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel);
1368 
1369 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1370 			resource_size_t min, resource_size_t max,
1371 			resource_size_t size, resource_size_t align,
1372 			bool fb_overlap_ok);
1373 void vmbus_free_mmio(resource_size_t start, resource_size_t size);
1374 
1375 /*
1376  * GUID definitions of various offer types - services offered to the guest.
1377  */
1378 
1379 /*
1380  * Network GUID
1381  * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1382  */
1383 #define HV_NIC_GUID \
1384 	.guid = GUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1385 			  0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1386 
1387 /*
1388  * IDE GUID
1389  * {32412632-86cb-44a2-9b5c-50d1417354f5}
1390  */
1391 #define HV_IDE_GUID \
1392 	.guid = GUID_INIT(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1393 			  0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1394 
1395 /*
1396  * SCSI GUID
1397  * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1398  */
1399 #define HV_SCSI_GUID \
1400 	.guid = GUID_INIT(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1401 			  0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1402 
1403 /*
1404  * Shutdown GUID
1405  * {0e0b6031-5213-4934-818b-38d90ced39db}
1406  */
1407 #define HV_SHUTDOWN_GUID \
1408 	.guid = GUID_INIT(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1409 			  0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1410 
1411 /*
1412  * Time Synch GUID
1413  * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1414  */
1415 #define HV_TS_GUID \
1416 	.guid = GUID_INIT(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1417 			  0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1418 
1419 /*
1420  * Heartbeat GUID
1421  * {57164f39-9115-4e78-ab55-382f3bd5422d}
1422  */
1423 #define HV_HEART_BEAT_GUID \
1424 	.guid = GUID_INIT(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1425 			  0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1426 
1427 /*
1428  * KVP GUID
1429  * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1430  */
1431 #define HV_KVP_GUID \
1432 	.guid = GUID_INIT(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1433 			  0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1434 
1435 /*
1436  * Dynamic memory GUID
1437  * {525074dc-8985-46e2-8057-a307dc18a502}
1438  */
1439 #define HV_DM_GUID \
1440 	.guid = GUID_INIT(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1441 			  0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1442 
1443 /*
1444  * Mouse GUID
1445  * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1446  */
1447 #define HV_MOUSE_GUID \
1448 	.guid = GUID_INIT(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1449 			  0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1450 
1451 /*
1452  * Keyboard GUID
1453  * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1454  */
1455 #define HV_KBD_GUID \
1456 	.guid = GUID_INIT(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1457 			  0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1458 
1459 /*
1460  * VSS (Backup/Restore) GUID
1461  */
1462 #define HV_VSS_GUID \
1463 	.guid = GUID_INIT(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1464 			  0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1465 /*
1466  * Synthetic Video GUID
1467  * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1468  */
1469 #define HV_SYNTHVID_GUID \
1470 	.guid = GUID_INIT(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1471 			  0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1472 
1473 /*
1474  * Synthetic FC GUID
1475  * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1476  */
1477 #define HV_SYNTHFC_GUID \
1478 	.guid = GUID_INIT(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1479 			  0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1480 
1481 /*
1482  * Guest File Copy Service
1483  * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1484  */
1485 
1486 #define HV_FCOPY_GUID \
1487 	.guid = GUID_INIT(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1488 			  0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1489 
1490 /*
1491  * NetworkDirect. This is the guest RDMA service.
1492  * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1493  */
1494 #define HV_ND_GUID \
1495 	.guid = GUID_INIT(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1496 			  0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1497 
1498 /*
1499  * PCI Express Pass Through
1500  * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1501  */
1502 
1503 #define HV_PCIE_GUID \
1504 	.guid = GUID_INIT(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1505 			  0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1506 
1507 /*
1508  * Linux doesn't support these 4 devices: the first two are for
1509  * Automatic Virtual Machine Activation, the third is for
1510  * Remote Desktop Virtualization, and the fourth is Initial
1511  * Machine Configuration (IMC) used only by Windows guests.
1512  * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1513  * {3375baf4-9e15-4b30-b765-67acb10d607b}
1514  * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1515  * {c376c1c3-d276-48d2-90a9-c04748072c60}
1516  */
1517 
1518 #define HV_AVMA1_GUID \
1519 	.guid = GUID_INIT(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1520 			  0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1521 
1522 #define HV_AVMA2_GUID \
1523 	.guid = GUID_INIT(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1524 			  0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1525 
1526 #define HV_RDV_GUID \
1527 	.guid = GUID_INIT(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1528 			  0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1529 
1530 #define HV_IMC_GUID \
1531 	.guid = GUID_INIT(0xc376c1c3, 0xd276, 0x48d2, 0x90, 0xa9, \
1532 			  0xc0, 0x47, 0x48, 0x07, 0x2c, 0x60)
1533 
1534 /*
1535  * Common header for Hyper-V ICs
1536  */
1537 
1538 #define ICMSGTYPE_NEGOTIATE		0
1539 #define ICMSGTYPE_HEARTBEAT		1
1540 #define ICMSGTYPE_KVPEXCHANGE		2
1541 #define ICMSGTYPE_SHUTDOWN		3
1542 #define ICMSGTYPE_TIMESYNC		4
1543 #define ICMSGTYPE_VSS			5
1544 #define ICMSGTYPE_FCOPY			7
1545 
1546 #define ICMSGHDRFLAG_TRANSACTION	1
1547 #define ICMSGHDRFLAG_REQUEST		2
1548 #define ICMSGHDRFLAG_RESPONSE		4
1549 
1550 
1551 /*
1552  * While we want to handle util services as regular devices,
1553  * there is only one instance of each of these services; so
1554  * we statically allocate the service specific state.
1555  */
1556 
1557 struct hv_util_service {
1558 	u8 *recv_buffer;
1559 	void *channel;
1560 	void (*util_cb)(void *);
1561 	int (*util_init)(struct hv_util_service *);
1562 	int (*util_init_transport)(void);
1563 	void (*util_deinit)(void);
1564 	int (*util_pre_suspend)(void);
1565 	int (*util_pre_resume)(void);
1566 };
1567 
1568 struct vmbuspipe_hdr {
1569 	u32 flags;
1570 	u32 msgsize;
1571 } __packed;
1572 
1573 struct ic_version {
1574 	u16 major;
1575 	u16 minor;
1576 } __packed;
1577 
1578 struct icmsg_hdr {
1579 	struct ic_version icverframe;
1580 	u16 icmsgtype;
1581 	struct ic_version icvermsg;
1582 	u16 icmsgsize;
1583 	u32 status;
1584 	u8 ictransaction_id;
1585 	u8 icflags;
1586 	u8 reserved[2];
1587 } __packed;
1588 
1589 #define IC_VERSION_NEGOTIATION_MAX_VER_COUNT 100
1590 #define ICMSG_HDR (sizeof(struct vmbuspipe_hdr) + sizeof(struct icmsg_hdr))
1591 #define ICMSG_NEGOTIATE_PKT_SIZE(icframe_vercnt, icmsg_vercnt) \
1592 	(ICMSG_HDR + sizeof(struct icmsg_negotiate) + \
1593 	 (((icframe_vercnt) + (icmsg_vercnt)) * sizeof(struct ic_version)))
1594 
1595 struct icmsg_negotiate {
1596 	u16 icframe_vercnt;
1597 	u16 icmsg_vercnt;
1598 	u32 reserved;
1599 	struct ic_version icversion_data[]; /* any size array */
1600 } __packed;
1601 
1602 struct shutdown_msg_data {
1603 	u32 reason_code;
1604 	u32 timeout_seconds;
1605 	u32 flags;
1606 	u8  display_message[2048];
1607 } __packed;
1608 
1609 struct heartbeat_msg_data {
1610 	u64 seq_num;
1611 	u32 reserved[8];
1612 } __packed;
1613 
1614 /* Time Sync IC defs */
1615 #define ICTIMESYNCFLAG_PROBE	0
1616 #define ICTIMESYNCFLAG_SYNC	1
1617 #define ICTIMESYNCFLAG_SAMPLE	2
1618 
1619 #ifdef __x86_64__
1620 #define WLTIMEDELTA	116444736000000000L	/* in 100ns unit */
1621 #else
1622 #define WLTIMEDELTA	116444736000000000LL
1623 #endif
1624 
1625 struct ictimesync_data {
1626 	u64 parenttime;
1627 	u64 childtime;
1628 	u64 roundtriptime;
1629 	u8 flags;
1630 } __packed;
1631 
1632 struct ictimesync_ref_data {
1633 	u64 parenttime;
1634 	u64 vmreferencetime;
1635 	u8 flags;
1636 	char leapflags;
1637 	char stratum;
1638 	u8 reserved[3];
1639 } __packed;
1640 
1641 struct hyperv_service_callback {
1642 	u8 msg_type;
1643 	char *log_msg;
1644 	guid_t data;
1645 	struct vmbus_channel *channel;
1646 	void (*callback)(void *context);
1647 };
1648 
1649 struct hv_dma_range {
1650 	dma_addr_t dma;
1651 	u32 mapping_size;
1652 };
1653 
1654 #define MAX_SRV_VER	0x7ffffff
1655 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf, u32 buflen,
1656 				const int *fw_version, int fw_vercnt,
1657 				const int *srv_version, int srv_vercnt,
1658 				int *nego_fw_version, int *nego_srv_version);
1659 
1660 void hv_process_channel_removal(struct vmbus_channel *channel);
1661 
1662 void vmbus_setevent(struct vmbus_channel *channel);
1663 /*
1664  * Negotiated version with the Host.
1665  */
1666 
1667 extern __u32 vmbus_proto_version;
1668 
1669 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
1670 				  const guid_t *shv_host_servie_id);
1671 int vmbus_send_modifychannel(struct vmbus_channel *channel, u32 target_vp);
1672 void vmbus_set_event(struct vmbus_channel *channel);
1673 
1674 /* Get the start of the ring buffer. */
1675 static inline void *
hv_get_ring_buffer(const struct hv_ring_buffer_info * ring_info)1676 hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info)
1677 {
1678 	return ring_info->ring_buffer->buffer;
1679 }
1680 
1681 /*
1682  * Mask off host interrupt callback notifications
1683  */
hv_begin_read(struct hv_ring_buffer_info * rbi)1684 static inline void hv_begin_read(struct hv_ring_buffer_info *rbi)
1685 {
1686 	rbi->ring_buffer->interrupt_mask = 1;
1687 
1688 	/* make sure mask update is not reordered */
1689 	virt_mb();
1690 }
1691 
1692 /*
1693  * Re-enable host callback and return number of outstanding bytes
1694  */
hv_end_read(struct hv_ring_buffer_info * rbi)1695 static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi)
1696 {
1697 
1698 	rbi->ring_buffer->interrupt_mask = 0;
1699 
1700 	/* make sure mask update is not reordered */
1701 	virt_mb();
1702 
1703 	/*
1704 	 * Now check to see if the ring buffer is still empty.
1705 	 * If it is not, we raced and we need to process new
1706 	 * incoming messages.
1707 	 */
1708 	return hv_get_bytes_to_read(rbi);
1709 }
1710 
1711 /*
1712  * An API to support in-place processing of incoming VMBUS packets.
1713  */
1714 
1715 /* Get data payload associated with descriptor */
hv_pkt_data(const struct vmpacket_descriptor * desc)1716 static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc)
1717 {
1718 	return (void *)((unsigned long)desc + (desc->offset8 << 3));
1719 }
1720 
1721 /* Get data size associated with descriptor */
hv_pkt_datalen(const struct vmpacket_descriptor * desc)1722 static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc)
1723 {
1724 	return (desc->len8 << 3) - (desc->offset8 << 3);
1725 }
1726 
1727 /* Get packet length associated with descriptor */
hv_pkt_len(const struct vmpacket_descriptor * desc)1728 static inline u32 hv_pkt_len(const struct vmpacket_descriptor *desc)
1729 {
1730 	return desc->len8 << 3;
1731 }
1732 
1733 struct vmpacket_descriptor *
1734 hv_pkt_iter_first(struct vmbus_channel *channel);
1735 
1736 struct vmpacket_descriptor *
1737 __hv_pkt_iter_next(struct vmbus_channel *channel,
1738 		   const struct vmpacket_descriptor *pkt);
1739 
1740 void hv_pkt_iter_close(struct vmbus_channel *channel);
1741 
1742 static inline struct vmpacket_descriptor *
hv_pkt_iter_next(struct vmbus_channel * channel,const struct vmpacket_descriptor * pkt)1743 hv_pkt_iter_next(struct vmbus_channel *channel,
1744 		 const struct vmpacket_descriptor *pkt)
1745 {
1746 	struct vmpacket_descriptor *nxt;
1747 
1748 	nxt = __hv_pkt_iter_next(channel, pkt);
1749 	if (!nxt)
1750 		hv_pkt_iter_close(channel);
1751 
1752 	return nxt;
1753 }
1754 
1755 #define foreach_vmbus_pkt(pkt, channel) \
1756 	for (pkt = hv_pkt_iter_first(channel); pkt; \
1757 	    pkt = hv_pkt_iter_next(channel, pkt))
1758 
1759 /*
1760  * Interface for passing data between SR-IOV PF and VF drivers. The VF driver
1761  * sends requests to read and write blocks. Each block must be 128 bytes or
1762  * smaller. Optionally, the VF driver can register a callback function which
1763  * will be invoked when the host says that one or more of the first 64 block
1764  * IDs is "invalid" which means that the VF driver should reread them.
1765  */
1766 #define HV_CONFIG_BLOCK_SIZE_MAX 128
1767 
1768 int hyperv_read_cfg_blk(struct pci_dev *dev, void *buf, unsigned int buf_len,
1769 			unsigned int block_id, unsigned int *bytes_returned);
1770 int hyperv_write_cfg_blk(struct pci_dev *dev, void *buf, unsigned int len,
1771 			 unsigned int block_id);
1772 int hyperv_reg_block_invalidate(struct pci_dev *dev, void *context,
1773 				void (*block_invalidate)(void *context,
1774 							 u64 block_mask));
1775 
1776 struct hyperv_pci_block_ops {
1777 	int (*read_block)(struct pci_dev *dev, void *buf, unsigned int buf_len,
1778 			  unsigned int block_id, unsigned int *bytes_returned);
1779 	int (*write_block)(struct pci_dev *dev, void *buf, unsigned int len,
1780 			   unsigned int block_id);
1781 	int (*reg_blk_invalidate)(struct pci_dev *dev, void *context,
1782 				  void (*block_invalidate)(void *context,
1783 							   u64 block_mask));
1784 };
1785 
1786 extern struct hyperv_pci_block_ops hvpci_block_ops;
1787 
virt_to_hvpfn(void * addr)1788 static inline unsigned long virt_to_hvpfn(void *addr)
1789 {
1790 	phys_addr_t paddr;
1791 
1792 	if (is_vmalloc_addr(addr))
1793 		paddr = page_to_phys(vmalloc_to_page(addr)) +
1794 				     offset_in_page(addr);
1795 	else
1796 		paddr = __pa(addr);
1797 
1798 	return  paddr >> HV_HYP_PAGE_SHIFT;
1799 }
1800 
1801 #define NR_HV_HYP_PAGES_IN_PAGE	(PAGE_SIZE / HV_HYP_PAGE_SIZE)
1802 #define offset_in_hvpage(ptr)	((unsigned long)(ptr) & ~HV_HYP_PAGE_MASK)
1803 #define HVPFN_UP(x)	(((x) + HV_HYP_PAGE_SIZE-1) >> HV_HYP_PAGE_SHIFT)
1804 #define HVPFN_DOWN(x)	((x) >> HV_HYP_PAGE_SHIFT)
1805 #define page_to_hvpfn(page)	(page_to_pfn(page) * NR_HV_HYP_PAGES_IN_PAGE)
1806 
1807 #endif /* _HYPERV_H */
1808