Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / include / linux / hyperv.h
diff --git a/kernel/include/linux/hyperv.h b/kernel/include/linux/hyperv.h
new file mode 100644 (file)
index 0000000..902c37a
--- /dev/null
@@ -0,0 +1,1256 @@
+/*
+ *
+ * Copyright (c) 2011, Microsoft Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+ * Place - Suite 330, Boston, MA 02111-1307 USA.
+ *
+ * Authors:
+ *   Haiyang Zhang <haiyangz@microsoft.com>
+ *   Hank Janssen  <hjanssen@microsoft.com>
+ *   K. Y. Srinivasan <kys@microsoft.com>
+ *
+ */
+
+#ifndef _HYPERV_H
+#define _HYPERV_H
+
+#include <uapi/linux/hyperv.h>
+
+#include <linux/types.h>
+#include <linux/scatterlist.h>
+#include <linux/list.h>
+#include <linux/timer.h>
+#include <linux/workqueue.h>
+#include <linux/completion.h>
+#include <linux/device.h>
+#include <linux/mod_devicetable.h>
+
+
+#define MAX_PAGE_BUFFER_COUNT                          32
+#define MAX_MULTIPAGE_BUFFER_COUNT                     32 /* 128K */
+
+#pragma pack(push, 1)
+
+/* Single-page buffer */
+struct hv_page_buffer {
+       u32 len;
+       u32 offset;
+       u64 pfn;
+};
+
+/* Multiple-page buffer */
+struct hv_multipage_buffer {
+       /* Length and Offset determines the # of pfns in the array */
+       u32 len;
+       u32 offset;
+       u64 pfn_array[MAX_MULTIPAGE_BUFFER_COUNT];
+};
+
+/*
+ * Multiple-page buffer array; the pfn array is variable size:
+ * The number of entries in the PFN array is determined by
+ * "len" and "offset".
+ */
+struct hv_mpb_array {
+       /* Length and Offset determines the # of pfns in the array */
+       u32 len;
+       u32 offset;
+       u64 pfn_array[];
+};
+
+/* 0x18 includes the proprietary packet header */
+#define MAX_PAGE_BUFFER_PACKET         (0x18 +                 \
+                                       (sizeof(struct hv_page_buffer) * \
+                                        MAX_PAGE_BUFFER_COUNT))
+#define MAX_MULTIPAGE_BUFFER_PACKET    (0x18 +                 \
+                                        sizeof(struct hv_multipage_buffer))
+
+
+#pragma pack(pop)
+
+struct hv_ring_buffer {
+       /* Offset in bytes from the start of ring data below */
+       u32 write_index;
+
+       /* Offset in bytes from the start of ring data below */
+       u32 read_index;
+
+       u32 interrupt_mask;
+
+       /*
+        * Win8 uses some of the reserved bits to implement
+        * interrupt driven flow management. On the send side
+        * we can request that the receiver interrupt the sender
+        * when the ring transitions from being full to being able
+        * to handle a message of size "pending_send_sz".
+        *
+        * Add necessary state for this enhancement.
+        */
+       u32 pending_send_sz;
+
+       u32 reserved1[12];
+
+       union {
+               struct {
+                       u32 feat_pending_send_sz:1;
+               };
+               u32 value;
+       } feature_bits;
+
+       /* Pad it to PAGE_SIZE so that data starts on page boundary */
+       u8      reserved2[4028];
+
+       /*
+        * Ring data starts here + RingDataStartOffset
+        * !!! DO NOT place any fields below this !!!
+        */
+       u8 buffer[0];
+} __packed;
+
+struct hv_ring_buffer_info {
+       struct hv_ring_buffer *ring_buffer;
+       u32 ring_size;                  /* Include the shared header */
+       spinlock_t ring_lock;
+
+       u32 ring_datasize;              /* < ring_size */
+       u32 ring_data_startoffset;
+};
+
+/*
+ *
+ * hv_get_ringbuffer_availbytes()
+ *
+ * Get number of bytes available to read and to write to
+ * for the specified ring buffer
+ */
+static inline void
+hv_get_ringbuffer_availbytes(struct hv_ring_buffer_info *rbi,
+                         u32 *read, u32 *write)
+{
+       u32 read_loc, write_loc, dsize;
+
+       smp_read_barrier_depends();
+
+       /* Capture the read/write indices before they changed */
+       read_loc = rbi->ring_buffer->read_index;
+       write_loc = rbi->ring_buffer->write_index;
+       dsize = rbi->ring_datasize;
+
+       *write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
+               read_loc - write_loc;
+       *read = dsize - *write;
+}
+
+/*
+ * VMBUS version is 32 bit entity broken up into
+ * two 16 bit quantities: major_number. minor_number.
+ *
+ * 0 . 13 (Windows Server 2008)
+ * 1 . 1  (Windows 7)
+ * 2 . 4  (Windows 8)
+ * 3 . 0  (Windows 8 R2)
+ */
+
+#define VERSION_WS2008  ((0 << 16) | (13))
+#define VERSION_WIN7    ((1 << 16) | (1))
+#define VERSION_WIN8    ((2 << 16) | (4))
+#define VERSION_WIN8_1    ((3 << 16) | (0))
+
+#define VERSION_INVAL -1
+
+#define VERSION_CURRENT VERSION_WIN8_1
+
+/* Make maximum size of pipe payload of 16K */
+#define MAX_PIPE_DATA_PAYLOAD          (sizeof(u8) * 16384)
+
+/* Define PipeMode values. */
+#define VMBUS_PIPE_TYPE_BYTE           0x00000000
+#define VMBUS_PIPE_TYPE_MESSAGE                0x00000004
+
+/* The size of the user defined data buffer for non-pipe offers. */
+#define MAX_USER_DEFINED_BYTES         120
+
+/* The size of the user defined data buffer for pipe offers. */
+#define MAX_PIPE_USER_DEFINED_BYTES    116
+
+/*
+ * At the center of the Channel Management library is the Channel Offer. This
+ * struct contains the fundamental information about an offer.
+ */
+struct vmbus_channel_offer {
+       uuid_le if_type;
+       uuid_le if_instance;
+
+       /*
+        * These two fields are not currently used.
+        */
+       u64 reserved1;
+       u64 reserved2;
+
+       u16 chn_flags;
+       u16 mmio_megabytes;             /* in bytes * 1024 * 1024 */
+
+       union {
+               /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
+               struct {
+                       unsigned char user_def[MAX_USER_DEFINED_BYTES];
+               } std;
+
+               /*
+                * Pipes:
+                * The following sructure is an integrated pipe protocol, which
+                * is implemented on top of standard user-defined data. Pipe
+                * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
+                * use.
+                */
+               struct {
+                       u32  pipe_mode;
+                       unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
+               } pipe;
+       } u;
+       /*
+        * The sub_channel_index is defined in win8.
+        */
+       u16 sub_channel_index;
+       u16 reserved3;
+} __packed;
+
+/* Server Flags */
+#define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE       1
+#define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES   2
+#define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS           4
+#define VMBUS_CHANNEL_NAMED_PIPE_MODE                  0x10
+#define VMBUS_CHANNEL_LOOPBACK_OFFER                   0x100
+#define VMBUS_CHANNEL_PARENT_OFFER                     0x200
+#define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION   0x400
+
+struct vmpacket_descriptor {
+       u16 type;
+       u16 offset8;
+       u16 len8;
+       u16 flags;
+       u64 trans_id;
+} __packed;
+
+struct vmpacket_header {
+       u32 prev_pkt_start_offset;
+       struct vmpacket_descriptor descriptor;
+} __packed;
+
+struct vmtransfer_page_range {
+       u32 byte_count;
+       u32 byte_offset;
+} __packed;
+
+struct vmtransfer_page_packet_header {
+       struct vmpacket_descriptor d;
+       u16 xfer_pageset_id;
+       u8  sender_owns_set;
+       u8 reserved;
+       u32 range_cnt;
+       struct vmtransfer_page_range ranges[1];
+} __packed;
+
+struct vmgpadl_packet_header {
+       struct vmpacket_descriptor d;
+       u32 gpadl;
+       u32 reserved;
+} __packed;
+
+struct vmadd_remove_transfer_page_set {
+       struct vmpacket_descriptor d;
+       u32 gpadl;
+       u16 xfer_pageset_id;
+       u16 reserved;
+} __packed;
+
+/*
+ * This structure defines a range in guest physical space that can be made to
+ * look virtually contiguous.
+ */
+struct gpa_range {
+       u32 byte_count;
+       u32 byte_offset;
+       u64 pfn_array[0];
+};
+
+/*
+ * This is the format for an Establish Gpadl packet, which contains a handle by
+ * which this GPADL will be known and a set of GPA ranges associated with it.
+ * This can be converted to a MDL by the guest OS.  If there are multiple GPA
+ * ranges, then the resulting MDL will be "chained," representing multiple VA
+ * ranges.
+ */
+struct vmestablish_gpadl {
+       struct vmpacket_descriptor d;
+       u32 gpadl;
+       u32 range_cnt;
+       struct gpa_range range[1];
+} __packed;
+
+/*
+ * This is the format for a Teardown Gpadl packet, which indicates that the
+ * GPADL handle in the Establish Gpadl packet will never be referenced again.
+ */
+struct vmteardown_gpadl {
+       struct vmpacket_descriptor d;
+       u32 gpadl;
+       u32 reserved;   /* for alignment to a 8-byte boundary */
+} __packed;
+
+/*
+ * This is the format for a GPA-Direct packet, which contains a set of GPA
+ * ranges, in addition to commands and/or data.
+ */
+struct vmdata_gpa_direct {
+       struct vmpacket_descriptor d;
+       u32 reserved;
+       u32 range_cnt;
+       struct gpa_range range[1];
+} __packed;
+
+/* This is the format for a Additional Data Packet. */
+struct vmadditional_data {
+       struct vmpacket_descriptor d;
+       u64 total_bytes;
+       u32 offset;
+       u32 byte_cnt;
+       unsigned char data[1];
+} __packed;
+
+union vmpacket_largest_possible_header {
+       struct vmpacket_descriptor simple_hdr;
+       struct vmtransfer_page_packet_header xfer_page_hdr;
+       struct vmgpadl_packet_header gpadl_hdr;
+       struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr;
+       struct vmestablish_gpadl establish_gpadl_hdr;
+       struct vmteardown_gpadl teardown_gpadl_hdr;
+       struct vmdata_gpa_direct data_gpa_direct_hdr;
+};
+
+#define VMPACKET_DATA_START_ADDRESS(__packet)  \
+       (void *)(((unsigned char *)__packet) +  \
+        ((struct vmpacket_descriptor)__packet)->offset8 * 8)
+
+#define VMPACKET_DATA_LENGTH(__packet)         \
+       ((((struct vmpacket_descriptor)__packet)->len8 -        \
+         ((struct vmpacket_descriptor)__packet)->offset8) * 8)
+
+#define VMPACKET_TRANSFER_MODE(__packet)       \
+       (((struct IMPACT)__packet)->type)
+
+enum vmbus_packet_type {
+       VM_PKT_INVALID                          = 0x0,
+       VM_PKT_SYNCH                            = 0x1,
+       VM_PKT_ADD_XFER_PAGESET                 = 0x2,
+       VM_PKT_RM_XFER_PAGESET                  = 0x3,
+       VM_PKT_ESTABLISH_GPADL                  = 0x4,
+       VM_PKT_TEARDOWN_GPADL                   = 0x5,
+       VM_PKT_DATA_INBAND                      = 0x6,
+       VM_PKT_DATA_USING_XFER_PAGES            = 0x7,
+       VM_PKT_DATA_USING_GPADL                 = 0x8,
+       VM_PKT_DATA_USING_GPA_DIRECT            = 0x9,
+       VM_PKT_CANCEL_REQUEST                   = 0xa,
+       VM_PKT_COMP                             = 0xb,
+       VM_PKT_DATA_USING_ADDITIONAL_PKT        = 0xc,
+       VM_PKT_ADDITIONAL_DATA                  = 0xd
+};
+
+#define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED    1
+
+
+/* Version 1 messages */
+enum vmbus_channel_message_type {
+       CHANNELMSG_INVALID                      =  0,
+       CHANNELMSG_OFFERCHANNEL         =  1,
+       CHANNELMSG_RESCIND_CHANNELOFFER =  2,
+       CHANNELMSG_REQUESTOFFERS                =  3,
+       CHANNELMSG_ALLOFFERS_DELIVERED  =  4,
+       CHANNELMSG_OPENCHANNEL          =  5,
+       CHANNELMSG_OPENCHANNEL_RESULT           =  6,
+       CHANNELMSG_CLOSECHANNEL         =  7,
+       CHANNELMSG_GPADL_HEADER         =  8,
+       CHANNELMSG_GPADL_BODY                   =  9,
+       CHANNELMSG_GPADL_CREATED                = 10,
+       CHANNELMSG_GPADL_TEARDOWN               = 11,
+       CHANNELMSG_GPADL_TORNDOWN               = 12,
+       CHANNELMSG_RELID_RELEASED               = 13,
+       CHANNELMSG_INITIATE_CONTACT             = 14,
+       CHANNELMSG_VERSION_RESPONSE             = 15,
+       CHANNELMSG_UNLOAD                       = 16,
+#ifdef VMBUS_FEATURE_PARENT_OR_PEER_MEMORY_MAPPED_INTO_A_CHILD
+       CHANNELMSG_VIEWRANGE_ADD                = 17,
+       CHANNELMSG_VIEWRANGE_REMOVE             = 18,
+#endif
+       CHANNELMSG_COUNT
+};
+
+struct vmbus_channel_message_header {
+       enum vmbus_channel_message_type msgtype;
+       u32 padding;
+} __packed;
+
+/* Query VMBus Version parameters */
+struct vmbus_channel_query_vmbus_version {
+       struct vmbus_channel_message_header header;
+       u32 version;
+} __packed;
+
+/* VMBus Version Supported parameters */
+struct vmbus_channel_version_supported {
+       struct vmbus_channel_message_header header;
+       u8 version_supported;
+} __packed;
+
+/* Offer Channel parameters */
+struct vmbus_channel_offer_channel {
+       struct vmbus_channel_message_header header;
+       struct vmbus_channel_offer offer;
+       u32 child_relid;
+       u8 monitorid;
+       /*
+        * win7 and beyond splits this field into a bit field.
+        */
+       u8 monitor_allocated:1;
+       u8 reserved:7;
+       /*
+        * These are new fields added in win7 and later.
+        * Do not access these fields without checking the
+        * negotiated protocol.
+        *
+        * If "is_dedicated_interrupt" is set, we must not set the
+        * associated bit in the channel bitmap while sending the
+        * interrupt to the host.
+        *
+        * connection_id is to be used in signaling the host.
+        */
+       u16 is_dedicated_interrupt:1;
+       u16 reserved1:15;
+       u32 connection_id;
+} __packed;
+
+/* Rescind Offer parameters */
+struct vmbus_channel_rescind_offer {
+       struct vmbus_channel_message_header header;
+       u32 child_relid;
+} __packed;
+
+/*
+ * Request Offer -- no parameters, SynIC message contains the partition ID
+ * Set Snoop -- no parameters, SynIC message contains the partition ID
+ * Clear Snoop -- no parameters, SynIC message contains the partition ID
+ * All Offers Delivered -- no parameters, SynIC message contains the partition
+ *                        ID
+ * Flush Client -- no parameters, SynIC message contains the partition ID
+ */
+
+/* Open Channel parameters */
+struct vmbus_channel_open_channel {
+       struct vmbus_channel_message_header header;
+
+       /* Identifies the specific VMBus channel that is being opened. */
+       u32 child_relid;
+
+       /* ID making a particular open request at a channel offer unique. */
+       u32 openid;
+
+       /* GPADL for the channel's ring buffer. */
+       u32 ringbuffer_gpadlhandle;
+
+       /*
+        * Starting with win8, this field will be used to specify
+        * the target virtual processor on which to deliver the interrupt for
+        * the host to guest communication.
+        * Prior to win8, incoming channel interrupts would only
+        * be delivered on cpu 0. Setting this value to 0 would
+        * preserve the earlier behavior.
+        */
+       u32 target_vp;
+
+       /*
+       * The upstream ring buffer begins at offset zero in the memory
+       * described by RingBufferGpadlHandle. The downstream ring buffer
+       * follows it at this offset (in pages).
+       */
+       u32 downstream_ringbuffer_pageoffset;
+
+       /* User-specific data to be passed along to the server endpoint. */
+       unsigned char userdata[MAX_USER_DEFINED_BYTES];
+} __packed;
+
+/* Open Channel Result parameters */
+struct vmbus_channel_open_result {
+       struct vmbus_channel_message_header header;
+       u32 child_relid;
+       u32 openid;
+       u32 status;
+} __packed;
+
+/* Close channel parameters; */
+struct vmbus_channel_close_channel {
+       struct vmbus_channel_message_header header;
+       u32 child_relid;
+} __packed;
+
+/* Channel Message GPADL */
+#define GPADL_TYPE_RING_BUFFER         1
+#define GPADL_TYPE_SERVER_SAVE_AREA    2
+#define GPADL_TYPE_TRANSACTION         8
+
+/*
+ * The number of PFNs in a GPADL message is defined by the number of
+ * pages that would be spanned by ByteCount and ByteOffset.  If the
+ * implied number of PFNs won't fit in this packet, there will be a
+ * follow-up packet that contains more.
+ */
+struct vmbus_channel_gpadl_header {
+       struct vmbus_channel_message_header header;
+       u32 child_relid;
+       u32 gpadl;
+       u16 range_buflen;
+       u16 rangecount;
+       struct gpa_range range[0];
+} __packed;
+
+/* This is the followup packet that contains more PFNs. */
+struct vmbus_channel_gpadl_body {
+       struct vmbus_channel_message_header header;
+       u32 msgnumber;
+       u32 gpadl;
+       u64 pfn[0];
+} __packed;
+
+struct vmbus_channel_gpadl_created {
+       struct vmbus_channel_message_header header;
+       u32 child_relid;
+       u32 gpadl;
+       u32 creation_status;
+} __packed;
+
+struct vmbus_channel_gpadl_teardown {
+       struct vmbus_channel_message_header header;
+       u32 child_relid;
+       u32 gpadl;
+} __packed;
+
+struct vmbus_channel_gpadl_torndown {
+       struct vmbus_channel_message_header header;
+       u32 gpadl;
+} __packed;
+
+#ifdef VMBUS_FEATURE_PARENT_OR_PEER_MEMORY_MAPPED_INTO_A_CHILD
+struct vmbus_channel_view_range_add {
+       struct vmbus_channel_message_header header;
+       PHYSICAL_ADDRESS viewrange_base;
+       u64 viewrange_length;
+       u32 child_relid;
+} __packed;
+
+struct vmbus_channel_view_range_remove {
+       struct vmbus_channel_message_header header;
+       PHYSICAL_ADDRESS viewrange_base;
+       u32 child_relid;
+} __packed;
+#endif
+
+struct vmbus_channel_relid_released {
+       struct vmbus_channel_message_header header;
+       u32 child_relid;
+} __packed;
+
+struct vmbus_channel_initiate_contact {
+       struct vmbus_channel_message_header header;
+       u32 vmbus_version_requested;
+       u32 target_vcpu; /* The VCPU the host should respond to */
+       u64 interrupt_page;
+       u64 monitor_page1;
+       u64 monitor_page2;
+} __packed;
+
+struct vmbus_channel_version_response {
+       struct vmbus_channel_message_header header;
+       u8 version_supported;
+} __packed;
+
+enum vmbus_channel_state {
+       CHANNEL_OFFER_STATE,
+       CHANNEL_OPENING_STATE,
+       CHANNEL_OPEN_STATE,
+       CHANNEL_OPENED_STATE,
+};
+
+/*
+ * Represents each channel msg on the vmbus connection This is a
+ * variable-size data structure depending on the msg type itself
+ */
+struct vmbus_channel_msginfo {
+       /* Bookkeeping stuff */
+       struct list_head msglistentry;
+
+       /* So far, this is only used to handle gpadl body message */
+       struct list_head submsglist;
+
+       /* Synchronize the request/response if needed */
+       struct completion  waitevent;
+       union {
+               struct vmbus_channel_version_supported version_supported;
+               struct vmbus_channel_open_result open_result;
+               struct vmbus_channel_gpadl_torndown gpadl_torndown;
+               struct vmbus_channel_gpadl_created gpadl_created;
+               struct vmbus_channel_version_response version_response;
+       } response;
+
+       u32 msgsize;
+       /*
+        * The channel message that goes out on the "wire".
+        * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
+        */
+       unsigned char msg[0];
+};
+
+struct vmbus_close_msg {
+       struct vmbus_channel_msginfo info;
+       struct vmbus_channel_close_channel msg;
+};
+
+/* Define connection identifier type. */
+union hv_connection_id {
+       u32 asu32;
+       struct {
+               u32 id:24;
+               u32 reserved:8;
+       } u;
+};
+
+/* Definition of the hv_signal_event hypercall input structure. */
+struct hv_input_signal_event {
+       union hv_connection_id connectionid;
+       u16 flag_number;
+       u16 rsvdz;
+};
+
+struct hv_input_signal_event_buffer {
+       u64 align8;
+       struct hv_input_signal_event event;
+};
+
+struct vmbus_channel {
+       /* Unique channel id */
+       int id;
+
+       struct list_head listentry;
+
+       struct hv_device *device_obj;
+
+       enum vmbus_channel_state state;
+
+       struct vmbus_channel_offer_channel offermsg;
+       /*
+        * These are based on the OfferMsg.MonitorId.
+        * Save it here for easy access.
+        */
+       u8 monitor_grp;
+       u8 monitor_bit;
+
+       bool rescind; /* got rescind msg */
+
+       u32 ringbuffer_gpadlhandle;
+
+       /* Allocated memory for ring buffer */
+       void *ringbuffer_pages;
+       u32 ringbuffer_pagecount;
+       struct hv_ring_buffer_info outbound;    /* send to parent */
+       struct hv_ring_buffer_info inbound;     /* receive from parent */
+       spinlock_t inbound_lock;
+
+       struct vmbus_close_msg close_msg;
+
+       /* Channel callback are invoked in this workqueue context */
+       /* HANDLE dataWorkQueue; */
+
+       void (*onchannel_callback)(void *context);
+       void *channel_callback_context;
+
+       /*
+        * A channel can be marked for efficient (batched)
+        * reading:
+        * If batched_reading is set to "true", we read until the
+        * channel is empty and hold off interrupts from the host
+        * during the entire read process.
+        * If batched_reading is set to "false", the client is not
+        * going to perform batched reading.
+        *
+        * By default we will enable batched reading; specific
+        * drivers that don't want this behavior can turn it off.
+        */
+
+       bool batched_reading;
+
+       bool is_dedicated_interrupt;
+       struct hv_input_signal_event_buffer sig_buf;
+       struct hv_input_signal_event *sig_event;
+
+       /*
+        * Starting with win8, this field will be used to specify
+        * the target virtual processor on which to deliver the interrupt for
+        * the host to guest communication.
+        * Prior to win8, incoming channel interrupts would only
+        * be delivered on cpu 0. Setting this value to 0 would
+        * preserve the earlier behavior.
+        */
+       u32 target_vp;
+       /* The corresponding CPUID in the guest */
+       u32 target_cpu;
+       /*
+        * Support for sub-channels. For high performance devices,
+        * it will be useful to have multiple sub-channels to support
+        * a scalable communication infrastructure with the host.
+        * The support for sub-channels is implemented as an extention
+        * to the current infrastructure.
+        * The initial offer is considered the primary channel and this
+        * offer message will indicate if the host supports sub-channels.
+        * The guest is free to ask for sub-channels to be offerred and can
+        * open these sub-channels as a normal "primary" channel. However,
+        * all sub-channels will have the same type and instance guids as the
+        * primary channel. Requests sent on a given channel will result in a
+        * response on the same channel.
+        */
+
+       /*
+        * Sub-channel creation callback. This callback will be called in
+        * process context when a sub-channel offer is received from the host.
+        * The guest can open the sub-channel in the context of this callback.
+        */
+       void (*sc_creation_callback)(struct vmbus_channel *new_sc);
+
+       /*
+        * The spinlock to protect the structure. It is being used to protect
+        * test-and-set access to various attributes of the structure as well
+        * as all sc_list operations.
+        */
+       spinlock_t lock;
+       /*
+        * All Sub-channels of a primary channel are linked here.
+        */
+       struct list_head sc_list;
+       /*
+        * The primary channel this sub-channel belongs to.
+        * This will be NULL for the primary channel.
+        */
+       struct vmbus_channel *primary_channel;
+       /*
+        * Support per-channel state for use by vmbus drivers.
+        */
+       void *per_channel_state;
+       /*
+        * To support per-cpu lookup mapping of relid to channel,
+        * link up channels based on their CPU affinity.
+        */
+       struct list_head percpu_list;
+
+       int num_sc;
+       int next_oc;
+};
+
+static inline void set_channel_read_state(struct vmbus_channel *c, bool state)
+{
+       c->batched_reading = state;
+}
+
+static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
+{
+       c->per_channel_state = s;
+}
+
+static inline void *get_per_channel_state(struct vmbus_channel *c)
+{
+       return c->per_channel_state;
+}
+
+void vmbus_onmessage(void *context);
+
+int vmbus_request_offers(void);
+
+/*
+ * APIs for managing sub-channels.
+ */
+
+void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
+                       void (*sc_cr_cb)(struct vmbus_channel *new_sc));
+
+/*
+ * Retrieve the (sub) channel on which to send an outgoing request.
+ * When a primary channel has multiple sub-channels, we choose a
+ * channel whose VCPU binding is closest to the VCPU on which
+ * this call is being made.
+ */
+struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary);
+
+/*
+ * Check if sub-channels have already been offerred. This API will be useful
+ * when the driver is unloaded after establishing sub-channels. In this case,
+ * when the driver is re-loaded, the driver would have to check if the
+ * subchannels have already been established before attempting to request
+ * the creation of sub-channels.
+ * This function returns TRUE to indicate that subchannels have already been
+ * created.
+ * This function should be invoked after setting the callback function for
+ * sub-channel creation.
+ */
+bool vmbus_are_subchannels_present(struct vmbus_channel *primary);
+
+/* The format must be the same as struct vmdata_gpa_direct */
+struct vmbus_channel_packet_page_buffer {
+       u16 type;
+       u16 dataoffset8;
+       u16 length8;
+       u16 flags;
+       u64 transactionid;
+       u32 reserved;
+       u32 rangecount;
+       struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
+} __packed;
+
+/* The format must be the same as struct vmdata_gpa_direct */
+struct vmbus_channel_packet_multipage_buffer {
+       u16 type;
+       u16 dataoffset8;
+       u16 length8;
+       u16 flags;
+       u64 transactionid;
+       u32 reserved;
+       u32 rangecount;         /* Always 1 in this case */
+       struct hv_multipage_buffer range;
+} __packed;
+
+/* The format must be the same as struct vmdata_gpa_direct */
+struct vmbus_packet_mpb_array {
+       u16 type;
+       u16 dataoffset8;
+       u16 length8;
+       u16 flags;
+       u64 transactionid;
+       u32 reserved;
+       u32 rangecount;         /* Always 1 in this case */
+       struct hv_mpb_array range;
+} __packed;
+
+
+extern int vmbus_open(struct vmbus_channel *channel,
+                           u32 send_ringbuffersize,
+                           u32 recv_ringbuffersize,
+                           void *userdata,
+                           u32 userdatalen,
+                           void(*onchannel_callback)(void *context),
+                           void *context);
+
+extern void vmbus_close(struct vmbus_channel *channel);
+
+extern int vmbus_sendpacket(struct vmbus_channel *channel,
+                                 void *buffer,
+                                 u32 bufferLen,
+                                 u64 requestid,
+                                 enum vmbus_packet_type type,
+                                 u32 flags);
+
+extern int vmbus_sendpacket_ctl(struct vmbus_channel *channel,
+                                 void *buffer,
+                                 u32 bufferLen,
+                                 u64 requestid,
+                                 enum vmbus_packet_type type,
+                                 u32 flags,
+                                 bool kick_q);
+
+extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
+                                           struct hv_page_buffer pagebuffers[],
+                                           u32 pagecount,
+                                           void *buffer,
+                                           u32 bufferlen,
+                                           u64 requestid);
+
+extern int vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel *channel,
+                                          struct hv_page_buffer pagebuffers[],
+                                          u32 pagecount,
+                                          void *buffer,
+                                          u32 bufferlen,
+                                          u64 requestid,
+                                          u32 flags,
+                                          bool kick_q);
+
+extern int vmbus_sendpacket_multipagebuffer(struct vmbus_channel *channel,
+                                       struct hv_multipage_buffer *mpb,
+                                       void *buffer,
+                                       u32 bufferlen,
+                                       u64 requestid);
+
+extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
+                                    struct vmbus_packet_mpb_array *mpb,
+                                    u32 desc_size,
+                                    void *buffer,
+                                    u32 bufferlen,
+                                    u64 requestid);
+
+extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
+                                     void *kbuffer,
+                                     u32 size,
+                                     u32 *gpadl_handle);
+
+extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
+                                    u32 gpadl_handle);
+
+extern int vmbus_recvpacket(struct vmbus_channel *channel,
+                                 void *buffer,
+                                 u32 bufferlen,
+                                 u32 *buffer_actual_len,
+                                 u64 *requestid);
+
+extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
+                                    void *buffer,
+                                    u32 bufferlen,
+                                    u32 *buffer_actual_len,
+                                    u64 *requestid);
+
+
+extern void vmbus_ontimer(unsigned long data);
+
+/* Base driver object */
+struct hv_driver {
+       const char *name;
+
+       /* the device type supported by this driver */
+       uuid_le dev_type;
+       const struct hv_vmbus_device_id *id_table;
+
+       struct device_driver driver;
+
+       int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
+       int (*remove)(struct hv_device *);
+       void (*shutdown)(struct hv_device *);
+
+};
+
+/* Base device object */
+struct hv_device {
+       /* the device type id of this device */
+       uuid_le dev_type;
+
+       /* the device instance id of this device */
+       uuid_le dev_instance;
+
+       struct device device;
+
+       struct vmbus_channel *channel;
+};
+
+
+static inline struct hv_device *device_to_hv_device(struct device *d)
+{
+       return container_of(d, struct hv_device, device);
+}
+
+static inline struct hv_driver *drv_to_hv_drv(struct device_driver *d)
+{
+       return container_of(d, struct hv_driver, driver);
+}
+
+static inline void hv_set_drvdata(struct hv_device *dev, void *data)
+{
+       dev_set_drvdata(&dev->device, data);
+}
+
+static inline void *hv_get_drvdata(struct hv_device *dev)
+{
+       return dev_get_drvdata(&dev->device);
+}
+
+/* Vmbus interface */
+#define vmbus_driver_register(driver)  \
+       __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
+int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
+                                        struct module *owner,
+                                        const char *mod_name);
+void vmbus_driver_unregister(struct hv_driver *hv_driver);
+
+/**
+ * VMBUS_DEVICE - macro used to describe a specific hyperv vmbus device
+ *
+ * This macro is used to create a struct hv_vmbus_device_id that matches a
+ * specific device.
+ */
+#define VMBUS_DEVICE(g0, g1, g2, g3, g4, g5, g6, g7,   \
+                    g8, g9, ga, gb, gc, gd, ge, gf)    \
+       .guid = { g0, g1, g2, g3, g4, g5, g6, g7,       \
+                 g8, g9, ga, gb, gc, gd, ge, gf },
+
+/*
+ * GUID definitions of various offer types - services offered to the guest.
+ */
+
+/*
+ * Network GUID
+ * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
+ */
+#define HV_NIC_GUID \
+       .guid = { \
+                       0x63, 0x51, 0x61, 0xf8, 0x3e, 0xdf, 0xc5, 0x46, \
+                       0x91, 0x3f, 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e \
+               }
+
+/*
+ * IDE GUID
+ * {32412632-86cb-44a2-9b5c-50d1417354f5}
+ */
+#define HV_IDE_GUID \
+       .guid = { \
+                       0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44, \
+                       0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5 \
+               }
+
+/*
+ * SCSI GUID
+ * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
+ */
+#define HV_SCSI_GUID \
+       .guid = { \
+                       0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d, \
+                       0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f \
+               }
+
+/*
+ * Shutdown GUID
+ * {0e0b6031-5213-4934-818b-38d90ced39db}
+ */
+#define HV_SHUTDOWN_GUID \
+       .guid = { \
+                       0x31, 0x60, 0x0b, 0x0e, 0x13, 0x52, 0x34, 0x49, \
+                       0x81, 0x8b, 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb \
+               }
+
+/*
+ * Time Synch GUID
+ * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
+ */
+#define HV_TS_GUID \
+       .guid = { \
+                       0x30, 0xe6, 0x27, 0x95, 0xae, 0xd0, 0x7b, 0x49, \
+                       0xad, 0xce, 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf \
+               }
+
+/*
+ * Heartbeat GUID
+ * {57164f39-9115-4e78-ab55-382f3bd5422d}
+ */
+#define HV_HEART_BEAT_GUID \
+       .guid = { \
+                       0x39, 0x4f, 0x16, 0x57, 0x15, 0x91, 0x78, 0x4e, \
+                       0xab, 0x55, 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d \
+               }
+
+/*
+ * KVP GUID
+ * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
+ */
+#define HV_KVP_GUID \
+       .guid = { \
+                       0xe7, 0xf4, 0xa0, 0xa9, 0x45, 0x5a, 0x96, 0x4d, \
+                       0xb8, 0x27, 0x8a, 0x84, 0x1e, 0x8c, 0x3,  0xe6 \
+               }
+
+/*
+ * Dynamic memory GUID
+ * {525074dc-8985-46e2-8057-a307dc18a502}
+ */
+#define HV_DM_GUID \
+       .guid = { \
+                       0xdc, 0x74, 0x50, 0X52, 0x85, 0x89, 0xe2, 0x46, \
+                       0x80, 0x57, 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02 \
+               }
+
+/*
+ * Mouse GUID
+ * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
+ */
+#define HV_MOUSE_GUID \
+       .guid = { \
+                       0x9e, 0xb6, 0xa8, 0xcf, 0x4a, 0x5b, 0xc0, 0x4c, \
+                       0xb9, 0x8b, 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a \
+               }
+
+/*
+ * VSS (Backup/Restore) GUID
+ */
+#define HV_VSS_GUID \
+       .guid = { \
+                       0x29, 0x2e, 0xfa, 0x35, 0x23, 0xea, 0x36, 0x42, \
+                       0x96, 0xae, 0x3a, 0x6e, 0xba, 0xcb, 0xa4,  0x40 \
+               }
+/*
+ * Synthetic Video GUID
+ * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
+ */
+#define HV_SYNTHVID_GUID \
+       .guid = { \
+                       0x02, 0x78, 0x0a, 0xda, 0x77, 0xe3, 0xac, 0x4a, \
+                       0x8e, 0x77, 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8 \
+               }
+
+/*
+ * Synthetic FC GUID
+ * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
+ */
+#define HV_SYNTHFC_GUID \
+       .guid = { \
+                       0x4A, 0xCC, 0x9B, 0x2F, 0x69, 0x00, 0xF3, 0x4A, \
+                       0xB7, 0x6B, 0x6F, 0xD0, 0xBE, 0x52, 0x8C, 0xDA \
+               }
+
+/*
+ * Guest File Copy Service
+ * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
+ */
+
+#define HV_FCOPY_GUID \
+       .guid = { \
+                       0xE3, 0x4B, 0xD1, 0x34, 0xE4, 0xDE, 0xC8, 0x41, \
+                       0x9A, 0xE7, 0x6B, 0x17, 0x49, 0x77, 0xC1, 0x92 \
+               }
+
+/*
+ * NetworkDirect. This is the guest RDMA service.
+ * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
+ */
+#define HV_ND_GUID \
+       .guid = { \
+                       0x3d, 0xaf, 0x2e, 0x8c, 0xa7, 0x32, 0x09, 0x4b, \
+                       0xab, 0x99, 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01 \
+               }
+
+/*
+ * Common header for Hyper-V ICs
+ */
+
+#define ICMSGTYPE_NEGOTIATE            0
+#define ICMSGTYPE_HEARTBEAT            1
+#define ICMSGTYPE_KVPEXCHANGE          2
+#define ICMSGTYPE_SHUTDOWN             3
+#define ICMSGTYPE_TIMESYNC             4
+#define ICMSGTYPE_VSS                  5
+
+#define ICMSGHDRFLAG_TRANSACTION       1
+#define ICMSGHDRFLAG_REQUEST           2
+#define ICMSGHDRFLAG_RESPONSE          4
+
+
+/*
+ * While we want to handle util services as regular devices,
+ * there is only one instance of each of these services; so
+ * we statically allocate the service specific state.
+ */
+
+struct hv_util_service {
+       u8 *recv_buffer;
+       void (*util_cb)(void *);
+       int (*util_init)(struct hv_util_service *);
+       void (*util_deinit)(void);
+};
+
+struct vmbuspipe_hdr {
+       u32 flags;
+       u32 msgsize;
+} __packed;
+
+struct ic_version {
+       u16 major;
+       u16 minor;
+} __packed;
+
+struct icmsg_hdr {
+       struct ic_version icverframe;
+       u16 icmsgtype;
+       struct ic_version icvermsg;
+       u16 icmsgsize;
+       u32 status;
+       u8 ictransaction_id;
+       u8 icflags;
+       u8 reserved[2];
+} __packed;
+
+struct icmsg_negotiate {
+       u16 icframe_vercnt;
+       u16 icmsg_vercnt;
+       u32 reserved;
+       struct ic_version icversion_data[1]; /* any size array */
+} __packed;
+
+struct shutdown_msg_data {
+       u32 reason_code;
+       u32 timeout_seconds;
+       u32 flags;
+       u8  display_message[2048];
+} __packed;
+
+struct heartbeat_msg_data {
+       u64 seq_num;
+       u32 reserved[8];
+} __packed;
+
+/* Time Sync IC defs */
+#define ICTIMESYNCFLAG_PROBE   0
+#define ICTIMESYNCFLAG_SYNC    1
+#define ICTIMESYNCFLAG_SAMPLE  2
+
+#ifdef __x86_64__
+#define WLTIMEDELTA    116444736000000000L     /* in 100ns unit */
+#else
+#define WLTIMEDELTA    116444736000000000LL
+#endif
+
+struct ictimesync_data {
+       u64 parenttime;
+       u64 childtime;
+       u64 roundtriptime;
+       u8 flags;
+} __packed;
+
+struct hyperv_service_callback {
+       u8 msg_type;
+       char *log_msg;
+       uuid_le data;
+       struct vmbus_channel *channel;
+       void (*callback) (void *context);
+};
+
+#define MAX_SRV_VER    0x7ffffff
+extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *,
+                                       struct icmsg_negotiate *, u8 *, int,
+                                       int);
+
+int hv_kvp_init(struct hv_util_service *);
+void hv_kvp_deinit(void);
+void hv_kvp_onchannelcallback(void *);
+
+int hv_vss_init(struct hv_util_service *);
+void hv_vss_deinit(void);
+void hv_vss_onchannelcallback(void *);
+void hv_process_channel_removal(struct vmbus_channel *channel, u32 relid);
+
+extern struct resource hyperv_mmio;
+
+/*
+ * Negotiated version with the Host.
+ */
+
+extern __u32 vmbus_proto_version;
+
+#endif /* _HYPERV_H */