1 /* SPDX-License-Identifier: GPL-2.0-only */
4 * Copyright (c) 2011, Microsoft Corporation.
15 #include <uapi/linux/hyperv.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>
29 #define MAX_PAGE_BUFFER_COUNT 32
30 #define MAX_MULTIPAGE_BUFFER_COUNT 32 /* 128K */
35 * Types for GPADL, decides is how GPADL header is created.
37 * It doesn't make much difference between BUFFER and RING if PAGE_SIZE is the
38 * same as HV_HYP_PAGE_SIZE.
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:
48 * gva: |-- 64k --|-- 64k --| ... |
49 * gpa: | 4k | 4k | ... | 4k | 4k | 4k | ... | 4k |
50 * index: 0 1 2 15 16 17 18 .. 31 32 ...
51 * | | ... | | | ... | ...
53 * gpadl: | 4k | 4k | ... | 4k | 4k | 4k | ... | 4k | ... |
54 * index: 0 1 2 ... 15 16 17 18 .. 31 32 ...
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
64 * | / / ... / ... | / ... /
68 * gpadl: | 4k | 4k | ... | ... | 4k | 4k | ... |
69 * index: 0 1 2 ... 16 ... n-15 n-14 n-13 ... 2n-30
76 /* Single-page buffer */
77 struct hv_page_buffer {
83 /* Multiple-page buffer */
84 struct hv_multipage_buffer {
85 /* Length and Offset determines the # of pfns in the array */
88 u64 pfn_array[MAX_MULTIPAGE_BUFFER_COUNT];
92 * Multiple-page buffer array; the pfn array is variable size:
93 * The number of entries in the PFN array is determined by
97 /* Length and Offset determines the # of pfns in the array */
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))
113 struct hv_ring_buffer {
114 /* Offset in bytes from the start of ring data below */
117 /* Offset in bytes from the start of ring data below */
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.
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
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
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.
152 u32 feat_pending_send_sz:1;
157 /* Pad it to PAGE_SIZE so that data starts on page boundary */
158 u8 reserved2[PAGE_SIZE - 68];
161 * Ring data starts here + RingDataStartOffset
162 * !!! DO NOT place any fields below this !!!
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) + \
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;
177 u32 ring_datasize; /* < ring_size */
180 * The ring buffer mutex lock. This lock prevents the ring buffer from
181 * being freed while the ring buffer is being accessed.
183 struct mutex ring_buffer_mutex;
187 static inline u32 hv_get_bytes_to_read(const struct hv_ring_buffer_info *rbi)
189 u32 read_loc, write_loc, dsize, read;
191 dsize = rbi->ring_datasize;
192 read_loc = rbi->ring_buffer->read_index;
193 write_loc = READ_ONCE(rbi->ring_buffer->write_index);
195 read = write_loc >= read_loc ? (write_loc - read_loc) :
196 (dsize - read_loc) + write_loc;
201 static inline u32 hv_get_bytes_to_write(const struct hv_ring_buffer_info *rbi)
203 u32 read_loc, write_loc, dsize, write;
205 dsize = rbi->ring_datasize;
206 read_loc = READ_ONCE(rbi->ring_buffer->read_index);
207 write_loc = rbi->ring_buffer->write_index;
209 write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
210 read_loc - write_loc;
214 static inline u32 hv_get_avail_to_write_percent(
215 const struct hv_ring_buffer_info *rbi)
217 u32 avail_write = hv_get_bytes_to_write(rbi);
219 return reciprocal_divide(
220 (avail_write << 3) + (avail_write << 1),
221 rbi->ring_size_div10_reciprocal);
225 * VMBUS version is 32 bit entity broken up into
226 * two 16 bit quantities: major_number. minor_number.
228 * 0 . 13 (Windows Server 2008)
231 * 3 . 0 (Windows 8 R2)
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)
239 #define VERSION_WS2008 ((0 << 16) | (13))
240 #define VERSION_WIN7 ((1 << 16) | (1))
241 #define VERSION_WIN8 ((2 << 16) | (4))
242 #define VERSION_WIN8_1 ((3 << 16) | (0))
243 #define VERSION_WIN10 ((4 << 16) | (0))
244 #define VERSION_WIN10_V4_1 ((4 << 16) | (1))
245 #define VERSION_WIN10_V5 ((5 << 16) | (0))
246 #define VERSION_WIN10_V5_1 ((5 << 16) | (1))
247 #define VERSION_WIN10_V5_2 ((5 << 16) | (2))
249 /* Make maximum size of pipe payload of 16K */
250 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
252 /* Define PipeMode values. */
253 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
254 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
256 /* The size of the user defined data buffer for non-pipe offers. */
257 #define MAX_USER_DEFINED_BYTES 120
259 /* The size of the user defined data buffer for pipe offers. */
260 #define MAX_PIPE_USER_DEFINED_BYTES 116
263 * At the center of the Channel Management library is the Channel Offer. This
264 * struct contains the fundamental information about an offer.
266 struct vmbus_channel_offer {
271 * These two fields are not currently used.
277 u16 mmio_megabytes; /* in bytes * 1024 * 1024 */
280 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
282 unsigned char user_def[MAX_USER_DEFINED_BYTES];
287 * The following sructure is an integrated pipe protocol, which
288 * is implemented on top of standard user-defined data. Pipe
289 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
294 unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
298 * The sub_channel_index is defined in Win8: a value of zero means a
299 * primary channel and a value of non-zero means a sub-channel.
301 * Before Win8, the field is reserved, meaning it's always zero.
303 u16 sub_channel_index;
308 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
309 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
310 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
311 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
312 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
313 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
314 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
315 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
317 struct vmpacket_descriptor {
325 struct vmpacket_header {
326 u32 prev_pkt_start_offset;
327 struct vmpacket_descriptor descriptor;
330 struct vmtransfer_page_range {
335 struct vmtransfer_page_packet_header {
336 struct vmpacket_descriptor d;
341 struct vmtransfer_page_range ranges[1];
344 struct vmgpadl_packet_header {
345 struct vmpacket_descriptor d;
350 struct vmadd_remove_transfer_page_set {
351 struct vmpacket_descriptor d;
358 * This structure defines a range in guest physical space that can be made to
359 * look virtually contiguous.
368 * This is the format for an Establish Gpadl packet, which contains a handle by
369 * which this GPADL will be known and a set of GPA ranges associated with it.
370 * This can be converted to a MDL by the guest OS. If there are multiple GPA
371 * ranges, then the resulting MDL will be "chained," representing multiple VA
374 struct vmestablish_gpadl {
375 struct vmpacket_descriptor d;
378 struct gpa_range range[1];
382 * This is the format for a Teardown Gpadl packet, which indicates that the
383 * GPADL handle in the Establish Gpadl packet will never be referenced again.
385 struct vmteardown_gpadl {
386 struct vmpacket_descriptor d;
388 u32 reserved; /* for alignment to a 8-byte boundary */
392 * This is the format for a GPA-Direct packet, which contains a set of GPA
393 * ranges, in addition to commands and/or data.
395 struct vmdata_gpa_direct {
396 struct vmpacket_descriptor d;
399 struct gpa_range range[1];
402 /* This is the format for a Additional Data Packet. */
403 struct vmadditional_data {
404 struct vmpacket_descriptor d;
408 unsigned char data[1];
411 union vmpacket_largest_possible_header {
412 struct vmpacket_descriptor simple_hdr;
413 struct vmtransfer_page_packet_header xfer_page_hdr;
414 struct vmgpadl_packet_header gpadl_hdr;
415 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr;
416 struct vmestablish_gpadl establish_gpadl_hdr;
417 struct vmteardown_gpadl teardown_gpadl_hdr;
418 struct vmdata_gpa_direct data_gpa_direct_hdr;
421 #define VMPACKET_DATA_START_ADDRESS(__packet) \
422 (void *)(((unsigned char *)__packet) + \
423 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
425 #define VMPACKET_DATA_LENGTH(__packet) \
426 ((((struct vmpacket_descriptor)__packet)->len8 - \
427 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
429 #define VMPACKET_TRANSFER_MODE(__packet) \
430 (((struct IMPACT)__packet)->type)
432 enum vmbus_packet_type {
433 VM_PKT_INVALID = 0x0,
435 VM_PKT_ADD_XFER_PAGESET = 0x2,
436 VM_PKT_RM_XFER_PAGESET = 0x3,
437 VM_PKT_ESTABLISH_GPADL = 0x4,
438 VM_PKT_TEARDOWN_GPADL = 0x5,
439 VM_PKT_DATA_INBAND = 0x6,
440 VM_PKT_DATA_USING_XFER_PAGES = 0x7,
441 VM_PKT_DATA_USING_GPADL = 0x8,
442 VM_PKT_DATA_USING_GPA_DIRECT = 0x9,
443 VM_PKT_CANCEL_REQUEST = 0xa,
445 VM_PKT_DATA_USING_ADDITIONAL_PKT = 0xc,
446 VM_PKT_ADDITIONAL_DATA = 0xd
449 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
452 /* Version 1 messages */
453 enum vmbus_channel_message_type {
454 CHANNELMSG_INVALID = 0,
455 CHANNELMSG_OFFERCHANNEL = 1,
456 CHANNELMSG_RESCIND_CHANNELOFFER = 2,
457 CHANNELMSG_REQUESTOFFERS = 3,
458 CHANNELMSG_ALLOFFERS_DELIVERED = 4,
459 CHANNELMSG_OPENCHANNEL = 5,
460 CHANNELMSG_OPENCHANNEL_RESULT = 6,
461 CHANNELMSG_CLOSECHANNEL = 7,
462 CHANNELMSG_GPADL_HEADER = 8,
463 CHANNELMSG_GPADL_BODY = 9,
464 CHANNELMSG_GPADL_CREATED = 10,
465 CHANNELMSG_GPADL_TEARDOWN = 11,
466 CHANNELMSG_GPADL_TORNDOWN = 12,
467 CHANNELMSG_RELID_RELEASED = 13,
468 CHANNELMSG_INITIATE_CONTACT = 14,
469 CHANNELMSG_VERSION_RESPONSE = 15,
470 CHANNELMSG_UNLOAD = 16,
471 CHANNELMSG_UNLOAD_RESPONSE = 17,
475 CHANNELMSG_TL_CONNECT_REQUEST = 21,
476 CHANNELMSG_MODIFYCHANNEL = 22,
477 CHANNELMSG_TL_CONNECT_RESULT = 23,
481 /* Hyper-V supports about 2048 channels, and the RELIDs start with 1. */
482 #define INVALID_RELID U32_MAX
484 struct vmbus_channel_message_header {
485 enum vmbus_channel_message_type msgtype;
489 /* Query VMBus Version parameters */
490 struct vmbus_channel_query_vmbus_version {
491 struct vmbus_channel_message_header header;
495 /* VMBus Version Supported parameters */
496 struct vmbus_channel_version_supported {
497 struct vmbus_channel_message_header header;
498 u8 version_supported;
501 /* Offer Channel parameters */
502 struct vmbus_channel_offer_channel {
503 struct vmbus_channel_message_header header;
504 struct vmbus_channel_offer offer;
508 * win7 and beyond splits this field into a bit field.
510 u8 monitor_allocated:1;
513 * These are new fields added in win7 and later.
514 * Do not access these fields without checking the
515 * negotiated protocol.
517 * If "is_dedicated_interrupt" is set, we must not set the
518 * associated bit in the channel bitmap while sending the
519 * interrupt to the host.
521 * connection_id is to be used in signaling the host.
523 u16 is_dedicated_interrupt:1;
528 /* Rescind Offer parameters */
529 struct vmbus_channel_rescind_offer {
530 struct vmbus_channel_message_header header;
535 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info *rbi)
537 return rbi->ring_buffer->pending_send_sz;
541 * Request Offer -- no parameters, SynIC message contains the partition ID
542 * Set Snoop -- no parameters, SynIC message contains the partition ID
543 * Clear Snoop -- no parameters, SynIC message contains the partition ID
544 * All Offers Delivered -- no parameters, SynIC message contains the partition
546 * Flush Client -- no parameters, SynIC message contains the partition ID
549 /* Open Channel parameters */
550 struct vmbus_channel_open_channel {
551 struct vmbus_channel_message_header header;
553 /* Identifies the specific VMBus channel that is being opened. */
556 /* ID making a particular open request at a channel offer unique. */
559 /* GPADL for the channel's ring buffer. */
560 u32 ringbuffer_gpadlhandle;
563 * Starting with win8, this field will be used to specify
564 * the target virtual processor on which to deliver the interrupt for
565 * the host to guest communication.
566 * Prior to win8, incoming channel interrupts would only
567 * be delivered on cpu 0. Setting this value to 0 would
568 * preserve the earlier behavior.
573 * The upstream ring buffer begins at offset zero in the memory
574 * described by RingBufferGpadlHandle. The downstream ring buffer
575 * follows it at this offset (in pages).
577 u32 downstream_ringbuffer_pageoffset;
579 /* User-specific data to be passed along to the server endpoint. */
580 unsigned char userdata[MAX_USER_DEFINED_BYTES];
583 /* Open Channel Result parameters */
584 struct vmbus_channel_open_result {
585 struct vmbus_channel_message_header header;
591 /* Close channel parameters; */
592 struct vmbus_channel_close_channel {
593 struct vmbus_channel_message_header header;
597 /* Channel Message GPADL */
598 #define GPADL_TYPE_RING_BUFFER 1
599 #define GPADL_TYPE_SERVER_SAVE_AREA 2
600 #define GPADL_TYPE_TRANSACTION 8
603 * The number of PFNs in a GPADL message is defined by the number of
604 * pages that would be spanned by ByteCount and ByteOffset. If the
605 * implied number of PFNs won't fit in this packet, there will be a
606 * follow-up packet that contains more.
608 struct vmbus_channel_gpadl_header {
609 struct vmbus_channel_message_header header;
614 struct gpa_range range[];
617 /* This is the followup packet that contains more PFNs. */
618 struct vmbus_channel_gpadl_body {
619 struct vmbus_channel_message_header header;
625 struct vmbus_channel_gpadl_created {
626 struct vmbus_channel_message_header header;
632 struct vmbus_channel_gpadl_teardown {
633 struct vmbus_channel_message_header header;
638 struct vmbus_channel_gpadl_torndown {
639 struct vmbus_channel_message_header header;
643 struct vmbus_channel_relid_released {
644 struct vmbus_channel_message_header header;
648 struct vmbus_channel_initiate_contact {
649 struct vmbus_channel_message_header header;
650 u32 vmbus_version_requested;
651 u32 target_vcpu; /* The VCPU the host should respond to */
664 /* Hyper-V socket: guest's connect()-ing to host */
665 struct vmbus_channel_tl_connect_request {
666 struct vmbus_channel_message_header header;
667 guid_t guest_endpoint_id;
668 guid_t host_service_id;
671 /* Modify Channel parameters, cf. vmbus_send_modifychannel() */
672 struct vmbus_channel_modifychannel {
673 struct vmbus_channel_message_header header;
678 struct vmbus_channel_version_response {
679 struct vmbus_channel_message_header header;
680 u8 version_supported;
686 * On new hosts that support VMBus protocol 5.0, we must use
687 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
688 * and for subsequent messages, we must use the Message Connection ID
689 * field in the host-returned Version Response Message.
691 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
696 enum vmbus_channel_state {
698 CHANNEL_OPENING_STATE,
700 CHANNEL_OPENED_STATE,
704 * Represents each channel msg on the vmbus connection This is a
705 * variable-size data structure depending on the msg type itself
707 struct vmbus_channel_msginfo {
708 /* Bookkeeping stuff */
709 struct list_head msglistentry;
711 /* So far, this is only used to handle gpadl body message */
712 struct list_head submsglist;
714 /* Synchronize the request/response if needed */
715 struct completion waitevent;
716 struct vmbus_channel *waiting_channel;
718 struct vmbus_channel_version_supported version_supported;
719 struct vmbus_channel_open_result open_result;
720 struct vmbus_channel_gpadl_torndown gpadl_torndown;
721 struct vmbus_channel_gpadl_created gpadl_created;
722 struct vmbus_channel_version_response version_response;
727 * The channel message that goes out on the "wire".
728 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
733 struct vmbus_close_msg {
734 struct vmbus_channel_msginfo info;
735 struct vmbus_channel_close_channel msg;
738 /* Define connection identifier type. */
739 union hv_connection_id {
747 enum vmbus_device_type {
767 struct vmbus_device {
773 struct vmbus_channel {
774 struct list_head listentry;
776 struct hv_device *device_obj;
778 enum vmbus_channel_state state;
780 struct vmbus_channel_offer_channel offermsg;
782 * These are based on the OfferMsg.MonitorId.
783 * Save it here for easy access.
788 bool rescind; /* got rescind msg */
789 struct completion rescind_event;
791 u32 ringbuffer_gpadlhandle;
793 /* Allocated memory for ring buffer */
794 struct page *ringbuffer_page;
795 u32 ringbuffer_pagecount;
796 u32 ringbuffer_send_offset;
797 struct hv_ring_buffer_info outbound; /* send to parent */
798 struct hv_ring_buffer_info inbound; /* receive from parent */
800 struct vmbus_close_msg close_msg;
803 u64 interrupts; /* Host to Guest interrupts */
804 u64 sig_events; /* Guest to Host events */
807 * Guest to host interrupts caused by the outbound ring buffer changing
808 * from empty to not empty.
813 * Indicates that a full outbound ring buffer was encountered. The flag
814 * is set to true when a full outbound ring buffer is encountered and
815 * set to false when a write to the outbound ring buffer is completed.
819 /* Channel callback's invoked in softirq context */
820 struct tasklet_struct callback_event;
821 void (*onchannel_callback)(void *context);
822 void *channel_callback_context;
824 void (*change_target_cpu_callback)(struct vmbus_channel *channel,
828 * Synchronize channel scheduling and channel removal; see the inline
829 * comments in vmbus_chan_sched() and vmbus_reset_channel_cb().
831 spinlock_t sched_lock;
834 * A channel can be marked for one of three modes of reading:
835 * BATCHED - callback called from taslket and should read
836 * channel until empty. Interrupts from the host
837 * are masked while read is in process (default).
838 * DIRECT - callback called from tasklet (softirq).
839 * ISR - callback called in interrupt context and must
840 * invoke its own deferred processing.
841 * Host interrupts are disabled and must be re-enabled
842 * when ring is empty.
844 enum hv_callback_mode {
850 bool is_dedicated_interrupt;
854 * Starting with win8, this field will be used to specify the
855 * target CPU on which to deliver the interrupt for the host
856 * to guest communication.
858 * Prior to win8, incoming channel interrupts would only be
859 * delivered on CPU 0. Setting this value to 0 would preserve
860 * the earlier behavior.
864 * Support for sub-channels. For high performance devices,
865 * it will be useful to have multiple sub-channels to support
866 * a scalable communication infrastructure with the host.
867 * The support for sub-channels is implemented as an extention
868 * to the current infrastructure.
869 * The initial offer is considered the primary channel and this
870 * offer message will indicate if the host supports sub-channels.
871 * The guest is free to ask for sub-channels to be offerred and can
872 * open these sub-channels as a normal "primary" channel. However,
873 * all sub-channels will have the same type and instance guids as the
874 * primary channel. Requests sent on a given channel will result in a
875 * response on the same channel.
879 * Sub-channel creation callback. This callback will be called in
880 * process context when a sub-channel offer is received from the host.
881 * The guest can open the sub-channel in the context of this callback.
883 void (*sc_creation_callback)(struct vmbus_channel *new_sc);
886 * Channel rescind callback. Some channels (the hvsock ones), need to
887 * register a callback which is invoked in vmbus_onoffer_rescind().
889 void (*chn_rescind_callback)(struct vmbus_channel *channel);
892 * All Sub-channels of a primary channel are linked here.
894 struct list_head sc_list;
896 * The primary channel this sub-channel belongs to.
897 * This will be NULL for the primary channel.
899 struct vmbus_channel *primary_channel;
901 * Support per-channel state for use by vmbus drivers.
903 void *per_channel_state;
906 * Defer freeing channel until after all cpu's have
907 * gone through grace period.
912 * For sysfs per-channel properties.
917 * For performance critical channels (storage, networking
918 * etc,), Hyper-V has a mechanism to enhance the throughput
919 * at the expense of latency:
920 * When the host is to be signaled, we just set a bit in a shared page
921 * and this bit will be inspected by the hypervisor within a certain
922 * window and if the bit is set, the host will be signaled. The window
923 * of time is the monitor latency - currently around 100 usecs. This
924 * mechanism improves throughput by:
926 * A) Making the host more efficient - each time it wakes up,
927 * potentially it will process morev number of packets. The
928 * monitor latency allows a batch to build up.
929 * B) By deferring the hypercall to signal, we will also minimize
932 * Clearly, these optimizations improve throughput at the expense of
933 * latency. Furthermore, since the channel is shared for both
934 * control and data messages, control messages currently suffer
935 * unnecessary latency adversley impacting performance and boot
936 * time. To fix this issue, permit tagging the channel as being
937 * in "low latency" mode. In this mode, we will bypass the monitor
945 * Cache the device ID here for easy access; this is useful, in
946 * particular, in situations where the channel's device_obj has
947 * not been allocated/initialized yet.
952 * We must offload the handling of the primary/sub channels
953 * from the single-threaded vmbus_connection.work_queue to
954 * two different workqueue, otherwise we can block
955 * vmbus_connection.work_queue and hang: see vmbus_process_offer().
957 struct work_struct add_channel_work;
960 * Guest to host interrupts caused by the inbound ring buffer changing
961 * from full to not full while a packet is waiting.
966 * The total number of write operations that encountered a full
967 * outbound ring buffer.
972 * The number of write operations that were the first to encounter a
973 * full outbound ring buffer.
977 /* enabling/disabling fuzz testing on the channel (default is false)*/
978 bool fuzz_testing_state;
981 * Interrupt delay will delay the guest from emptying the ring buffer
982 * for a specific amount of time. The delay is in microseconds and will
983 * be between 1 to a maximum of 1000, its default is 0 (no delay).
984 * The Message delay will delay guest reading on a per message basis
985 * in microseconds between 1 to 1000 with the default being 0
988 u32 fuzz_testing_interrupt_delay;
989 u32 fuzz_testing_message_delay;
993 static inline bool is_hvsock_channel(const struct vmbus_channel *c)
995 return !!(c->offermsg.offer.chn_flags &
996 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER);
999 static inline bool is_sub_channel(const struct vmbus_channel *c)
1001 return c->offermsg.offer.sub_channel_index != 0;
1004 static inline void set_channel_read_mode(struct vmbus_channel *c,
1005 enum hv_callback_mode mode)
1007 c->callback_mode = mode;
1010 static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
1012 c->per_channel_state = s;
1015 static inline void *get_per_channel_state(struct vmbus_channel *c)
1017 return c->per_channel_state;
1020 static inline void set_channel_pending_send_size(struct vmbus_channel *c,
1023 unsigned long flags;
1026 spin_lock_irqsave(&c->outbound.ring_lock, flags);
1027 ++c->out_full_total;
1029 if (!c->out_full_flag) {
1030 ++c->out_full_first;
1031 c->out_full_flag = true;
1033 spin_unlock_irqrestore(&c->outbound.ring_lock, flags);
1035 c->out_full_flag = false;
1038 c->outbound.ring_buffer->pending_send_sz = size;
1041 static inline void set_low_latency_mode(struct vmbus_channel *c)
1043 c->low_latency = true;
1046 static inline void clear_low_latency_mode(struct vmbus_channel *c)
1048 c->low_latency = false;
1051 void vmbus_onmessage(struct vmbus_channel_message_header *hdr);
1053 int vmbus_request_offers(void);
1056 * APIs for managing sub-channels.
1059 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1060 void (*sc_cr_cb)(struct vmbus_channel *new_sc));
1062 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1063 void (*chn_rescind_cb)(struct vmbus_channel *));
1066 * Check if sub-channels have already been offerred. This API will be useful
1067 * when the driver is unloaded after establishing sub-channels. In this case,
1068 * when the driver is re-loaded, the driver would have to check if the
1069 * subchannels have already been established before attempting to request
1070 * the creation of sub-channels.
1071 * This function returns TRUE to indicate that subchannels have already been
1073 * This function should be invoked after setting the callback function for
1074 * sub-channel creation.
1076 bool vmbus_are_subchannels_present(struct vmbus_channel *primary);
1078 /* The format must be the same as struct vmdata_gpa_direct */
1079 struct vmbus_channel_packet_page_buffer {
1087 struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
1090 /* The format must be the same as struct vmdata_gpa_direct */
1091 struct vmbus_channel_packet_multipage_buffer {
1098 u32 rangecount; /* Always 1 in this case */
1099 struct hv_multipage_buffer range;
1102 /* The format must be the same as struct vmdata_gpa_direct */
1103 struct vmbus_packet_mpb_array {
1110 u32 rangecount; /* Always 1 in this case */
1111 struct hv_mpb_array range;
1114 int vmbus_alloc_ring(struct vmbus_channel *channel,
1115 u32 send_size, u32 recv_size);
1116 void vmbus_free_ring(struct vmbus_channel *channel);
1118 int vmbus_connect_ring(struct vmbus_channel *channel,
1119 void (*onchannel_callback)(void *context),
1121 int vmbus_disconnect_ring(struct vmbus_channel *channel);
1123 extern int vmbus_open(struct vmbus_channel *channel,
1124 u32 send_ringbuffersize,
1125 u32 recv_ringbuffersize,
1128 void (*onchannel_callback)(void *context),
1131 extern void vmbus_close(struct vmbus_channel *channel);
1133 extern int vmbus_sendpacket(struct vmbus_channel *channel,
1137 enum vmbus_packet_type type,
1140 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1141 struct hv_page_buffer pagebuffers[],
1147 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1148 struct vmbus_packet_mpb_array *mpb,
1154 extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
1159 extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
1162 void vmbus_reset_channel_cb(struct vmbus_channel *channel);
1164 extern int vmbus_recvpacket(struct vmbus_channel *channel,
1167 u32 *buffer_actual_len,
1170 extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
1173 u32 *buffer_actual_len,
1177 extern void vmbus_ontimer(unsigned long data);
1179 /* Base driver object */
1184 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1185 * channel flag, actually doesn't mean a synthetic device because the
1186 * offer's if_type/if_instance can change for every new hvsock
1189 * However, to facilitate the notification of new-offer/rescind-offer
1190 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1191 * a special vmbus device, and hence we need the below flag to
1192 * indicate if the driver is the hvsock driver or not: we need to
1193 * specially treat the hvosck offer & driver in vmbus_match().
1197 /* the device type supported by this driver */
1199 const struct hv_vmbus_device_id *id_table;
1201 struct device_driver driver;
1203 /* dynamic device GUID's */
1206 struct list_head list;
1209 int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
1210 int (*remove)(struct hv_device *);
1211 void (*shutdown)(struct hv_device *);
1213 int (*suspend)(struct hv_device *);
1214 int (*resume)(struct hv_device *);
1218 /* Base device object */
1220 /* the device type id of this device */
1223 /* the device instance id of this device */
1224 guid_t dev_instance;
1228 struct device device;
1229 char *driver_override; /* Driver name to force a match */
1231 struct vmbus_channel *channel;
1232 struct kset *channels_kset;
1234 /* place holder to keep track of the dir for hv device in debugfs */
1235 struct dentry *debug_dir;
1240 static inline struct hv_device *device_to_hv_device(struct device *d)
1242 return container_of(d, struct hv_device, device);
1245 static inline struct hv_driver *drv_to_hv_drv(struct device_driver *d)
1247 return container_of(d, struct hv_driver, driver);
1250 static inline void hv_set_drvdata(struct hv_device *dev, void *data)
1252 dev_set_drvdata(&dev->device, data);
1255 static inline void *hv_get_drvdata(struct hv_device *dev)
1257 return dev_get_drvdata(&dev->device);
1260 struct hv_ring_buffer_debug_info {
1261 u32 current_interrupt_mask;
1262 u32 current_read_index;
1263 u32 current_write_index;
1264 u32 bytes_avail_toread;
1265 u32 bytes_avail_towrite;
1269 int hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
1270 struct hv_ring_buffer_debug_info *debug_info);
1272 /* Vmbus interface */
1273 #define vmbus_driver_register(driver) \
1274 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1275 int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
1276 struct module *owner,
1277 const char *mod_name);
1278 void vmbus_driver_unregister(struct hv_driver *hv_driver);
1280 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel);
1282 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1283 resource_size_t min, resource_size_t max,
1284 resource_size_t size, resource_size_t align,
1285 bool fb_overlap_ok);
1286 void vmbus_free_mmio(resource_size_t start, resource_size_t size);
1289 * GUID definitions of various offer types - services offered to the guest.
1294 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1296 #define HV_NIC_GUID \
1297 .guid = GUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1298 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1302 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1304 #define HV_IDE_GUID \
1305 .guid = GUID_INIT(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1306 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1310 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1312 #define HV_SCSI_GUID \
1313 .guid = GUID_INIT(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1314 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1318 * {0e0b6031-5213-4934-818b-38d90ced39db}
1320 #define HV_SHUTDOWN_GUID \
1321 .guid = GUID_INIT(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1322 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1326 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1328 #define HV_TS_GUID \
1329 .guid = GUID_INIT(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1330 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1334 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1336 #define HV_HEART_BEAT_GUID \
1337 .guid = GUID_INIT(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1338 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1342 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1344 #define HV_KVP_GUID \
1345 .guid = GUID_INIT(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1346 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1349 * Dynamic memory GUID
1350 * {525074dc-8985-46e2-8057-a307dc18a502}
1352 #define HV_DM_GUID \
1353 .guid = GUID_INIT(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1354 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1358 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1360 #define HV_MOUSE_GUID \
1361 .guid = GUID_INIT(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1362 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1366 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1368 #define HV_KBD_GUID \
1369 .guid = GUID_INIT(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1370 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1373 * VSS (Backup/Restore) GUID
1375 #define HV_VSS_GUID \
1376 .guid = GUID_INIT(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1377 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1379 * Synthetic Video GUID
1380 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1382 #define HV_SYNTHVID_GUID \
1383 .guid = GUID_INIT(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1384 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1388 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1390 #define HV_SYNTHFC_GUID \
1391 .guid = GUID_INIT(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1392 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1395 * Guest File Copy Service
1396 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1399 #define HV_FCOPY_GUID \
1400 .guid = GUID_INIT(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1401 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1404 * NetworkDirect. This is the guest RDMA service.
1405 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1407 #define HV_ND_GUID \
1408 .guid = GUID_INIT(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1409 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1412 * PCI Express Pass Through
1413 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1416 #define HV_PCIE_GUID \
1417 .guid = GUID_INIT(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1418 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1421 * Linux doesn't support the 3 devices: the first two are for
1422 * Automatic Virtual Machine Activation, and the third is for
1423 * Remote Desktop Virtualization.
1424 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1425 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1426 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1429 #define HV_AVMA1_GUID \
1430 .guid = GUID_INIT(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1431 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1433 #define HV_AVMA2_GUID \
1434 .guid = GUID_INIT(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1435 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1437 #define HV_RDV_GUID \
1438 .guid = GUID_INIT(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1439 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1442 * Common header for Hyper-V ICs
1445 #define ICMSGTYPE_NEGOTIATE 0
1446 #define ICMSGTYPE_HEARTBEAT 1
1447 #define ICMSGTYPE_KVPEXCHANGE 2
1448 #define ICMSGTYPE_SHUTDOWN 3
1449 #define ICMSGTYPE_TIMESYNC 4
1450 #define ICMSGTYPE_VSS 5
1452 #define ICMSGHDRFLAG_TRANSACTION 1
1453 #define ICMSGHDRFLAG_REQUEST 2
1454 #define ICMSGHDRFLAG_RESPONSE 4
1458 * While we want to handle util services as regular devices,
1459 * there is only one instance of each of these services; so
1460 * we statically allocate the service specific state.
1463 struct hv_util_service {
1466 void (*util_cb)(void *);
1467 int (*util_init)(struct hv_util_service *);
1468 void (*util_deinit)(void);
1469 int (*util_pre_suspend)(void);
1470 int (*util_pre_resume)(void);
1473 struct vmbuspipe_hdr {
1484 struct ic_version icverframe;
1486 struct ic_version icvermsg;
1489 u8 ictransaction_id;
1494 struct icmsg_negotiate {
1498 struct ic_version icversion_data[1]; /* any size array */
1501 struct shutdown_msg_data {
1503 u32 timeout_seconds;
1505 u8 display_message[2048];
1508 struct heartbeat_msg_data {
1513 /* Time Sync IC defs */
1514 #define ICTIMESYNCFLAG_PROBE 0
1515 #define ICTIMESYNCFLAG_SYNC 1
1516 #define ICTIMESYNCFLAG_SAMPLE 2
1519 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1521 #define WLTIMEDELTA 116444736000000000LL
1524 struct ictimesync_data {
1531 struct ictimesync_ref_data {
1533 u64 vmreferencetime;
1540 struct hyperv_service_callback {
1544 struct vmbus_channel *channel;
1545 void (*callback)(void *context);
1548 #define MAX_SRV_VER 0x7ffffff
1549 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf,
1550 const int *fw_version, int fw_vercnt,
1551 const int *srv_version, int srv_vercnt,
1552 int *nego_fw_version, int *nego_srv_version);
1554 void hv_process_channel_removal(struct vmbus_channel *channel);
1556 void vmbus_setevent(struct vmbus_channel *channel);
1558 * Negotiated version with the Host.
1561 extern __u32 vmbus_proto_version;
1563 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
1564 const guid_t *shv_host_servie_id);
1565 int vmbus_send_modifychannel(u32 child_relid, u32 target_vp);
1566 void vmbus_set_event(struct vmbus_channel *channel);
1568 /* Get the start of the ring buffer. */
1569 static inline void *
1570 hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info)
1572 return ring_info->ring_buffer->buffer;
1576 * Mask off host interrupt callback notifications
1578 static inline void hv_begin_read(struct hv_ring_buffer_info *rbi)
1580 rbi->ring_buffer->interrupt_mask = 1;
1582 /* make sure mask update is not reordered */
1587 * Re-enable host callback and return number of outstanding bytes
1589 static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi)
1592 rbi->ring_buffer->interrupt_mask = 0;
1594 /* make sure mask update is not reordered */
1598 * Now check to see if the ring buffer is still empty.
1599 * If it is not, we raced and we need to process new
1600 * incoming messages.
1602 return hv_get_bytes_to_read(rbi);
1606 * An API to support in-place processing of incoming VMBUS packets.
1609 /* Get data payload associated with descriptor */
1610 static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc)
1612 return (void *)((unsigned long)desc + (desc->offset8 << 3));
1615 /* Get data size associated with descriptor */
1616 static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc)
1618 return (desc->len8 << 3) - (desc->offset8 << 3);
1622 struct vmpacket_descriptor *
1623 hv_pkt_iter_first(struct vmbus_channel *channel);
1625 struct vmpacket_descriptor *
1626 __hv_pkt_iter_next(struct vmbus_channel *channel,
1627 const struct vmpacket_descriptor *pkt);
1629 void hv_pkt_iter_close(struct vmbus_channel *channel);
1632 * Get next packet descriptor from iterator
1633 * If at end of list, return NULL and update host.
1635 static inline struct vmpacket_descriptor *
1636 hv_pkt_iter_next(struct vmbus_channel *channel,
1637 const struct vmpacket_descriptor *pkt)
1639 struct vmpacket_descriptor *nxt;
1641 nxt = __hv_pkt_iter_next(channel, pkt);
1643 hv_pkt_iter_close(channel);
1648 #define foreach_vmbus_pkt(pkt, channel) \
1649 for (pkt = hv_pkt_iter_first(channel); pkt; \
1650 pkt = hv_pkt_iter_next(channel, pkt))
1653 * Interface for passing data between SR-IOV PF and VF drivers. The VF driver
1654 * sends requests to read and write blocks. Each block must be 128 bytes or
1655 * smaller. Optionally, the VF driver can register a callback function which
1656 * will be invoked when the host says that one or more of the first 64 block
1657 * IDs is "invalid" which means that the VF driver should reread them.
1659 #define HV_CONFIG_BLOCK_SIZE_MAX 128
1661 int hyperv_read_cfg_blk(struct pci_dev *dev, void *buf, unsigned int buf_len,
1662 unsigned int block_id, unsigned int *bytes_returned);
1663 int hyperv_write_cfg_blk(struct pci_dev *dev, void *buf, unsigned int len,
1664 unsigned int block_id);
1665 int hyperv_reg_block_invalidate(struct pci_dev *dev, void *context,
1666 void (*block_invalidate)(void *context,
1669 struct hyperv_pci_block_ops {
1670 int (*read_block)(struct pci_dev *dev, void *buf, unsigned int buf_len,
1671 unsigned int block_id, unsigned int *bytes_returned);
1672 int (*write_block)(struct pci_dev *dev, void *buf, unsigned int len,
1673 unsigned int block_id);
1674 int (*reg_blk_invalidate)(struct pci_dev *dev, void *context,
1675 void (*block_invalidate)(void *context,
1679 extern struct hyperv_pci_block_ops hvpci_block_ops;
1681 static inline unsigned long virt_to_hvpfn(void *addr)
1685 if (is_vmalloc_addr(addr))
1686 paddr = page_to_phys(vmalloc_to_page(addr)) +
1687 offset_in_page(addr);
1691 return paddr >> HV_HYP_PAGE_SHIFT;
1694 #define NR_HV_HYP_PAGES_IN_PAGE (PAGE_SIZE / HV_HYP_PAGE_SIZE)
1695 #define offset_in_hvpage(ptr) ((unsigned long)(ptr) & ~HV_HYP_PAGE_MASK)
1696 #define HVPFN_UP(x) (((x) + HV_HYP_PAGE_SIZE-1) >> HV_HYP_PAGE_SHIFT)
1697 #define page_to_hvpfn(page) (page_to_pfn(page) * NR_HV_HYP_PAGES_IN_PAGE)
1699 #endif /* _HYPERV_H */