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
186
hv_get_bytes_to_read(const struct hv_ring_buffer_info * rbi)187 static inline u32 hv_get_bytes_to_read(const struct hv_ring_buffer_info *rbi)
188 {
189 u32 read_loc, write_loc, dsize, read;
190
191 dsize = rbi->ring_datasize;
192 read_loc = rbi->ring_buffer->read_index;
193 write_loc = READ_ONCE(rbi->ring_buffer->write_index);
194
195 read = write_loc >= read_loc ? (write_loc - read_loc) :
196 (dsize - read_loc) + write_loc;
197
198 return read;
199 }
200
hv_get_bytes_to_write(const struct hv_ring_buffer_info * rbi)201 static inline u32 hv_get_bytes_to_write(const struct hv_ring_buffer_info *rbi)
202 {
203 u32 read_loc, write_loc, dsize, write;
204
205 dsize = rbi->ring_datasize;
206 read_loc = READ_ONCE(rbi->ring_buffer->read_index);
207 write_loc = rbi->ring_buffer->write_index;
208
209 write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
210 read_loc - write_loc;
211 return write;
212 }
213
hv_get_avail_to_write_percent(const struct hv_ring_buffer_info * rbi)214 static inline u32 hv_get_avail_to_write_percent(
215 const struct hv_ring_buffer_info *rbi)
216 {
217 u32 avail_write = hv_get_bytes_to_write(rbi);
218
219 return reciprocal_divide(
220 (avail_write << 3) + (avail_write << 1),
221 rbi->ring_size_div10_reciprocal);
222 }
223
224 /*
225 * VMBUS version is 32 bit entity broken up into
226 * two 16 bit quantities: major_number. minor_number.
227 *
228 * 0 . 13 (Windows Server 2008)
229 * 1 . 1 (Windows 7)
230 * 2 . 4 (Windows 8)
231 * 3 . 0 (Windows 8 R2)
232 * 4 . 0 (Windows 10)
233 * 4 . 1 (Windows 10 RS3)
234 * 5 . 0 (Newer Windows 10)
235 * 5 . 1 (Windows 10 RS4)
236 * 5 . 2 (Windows Server 2019, RS5)
237 * 5 . 3 (Windows Server 2022)
238 */
239
240 #define VERSION_WS2008 ((0 << 16) | (13))
241 #define VERSION_WIN7 ((1 << 16) | (1))
242 #define VERSION_WIN8 ((2 << 16) | (4))
243 #define VERSION_WIN8_1 ((3 << 16) | (0))
244 #define VERSION_WIN10 ((4 << 16) | (0))
245 #define VERSION_WIN10_V4_1 ((4 << 16) | (1))
246 #define VERSION_WIN10_V5 ((5 << 16) | (0))
247 #define VERSION_WIN10_V5_1 ((5 << 16) | (1))
248 #define VERSION_WIN10_V5_2 ((5 << 16) | (2))
249 #define VERSION_WIN10_V5_3 ((5 << 16) | (3))
250
251 /* Make maximum size of pipe payload of 16K */
252 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
253
254 /* Define PipeMode values. */
255 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
256 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
257
258 /* The size of the user defined data buffer for non-pipe offers. */
259 #define MAX_USER_DEFINED_BYTES 120
260
261 /* The size of the user defined data buffer for pipe offers. */
262 #define MAX_PIPE_USER_DEFINED_BYTES 116
263
264 /*
265 * At the center of the Channel Management library is the Channel Offer. This
266 * struct contains the fundamental information about an offer.
267 */
268 struct vmbus_channel_offer {
269 guid_t if_type;
270 guid_t if_instance;
271
272 /*
273 * These two fields are not currently used.
274 */
275 u64 reserved1;
276 u64 reserved2;
277
278 u16 chn_flags;
279 u16 mmio_megabytes; /* in bytes * 1024 * 1024 */
280
281 union {
282 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
283 struct {
284 unsigned char user_def[MAX_USER_DEFINED_BYTES];
285 } std;
286
287 /*
288 * Pipes:
289 * The following structure is an integrated pipe protocol, which
290 * is implemented on top of standard user-defined data. Pipe
291 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
292 * use.
293 */
294 struct {
295 u32 pipe_mode;
296 unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
297 } pipe;
298 } u;
299 /*
300 * The sub_channel_index is defined in Win8: a value of zero means a
301 * primary channel and a value of non-zero means a sub-channel.
302 *
303 * Before Win8, the field is reserved, meaning it's always zero.
304 */
305 u16 sub_channel_index;
306 u16 reserved3;
307 } __packed;
308
309 /* Server Flags */
310 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
311 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
312 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
313 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
314 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
315 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
316 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
317 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
318
319 struct vmpacket_descriptor {
320 u16 type;
321 u16 offset8;
322 u16 len8;
323 u16 flags;
324 u64 trans_id;
325 } __packed;
326
327 struct vmpacket_header {
328 u32 prev_pkt_start_offset;
329 struct vmpacket_descriptor descriptor;
330 } __packed;
331
332 struct vmtransfer_page_range {
333 u32 byte_count;
334 u32 byte_offset;
335 } __packed;
336
337 struct vmtransfer_page_packet_header {
338 struct vmpacket_descriptor d;
339 u16 xfer_pageset_id;
340 u8 sender_owns_set;
341 u8 reserved;
342 u32 range_cnt;
343 struct vmtransfer_page_range ranges[1];
344 } __packed;
345
346 struct vmgpadl_packet_header {
347 struct vmpacket_descriptor d;
348 u32 gpadl;
349 u32 reserved;
350 } __packed;
351
352 struct vmadd_remove_transfer_page_set {
353 struct vmpacket_descriptor d;
354 u32 gpadl;
355 u16 xfer_pageset_id;
356 u16 reserved;
357 } __packed;
358
359 /*
360 * This structure defines a range in guest physical space that can be made to
361 * look virtually contiguous.
362 */
363 struct gpa_range {
364 u32 byte_count;
365 u32 byte_offset;
366 u64 pfn_array[];
367 };
368
369 /*
370 * This is the format for an Establish Gpadl packet, which contains a handle by
371 * which this GPADL will be known and a set of GPA ranges associated with it.
372 * This can be converted to a MDL by the guest OS. If there are multiple GPA
373 * ranges, then the resulting MDL will be "chained," representing multiple VA
374 * ranges.
375 */
376 struct vmestablish_gpadl {
377 struct vmpacket_descriptor d;
378 u32 gpadl;
379 u32 range_cnt;
380 struct gpa_range range[1];
381 } __packed;
382
383 /*
384 * This is the format for a Teardown Gpadl packet, which indicates that the
385 * GPADL handle in the Establish Gpadl packet will never be referenced again.
386 */
387 struct vmteardown_gpadl {
388 struct vmpacket_descriptor d;
389 u32 gpadl;
390 u32 reserved; /* for alignment to a 8-byte boundary */
391 } __packed;
392
393 /*
394 * This is the format for a GPA-Direct packet, which contains a set of GPA
395 * ranges, in addition to commands and/or data.
396 */
397 struct vmdata_gpa_direct {
398 struct vmpacket_descriptor d;
399 u32 reserved;
400 u32 range_cnt;
401 struct gpa_range range[1];
402 } __packed;
403
404 /* This is the format for a Additional Data Packet. */
405 struct vmadditional_data {
406 struct vmpacket_descriptor d;
407 u64 total_bytes;
408 u32 offset;
409 u32 byte_cnt;
410 unsigned char data[1];
411 } __packed;
412
413 union vmpacket_largest_possible_header {
414 struct vmpacket_descriptor simple_hdr;
415 struct vmtransfer_page_packet_header xfer_page_hdr;
416 struct vmgpadl_packet_header gpadl_hdr;
417 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr;
418 struct vmestablish_gpadl establish_gpadl_hdr;
419 struct vmteardown_gpadl teardown_gpadl_hdr;
420 struct vmdata_gpa_direct data_gpa_direct_hdr;
421 };
422
423 #define VMPACKET_DATA_START_ADDRESS(__packet) \
424 (void *)(((unsigned char *)__packet) + \
425 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
426
427 #define VMPACKET_DATA_LENGTH(__packet) \
428 ((((struct vmpacket_descriptor)__packet)->len8 - \
429 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
430
431 #define VMPACKET_TRANSFER_MODE(__packet) \
432 (((struct IMPACT)__packet)->type)
433
434 enum vmbus_packet_type {
435 VM_PKT_INVALID = 0x0,
436 VM_PKT_SYNCH = 0x1,
437 VM_PKT_ADD_XFER_PAGESET = 0x2,
438 VM_PKT_RM_XFER_PAGESET = 0x3,
439 VM_PKT_ESTABLISH_GPADL = 0x4,
440 VM_PKT_TEARDOWN_GPADL = 0x5,
441 VM_PKT_DATA_INBAND = 0x6,
442 VM_PKT_DATA_USING_XFER_PAGES = 0x7,
443 VM_PKT_DATA_USING_GPADL = 0x8,
444 VM_PKT_DATA_USING_GPA_DIRECT = 0x9,
445 VM_PKT_CANCEL_REQUEST = 0xa,
446 VM_PKT_COMP = 0xb,
447 VM_PKT_DATA_USING_ADDITIONAL_PKT = 0xc,
448 VM_PKT_ADDITIONAL_DATA = 0xd
449 };
450
451 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
452
453
454 /* Version 1 messages */
455 enum vmbus_channel_message_type {
456 CHANNELMSG_INVALID = 0,
457 CHANNELMSG_OFFERCHANNEL = 1,
458 CHANNELMSG_RESCIND_CHANNELOFFER = 2,
459 CHANNELMSG_REQUESTOFFERS = 3,
460 CHANNELMSG_ALLOFFERS_DELIVERED = 4,
461 CHANNELMSG_OPENCHANNEL = 5,
462 CHANNELMSG_OPENCHANNEL_RESULT = 6,
463 CHANNELMSG_CLOSECHANNEL = 7,
464 CHANNELMSG_GPADL_HEADER = 8,
465 CHANNELMSG_GPADL_BODY = 9,
466 CHANNELMSG_GPADL_CREATED = 10,
467 CHANNELMSG_GPADL_TEARDOWN = 11,
468 CHANNELMSG_GPADL_TORNDOWN = 12,
469 CHANNELMSG_RELID_RELEASED = 13,
470 CHANNELMSG_INITIATE_CONTACT = 14,
471 CHANNELMSG_VERSION_RESPONSE = 15,
472 CHANNELMSG_UNLOAD = 16,
473 CHANNELMSG_UNLOAD_RESPONSE = 17,
474 CHANNELMSG_18 = 18,
475 CHANNELMSG_19 = 19,
476 CHANNELMSG_20 = 20,
477 CHANNELMSG_TL_CONNECT_REQUEST = 21,
478 CHANNELMSG_MODIFYCHANNEL = 22,
479 CHANNELMSG_TL_CONNECT_RESULT = 23,
480 CHANNELMSG_MODIFYCHANNEL_RESPONSE = 24,
481 CHANNELMSG_COUNT
482 };
483
484 /* Hyper-V supports about 2048 channels, and the RELIDs start with 1. */
485 #define INVALID_RELID U32_MAX
486
487 struct vmbus_channel_message_header {
488 enum vmbus_channel_message_type msgtype;
489 u32 padding;
490 } __packed;
491
492 /* Query VMBus Version parameters */
493 struct vmbus_channel_query_vmbus_version {
494 struct vmbus_channel_message_header header;
495 u32 version;
496 } __packed;
497
498 /* VMBus Version Supported parameters */
499 struct vmbus_channel_version_supported {
500 struct vmbus_channel_message_header header;
501 u8 version_supported;
502 } __packed;
503
504 /* Offer Channel parameters */
505 struct vmbus_channel_offer_channel {
506 struct vmbus_channel_message_header header;
507 struct vmbus_channel_offer offer;
508 u32 child_relid;
509 u8 monitorid;
510 /*
511 * win7 and beyond splits this field into a bit field.
512 */
513 u8 monitor_allocated:1;
514 u8 reserved:7;
515 /*
516 * These are new fields added in win7 and later.
517 * Do not access these fields without checking the
518 * negotiated protocol.
519 *
520 * If "is_dedicated_interrupt" is set, we must not set the
521 * associated bit in the channel bitmap while sending the
522 * interrupt to the host.
523 *
524 * connection_id is to be used in signaling the host.
525 */
526 u16 is_dedicated_interrupt:1;
527 u16 reserved1:15;
528 u32 connection_id;
529 } __packed;
530
531 /* Rescind Offer parameters */
532 struct vmbus_channel_rescind_offer {
533 struct vmbus_channel_message_header header;
534 u32 child_relid;
535 } __packed;
536
537 static inline u32
hv_ringbuffer_pending_size(const struct hv_ring_buffer_info * rbi)538 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info *rbi)
539 {
540 return rbi->ring_buffer->pending_send_sz;
541 }
542
543 /*
544 * Request Offer -- no parameters, SynIC message contains the partition ID
545 * Set Snoop -- no parameters, SynIC message contains the partition ID
546 * Clear Snoop -- no parameters, SynIC message contains the partition ID
547 * All Offers Delivered -- no parameters, SynIC message contains the partition
548 * ID
549 * Flush Client -- no parameters, SynIC message contains the partition ID
550 */
551
552 /* Open Channel parameters */
553 struct vmbus_channel_open_channel {
554 struct vmbus_channel_message_header header;
555
556 /* Identifies the specific VMBus channel that is being opened. */
557 u32 child_relid;
558
559 /* ID making a particular open request at a channel offer unique. */
560 u32 openid;
561
562 /* GPADL for the channel's ring buffer. */
563 u32 ringbuffer_gpadlhandle;
564
565 /*
566 * Starting with win8, this field will be used to specify
567 * the target virtual processor on which to deliver the interrupt for
568 * the host to guest communication.
569 * Prior to win8, incoming channel interrupts would only
570 * be delivered on cpu 0. Setting this value to 0 would
571 * preserve the earlier behavior.
572 */
573 u32 target_vp;
574
575 /*
576 * The upstream ring buffer begins at offset zero in the memory
577 * described by RingBufferGpadlHandle. The downstream ring buffer
578 * follows it at this offset (in pages).
579 */
580 u32 downstream_ringbuffer_pageoffset;
581
582 /* User-specific data to be passed along to the server endpoint. */
583 unsigned char userdata[MAX_USER_DEFINED_BYTES];
584 } __packed;
585
586 /* Open Channel Result parameters */
587 struct vmbus_channel_open_result {
588 struct vmbus_channel_message_header header;
589 u32 child_relid;
590 u32 openid;
591 u32 status;
592 } __packed;
593
594 /* Modify Channel Result parameters */
595 struct vmbus_channel_modifychannel_response {
596 struct vmbus_channel_message_header header;
597 u32 child_relid;
598 u32 status;
599 } __packed;
600
601 /* Close channel parameters; */
602 struct vmbus_channel_close_channel {
603 struct vmbus_channel_message_header header;
604 u32 child_relid;
605 } __packed;
606
607 /* Channel Message GPADL */
608 #define GPADL_TYPE_RING_BUFFER 1
609 #define GPADL_TYPE_SERVER_SAVE_AREA 2
610 #define GPADL_TYPE_TRANSACTION 8
611
612 /*
613 * The number of PFNs in a GPADL message is defined by the number of
614 * pages that would be spanned by ByteCount and ByteOffset. If the
615 * implied number of PFNs won't fit in this packet, there will be a
616 * follow-up packet that contains more.
617 */
618 struct vmbus_channel_gpadl_header {
619 struct vmbus_channel_message_header header;
620 u32 child_relid;
621 u32 gpadl;
622 u16 range_buflen;
623 u16 rangecount;
624 struct gpa_range range[];
625 } __packed;
626
627 /* This is the followup packet that contains more PFNs. */
628 struct vmbus_channel_gpadl_body {
629 struct vmbus_channel_message_header header;
630 u32 msgnumber;
631 u32 gpadl;
632 u64 pfn[];
633 } __packed;
634
635 struct vmbus_channel_gpadl_created {
636 struct vmbus_channel_message_header header;
637 u32 child_relid;
638 u32 gpadl;
639 u32 creation_status;
640 } __packed;
641
642 struct vmbus_channel_gpadl_teardown {
643 struct vmbus_channel_message_header header;
644 u32 child_relid;
645 u32 gpadl;
646 } __packed;
647
648 struct vmbus_channel_gpadl_torndown {
649 struct vmbus_channel_message_header header;
650 u32 gpadl;
651 } __packed;
652
653 struct vmbus_channel_relid_released {
654 struct vmbus_channel_message_header header;
655 u32 child_relid;
656 } __packed;
657
658 struct vmbus_channel_initiate_contact {
659 struct vmbus_channel_message_header header;
660 u32 vmbus_version_requested;
661 u32 target_vcpu; /* The VCPU the host should respond to */
662 union {
663 u64 interrupt_page;
664 struct {
665 u8 msg_sint;
666 u8 padding1[3];
667 u32 padding2;
668 };
669 };
670 u64 monitor_page1;
671 u64 monitor_page2;
672 } __packed;
673
674 /* Hyper-V socket: guest's connect()-ing to host */
675 struct vmbus_channel_tl_connect_request {
676 struct vmbus_channel_message_header header;
677 guid_t guest_endpoint_id;
678 guid_t host_service_id;
679 } __packed;
680
681 /* Modify Channel parameters, cf. vmbus_send_modifychannel() */
682 struct vmbus_channel_modifychannel {
683 struct vmbus_channel_message_header header;
684 u32 child_relid;
685 u32 target_vp;
686 } __packed;
687
688 struct vmbus_channel_version_response {
689 struct vmbus_channel_message_header header;
690 u8 version_supported;
691
692 u8 connection_state;
693 u16 padding;
694
695 /*
696 * On new hosts that support VMBus protocol 5.0, we must use
697 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
698 * and for subsequent messages, we must use the Message Connection ID
699 * field in the host-returned Version Response Message.
700 *
701 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
702 */
703 u32 msg_conn_id;
704 } __packed;
705
706 enum vmbus_channel_state {
707 CHANNEL_OFFER_STATE,
708 CHANNEL_OPENING_STATE,
709 CHANNEL_OPEN_STATE,
710 CHANNEL_OPENED_STATE,
711 };
712
713 /*
714 * Represents each channel msg on the vmbus connection This is a
715 * variable-size data structure depending on the msg type itself
716 */
717 struct vmbus_channel_msginfo {
718 /* Bookkeeping stuff */
719 struct list_head msglistentry;
720
721 /* So far, this is only used to handle gpadl body message */
722 struct list_head submsglist;
723
724 /* Synchronize the request/response if needed */
725 struct completion waitevent;
726 struct vmbus_channel *waiting_channel;
727 union {
728 struct vmbus_channel_version_supported version_supported;
729 struct vmbus_channel_open_result open_result;
730 struct vmbus_channel_gpadl_torndown gpadl_torndown;
731 struct vmbus_channel_gpadl_created gpadl_created;
732 struct vmbus_channel_version_response version_response;
733 struct vmbus_channel_modifychannel_response modify_response;
734 } response;
735
736 u32 msgsize;
737 /*
738 * The channel message that goes out on the "wire".
739 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
740 */
741 unsigned char msg[];
742 };
743
744 struct vmbus_close_msg {
745 struct vmbus_channel_msginfo info;
746 struct vmbus_channel_close_channel msg;
747 };
748
749 /* Define connection identifier type. */
750 union hv_connection_id {
751 u32 asu32;
752 struct {
753 u32 id:24;
754 u32 reserved:8;
755 } u;
756 };
757
758 enum vmbus_device_type {
759 HV_IDE = 0,
760 HV_SCSI,
761 HV_FC,
762 HV_NIC,
763 HV_ND,
764 HV_PCIE,
765 HV_FB,
766 HV_KBD,
767 HV_MOUSE,
768 HV_KVP,
769 HV_TS,
770 HV_HB,
771 HV_SHUTDOWN,
772 HV_FCOPY,
773 HV_BACKUP,
774 HV_DM,
775 HV_UNKNOWN,
776 };
777
778 /*
779 * Provides request ids for VMBus. Encapsulates guest memory
780 * addresses and stores the next available slot in req_arr
781 * to generate new ids in constant time.
782 */
783 struct vmbus_requestor {
784 u64 *req_arr;
785 unsigned long *req_bitmap; /* is a given slot available? */
786 u32 size;
787 u64 next_request_id;
788 spinlock_t req_lock; /* provides atomicity */
789 };
790
791 #define VMBUS_NO_RQSTOR U64_MAX
792 #define VMBUS_RQST_ERROR (U64_MAX - 1)
793 #define VMBUS_RQST_ID_NO_RESPONSE (U64_MAX - 2)
794
795 struct vmbus_device {
796 u16 dev_type;
797 guid_t guid;
798 bool perf_device;
799 bool allowed_in_isolated;
800 };
801
802 struct vmbus_channel {
803 struct list_head listentry;
804
805 struct hv_device *device_obj;
806
807 enum vmbus_channel_state state;
808
809 struct vmbus_channel_offer_channel offermsg;
810 /*
811 * These are based on the OfferMsg.MonitorId.
812 * Save it here for easy access.
813 */
814 u8 monitor_grp;
815 u8 monitor_bit;
816
817 bool rescind; /* got rescind msg */
818 bool rescind_ref; /* got rescind msg, got channel reference */
819 struct completion rescind_event;
820
821 u32 ringbuffer_gpadlhandle;
822
823 /* Allocated memory for ring buffer */
824 struct page *ringbuffer_page;
825 u32 ringbuffer_pagecount;
826 u32 ringbuffer_send_offset;
827 struct hv_ring_buffer_info outbound; /* send to parent */
828 struct hv_ring_buffer_info inbound; /* receive from parent */
829
830 struct vmbus_close_msg close_msg;
831
832 /* Statistics */
833 u64 interrupts; /* Host to Guest interrupts */
834 u64 sig_events; /* Guest to Host events */
835
836 /*
837 * Guest to host interrupts caused by the outbound ring buffer changing
838 * from empty to not empty.
839 */
840 u64 intr_out_empty;
841
842 /*
843 * Indicates that a full outbound ring buffer was encountered. The flag
844 * is set to true when a full outbound ring buffer is encountered and
845 * set to false when a write to the outbound ring buffer is completed.
846 */
847 bool out_full_flag;
848
849 /* Channel callback's invoked in softirq context */
850 struct tasklet_struct callback_event;
851 void (*onchannel_callback)(void *context);
852 void *channel_callback_context;
853
854 void (*change_target_cpu_callback)(struct vmbus_channel *channel,
855 u32 old, u32 new);
856
857 /*
858 * Synchronize channel scheduling and channel removal; see the inline
859 * comments in vmbus_chan_sched() and vmbus_reset_channel_cb().
860 */
861 spinlock_t sched_lock;
862
863 /*
864 * A channel can be marked for one of three modes of reading:
865 * BATCHED - callback called from taslket and should read
866 * channel until empty. Interrupts from the host
867 * are masked while read is in process (default).
868 * DIRECT - callback called from tasklet (softirq).
869 * ISR - callback called in interrupt context and must
870 * invoke its own deferred processing.
871 * Host interrupts are disabled and must be re-enabled
872 * when ring is empty.
873 */
874 enum hv_callback_mode {
875 HV_CALL_BATCHED,
876 HV_CALL_DIRECT,
877 HV_CALL_ISR
878 } callback_mode;
879
880 bool is_dedicated_interrupt;
881 u64 sig_event;
882
883 /*
884 * Starting with win8, this field will be used to specify the
885 * target CPU on which to deliver the interrupt for the host
886 * to guest communication.
887 *
888 * Prior to win8, incoming channel interrupts would only be
889 * delivered on CPU 0. Setting this value to 0 would preserve
890 * the earlier behavior.
891 */
892 u32 target_cpu;
893 /*
894 * Support for sub-channels. For high performance devices,
895 * it will be useful to have multiple sub-channels to support
896 * a scalable communication infrastructure with the host.
897 * The support for sub-channels is implemented as an extension
898 * to the current infrastructure.
899 * The initial offer is considered the primary channel and this
900 * offer message will indicate if the host supports sub-channels.
901 * The guest is free to ask for sub-channels to be offered and can
902 * open these sub-channels as a normal "primary" channel. However,
903 * all sub-channels will have the same type and instance guids as the
904 * primary channel. Requests sent on a given channel will result in a
905 * response on the same channel.
906 */
907
908 /*
909 * Sub-channel creation callback. This callback will be called in
910 * process context when a sub-channel offer is received from the host.
911 * The guest can open the sub-channel in the context of this callback.
912 */
913 void (*sc_creation_callback)(struct vmbus_channel *new_sc);
914
915 /*
916 * Channel rescind callback. Some channels (the hvsock ones), need to
917 * register a callback which is invoked in vmbus_onoffer_rescind().
918 */
919 void (*chn_rescind_callback)(struct vmbus_channel *channel);
920
921 /*
922 * All Sub-channels of a primary channel are linked here.
923 */
924 struct list_head sc_list;
925 /*
926 * The primary channel this sub-channel belongs to.
927 * This will be NULL for the primary channel.
928 */
929 struct vmbus_channel *primary_channel;
930 /*
931 * Support per-channel state for use by vmbus drivers.
932 */
933 void *per_channel_state;
934
935 /*
936 * Defer freeing channel until after all cpu's have
937 * gone through grace period.
938 */
939 struct rcu_head rcu;
940
941 /*
942 * For sysfs per-channel properties.
943 */
944 struct kobject kobj;
945
946 /*
947 * For performance critical channels (storage, networking
948 * etc,), Hyper-V has a mechanism to enhance the throughput
949 * at the expense of latency:
950 * When the host is to be signaled, we just set a bit in a shared page
951 * and this bit will be inspected by the hypervisor within a certain
952 * window and if the bit is set, the host will be signaled. The window
953 * of time is the monitor latency - currently around 100 usecs. This
954 * mechanism improves throughput by:
955 *
956 * A) Making the host more efficient - each time it wakes up,
957 * potentially it will process morev number of packets. The
958 * monitor latency allows a batch to build up.
959 * B) By deferring the hypercall to signal, we will also minimize
960 * the interrupts.
961 *
962 * Clearly, these optimizations improve throughput at the expense of
963 * latency. Furthermore, since the channel is shared for both
964 * control and data messages, control messages currently suffer
965 * unnecessary latency adversely impacting performance and boot
966 * time. To fix this issue, permit tagging the channel as being
967 * in "low latency" mode. In this mode, we will bypass the monitor
968 * mechanism.
969 */
970 bool low_latency;
971
972 bool probe_done;
973
974 /*
975 * Cache the device ID here for easy access; this is useful, in
976 * particular, in situations where the channel's device_obj has
977 * not been allocated/initialized yet.
978 */
979 u16 device_id;
980
981 /*
982 * We must offload the handling of the primary/sub channels
983 * from the single-threaded vmbus_connection.work_queue to
984 * two different workqueue, otherwise we can block
985 * vmbus_connection.work_queue and hang: see vmbus_process_offer().
986 */
987 struct work_struct add_channel_work;
988
989 /*
990 * Guest to host interrupts caused by the inbound ring buffer changing
991 * from full to not full while a packet is waiting.
992 */
993 u64 intr_in_full;
994
995 /*
996 * The total number of write operations that encountered a full
997 * outbound ring buffer.
998 */
999 u64 out_full_total;
1000
1001 /*
1002 * The number of write operations that were the first to encounter a
1003 * full outbound ring buffer.
1004 */
1005 u64 out_full_first;
1006
1007 /* enabling/disabling fuzz testing on the channel (default is false)*/
1008 bool fuzz_testing_state;
1009
1010 /*
1011 * Interrupt delay will delay the guest from emptying the ring buffer
1012 * for a specific amount of time. The delay is in microseconds and will
1013 * be between 1 to a maximum of 1000, its default is 0 (no delay).
1014 * The Message delay will delay guest reading on a per message basis
1015 * in microseconds between 1 to 1000 with the default being 0
1016 * (no delay).
1017 */
1018 u32 fuzz_testing_interrupt_delay;
1019 u32 fuzz_testing_message_delay;
1020
1021 /* request/transaction ids for VMBus */
1022 struct vmbus_requestor requestor;
1023 u32 rqstor_size;
1024 };
1025
1026 u64 vmbus_next_request_id(struct vmbus_requestor *rqstor, u64 rqst_addr);
1027 u64 vmbus_request_addr(struct vmbus_requestor *rqstor, u64 trans_id);
1028
is_hvsock_channel(const struct vmbus_channel * c)1029 static inline bool is_hvsock_channel(const struct vmbus_channel *c)
1030 {
1031 return !!(c->offermsg.offer.chn_flags &
1032 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER);
1033 }
1034
is_sub_channel(const struct vmbus_channel * c)1035 static inline bool is_sub_channel(const struct vmbus_channel *c)
1036 {
1037 return c->offermsg.offer.sub_channel_index != 0;
1038 }
1039
set_channel_read_mode(struct vmbus_channel * c,enum hv_callback_mode mode)1040 static inline void set_channel_read_mode(struct vmbus_channel *c,
1041 enum hv_callback_mode mode)
1042 {
1043 c->callback_mode = mode;
1044 }
1045
set_per_channel_state(struct vmbus_channel * c,void * s)1046 static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
1047 {
1048 c->per_channel_state = s;
1049 }
1050
get_per_channel_state(struct vmbus_channel * c)1051 static inline void *get_per_channel_state(struct vmbus_channel *c)
1052 {
1053 return c->per_channel_state;
1054 }
1055
set_channel_pending_send_size(struct vmbus_channel * c,u32 size)1056 static inline void set_channel_pending_send_size(struct vmbus_channel *c,
1057 u32 size)
1058 {
1059 unsigned long flags;
1060
1061 if (size) {
1062 spin_lock_irqsave(&c->outbound.ring_lock, flags);
1063 ++c->out_full_total;
1064
1065 if (!c->out_full_flag) {
1066 ++c->out_full_first;
1067 c->out_full_flag = true;
1068 }
1069 spin_unlock_irqrestore(&c->outbound.ring_lock, flags);
1070 } else {
1071 c->out_full_flag = false;
1072 }
1073
1074 c->outbound.ring_buffer->pending_send_sz = size;
1075 }
1076
set_low_latency_mode(struct vmbus_channel * c)1077 static inline void set_low_latency_mode(struct vmbus_channel *c)
1078 {
1079 c->low_latency = true;
1080 }
1081
clear_low_latency_mode(struct vmbus_channel * c)1082 static inline void clear_low_latency_mode(struct vmbus_channel *c)
1083 {
1084 c->low_latency = false;
1085 }
1086
1087 void vmbus_onmessage(struct vmbus_channel_message_header *hdr);
1088
1089 int vmbus_request_offers(void);
1090
1091 /*
1092 * APIs for managing sub-channels.
1093 */
1094
1095 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1096 void (*sc_cr_cb)(struct vmbus_channel *new_sc));
1097
1098 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1099 void (*chn_rescind_cb)(struct vmbus_channel *));
1100
1101 /*
1102 * Check if sub-channels have already been offerred. This API will be useful
1103 * when the driver is unloaded after establishing sub-channels. In this case,
1104 * when the driver is re-loaded, the driver would have to check if the
1105 * subchannels have already been established before attempting to request
1106 * the creation of sub-channels.
1107 * This function returns TRUE to indicate that subchannels have already been
1108 * created.
1109 * This function should be invoked after setting the callback function for
1110 * sub-channel creation.
1111 */
1112 bool vmbus_are_subchannels_present(struct vmbus_channel *primary);
1113
1114 /* The format must be the same as struct vmdata_gpa_direct */
1115 struct vmbus_channel_packet_page_buffer {
1116 u16 type;
1117 u16 dataoffset8;
1118 u16 length8;
1119 u16 flags;
1120 u64 transactionid;
1121 u32 reserved;
1122 u32 rangecount;
1123 struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
1124 } __packed;
1125
1126 /* The format must be the same as struct vmdata_gpa_direct */
1127 struct vmbus_channel_packet_multipage_buffer {
1128 u16 type;
1129 u16 dataoffset8;
1130 u16 length8;
1131 u16 flags;
1132 u64 transactionid;
1133 u32 reserved;
1134 u32 rangecount; /* Always 1 in this case */
1135 struct hv_multipage_buffer range;
1136 } __packed;
1137
1138 /* The format must be the same as struct vmdata_gpa_direct */
1139 struct vmbus_packet_mpb_array {
1140 u16 type;
1141 u16 dataoffset8;
1142 u16 length8;
1143 u16 flags;
1144 u64 transactionid;
1145 u32 reserved;
1146 u32 rangecount; /* Always 1 in this case */
1147 struct hv_mpb_array range;
1148 } __packed;
1149
1150 int vmbus_alloc_ring(struct vmbus_channel *channel,
1151 u32 send_size, u32 recv_size);
1152 void vmbus_free_ring(struct vmbus_channel *channel);
1153
1154 int vmbus_connect_ring(struct vmbus_channel *channel,
1155 void (*onchannel_callback)(void *context),
1156 void *context);
1157 int vmbus_disconnect_ring(struct vmbus_channel *channel);
1158
1159 extern int vmbus_open(struct vmbus_channel *channel,
1160 u32 send_ringbuffersize,
1161 u32 recv_ringbuffersize,
1162 void *userdata,
1163 u32 userdatalen,
1164 void (*onchannel_callback)(void *context),
1165 void *context);
1166
1167 extern void vmbus_close(struct vmbus_channel *channel);
1168
1169 extern int vmbus_sendpacket(struct vmbus_channel *channel,
1170 void *buffer,
1171 u32 bufferLen,
1172 u64 requestid,
1173 enum vmbus_packet_type type,
1174 u32 flags);
1175
1176 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1177 struct hv_page_buffer pagebuffers[],
1178 u32 pagecount,
1179 void *buffer,
1180 u32 bufferlen,
1181 u64 requestid);
1182
1183 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1184 struct vmbus_packet_mpb_array *mpb,
1185 u32 desc_size,
1186 void *buffer,
1187 u32 bufferlen,
1188 u64 requestid);
1189
1190 extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
1191 void *kbuffer,
1192 u32 size,
1193 u32 *gpadl_handle);
1194
1195 extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
1196 u32 gpadl_handle);
1197
1198 void vmbus_reset_channel_cb(struct vmbus_channel *channel);
1199
1200 extern int vmbus_recvpacket(struct vmbus_channel *channel,
1201 void *buffer,
1202 u32 bufferlen,
1203 u32 *buffer_actual_len,
1204 u64 *requestid);
1205
1206 extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
1207 void *buffer,
1208 u32 bufferlen,
1209 u32 *buffer_actual_len,
1210 u64 *requestid);
1211
1212
1213 extern void vmbus_ontimer(unsigned long data);
1214
1215 /* Base driver object */
1216 struct hv_driver {
1217 const char *name;
1218
1219 /*
1220 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1221 * channel flag, actually doesn't mean a synthetic device because the
1222 * offer's if_type/if_instance can change for every new hvsock
1223 * connection.
1224 *
1225 * However, to facilitate the notification of new-offer/rescind-offer
1226 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1227 * a special vmbus device, and hence we need the below flag to
1228 * indicate if the driver is the hvsock driver or not: we need to
1229 * specially treat the hvosck offer & driver in vmbus_match().
1230 */
1231 bool hvsock;
1232
1233 /* the device type supported by this driver */
1234 guid_t dev_type;
1235 const struct hv_vmbus_device_id *id_table;
1236
1237 struct device_driver driver;
1238
1239 /* dynamic device GUID's */
1240 struct {
1241 spinlock_t lock;
1242 struct list_head list;
1243 } dynids;
1244
1245 int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
1246 int (*remove)(struct hv_device *);
1247 void (*shutdown)(struct hv_device *);
1248
1249 int (*suspend)(struct hv_device *);
1250 int (*resume)(struct hv_device *);
1251
1252 };
1253
1254 /* Base device object */
1255 struct hv_device {
1256 /* the device type id of this device */
1257 guid_t dev_type;
1258
1259 /* the device instance id of this device */
1260 guid_t dev_instance;
1261 u16 vendor_id;
1262 u16 device_id;
1263
1264 struct device device;
1265 char *driver_override; /* Driver name to force a match */
1266
1267 struct vmbus_channel *channel;
1268 struct kset *channels_kset;
1269
1270 /* place holder to keep track of the dir for hv device in debugfs */
1271 struct dentry *debug_dir;
1272
1273 };
1274
1275
device_to_hv_device(struct device * d)1276 static inline struct hv_device *device_to_hv_device(struct device *d)
1277 {
1278 return container_of(d, struct hv_device, device);
1279 }
1280
drv_to_hv_drv(struct device_driver * d)1281 static inline struct hv_driver *drv_to_hv_drv(struct device_driver *d)
1282 {
1283 return container_of(d, struct hv_driver, driver);
1284 }
1285
hv_set_drvdata(struct hv_device * dev,void * data)1286 static inline void hv_set_drvdata(struct hv_device *dev, void *data)
1287 {
1288 dev_set_drvdata(&dev->device, data);
1289 }
1290
hv_get_drvdata(struct hv_device * dev)1291 static inline void *hv_get_drvdata(struct hv_device *dev)
1292 {
1293 return dev_get_drvdata(&dev->device);
1294 }
1295
1296 struct hv_ring_buffer_debug_info {
1297 u32 current_interrupt_mask;
1298 u32 current_read_index;
1299 u32 current_write_index;
1300 u32 bytes_avail_toread;
1301 u32 bytes_avail_towrite;
1302 };
1303
1304
1305 int hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
1306 struct hv_ring_buffer_debug_info *debug_info);
1307
1308 /* Vmbus interface */
1309 #define vmbus_driver_register(driver) \
1310 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1311 int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
1312 struct module *owner,
1313 const char *mod_name);
1314 void vmbus_driver_unregister(struct hv_driver *hv_driver);
1315
1316 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel);
1317
1318 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1319 resource_size_t min, resource_size_t max,
1320 resource_size_t size, resource_size_t align,
1321 bool fb_overlap_ok);
1322 void vmbus_free_mmio(resource_size_t start, resource_size_t size);
1323
1324 /*
1325 * GUID definitions of various offer types - services offered to the guest.
1326 */
1327
1328 /*
1329 * Network GUID
1330 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1331 */
1332 #define HV_NIC_GUID \
1333 .guid = GUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1334 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1335
1336 /*
1337 * IDE GUID
1338 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1339 */
1340 #define HV_IDE_GUID \
1341 .guid = GUID_INIT(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1342 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1343
1344 /*
1345 * SCSI GUID
1346 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1347 */
1348 #define HV_SCSI_GUID \
1349 .guid = GUID_INIT(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1350 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1351
1352 /*
1353 * Shutdown GUID
1354 * {0e0b6031-5213-4934-818b-38d90ced39db}
1355 */
1356 #define HV_SHUTDOWN_GUID \
1357 .guid = GUID_INIT(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1358 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1359
1360 /*
1361 * Time Synch GUID
1362 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1363 */
1364 #define HV_TS_GUID \
1365 .guid = GUID_INIT(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1366 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1367
1368 /*
1369 * Heartbeat GUID
1370 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1371 */
1372 #define HV_HEART_BEAT_GUID \
1373 .guid = GUID_INIT(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1374 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1375
1376 /*
1377 * KVP GUID
1378 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1379 */
1380 #define HV_KVP_GUID \
1381 .guid = GUID_INIT(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1382 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1383
1384 /*
1385 * Dynamic memory GUID
1386 * {525074dc-8985-46e2-8057-a307dc18a502}
1387 */
1388 #define HV_DM_GUID \
1389 .guid = GUID_INIT(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1390 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1391
1392 /*
1393 * Mouse GUID
1394 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1395 */
1396 #define HV_MOUSE_GUID \
1397 .guid = GUID_INIT(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1398 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1399
1400 /*
1401 * Keyboard GUID
1402 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1403 */
1404 #define HV_KBD_GUID \
1405 .guid = GUID_INIT(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1406 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1407
1408 /*
1409 * VSS (Backup/Restore) GUID
1410 */
1411 #define HV_VSS_GUID \
1412 .guid = GUID_INIT(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1413 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1414 /*
1415 * Synthetic Video GUID
1416 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1417 */
1418 #define HV_SYNTHVID_GUID \
1419 .guid = GUID_INIT(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1420 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1421
1422 /*
1423 * Synthetic FC GUID
1424 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1425 */
1426 #define HV_SYNTHFC_GUID \
1427 .guid = GUID_INIT(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1428 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1429
1430 /*
1431 * Guest File Copy Service
1432 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1433 */
1434
1435 #define HV_FCOPY_GUID \
1436 .guid = GUID_INIT(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1437 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1438
1439 /*
1440 * NetworkDirect. This is the guest RDMA service.
1441 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1442 */
1443 #define HV_ND_GUID \
1444 .guid = GUID_INIT(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1445 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1446
1447 /*
1448 * PCI Express Pass Through
1449 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1450 */
1451
1452 #define HV_PCIE_GUID \
1453 .guid = GUID_INIT(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1454 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1455
1456 /*
1457 * Linux doesn't support the 3 devices: the first two are for
1458 * Automatic Virtual Machine Activation, and the third is for
1459 * Remote Desktop Virtualization.
1460 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1461 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1462 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1463 */
1464
1465 #define HV_AVMA1_GUID \
1466 .guid = GUID_INIT(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1467 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1468
1469 #define HV_AVMA2_GUID \
1470 .guid = GUID_INIT(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1471 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1472
1473 #define HV_RDV_GUID \
1474 .guid = GUID_INIT(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1475 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1476
1477 /*
1478 * Common header for Hyper-V ICs
1479 */
1480
1481 #define ICMSGTYPE_NEGOTIATE 0
1482 #define ICMSGTYPE_HEARTBEAT 1
1483 #define ICMSGTYPE_KVPEXCHANGE 2
1484 #define ICMSGTYPE_SHUTDOWN 3
1485 #define ICMSGTYPE_TIMESYNC 4
1486 #define ICMSGTYPE_VSS 5
1487 #define ICMSGTYPE_FCOPY 7
1488
1489 #define ICMSGHDRFLAG_TRANSACTION 1
1490 #define ICMSGHDRFLAG_REQUEST 2
1491 #define ICMSGHDRFLAG_RESPONSE 4
1492
1493
1494 /*
1495 * While we want to handle util services as regular devices,
1496 * there is only one instance of each of these services; so
1497 * we statically allocate the service specific state.
1498 */
1499
1500 struct hv_util_service {
1501 u8 *recv_buffer;
1502 void *channel;
1503 void (*util_cb)(void *);
1504 int (*util_init)(struct hv_util_service *);
1505 void (*util_deinit)(void);
1506 int (*util_pre_suspend)(void);
1507 int (*util_pre_resume)(void);
1508 };
1509
1510 struct vmbuspipe_hdr {
1511 u32 flags;
1512 u32 msgsize;
1513 } __packed;
1514
1515 struct ic_version {
1516 u16 major;
1517 u16 minor;
1518 } __packed;
1519
1520 struct icmsg_hdr {
1521 struct ic_version icverframe;
1522 u16 icmsgtype;
1523 struct ic_version icvermsg;
1524 u16 icmsgsize;
1525 u32 status;
1526 u8 ictransaction_id;
1527 u8 icflags;
1528 u8 reserved[2];
1529 } __packed;
1530
1531 #define IC_VERSION_NEGOTIATION_MAX_VER_COUNT 100
1532 #define ICMSG_HDR (sizeof(struct vmbuspipe_hdr) + sizeof(struct icmsg_hdr))
1533 #define ICMSG_NEGOTIATE_PKT_SIZE(icframe_vercnt, icmsg_vercnt) \
1534 (ICMSG_HDR + sizeof(struct icmsg_negotiate) + \
1535 (((icframe_vercnt) + (icmsg_vercnt)) * sizeof(struct ic_version)))
1536
1537 struct icmsg_negotiate {
1538 u16 icframe_vercnt;
1539 u16 icmsg_vercnt;
1540 u32 reserved;
1541 struct ic_version icversion_data[]; /* any size array */
1542 } __packed;
1543
1544 struct shutdown_msg_data {
1545 u32 reason_code;
1546 u32 timeout_seconds;
1547 u32 flags;
1548 u8 display_message[2048];
1549 } __packed;
1550
1551 struct heartbeat_msg_data {
1552 u64 seq_num;
1553 u32 reserved[8];
1554 } __packed;
1555
1556 /* Time Sync IC defs */
1557 #define ICTIMESYNCFLAG_PROBE 0
1558 #define ICTIMESYNCFLAG_SYNC 1
1559 #define ICTIMESYNCFLAG_SAMPLE 2
1560
1561 #ifdef __x86_64__
1562 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1563 #else
1564 #define WLTIMEDELTA 116444736000000000LL
1565 #endif
1566
1567 struct ictimesync_data {
1568 u64 parenttime;
1569 u64 childtime;
1570 u64 roundtriptime;
1571 u8 flags;
1572 } __packed;
1573
1574 struct ictimesync_ref_data {
1575 u64 parenttime;
1576 u64 vmreferencetime;
1577 u8 flags;
1578 char leapflags;
1579 char stratum;
1580 u8 reserved[3];
1581 } __packed;
1582
1583 struct hyperv_service_callback {
1584 u8 msg_type;
1585 char *log_msg;
1586 guid_t data;
1587 struct vmbus_channel *channel;
1588 void (*callback)(void *context);
1589 };
1590
1591 #define MAX_SRV_VER 0x7ffffff
1592 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf, u32 buflen,
1593 const int *fw_version, int fw_vercnt,
1594 const int *srv_version, int srv_vercnt,
1595 int *nego_fw_version, int *nego_srv_version);
1596
1597 void hv_process_channel_removal(struct vmbus_channel *channel);
1598
1599 void vmbus_setevent(struct vmbus_channel *channel);
1600 /*
1601 * Negotiated version with the Host.
1602 */
1603
1604 extern __u32 vmbus_proto_version;
1605
1606 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
1607 const guid_t *shv_host_servie_id);
1608 int vmbus_send_modifychannel(struct vmbus_channel *channel, u32 target_vp);
1609 void vmbus_set_event(struct vmbus_channel *channel);
1610
1611 /* Get the start of the ring buffer. */
1612 static inline void *
hv_get_ring_buffer(const struct hv_ring_buffer_info * ring_info)1613 hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info)
1614 {
1615 return ring_info->ring_buffer->buffer;
1616 }
1617
1618 /*
1619 * Mask off host interrupt callback notifications
1620 */
hv_begin_read(struct hv_ring_buffer_info * rbi)1621 static inline void hv_begin_read(struct hv_ring_buffer_info *rbi)
1622 {
1623 rbi->ring_buffer->interrupt_mask = 1;
1624
1625 /* make sure mask update is not reordered */
1626 virt_mb();
1627 }
1628
1629 /*
1630 * Re-enable host callback and return number of outstanding bytes
1631 */
hv_end_read(struct hv_ring_buffer_info * rbi)1632 static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi)
1633 {
1634
1635 rbi->ring_buffer->interrupt_mask = 0;
1636
1637 /* make sure mask update is not reordered */
1638 virt_mb();
1639
1640 /*
1641 * Now check to see if the ring buffer is still empty.
1642 * If it is not, we raced and we need to process new
1643 * incoming messages.
1644 */
1645 return hv_get_bytes_to_read(rbi);
1646 }
1647
1648 /*
1649 * An API to support in-place processing of incoming VMBUS packets.
1650 */
1651
1652 /* Get data payload associated with descriptor */
hv_pkt_data(const struct vmpacket_descriptor * desc)1653 static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc)
1654 {
1655 return (void *)((unsigned long)desc + (desc->offset8 << 3));
1656 }
1657
1658 /* Get data size associated with descriptor */
hv_pkt_datalen(const struct vmpacket_descriptor * desc)1659 static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc)
1660 {
1661 return (desc->len8 << 3) - (desc->offset8 << 3);
1662 }
1663
1664
1665 struct vmpacket_descriptor *
1666 hv_pkt_iter_first(struct vmbus_channel *channel);
1667
1668 struct vmpacket_descriptor *
1669 __hv_pkt_iter_next(struct vmbus_channel *channel,
1670 const struct vmpacket_descriptor *pkt);
1671
1672 void hv_pkt_iter_close(struct vmbus_channel *channel);
1673
1674 /*
1675 * Get next packet descriptor from iterator
1676 * If at end of list, return NULL and update host.
1677 */
1678 static inline struct vmpacket_descriptor *
hv_pkt_iter_next(struct vmbus_channel * channel,const struct vmpacket_descriptor * pkt)1679 hv_pkt_iter_next(struct vmbus_channel *channel,
1680 const struct vmpacket_descriptor *pkt)
1681 {
1682 struct vmpacket_descriptor *nxt;
1683
1684 nxt = __hv_pkt_iter_next(channel, pkt);
1685 if (!nxt)
1686 hv_pkt_iter_close(channel);
1687
1688 return nxt;
1689 }
1690
1691 #define foreach_vmbus_pkt(pkt, channel) \
1692 for (pkt = hv_pkt_iter_first(channel); pkt; \
1693 pkt = hv_pkt_iter_next(channel, pkt))
1694
1695 /*
1696 * Interface for passing data between SR-IOV PF and VF drivers. The VF driver
1697 * sends requests to read and write blocks. Each block must be 128 bytes or
1698 * smaller. Optionally, the VF driver can register a callback function which
1699 * will be invoked when the host says that one or more of the first 64 block
1700 * IDs is "invalid" which means that the VF driver should reread them.
1701 */
1702 #define HV_CONFIG_BLOCK_SIZE_MAX 128
1703
1704 int hyperv_read_cfg_blk(struct pci_dev *dev, void *buf, unsigned int buf_len,
1705 unsigned int block_id, unsigned int *bytes_returned);
1706 int hyperv_write_cfg_blk(struct pci_dev *dev, void *buf, unsigned int len,
1707 unsigned int block_id);
1708 int hyperv_reg_block_invalidate(struct pci_dev *dev, void *context,
1709 void (*block_invalidate)(void *context,
1710 u64 block_mask));
1711
1712 struct hyperv_pci_block_ops {
1713 int (*read_block)(struct pci_dev *dev, void *buf, unsigned int buf_len,
1714 unsigned int block_id, unsigned int *bytes_returned);
1715 int (*write_block)(struct pci_dev *dev, void *buf, unsigned int len,
1716 unsigned int block_id);
1717 int (*reg_blk_invalidate)(struct pci_dev *dev, void *context,
1718 void (*block_invalidate)(void *context,
1719 u64 block_mask));
1720 };
1721
1722 extern struct hyperv_pci_block_ops hvpci_block_ops;
1723
virt_to_hvpfn(void * addr)1724 static inline unsigned long virt_to_hvpfn(void *addr)
1725 {
1726 phys_addr_t paddr;
1727
1728 if (is_vmalloc_addr(addr))
1729 paddr = page_to_phys(vmalloc_to_page(addr)) +
1730 offset_in_page(addr);
1731 else
1732 paddr = __pa(addr);
1733
1734 return paddr >> HV_HYP_PAGE_SHIFT;
1735 }
1736
1737 #define NR_HV_HYP_PAGES_IN_PAGE (PAGE_SIZE / HV_HYP_PAGE_SIZE)
1738 #define offset_in_hvpage(ptr) ((unsigned long)(ptr) & ~HV_HYP_PAGE_MASK)
1739 #define HVPFN_UP(x) (((x) + HV_HYP_PAGE_SIZE-1) >> HV_HYP_PAGE_SHIFT)
1740 #define HVPFN_DOWN(x) ((x) >> HV_HYP_PAGE_SHIFT)
1741 #define page_to_hvpfn(page) (page_to_pfn(page) * NR_HV_HYP_PAGES_IN_PAGE)
1742
1743 #endif /* _HYPERV_H */
1744