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>
29 #include <uapi/asm/hyperv.h>
31 #include <linux/types.h>
32 #include <linux/scatterlist.h>
33 #include <linux/list.h>
34 #include <linux/timer.h>
35 #include <linux/completion.h>
36 #include <linux/device.h>
37 #include <linux/mod_devicetable.h>
38 #include <linux/interrupt.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 spinlock_t ring_lock;
126 u32 ring_datasize; /* < ring_size */
132 * hv_get_ringbuffer_availbytes()
134 * Get number of bytes available to read and to write to
135 * for the specified ring buffer
138 hv_get_ringbuffer_availbytes(const struct hv_ring_buffer_info *rbi,
139 u32 *read, u32 *write)
141 u32 read_loc, write_loc, dsize;
143 /* Capture the read/write indices before they changed */
144 read_loc = rbi->ring_buffer->read_index;
145 write_loc = rbi->ring_buffer->write_index;
146 dsize = rbi->ring_datasize;
148 *write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
149 read_loc - write_loc;
150 *read = dsize - *write;
153 static inline u32 hv_get_bytes_to_read(const struct hv_ring_buffer_info *rbi)
155 u32 read_loc, write_loc, dsize, read;
157 dsize = rbi->ring_datasize;
158 read_loc = rbi->ring_buffer->read_index;
159 write_loc = READ_ONCE(rbi->ring_buffer->write_index);
161 read = write_loc >= read_loc ? (write_loc - read_loc) :
162 (dsize - read_loc) + write_loc;
167 static inline u32 hv_get_bytes_to_write(const struct hv_ring_buffer_info *rbi)
169 u32 read_loc, write_loc, dsize, write;
171 dsize = rbi->ring_datasize;
172 read_loc = READ_ONCE(rbi->ring_buffer->read_index);
173 write_loc = rbi->ring_buffer->write_index;
175 write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
176 read_loc - write_loc;
181 * VMBUS version is 32 bit entity broken up into
182 * two 16 bit quantities: major_number. minor_number.
184 * 0 . 13 (Windows Server 2008)
187 * 3 . 0 (Windows 8 R2)
191 #define VERSION_WS2008 ((0 << 16) | (13))
192 #define VERSION_WIN7 ((1 << 16) | (1))
193 #define VERSION_WIN8 ((2 << 16) | (4))
194 #define VERSION_WIN8_1 ((3 << 16) | (0))
195 #define VERSION_WIN10 ((4 << 16) | (0))
197 #define VERSION_INVAL -1
199 #define VERSION_CURRENT VERSION_WIN10
201 /* Make maximum size of pipe payload of 16K */
202 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
204 /* Define PipeMode values. */
205 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
206 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
208 /* The size of the user defined data buffer for non-pipe offers. */
209 #define MAX_USER_DEFINED_BYTES 120
211 /* The size of the user defined data buffer for pipe offers. */
212 #define MAX_PIPE_USER_DEFINED_BYTES 116
215 * At the center of the Channel Management library is the Channel Offer. This
216 * struct contains the fundamental information about an offer.
218 struct vmbus_channel_offer {
223 * These two fields are not currently used.
229 u16 mmio_megabytes; /* in bytes * 1024 * 1024 */
232 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
234 unsigned char user_def[MAX_USER_DEFINED_BYTES];
239 * The following sructure is an integrated pipe protocol, which
240 * is implemented on top of standard user-defined data. Pipe
241 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
246 unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
250 * The sub_channel_index is defined in win8.
252 u16 sub_channel_index;
257 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
258 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
259 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
260 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
261 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
262 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
263 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
264 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
266 struct vmpacket_descriptor {
274 struct vmpacket_header {
275 u32 prev_pkt_start_offset;
276 struct vmpacket_descriptor descriptor;
279 struct vmtransfer_page_range {
284 struct vmtransfer_page_packet_header {
285 struct vmpacket_descriptor d;
290 struct vmtransfer_page_range ranges[1];
293 struct vmgpadl_packet_header {
294 struct vmpacket_descriptor d;
299 struct vmadd_remove_transfer_page_set {
300 struct vmpacket_descriptor d;
307 * This structure defines a range in guest physical space that can be made to
308 * look virtually contiguous.
317 * This is the format for an Establish Gpadl packet, which contains a handle by
318 * which this GPADL will be known and a set of GPA ranges associated with it.
319 * This can be converted to a MDL by the guest OS. If there are multiple GPA
320 * ranges, then the resulting MDL will be "chained," representing multiple VA
323 struct vmestablish_gpadl {
324 struct vmpacket_descriptor d;
327 struct gpa_range range[1];
331 * This is the format for a Teardown Gpadl packet, which indicates that the
332 * GPADL handle in the Establish Gpadl packet will never be referenced again.
334 struct vmteardown_gpadl {
335 struct vmpacket_descriptor d;
337 u32 reserved; /* for alignment to a 8-byte boundary */
341 * This is the format for a GPA-Direct packet, which contains a set of GPA
342 * ranges, in addition to commands and/or data.
344 struct vmdata_gpa_direct {
345 struct vmpacket_descriptor d;
348 struct gpa_range range[1];
351 /* This is the format for a Additional Data Packet. */
352 struct vmadditional_data {
353 struct vmpacket_descriptor d;
357 unsigned char data[1];
360 union vmpacket_largest_possible_header {
361 struct vmpacket_descriptor simple_hdr;
362 struct vmtransfer_page_packet_header xfer_page_hdr;
363 struct vmgpadl_packet_header gpadl_hdr;
364 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr;
365 struct vmestablish_gpadl establish_gpadl_hdr;
366 struct vmteardown_gpadl teardown_gpadl_hdr;
367 struct vmdata_gpa_direct data_gpa_direct_hdr;
370 #define VMPACKET_DATA_START_ADDRESS(__packet) \
371 (void *)(((unsigned char *)__packet) + \
372 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
374 #define VMPACKET_DATA_LENGTH(__packet) \
375 ((((struct vmpacket_descriptor)__packet)->len8 - \
376 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
378 #define VMPACKET_TRANSFER_MODE(__packet) \
379 (((struct IMPACT)__packet)->type)
381 enum vmbus_packet_type {
382 VM_PKT_INVALID = 0x0,
384 VM_PKT_ADD_XFER_PAGESET = 0x2,
385 VM_PKT_RM_XFER_PAGESET = 0x3,
386 VM_PKT_ESTABLISH_GPADL = 0x4,
387 VM_PKT_TEARDOWN_GPADL = 0x5,
388 VM_PKT_DATA_INBAND = 0x6,
389 VM_PKT_DATA_USING_XFER_PAGES = 0x7,
390 VM_PKT_DATA_USING_GPADL = 0x8,
391 VM_PKT_DATA_USING_GPA_DIRECT = 0x9,
392 VM_PKT_CANCEL_REQUEST = 0xa,
394 VM_PKT_DATA_USING_ADDITIONAL_PKT = 0xc,
395 VM_PKT_ADDITIONAL_DATA = 0xd
398 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
401 /* Version 1 messages */
402 enum vmbus_channel_message_type {
403 CHANNELMSG_INVALID = 0,
404 CHANNELMSG_OFFERCHANNEL = 1,
405 CHANNELMSG_RESCIND_CHANNELOFFER = 2,
406 CHANNELMSG_REQUESTOFFERS = 3,
407 CHANNELMSG_ALLOFFERS_DELIVERED = 4,
408 CHANNELMSG_OPENCHANNEL = 5,
409 CHANNELMSG_OPENCHANNEL_RESULT = 6,
410 CHANNELMSG_CLOSECHANNEL = 7,
411 CHANNELMSG_GPADL_HEADER = 8,
412 CHANNELMSG_GPADL_BODY = 9,
413 CHANNELMSG_GPADL_CREATED = 10,
414 CHANNELMSG_GPADL_TEARDOWN = 11,
415 CHANNELMSG_GPADL_TORNDOWN = 12,
416 CHANNELMSG_RELID_RELEASED = 13,
417 CHANNELMSG_INITIATE_CONTACT = 14,
418 CHANNELMSG_VERSION_RESPONSE = 15,
419 CHANNELMSG_UNLOAD = 16,
420 CHANNELMSG_UNLOAD_RESPONSE = 17,
424 CHANNELMSG_TL_CONNECT_REQUEST = 21,
428 struct vmbus_channel_message_header {
429 enum vmbus_channel_message_type msgtype;
433 /* Query VMBus Version parameters */
434 struct vmbus_channel_query_vmbus_version {
435 struct vmbus_channel_message_header header;
439 /* VMBus Version Supported parameters */
440 struct vmbus_channel_version_supported {
441 struct vmbus_channel_message_header header;
442 u8 version_supported;
445 /* Offer Channel parameters */
446 struct vmbus_channel_offer_channel {
447 struct vmbus_channel_message_header header;
448 struct vmbus_channel_offer offer;
452 * win7 and beyond splits this field into a bit field.
454 u8 monitor_allocated:1;
457 * These are new fields added in win7 and later.
458 * Do not access these fields without checking the
459 * negotiated protocol.
461 * If "is_dedicated_interrupt" is set, we must not set the
462 * associated bit in the channel bitmap while sending the
463 * interrupt to the host.
465 * connection_id is to be used in signaling the host.
467 u16 is_dedicated_interrupt:1;
472 /* Rescind Offer parameters */
473 struct vmbus_channel_rescind_offer {
474 struct vmbus_channel_message_header header;
479 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info *rbi)
481 return rbi->ring_buffer->pending_send_sz;
485 * Request Offer -- no parameters, SynIC message contains the partition ID
486 * Set Snoop -- no parameters, SynIC message contains the partition ID
487 * Clear Snoop -- no parameters, SynIC message contains the partition ID
488 * All Offers Delivered -- no parameters, SynIC message contains the partition
490 * Flush Client -- no parameters, SynIC message contains the partition ID
493 /* Open Channel parameters */
494 struct vmbus_channel_open_channel {
495 struct vmbus_channel_message_header header;
497 /* Identifies the specific VMBus channel that is being opened. */
500 /* ID making a particular open request at a channel offer unique. */
503 /* GPADL for the channel's ring buffer. */
504 u32 ringbuffer_gpadlhandle;
507 * Starting with win8, this field will be used to specify
508 * the target virtual processor on which to deliver the interrupt for
509 * the host to guest communication.
510 * Prior to win8, incoming channel interrupts would only
511 * be delivered on cpu 0. Setting this value to 0 would
512 * preserve the earlier behavior.
517 * The upstream ring buffer begins at offset zero in the memory
518 * described by RingBufferGpadlHandle. The downstream ring buffer
519 * follows it at this offset (in pages).
521 u32 downstream_ringbuffer_pageoffset;
523 /* User-specific data to be passed along to the server endpoint. */
524 unsigned char userdata[MAX_USER_DEFINED_BYTES];
527 /* Open Channel Result parameters */
528 struct vmbus_channel_open_result {
529 struct vmbus_channel_message_header header;
535 /* Close channel parameters; */
536 struct vmbus_channel_close_channel {
537 struct vmbus_channel_message_header header;
541 /* Channel Message GPADL */
542 #define GPADL_TYPE_RING_BUFFER 1
543 #define GPADL_TYPE_SERVER_SAVE_AREA 2
544 #define GPADL_TYPE_TRANSACTION 8
547 * The number of PFNs in a GPADL message is defined by the number of
548 * pages that would be spanned by ByteCount and ByteOffset. If the
549 * implied number of PFNs won't fit in this packet, there will be a
550 * follow-up packet that contains more.
552 struct vmbus_channel_gpadl_header {
553 struct vmbus_channel_message_header header;
558 struct gpa_range range[0];
561 /* This is the followup packet that contains more PFNs. */
562 struct vmbus_channel_gpadl_body {
563 struct vmbus_channel_message_header header;
569 struct vmbus_channel_gpadl_created {
570 struct vmbus_channel_message_header header;
576 struct vmbus_channel_gpadl_teardown {
577 struct vmbus_channel_message_header header;
582 struct vmbus_channel_gpadl_torndown {
583 struct vmbus_channel_message_header header;
587 struct vmbus_channel_relid_released {
588 struct vmbus_channel_message_header header;
592 struct vmbus_channel_initiate_contact {
593 struct vmbus_channel_message_header header;
594 u32 vmbus_version_requested;
595 u32 target_vcpu; /* The VCPU the host should respond to */
601 /* Hyper-V socket: guest's connect()-ing to host */
602 struct vmbus_channel_tl_connect_request {
603 struct vmbus_channel_message_header header;
604 uuid_le guest_endpoint_id;
605 uuid_le host_service_id;
608 struct vmbus_channel_version_response {
609 struct vmbus_channel_message_header header;
610 u8 version_supported;
613 enum vmbus_channel_state {
615 CHANNEL_OPENING_STATE,
617 CHANNEL_OPENED_STATE,
621 * Represents each channel msg on the vmbus connection This is a
622 * variable-size data structure depending on the msg type itself
624 struct vmbus_channel_msginfo {
625 /* Bookkeeping stuff */
626 struct list_head msglistentry;
628 /* So far, this is only used to handle gpadl body message */
629 struct list_head submsglist;
631 /* Synchronize the request/response if needed */
632 struct completion waitevent;
633 struct vmbus_channel *waiting_channel;
635 struct vmbus_channel_version_supported version_supported;
636 struct vmbus_channel_open_result open_result;
637 struct vmbus_channel_gpadl_torndown gpadl_torndown;
638 struct vmbus_channel_gpadl_created gpadl_created;
639 struct vmbus_channel_version_response version_response;
644 * The channel message that goes out on the "wire".
645 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
647 unsigned char msg[0];
650 struct vmbus_close_msg {
651 struct vmbus_channel_msginfo info;
652 struct vmbus_channel_close_channel msg;
655 /* Define connection identifier type. */
656 union hv_connection_id {
664 /* Definition of the hv_signal_event hypercall input structure. */
665 struct hv_input_signal_event {
666 union hv_connection_id connectionid;
671 struct hv_input_signal_event_buffer {
673 struct hv_input_signal_event event;
676 enum hv_numa_policy {
681 enum vmbus_device_type {
701 struct vmbus_device {
707 struct vmbus_channel {
708 struct list_head listentry;
710 struct hv_device *device_obj;
712 enum vmbus_channel_state state;
714 struct vmbus_channel_offer_channel offermsg;
716 * These are based on the OfferMsg.MonitorId.
717 * Save it here for easy access.
722 bool rescind; /* got rescind msg */
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;
734 /* Channel callback's invoked in softirq context */
735 struct tasklet_struct callback_event;
736 void (*onchannel_callback)(void *context);
737 void *channel_callback_context;
740 * A channel can be marked for one of three modes of reading:
741 * BATCHED - callback called from taslket and should read
742 * channel until empty. Interrupts from the host
743 * are masked while read is in process (default).
744 * DIRECT - callback called from tasklet (softirq).
745 * ISR - callback called in interrupt context and must
746 * invoke its own deferred processing.
747 * Host interrupts are disabled and must be re-enabled
748 * when ring is empty.
750 enum hv_callback_mode {
756 bool is_dedicated_interrupt;
757 struct hv_input_signal_event_buffer sig_buf;
758 struct hv_input_signal_event *sig_event;
761 * Starting with win8, this field will be used to specify
762 * the target virtual processor on which to deliver the interrupt for
763 * the host to guest communication.
764 * Prior to win8, incoming channel interrupts would only
765 * be delivered on cpu 0. Setting this value to 0 would
766 * preserve the earlier behavior.
769 /* The corresponding CPUID in the guest */
772 * State to manage the CPU affiliation of channels.
774 struct cpumask alloced_cpus_in_node;
777 * Support for sub-channels. For high performance devices,
778 * it will be useful to have multiple sub-channels to support
779 * a scalable communication infrastructure with the host.
780 * The support for sub-channels is implemented as an extention
781 * to the current infrastructure.
782 * The initial offer is considered the primary channel and this
783 * offer message will indicate if the host supports sub-channels.
784 * The guest is free to ask for sub-channels to be offerred and can
785 * open these sub-channels as a normal "primary" channel. However,
786 * all sub-channels will have the same type and instance guids as the
787 * primary channel. Requests sent on a given channel will result in a
788 * response on the same channel.
792 * Sub-channel creation callback. This callback will be called in
793 * process context when a sub-channel offer is received from the host.
794 * The guest can open the sub-channel in the context of this callback.
796 void (*sc_creation_callback)(struct vmbus_channel *new_sc);
799 * Channel rescind callback. Some channels (the hvsock ones), need to
800 * register a callback which is invoked in vmbus_onoffer_rescind().
802 void (*chn_rescind_callback)(struct vmbus_channel *channel);
805 * The spinlock to protect the structure. It is being used to protect
806 * test-and-set access to various attributes of the structure as well
807 * as all sc_list operations.
811 * All Sub-channels of a primary channel are linked here.
813 struct list_head sc_list;
815 * Current number of sub-channels.
819 * Number of a sub-channel (position within sc_list) which is supposed
820 * to be used as the next outgoing channel.
824 * The primary channel this sub-channel belongs to.
825 * This will be NULL for the primary channel.
827 struct vmbus_channel *primary_channel;
829 * Support per-channel state for use by vmbus drivers.
831 void *per_channel_state;
833 * To support per-cpu lookup mapping of relid to channel,
834 * link up channels based on their CPU affinity.
836 struct list_head percpu_list;
839 * Defer freeing channel until after all cpu's have
840 * gone through grace period.
845 * For performance critical channels (storage, networking
846 * etc,), Hyper-V has a mechanism to enhance the throughput
847 * at the expense of latency:
848 * When the host is to be signaled, we just set a bit in a shared page
849 * and this bit will be inspected by the hypervisor within a certain
850 * window and if the bit is set, the host will be signaled. The window
851 * of time is the monitor latency - currently around 100 usecs. This
852 * mechanism improves throughput by:
854 * A) Making the host more efficient - each time it wakes up,
855 * potentially it will process morev number of packets. The
856 * monitor latency allows a batch to build up.
857 * B) By deferring the hypercall to signal, we will also minimize
860 * Clearly, these optimizations improve throughput at the expense of
861 * latency. Furthermore, since the channel is shared for both
862 * control and data messages, control messages currently suffer
863 * unnecessary latency adversley impacting performance and boot
864 * time. To fix this issue, permit tagging the channel as being
865 * in "low latency" mode. In this mode, we will bypass the monitor
871 * NUMA distribution policy:
872 * We support teo policies:
873 * 1) Balanced: Here all performance critical channels are
874 * distributed evenly amongst all the NUMA nodes.
875 * This policy will be the default policy.
876 * 2) Localized: All channels of a given instance of a
877 * performance critical service will be assigned CPUs
878 * within a selected NUMA node.
880 enum hv_numa_policy affinity_policy;
886 static inline bool is_hvsock_channel(const struct vmbus_channel *c)
888 return !!(c->offermsg.offer.chn_flags &
889 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER);
892 static inline void set_channel_affinity_state(struct vmbus_channel *c,
893 enum hv_numa_policy policy)
895 c->affinity_policy = policy;
898 static inline void set_channel_read_mode(struct vmbus_channel *c,
899 enum hv_callback_mode mode)
901 c->callback_mode = mode;
904 static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
906 c->per_channel_state = s;
909 static inline void *get_per_channel_state(struct vmbus_channel *c)
911 return c->per_channel_state;
914 static inline void set_channel_pending_send_size(struct vmbus_channel *c,
917 c->outbound.ring_buffer->pending_send_sz = size;
920 static inline void set_low_latency_mode(struct vmbus_channel *c)
922 c->low_latency = true;
925 static inline void clear_low_latency_mode(struct vmbus_channel *c)
927 c->low_latency = false;
930 void vmbus_onmessage(void *context);
932 int vmbus_request_offers(void);
935 * APIs for managing sub-channels.
938 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
939 void (*sc_cr_cb)(struct vmbus_channel *new_sc));
941 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
942 void (*chn_rescind_cb)(struct vmbus_channel *));
945 * Retrieve the (sub) channel on which to send an outgoing request.
946 * When a primary channel has multiple sub-channels, we choose a
947 * channel whose VCPU binding is closest to the VCPU on which
948 * this call is being made.
950 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary);
953 * Check if sub-channels have already been offerred. This API will be useful
954 * when the driver is unloaded after establishing sub-channels. In this case,
955 * when the driver is re-loaded, the driver would have to check if the
956 * subchannels have already been established before attempting to request
957 * the creation of sub-channels.
958 * This function returns TRUE to indicate that subchannels have already been
960 * This function should be invoked after setting the callback function for
961 * sub-channel creation.
963 bool vmbus_are_subchannels_present(struct vmbus_channel *primary);
965 /* The format must be the same as struct vmdata_gpa_direct */
966 struct vmbus_channel_packet_page_buffer {
974 struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
977 /* The format must be the same as struct vmdata_gpa_direct */
978 struct vmbus_channel_packet_multipage_buffer {
985 u32 rangecount; /* Always 1 in this case */
986 struct hv_multipage_buffer range;
989 /* The format must be the same as struct vmdata_gpa_direct */
990 struct vmbus_packet_mpb_array {
997 u32 rangecount; /* Always 1 in this case */
998 struct hv_mpb_array range;
1002 extern int vmbus_open(struct vmbus_channel *channel,
1003 u32 send_ringbuffersize,
1004 u32 recv_ringbuffersize,
1007 void (*onchannel_callback)(void *context),
1010 extern void vmbus_close(struct vmbus_channel *channel);
1012 extern int vmbus_sendpacket(struct vmbus_channel *channel,
1016 enum vmbus_packet_type type,
1019 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1020 struct hv_page_buffer pagebuffers[],
1026 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1027 struct vmbus_packet_mpb_array *mpb,
1033 extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
1038 extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
1041 extern int vmbus_recvpacket(struct vmbus_channel *channel,
1044 u32 *buffer_actual_len,
1047 extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
1050 u32 *buffer_actual_len,
1054 extern void vmbus_ontimer(unsigned long data);
1056 /* Base driver object */
1061 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1062 * channel flag, actually doesn't mean a synthetic device because the
1063 * offer's if_type/if_instance can change for every new hvsock
1066 * However, to facilitate the notification of new-offer/rescind-offer
1067 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1068 * a special vmbus device, and hence we need the below flag to
1069 * indicate if the driver is the hvsock driver or not: we need to
1070 * specially treat the hvosck offer & driver in vmbus_match().
1074 /* the device type supported by this driver */
1076 const struct hv_vmbus_device_id *id_table;
1078 struct device_driver driver;
1080 /* dynamic device GUID's */
1083 struct list_head list;
1086 int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
1087 int (*remove)(struct hv_device *);
1088 void (*shutdown)(struct hv_device *);
1092 /* Base device object */
1094 /* the device type id of this device */
1097 /* the device instance id of this device */
1098 uuid_le dev_instance;
1102 struct device device;
1104 struct vmbus_channel *channel;
1108 static inline struct hv_device *device_to_hv_device(struct device *d)
1110 return container_of(d, struct hv_device, device);
1113 static inline struct hv_driver *drv_to_hv_drv(struct device_driver *d)
1115 return container_of(d, struct hv_driver, driver);
1118 static inline void hv_set_drvdata(struct hv_device *dev, void *data)
1120 dev_set_drvdata(&dev->device, data);
1123 static inline void *hv_get_drvdata(struct hv_device *dev)
1125 return dev_get_drvdata(&dev->device);
1128 struct hv_ring_buffer_debug_info {
1129 u32 current_interrupt_mask;
1130 u32 current_read_index;
1131 u32 current_write_index;
1132 u32 bytes_avail_toread;
1133 u32 bytes_avail_towrite;
1136 void hv_ringbuffer_get_debuginfo(const struct hv_ring_buffer_info *ring_info,
1137 struct hv_ring_buffer_debug_info *debug_info);
1139 /* Vmbus interface */
1140 #define vmbus_driver_register(driver) \
1141 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1142 int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
1143 struct module *owner,
1144 const char *mod_name);
1145 void vmbus_driver_unregister(struct hv_driver *hv_driver);
1147 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel);
1149 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1150 resource_size_t min, resource_size_t max,
1151 resource_size_t size, resource_size_t align,
1152 bool fb_overlap_ok);
1153 void vmbus_free_mmio(resource_size_t start, resource_size_t size);
1154 int vmbus_cpu_number_to_vp_number(int cpu_number);
1155 u64 hv_do_hypercall(u64 control, void *input, void *output);
1158 * GUID definitions of various offer types - services offered to the guest.
1163 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1165 #define HV_NIC_GUID \
1166 .guid = UUID_LE(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1167 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1171 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1173 #define HV_IDE_GUID \
1174 .guid = UUID_LE(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1175 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1179 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1181 #define HV_SCSI_GUID \
1182 .guid = UUID_LE(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1183 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1187 * {0e0b6031-5213-4934-818b-38d90ced39db}
1189 #define HV_SHUTDOWN_GUID \
1190 .guid = UUID_LE(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1191 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1195 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1197 #define HV_TS_GUID \
1198 .guid = UUID_LE(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1199 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1203 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1205 #define HV_HEART_BEAT_GUID \
1206 .guid = UUID_LE(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1207 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1211 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1213 #define HV_KVP_GUID \
1214 .guid = UUID_LE(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1215 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1218 * Dynamic memory GUID
1219 * {525074dc-8985-46e2-8057-a307dc18a502}
1221 #define HV_DM_GUID \
1222 .guid = UUID_LE(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1223 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1227 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1229 #define HV_MOUSE_GUID \
1230 .guid = UUID_LE(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1231 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1235 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1237 #define HV_KBD_GUID \
1238 .guid = UUID_LE(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1239 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1242 * VSS (Backup/Restore) GUID
1244 #define HV_VSS_GUID \
1245 .guid = UUID_LE(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1246 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1248 * Synthetic Video GUID
1249 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1251 #define HV_SYNTHVID_GUID \
1252 .guid = UUID_LE(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1253 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1257 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1259 #define HV_SYNTHFC_GUID \
1260 .guid = UUID_LE(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1261 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1264 * Guest File Copy Service
1265 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1268 #define HV_FCOPY_GUID \
1269 .guid = UUID_LE(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1270 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1273 * NetworkDirect. This is the guest RDMA service.
1274 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1276 #define HV_ND_GUID \
1277 .guid = UUID_LE(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1278 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1281 * PCI Express Pass Through
1282 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1285 #define HV_PCIE_GUID \
1286 .guid = UUID_LE(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1287 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1290 * Linux doesn't support the 3 devices: the first two are for
1291 * Automatic Virtual Machine Activation, and the third is for
1292 * Remote Desktop Virtualization.
1293 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1294 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1295 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1298 #define HV_AVMA1_GUID \
1299 .guid = UUID_LE(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1300 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1302 #define HV_AVMA2_GUID \
1303 .guid = UUID_LE(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1304 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1306 #define HV_RDV_GUID \
1307 .guid = UUID_LE(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1308 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1311 * Common header for Hyper-V ICs
1314 #define ICMSGTYPE_NEGOTIATE 0
1315 #define ICMSGTYPE_HEARTBEAT 1
1316 #define ICMSGTYPE_KVPEXCHANGE 2
1317 #define ICMSGTYPE_SHUTDOWN 3
1318 #define ICMSGTYPE_TIMESYNC 4
1319 #define ICMSGTYPE_VSS 5
1321 #define ICMSGHDRFLAG_TRANSACTION 1
1322 #define ICMSGHDRFLAG_REQUEST 2
1323 #define ICMSGHDRFLAG_RESPONSE 4
1327 * While we want to handle util services as regular devices,
1328 * there is only one instance of each of these services; so
1329 * we statically allocate the service specific state.
1332 struct hv_util_service {
1335 void (*util_cb)(void *);
1336 int (*util_init)(struct hv_util_service *);
1337 void (*util_deinit)(void);
1340 struct vmbuspipe_hdr {
1351 struct ic_version icverframe;
1353 struct ic_version icvermsg;
1356 u8 ictransaction_id;
1361 struct icmsg_negotiate {
1365 struct ic_version icversion_data[1]; /* any size array */
1368 struct shutdown_msg_data {
1370 u32 timeout_seconds;
1372 u8 display_message[2048];
1375 struct heartbeat_msg_data {
1380 /* Time Sync IC defs */
1381 #define ICTIMESYNCFLAG_PROBE 0
1382 #define ICTIMESYNCFLAG_SYNC 1
1383 #define ICTIMESYNCFLAG_SAMPLE 2
1386 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1388 #define WLTIMEDELTA 116444736000000000LL
1391 struct ictimesync_data {
1398 struct ictimesync_ref_data {
1400 u64 vmreferencetime;
1407 struct hyperv_service_callback {
1411 struct vmbus_channel *channel;
1412 void (*callback)(void *context);
1415 #define MAX_SRV_VER 0x7ffffff
1416 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf,
1417 const int *fw_version, int fw_vercnt,
1418 const int *srv_version, int srv_vercnt,
1419 int *nego_fw_version, int *nego_srv_version);
1421 void hv_process_channel_removal(struct vmbus_channel *channel, u32 relid);
1423 void vmbus_setevent(struct vmbus_channel *channel);
1425 * Negotiated version with the Host.
1428 extern __u32 vmbus_proto_version;
1430 int vmbus_send_tl_connect_request(const uuid_le *shv_guest_servie_id,
1431 const uuid_le *shv_host_servie_id);
1432 void vmbus_set_event(struct vmbus_channel *channel);
1434 /* Get the start of the ring buffer. */
1435 static inline void *
1436 hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info)
1438 return ring_info->ring_buffer->buffer;
1442 * Mask off host interrupt callback notifications
1444 static inline void hv_begin_read(struct hv_ring_buffer_info *rbi)
1446 rbi->ring_buffer->interrupt_mask = 1;
1448 /* make sure mask update is not reordered */
1453 * Re-enable host callback and return number of outstanding bytes
1455 static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi)
1458 rbi->ring_buffer->interrupt_mask = 0;
1460 /* make sure mask update is not reordered */
1464 * Now check to see if the ring buffer is still empty.
1465 * If it is not, we raced and we need to process new
1466 * incoming messages.
1468 return hv_get_bytes_to_read(rbi);
1472 * An API to support in-place processing of incoming VMBUS packets.
1475 /* Get data payload associated with descriptor */
1476 static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc)
1478 return (void *)((unsigned long)desc + (desc->offset8 << 3));
1481 /* Get data size associated with descriptor */
1482 static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc)
1484 return (desc->len8 << 3) - (desc->offset8 << 3);
1488 struct vmpacket_descriptor *
1489 hv_pkt_iter_first(struct vmbus_channel *channel);
1491 struct vmpacket_descriptor *
1492 __hv_pkt_iter_next(struct vmbus_channel *channel,
1493 const struct vmpacket_descriptor *pkt);
1495 void hv_pkt_iter_close(struct vmbus_channel *channel);
1498 * Get next packet descriptor from iterator
1499 * If at end of list, return NULL and update host.
1501 static inline struct vmpacket_descriptor *
1502 hv_pkt_iter_next(struct vmbus_channel *channel,
1503 const struct vmpacket_descriptor *pkt)
1505 struct vmpacket_descriptor *nxt;
1507 nxt = __hv_pkt_iter_next(channel, pkt);
1509 hv_pkt_iter_close(channel);
1514 #define foreach_vmbus_pkt(pkt, channel) \
1515 for (pkt = hv_pkt_iter_first(channel); pkt; \
1516 pkt = hv_pkt_iter_next(channel, pkt))
1518 #endif /* _HYPERV_H */