3 * Copyright (c) 2011, Microsoft Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
16 * Place - Suite 330, Boston, MA 02111-1307 USA.
28 #include <uapi/linux/hyperv.h>
30 #include <linux/types.h>
31 #include <linux/scatterlist.h>
32 #include <linux/list.h>
33 #include <linux/timer.h>
34 #include <linux/completion.h>
35 #include <linux/device.h>
36 #include <linux/mod_devicetable.h>
37 #include <linux/interrupt.h>
38 #include <linux/reciprocal_div.h>
40 #define MAX_PAGE_BUFFER_COUNT 32
41 #define MAX_MULTIPAGE_BUFFER_COUNT 32 /* 128K */
45 /* Single-page buffer */
46 struct hv_page_buffer {
52 /* Multiple-page buffer */
53 struct hv_multipage_buffer {
54 /* Length and Offset determines the # of pfns in the array */
57 u64 pfn_array[MAX_MULTIPAGE_BUFFER_COUNT];
61 * Multiple-page buffer array; the pfn array is variable size:
62 * The number of entries in the PFN array is determined by
66 /* Length and Offset determines the # of pfns in the array */
72 /* 0x18 includes the proprietary packet header */
73 #define MAX_PAGE_BUFFER_PACKET (0x18 + \
74 (sizeof(struct hv_page_buffer) * \
75 MAX_PAGE_BUFFER_COUNT))
76 #define MAX_MULTIPAGE_BUFFER_PACKET (0x18 + \
77 sizeof(struct hv_multipage_buffer))
82 struct hv_ring_buffer {
83 /* Offset in bytes from the start of ring data below */
86 /* Offset in bytes from the start of ring data below */
92 * Win8 uses some of the reserved bits to implement
93 * interrupt driven flow management. On the send side
94 * we can request that the receiver interrupt the sender
95 * when the ring transitions from being full to being able
96 * to handle a message of size "pending_send_sz".
98 * Add necessary state for this enhancement.
106 u32 feat_pending_send_sz:1;
111 /* Pad it to PAGE_SIZE so that data starts on page boundary */
115 * Ring data starts here + RingDataStartOffset
116 * !!! DO NOT place any fields below this !!!
121 struct hv_ring_buffer_info {
122 struct hv_ring_buffer *ring_buffer;
123 u32 ring_size; /* Include the shared header */
124 struct reciprocal_value ring_size_div10_reciprocal;
125 spinlock_t ring_lock;
127 u32 ring_datasize; /* < ring_size */
132 static inline u32 hv_get_bytes_to_read(const struct hv_ring_buffer_info *rbi)
134 u32 read_loc, write_loc, dsize, read;
136 dsize = rbi->ring_datasize;
137 read_loc = rbi->ring_buffer->read_index;
138 write_loc = READ_ONCE(rbi->ring_buffer->write_index);
140 read = write_loc >= read_loc ? (write_loc - read_loc) :
141 (dsize - read_loc) + write_loc;
146 static inline u32 hv_get_bytes_to_write(const struct hv_ring_buffer_info *rbi)
148 u32 read_loc, write_loc, dsize, write;
150 dsize = rbi->ring_datasize;
151 read_loc = READ_ONCE(rbi->ring_buffer->read_index);
152 write_loc = rbi->ring_buffer->write_index;
154 write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
155 read_loc - write_loc;
159 static inline u32 hv_get_avail_to_write_percent(
160 const struct hv_ring_buffer_info *rbi)
162 u32 avail_write = hv_get_bytes_to_write(rbi);
164 return reciprocal_divide(
165 (avail_write << 3) + (avail_write << 1),
166 rbi->ring_size_div10_reciprocal);
170 * VMBUS version is 32 bit entity broken up into
171 * two 16 bit quantities: major_number. minor_number.
173 * 0 . 13 (Windows Server 2008)
176 * 3 . 0 (Windows 8 R2)
178 * 5 . 0 (Newer Windows 10)
181 #define VERSION_WS2008 ((0 << 16) | (13))
182 #define VERSION_WIN7 ((1 << 16) | (1))
183 #define VERSION_WIN8 ((2 << 16) | (4))
184 #define VERSION_WIN8_1 ((3 << 16) | (0))
185 #define VERSION_WIN10 ((4 << 16) | (0))
186 #define VERSION_WIN10_V5 ((5 << 16) | (0))
188 #define VERSION_INVAL -1
190 #define VERSION_CURRENT VERSION_WIN10_V5
192 /* Make maximum size of pipe payload of 16K */
193 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
195 /* Define PipeMode values. */
196 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
197 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
199 /* The size of the user defined data buffer for non-pipe offers. */
200 #define MAX_USER_DEFINED_BYTES 120
202 /* The size of the user defined data buffer for pipe offers. */
203 #define MAX_PIPE_USER_DEFINED_BYTES 116
206 * At the center of the Channel Management library is the Channel Offer. This
207 * struct contains the fundamental information about an offer.
209 struct vmbus_channel_offer {
214 * These two fields are not currently used.
220 u16 mmio_megabytes; /* in bytes * 1024 * 1024 */
223 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
225 unsigned char user_def[MAX_USER_DEFINED_BYTES];
230 * The following sructure is an integrated pipe protocol, which
231 * is implemented on top of standard user-defined data. Pipe
232 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
237 unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
241 * The sub_channel_index is defined in win8.
243 u16 sub_channel_index;
248 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
249 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
250 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
251 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
252 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
253 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
254 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
255 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
257 struct vmpacket_descriptor {
265 struct vmpacket_header {
266 u32 prev_pkt_start_offset;
267 struct vmpacket_descriptor descriptor;
270 struct vmtransfer_page_range {
275 struct vmtransfer_page_packet_header {
276 struct vmpacket_descriptor d;
281 struct vmtransfer_page_range ranges[1];
284 struct vmgpadl_packet_header {
285 struct vmpacket_descriptor d;
290 struct vmadd_remove_transfer_page_set {
291 struct vmpacket_descriptor d;
298 * This structure defines a range in guest physical space that can be made to
299 * look virtually contiguous.
308 * This is the format for an Establish Gpadl packet, which contains a handle by
309 * which this GPADL will be known and a set of GPA ranges associated with it.
310 * This can be converted to a MDL by the guest OS. If there are multiple GPA
311 * ranges, then the resulting MDL will be "chained," representing multiple VA
314 struct vmestablish_gpadl {
315 struct vmpacket_descriptor d;
318 struct gpa_range range[1];
322 * This is the format for a Teardown Gpadl packet, which indicates that the
323 * GPADL handle in the Establish Gpadl packet will never be referenced again.
325 struct vmteardown_gpadl {
326 struct vmpacket_descriptor d;
328 u32 reserved; /* for alignment to a 8-byte boundary */
332 * This is the format for a GPA-Direct packet, which contains a set of GPA
333 * ranges, in addition to commands and/or data.
335 struct vmdata_gpa_direct {
336 struct vmpacket_descriptor d;
339 struct gpa_range range[1];
342 /* This is the format for a Additional Data Packet. */
343 struct vmadditional_data {
344 struct vmpacket_descriptor d;
348 unsigned char data[1];
351 union vmpacket_largest_possible_header {
352 struct vmpacket_descriptor simple_hdr;
353 struct vmtransfer_page_packet_header xfer_page_hdr;
354 struct vmgpadl_packet_header gpadl_hdr;
355 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr;
356 struct vmestablish_gpadl establish_gpadl_hdr;
357 struct vmteardown_gpadl teardown_gpadl_hdr;
358 struct vmdata_gpa_direct data_gpa_direct_hdr;
361 #define VMPACKET_DATA_START_ADDRESS(__packet) \
362 (void *)(((unsigned char *)__packet) + \
363 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
365 #define VMPACKET_DATA_LENGTH(__packet) \
366 ((((struct vmpacket_descriptor)__packet)->len8 - \
367 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
369 #define VMPACKET_TRANSFER_MODE(__packet) \
370 (((struct IMPACT)__packet)->type)
372 enum vmbus_packet_type {
373 VM_PKT_INVALID = 0x0,
375 VM_PKT_ADD_XFER_PAGESET = 0x2,
376 VM_PKT_RM_XFER_PAGESET = 0x3,
377 VM_PKT_ESTABLISH_GPADL = 0x4,
378 VM_PKT_TEARDOWN_GPADL = 0x5,
379 VM_PKT_DATA_INBAND = 0x6,
380 VM_PKT_DATA_USING_XFER_PAGES = 0x7,
381 VM_PKT_DATA_USING_GPADL = 0x8,
382 VM_PKT_DATA_USING_GPA_DIRECT = 0x9,
383 VM_PKT_CANCEL_REQUEST = 0xa,
385 VM_PKT_DATA_USING_ADDITIONAL_PKT = 0xc,
386 VM_PKT_ADDITIONAL_DATA = 0xd
389 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
392 /* Version 1 messages */
393 enum vmbus_channel_message_type {
394 CHANNELMSG_INVALID = 0,
395 CHANNELMSG_OFFERCHANNEL = 1,
396 CHANNELMSG_RESCIND_CHANNELOFFER = 2,
397 CHANNELMSG_REQUESTOFFERS = 3,
398 CHANNELMSG_ALLOFFERS_DELIVERED = 4,
399 CHANNELMSG_OPENCHANNEL = 5,
400 CHANNELMSG_OPENCHANNEL_RESULT = 6,
401 CHANNELMSG_CLOSECHANNEL = 7,
402 CHANNELMSG_GPADL_HEADER = 8,
403 CHANNELMSG_GPADL_BODY = 9,
404 CHANNELMSG_GPADL_CREATED = 10,
405 CHANNELMSG_GPADL_TEARDOWN = 11,
406 CHANNELMSG_GPADL_TORNDOWN = 12,
407 CHANNELMSG_RELID_RELEASED = 13,
408 CHANNELMSG_INITIATE_CONTACT = 14,
409 CHANNELMSG_VERSION_RESPONSE = 15,
410 CHANNELMSG_UNLOAD = 16,
411 CHANNELMSG_UNLOAD_RESPONSE = 17,
415 CHANNELMSG_TL_CONNECT_REQUEST = 21,
419 struct vmbus_channel_message_header {
420 enum vmbus_channel_message_type msgtype;
424 /* Query VMBus Version parameters */
425 struct vmbus_channel_query_vmbus_version {
426 struct vmbus_channel_message_header header;
430 /* VMBus Version Supported parameters */
431 struct vmbus_channel_version_supported {
432 struct vmbus_channel_message_header header;
433 u8 version_supported;
436 /* Offer Channel parameters */
437 struct vmbus_channel_offer_channel {
438 struct vmbus_channel_message_header header;
439 struct vmbus_channel_offer offer;
443 * win7 and beyond splits this field into a bit field.
445 u8 monitor_allocated:1;
448 * These are new fields added in win7 and later.
449 * Do not access these fields without checking the
450 * negotiated protocol.
452 * If "is_dedicated_interrupt" is set, we must not set the
453 * associated bit in the channel bitmap while sending the
454 * interrupt to the host.
456 * connection_id is to be used in signaling the host.
458 u16 is_dedicated_interrupt:1;
463 /* Rescind Offer parameters */
464 struct vmbus_channel_rescind_offer {
465 struct vmbus_channel_message_header header;
470 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info *rbi)
472 return rbi->ring_buffer->pending_send_sz;
476 * Request Offer -- no parameters, SynIC message contains the partition ID
477 * Set Snoop -- no parameters, SynIC message contains the partition ID
478 * Clear Snoop -- no parameters, SynIC message contains the partition ID
479 * All Offers Delivered -- no parameters, SynIC message contains the partition
481 * Flush Client -- no parameters, SynIC message contains the partition ID
484 /* Open Channel parameters */
485 struct vmbus_channel_open_channel {
486 struct vmbus_channel_message_header header;
488 /* Identifies the specific VMBus channel that is being opened. */
491 /* ID making a particular open request at a channel offer unique. */
494 /* GPADL for the channel's ring buffer. */
495 u32 ringbuffer_gpadlhandle;
498 * Starting with win8, this field will be used to specify
499 * the target virtual processor on which to deliver the interrupt for
500 * the host to guest communication.
501 * Prior to win8, incoming channel interrupts would only
502 * be delivered on cpu 0. Setting this value to 0 would
503 * preserve the earlier behavior.
508 * The upstream ring buffer begins at offset zero in the memory
509 * described by RingBufferGpadlHandle. The downstream ring buffer
510 * follows it at this offset (in pages).
512 u32 downstream_ringbuffer_pageoffset;
514 /* User-specific data to be passed along to the server endpoint. */
515 unsigned char userdata[MAX_USER_DEFINED_BYTES];
518 /* Open Channel Result parameters */
519 struct vmbus_channel_open_result {
520 struct vmbus_channel_message_header header;
526 /* Close channel parameters; */
527 struct vmbus_channel_close_channel {
528 struct vmbus_channel_message_header header;
532 /* Channel Message GPADL */
533 #define GPADL_TYPE_RING_BUFFER 1
534 #define GPADL_TYPE_SERVER_SAVE_AREA 2
535 #define GPADL_TYPE_TRANSACTION 8
538 * The number of PFNs in a GPADL message is defined by the number of
539 * pages that would be spanned by ByteCount and ByteOffset. If the
540 * implied number of PFNs won't fit in this packet, there will be a
541 * follow-up packet that contains more.
543 struct vmbus_channel_gpadl_header {
544 struct vmbus_channel_message_header header;
549 struct gpa_range range[0];
552 /* This is the followup packet that contains more PFNs. */
553 struct vmbus_channel_gpadl_body {
554 struct vmbus_channel_message_header header;
560 struct vmbus_channel_gpadl_created {
561 struct vmbus_channel_message_header header;
567 struct vmbus_channel_gpadl_teardown {
568 struct vmbus_channel_message_header header;
573 struct vmbus_channel_gpadl_torndown {
574 struct vmbus_channel_message_header header;
578 struct vmbus_channel_relid_released {
579 struct vmbus_channel_message_header header;
583 struct vmbus_channel_initiate_contact {
584 struct vmbus_channel_message_header header;
585 u32 vmbus_version_requested;
586 u32 target_vcpu; /* The VCPU the host should respond to */
599 /* Hyper-V socket: guest's connect()-ing to host */
600 struct vmbus_channel_tl_connect_request {
601 struct vmbus_channel_message_header header;
602 uuid_le guest_endpoint_id;
603 uuid_le host_service_id;
606 struct vmbus_channel_version_response {
607 struct vmbus_channel_message_header header;
608 u8 version_supported;
614 * On new hosts that support VMBus protocol 5.0, we must use
615 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
616 * and for subsequent messages, we must use the Message Connection ID
617 * field in the host-returned Version Response Message.
619 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
624 enum vmbus_channel_state {
626 CHANNEL_OPENING_STATE,
628 CHANNEL_OPENED_STATE,
632 * Represents each channel msg on the vmbus connection This is a
633 * variable-size data structure depending on the msg type itself
635 struct vmbus_channel_msginfo {
636 /* Bookkeeping stuff */
637 struct list_head msglistentry;
639 /* So far, this is only used to handle gpadl body message */
640 struct list_head submsglist;
642 /* Synchronize the request/response if needed */
643 struct completion waitevent;
644 struct vmbus_channel *waiting_channel;
646 struct vmbus_channel_version_supported version_supported;
647 struct vmbus_channel_open_result open_result;
648 struct vmbus_channel_gpadl_torndown gpadl_torndown;
649 struct vmbus_channel_gpadl_created gpadl_created;
650 struct vmbus_channel_version_response version_response;
655 * The channel message that goes out on the "wire".
656 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
658 unsigned char msg[0];
661 struct vmbus_close_msg {
662 struct vmbus_channel_msginfo info;
663 struct vmbus_channel_close_channel msg;
666 /* Define connection identifier type. */
667 union hv_connection_id {
675 enum hv_numa_policy {
680 enum vmbus_device_type {
700 struct vmbus_device {
706 struct vmbus_channel {
707 struct list_head listentry;
709 struct hv_device *device_obj;
711 enum vmbus_channel_state state;
713 struct vmbus_channel_offer_channel offermsg;
715 * These are based on the OfferMsg.MonitorId.
716 * Save it here for easy access.
721 bool rescind; /* got rescind msg */
722 struct completion rescind_event;
724 u32 ringbuffer_gpadlhandle;
726 /* Allocated memory for ring buffer */
727 void *ringbuffer_pages;
728 u32 ringbuffer_pagecount;
729 struct hv_ring_buffer_info outbound; /* send to parent */
730 struct hv_ring_buffer_info inbound; /* receive from parent */
732 struct vmbus_close_msg close_msg;
735 u64 interrupts; /* Host to Guest interrupts */
736 u64 sig_events; /* Guest to Host events */
738 /* Channel callback's invoked in softirq context */
739 struct tasklet_struct callback_event;
740 void (*onchannel_callback)(void *context);
741 void *channel_callback_context;
744 * A channel can be marked for one of three modes of reading:
745 * BATCHED - callback called from taslket and should read
746 * channel until empty. Interrupts from the host
747 * are masked while read is in process (default).
748 * DIRECT - callback called from tasklet (softirq).
749 * ISR - callback called in interrupt context and must
750 * invoke its own deferred processing.
751 * Host interrupts are disabled and must be re-enabled
752 * when ring is empty.
754 enum hv_callback_mode {
760 bool is_dedicated_interrupt;
764 * Starting with win8, this field will be used to specify
765 * the target virtual processor on which to deliver the interrupt for
766 * the host to guest communication.
767 * Prior to win8, incoming channel interrupts would only
768 * be delivered on cpu 0. Setting this value to 0 would
769 * preserve the earlier behavior.
772 /* The corresponding CPUID in the guest */
775 * State to manage the CPU affiliation of channels.
777 struct cpumask alloced_cpus_in_node;
780 * Support for sub-channels. For high performance devices,
781 * it will be useful to have multiple sub-channels to support
782 * a scalable communication infrastructure with the host.
783 * The support for sub-channels is implemented as an extention
784 * to the current infrastructure.
785 * The initial offer is considered the primary channel and this
786 * offer message will indicate if the host supports sub-channels.
787 * The guest is free to ask for sub-channels to be offerred and can
788 * open these sub-channels as a normal "primary" channel. However,
789 * all sub-channels will have the same type and instance guids as the
790 * primary channel. Requests sent on a given channel will result in a
791 * response on the same channel.
795 * Sub-channel creation callback. This callback will be called in
796 * process context when a sub-channel offer is received from the host.
797 * The guest can open the sub-channel in the context of this callback.
799 void (*sc_creation_callback)(struct vmbus_channel *new_sc);
802 * Channel rescind callback. Some channels (the hvsock ones), need to
803 * register a callback which is invoked in vmbus_onoffer_rescind().
805 void (*chn_rescind_callback)(struct vmbus_channel *channel);
808 * The spinlock to protect the structure. It is being used to protect
809 * test-and-set access to various attributes of the structure as well
810 * as all sc_list operations.
814 * All Sub-channels of a primary channel are linked here.
816 struct list_head sc_list;
818 * Current number of sub-channels.
822 * Number of a sub-channel (position within sc_list) which is supposed
823 * to be used as the next outgoing channel.
827 * The primary channel this sub-channel belongs to.
828 * This will be NULL for the primary channel.
830 struct vmbus_channel *primary_channel;
832 * Support per-channel state for use by vmbus drivers.
834 void *per_channel_state;
836 * To support per-cpu lookup mapping of relid to channel,
837 * link up channels based on their CPU affinity.
839 struct list_head percpu_list;
842 * Defer freeing channel until after all cpu's have
843 * gone through grace period.
848 * For sysfs per-channel properties.
853 * For performance critical channels (storage, networking
854 * etc,), Hyper-V has a mechanism to enhance the throughput
855 * at the expense of latency:
856 * When the host is to be signaled, we just set a bit in a shared page
857 * and this bit will be inspected by the hypervisor within a certain
858 * window and if the bit is set, the host will be signaled. The window
859 * of time is the monitor latency - currently around 100 usecs. This
860 * mechanism improves throughput by:
862 * A) Making the host more efficient - each time it wakes up,
863 * potentially it will process morev number of packets. The
864 * monitor latency allows a batch to build up.
865 * B) By deferring the hypercall to signal, we will also minimize
868 * Clearly, these optimizations improve throughput at the expense of
869 * latency. Furthermore, since the channel is shared for both
870 * control and data messages, control messages currently suffer
871 * unnecessary latency adversley impacting performance and boot
872 * time. To fix this issue, permit tagging the channel as being
873 * in "low latency" mode. In this mode, we will bypass the monitor
879 * NUMA distribution policy:
880 * We support two policies:
881 * 1) Balanced: Here all performance critical channels are
882 * distributed evenly amongst all the NUMA nodes.
883 * This policy will be the default policy.
884 * 2) Localized: All channels of a given instance of a
885 * performance critical service will be assigned CPUs
886 * within a selected NUMA node.
888 enum hv_numa_policy affinity_policy;
894 static inline bool is_hvsock_channel(const struct vmbus_channel *c)
896 return !!(c->offermsg.offer.chn_flags &
897 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER);
900 static inline void set_channel_affinity_state(struct vmbus_channel *c,
901 enum hv_numa_policy policy)
903 c->affinity_policy = policy;
906 static inline void set_channel_read_mode(struct vmbus_channel *c,
907 enum hv_callback_mode mode)
909 c->callback_mode = mode;
912 static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
914 c->per_channel_state = s;
917 static inline void *get_per_channel_state(struct vmbus_channel *c)
919 return c->per_channel_state;
922 static inline void set_channel_pending_send_size(struct vmbus_channel *c,
925 c->outbound.ring_buffer->pending_send_sz = size;
928 static inline void set_low_latency_mode(struct vmbus_channel *c)
930 c->low_latency = true;
933 static inline void clear_low_latency_mode(struct vmbus_channel *c)
935 c->low_latency = false;
938 void vmbus_onmessage(void *context);
940 int vmbus_request_offers(void);
943 * APIs for managing sub-channels.
946 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
947 void (*sc_cr_cb)(struct vmbus_channel *new_sc));
949 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
950 void (*chn_rescind_cb)(struct vmbus_channel *));
953 * Retrieve the (sub) channel on which to send an outgoing request.
954 * When a primary channel has multiple sub-channels, we choose a
955 * channel whose VCPU binding is closest to the VCPU on which
956 * this call is being made.
958 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary);
961 * Check if sub-channels have already been offerred. This API will be useful
962 * when the driver is unloaded after establishing sub-channels. In this case,
963 * when the driver is re-loaded, the driver would have to check if the
964 * subchannels have already been established before attempting to request
965 * the creation of sub-channels.
966 * This function returns TRUE to indicate that subchannels have already been
968 * This function should be invoked after setting the callback function for
969 * sub-channel creation.
971 bool vmbus_are_subchannels_present(struct vmbus_channel *primary);
973 /* The format must be the same as struct vmdata_gpa_direct */
974 struct vmbus_channel_packet_page_buffer {
982 struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
985 /* The format must be the same as struct vmdata_gpa_direct */
986 struct vmbus_channel_packet_multipage_buffer {
993 u32 rangecount; /* Always 1 in this case */
994 struct hv_multipage_buffer range;
997 /* The format must be the same as struct vmdata_gpa_direct */
998 struct vmbus_packet_mpb_array {
1005 u32 rangecount; /* Always 1 in this case */
1006 struct hv_mpb_array range;
1010 extern int vmbus_open(struct vmbus_channel *channel,
1011 u32 send_ringbuffersize,
1012 u32 recv_ringbuffersize,
1015 void (*onchannel_callback)(void *context),
1018 extern void vmbus_close(struct vmbus_channel *channel);
1020 extern int vmbus_sendpacket(struct vmbus_channel *channel,
1024 enum vmbus_packet_type type,
1027 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1028 struct hv_page_buffer pagebuffers[],
1034 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1035 struct vmbus_packet_mpb_array *mpb,
1041 extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
1046 extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
1049 extern int vmbus_recvpacket(struct vmbus_channel *channel,
1052 u32 *buffer_actual_len,
1055 extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
1058 u32 *buffer_actual_len,
1062 extern void vmbus_ontimer(unsigned long data);
1064 /* Base driver object */
1069 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1070 * channel flag, actually doesn't mean a synthetic device because the
1071 * offer's if_type/if_instance can change for every new hvsock
1074 * However, to facilitate the notification of new-offer/rescind-offer
1075 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1076 * a special vmbus device, and hence we need the below flag to
1077 * indicate if the driver is the hvsock driver or not: we need to
1078 * specially treat the hvosck offer & driver in vmbus_match().
1082 /* the device type supported by this driver */
1084 const struct hv_vmbus_device_id *id_table;
1086 struct device_driver driver;
1088 /* dynamic device GUID's */
1091 struct list_head list;
1094 int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
1095 int (*remove)(struct hv_device *);
1096 void (*shutdown)(struct hv_device *);
1100 /* Base device object */
1102 /* the device type id of this device */
1105 /* the device instance id of this device */
1106 uuid_le dev_instance;
1110 struct device device;
1112 struct vmbus_channel *channel;
1113 struct kset *channels_kset;
1117 static inline struct hv_device *device_to_hv_device(struct device *d)
1119 return container_of(d, struct hv_device, device);
1122 static inline struct hv_driver *drv_to_hv_drv(struct device_driver *d)
1124 return container_of(d, struct hv_driver, driver);
1127 static inline void hv_set_drvdata(struct hv_device *dev, void *data)
1129 dev_set_drvdata(&dev->device, data);
1132 static inline void *hv_get_drvdata(struct hv_device *dev)
1134 return dev_get_drvdata(&dev->device);
1137 struct hv_ring_buffer_debug_info {
1138 u32 current_interrupt_mask;
1139 u32 current_read_index;
1140 u32 current_write_index;
1141 u32 bytes_avail_toread;
1142 u32 bytes_avail_towrite;
1145 void hv_ringbuffer_get_debuginfo(const struct hv_ring_buffer_info *ring_info,
1146 struct hv_ring_buffer_debug_info *debug_info);
1148 /* Vmbus interface */
1149 #define vmbus_driver_register(driver) \
1150 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1151 int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
1152 struct module *owner,
1153 const char *mod_name);
1154 void vmbus_driver_unregister(struct hv_driver *hv_driver);
1156 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel);
1158 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1159 resource_size_t min, resource_size_t max,
1160 resource_size_t size, resource_size_t align,
1161 bool fb_overlap_ok);
1162 void vmbus_free_mmio(resource_size_t start, resource_size_t size);
1165 * GUID definitions of various offer types - services offered to the guest.
1170 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1172 #define HV_NIC_GUID \
1173 .guid = UUID_LE(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1174 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1178 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1180 #define HV_IDE_GUID \
1181 .guid = UUID_LE(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1182 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1186 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1188 #define HV_SCSI_GUID \
1189 .guid = UUID_LE(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1190 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1194 * {0e0b6031-5213-4934-818b-38d90ced39db}
1196 #define HV_SHUTDOWN_GUID \
1197 .guid = UUID_LE(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1198 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1202 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1204 #define HV_TS_GUID \
1205 .guid = UUID_LE(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1206 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1210 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1212 #define HV_HEART_BEAT_GUID \
1213 .guid = UUID_LE(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1214 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1218 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1220 #define HV_KVP_GUID \
1221 .guid = UUID_LE(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1222 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1225 * Dynamic memory GUID
1226 * {525074dc-8985-46e2-8057-a307dc18a502}
1228 #define HV_DM_GUID \
1229 .guid = UUID_LE(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1230 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1234 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1236 #define HV_MOUSE_GUID \
1237 .guid = UUID_LE(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1238 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1242 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1244 #define HV_KBD_GUID \
1245 .guid = UUID_LE(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1246 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1249 * VSS (Backup/Restore) GUID
1251 #define HV_VSS_GUID \
1252 .guid = UUID_LE(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1253 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1255 * Synthetic Video GUID
1256 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1258 #define HV_SYNTHVID_GUID \
1259 .guid = UUID_LE(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1260 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1264 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1266 #define HV_SYNTHFC_GUID \
1267 .guid = UUID_LE(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1268 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1271 * Guest File Copy Service
1272 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1275 #define HV_FCOPY_GUID \
1276 .guid = UUID_LE(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1277 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1280 * NetworkDirect. This is the guest RDMA service.
1281 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1283 #define HV_ND_GUID \
1284 .guid = UUID_LE(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1285 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1288 * PCI Express Pass Through
1289 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1292 #define HV_PCIE_GUID \
1293 .guid = UUID_LE(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1294 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1297 * Linux doesn't support the 3 devices: the first two are for
1298 * Automatic Virtual Machine Activation, and the third is for
1299 * Remote Desktop Virtualization.
1300 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1301 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1302 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1305 #define HV_AVMA1_GUID \
1306 .guid = UUID_LE(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1307 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1309 #define HV_AVMA2_GUID \
1310 .guid = UUID_LE(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1311 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1313 #define HV_RDV_GUID \
1314 .guid = UUID_LE(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1315 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1318 * Common header for Hyper-V ICs
1321 #define ICMSGTYPE_NEGOTIATE 0
1322 #define ICMSGTYPE_HEARTBEAT 1
1323 #define ICMSGTYPE_KVPEXCHANGE 2
1324 #define ICMSGTYPE_SHUTDOWN 3
1325 #define ICMSGTYPE_TIMESYNC 4
1326 #define ICMSGTYPE_VSS 5
1328 #define ICMSGHDRFLAG_TRANSACTION 1
1329 #define ICMSGHDRFLAG_REQUEST 2
1330 #define ICMSGHDRFLAG_RESPONSE 4
1334 * While we want to handle util services as regular devices,
1335 * there is only one instance of each of these services; so
1336 * we statically allocate the service specific state.
1339 struct hv_util_service {
1342 void (*util_cb)(void *);
1343 int (*util_init)(struct hv_util_service *);
1344 void (*util_deinit)(void);
1347 struct vmbuspipe_hdr {
1358 struct ic_version icverframe;
1360 struct ic_version icvermsg;
1363 u8 ictransaction_id;
1368 struct icmsg_negotiate {
1372 struct ic_version icversion_data[1]; /* any size array */
1375 struct shutdown_msg_data {
1377 u32 timeout_seconds;
1379 u8 display_message[2048];
1382 struct heartbeat_msg_data {
1387 /* Time Sync IC defs */
1388 #define ICTIMESYNCFLAG_PROBE 0
1389 #define ICTIMESYNCFLAG_SYNC 1
1390 #define ICTIMESYNCFLAG_SAMPLE 2
1393 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1395 #define WLTIMEDELTA 116444736000000000LL
1398 struct ictimesync_data {
1405 struct ictimesync_ref_data {
1407 u64 vmreferencetime;
1414 struct hyperv_service_callback {
1418 struct vmbus_channel *channel;
1419 void (*callback)(void *context);
1422 #define MAX_SRV_VER 0x7ffffff
1423 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf,
1424 const int *fw_version, int fw_vercnt,
1425 const int *srv_version, int srv_vercnt,
1426 int *nego_fw_version, int *nego_srv_version);
1428 void hv_process_channel_removal(u32 relid);
1430 void vmbus_setevent(struct vmbus_channel *channel);
1432 * Negotiated version with the Host.
1435 extern __u32 vmbus_proto_version;
1437 int vmbus_send_tl_connect_request(const uuid_le *shv_guest_servie_id,
1438 const uuid_le *shv_host_servie_id);
1439 void vmbus_set_event(struct vmbus_channel *channel);
1441 /* Get the start of the ring buffer. */
1442 static inline void *
1443 hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info)
1445 return ring_info->ring_buffer->buffer;
1449 * Mask off host interrupt callback notifications
1451 static inline void hv_begin_read(struct hv_ring_buffer_info *rbi)
1453 rbi->ring_buffer->interrupt_mask = 1;
1455 /* make sure mask update is not reordered */
1460 * Re-enable host callback and return number of outstanding bytes
1462 static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi)
1465 rbi->ring_buffer->interrupt_mask = 0;
1467 /* make sure mask update is not reordered */
1471 * Now check to see if the ring buffer is still empty.
1472 * If it is not, we raced and we need to process new
1473 * incoming messages.
1475 return hv_get_bytes_to_read(rbi);
1479 * An API to support in-place processing of incoming VMBUS packets.
1482 /* Get data payload associated with descriptor */
1483 static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc)
1485 return (void *)((unsigned long)desc + (desc->offset8 << 3));
1488 /* Get data size associated with descriptor */
1489 static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc)
1491 return (desc->len8 << 3) - (desc->offset8 << 3);
1495 struct vmpacket_descriptor *
1496 hv_pkt_iter_first(struct vmbus_channel *channel);
1498 struct vmpacket_descriptor *
1499 __hv_pkt_iter_next(struct vmbus_channel *channel,
1500 const struct vmpacket_descriptor *pkt);
1502 void hv_pkt_iter_close(struct vmbus_channel *channel);
1505 * Get next packet descriptor from iterator
1506 * If at end of list, return NULL and update host.
1508 static inline struct vmpacket_descriptor *
1509 hv_pkt_iter_next(struct vmbus_channel *channel,
1510 const struct vmpacket_descriptor *pkt)
1512 struct vmpacket_descriptor *nxt;
1514 nxt = __hv_pkt_iter_next(channel, pkt);
1516 hv_pkt_iter_close(channel);
1521 #define foreach_vmbus_pkt(pkt, channel) \
1522 for (pkt = hv_pkt_iter_first(channel); pkt; \
1523 pkt = hv_pkt_iter_next(channel, pkt))
1525 #endif /* _HYPERV_H */