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.
19 * Haiyang Zhang <haiyangz@microsoft.com>
20 * Hank Janssen <hjanssen@microsoft.com>
21 * K. Y. Srinivasan <kys@microsoft.com>
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 */
127 u32 ring_data_startoffset
;
128 u32 priv_write_index
;
134 * hv_get_ringbuffer_availbytes()
136 * Get number of bytes available to read and to write to
137 * for the specified ring buffer
140 hv_get_ringbuffer_availbytes(const struct hv_ring_buffer_info
*rbi
,
141 u32
*read
, u32
*write
)
143 u32 read_loc
, write_loc
, dsize
;
145 /* Capture the read/write indices before they changed */
146 read_loc
= rbi
->ring_buffer
->read_index
;
147 write_loc
= rbi
->ring_buffer
->write_index
;
148 dsize
= rbi
->ring_datasize
;
150 *write
= write_loc
>= read_loc
? dsize
- (write_loc
- read_loc
) :
151 read_loc
- write_loc
;
152 *read
= dsize
- *write
;
155 static inline u32
hv_get_bytes_to_read(const struct hv_ring_buffer_info
*rbi
)
157 u32 read_loc
, write_loc
, dsize
, read
;
159 dsize
= rbi
->ring_datasize
;
160 read_loc
= rbi
->ring_buffer
->read_index
;
161 write_loc
= READ_ONCE(rbi
->ring_buffer
->write_index
);
163 read
= write_loc
>= read_loc
? (write_loc
- read_loc
) :
164 (dsize
- read_loc
) + write_loc
;
169 static inline u32
hv_get_bytes_to_write(const struct hv_ring_buffer_info
*rbi
)
171 u32 read_loc
, write_loc
, dsize
, write
;
173 dsize
= rbi
->ring_datasize
;
174 read_loc
= READ_ONCE(rbi
->ring_buffer
->read_index
);
175 write_loc
= rbi
->ring_buffer
->write_index
;
177 write
= write_loc
>= read_loc
? dsize
- (write_loc
- read_loc
) :
178 read_loc
- write_loc
;
183 * VMBUS version is 32 bit entity broken up into
184 * two 16 bit quantities: major_number. minor_number.
186 * 0 . 13 (Windows Server 2008)
189 * 3 . 0 (Windows 8 R2)
193 #define VERSION_WS2008 ((0 << 16) | (13))
194 #define VERSION_WIN7 ((1 << 16) | (1))
195 #define VERSION_WIN8 ((2 << 16) | (4))
196 #define VERSION_WIN8_1 ((3 << 16) | (0))
197 #define VERSION_WIN10 ((4 << 16) | (0))
199 #define VERSION_INVAL -1
201 #define VERSION_CURRENT VERSION_WIN10
203 /* Make maximum size of pipe payload of 16K */
204 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
206 /* Define PipeMode values. */
207 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
208 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
210 /* The size of the user defined data buffer for non-pipe offers. */
211 #define MAX_USER_DEFINED_BYTES 120
213 /* The size of the user defined data buffer for pipe offers. */
214 #define MAX_PIPE_USER_DEFINED_BYTES 116
217 * At the center of the Channel Management library is the Channel Offer. This
218 * struct contains the fundamental information about an offer.
220 struct vmbus_channel_offer
{
225 * These two fields are not currently used.
231 u16 mmio_megabytes
; /* in bytes * 1024 * 1024 */
234 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
236 unsigned char user_def
[MAX_USER_DEFINED_BYTES
];
241 * The following sructure is an integrated pipe protocol, which
242 * is implemented on top of standard user-defined data. Pipe
243 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
248 unsigned char user_def
[MAX_PIPE_USER_DEFINED_BYTES
];
252 * The sub_channel_index is defined in win8.
254 u16 sub_channel_index
;
259 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
260 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
261 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
262 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
263 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
264 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
265 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
266 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
268 struct vmpacket_descriptor
{
276 struct vmpacket_header
{
277 u32 prev_pkt_start_offset
;
278 struct vmpacket_descriptor descriptor
;
281 struct vmtransfer_page_range
{
286 struct vmtransfer_page_packet_header
{
287 struct vmpacket_descriptor d
;
292 struct vmtransfer_page_range ranges
[1];
295 struct vmgpadl_packet_header
{
296 struct vmpacket_descriptor d
;
301 struct vmadd_remove_transfer_page_set
{
302 struct vmpacket_descriptor d
;
309 * This structure defines a range in guest physical space that can be made to
310 * look virtually contiguous.
319 * This is the format for an Establish Gpadl packet, which contains a handle by
320 * which this GPADL will be known and a set of GPA ranges associated with it.
321 * This can be converted to a MDL by the guest OS. If there are multiple GPA
322 * ranges, then the resulting MDL will be "chained," representing multiple VA
325 struct vmestablish_gpadl
{
326 struct vmpacket_descriptor d
;
329 struct gpa_range range
[1];
333 * This is the format for a Teardown Gpadl packet, which indicates that the
334 * GPADL handle in the Establish Gpadl packet will never be referenced again.
336 struct vmteardown_gpadl
{
337 struct vmpacket_descriptor d
;
339 u32 reserved
; /* for alignment to a 8-byte boundary */
343 * This is the format for a GPA-Direct packet, which contains a set of GPA
344 * ranges, in addition to commands and/or data.
346 struct vmdata_gpa_direct
{
347 struct vmpacket_descriptor d
;
350 struct gpa_range range
[1];
353 /* This is the format for a Additional Data Packet. */
354 struct vmadditional_data
{
355 struct vmpacket_descriptor d
;
359 unsigned char data
[1];
362 union vmpacket_largest_possible_header
{
363 struct vmpacket_descriptor simple_hdr
;
364 struct vmtransfer_page_packet_header xfer_page_hdr
;
365 struct vmgpadl_packet_header gpadl_hdr
;
366 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr
;
367 struct vmestablish_gpadl establish_gpadl_hdr
;
368 struct vmteardown_gpadl teardown_gpadl_hdr
;
369 struct vmdata_gpa_direct data_gpa_direct_hdr
;
372 #define VMPACKET_DATA_START_ADDRESS(__packet) \
373 (void *)(((unsigned char *)__packet) + \
374 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
376 #define VMPACKET_DATA_LENGTH(__packet) \
377 ((((struct vmpacket_descriptor)__packet)->len8 - \
378 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
380 #define VMPACKET_TRANSFER_MODE(__packet) \
381 (((struct IMPACT)__packet)->type)
383 enum vmbus_packet_type
{
384 VM_PKT_INVALID
= 0x0,
386 VM_PKT_ADD_XFER_PAGESET
= 0x2,
387 VM_PKT_RM_XFER_PAGESET
= 0x3,
388 VM_PKT_ESTABLISH_GPADL
= 0x4,
389 VM_PKT_TEARDOWN_GPADL
= 0x5,
390 VM_PKT_DATA_INBAND
= 0x6,
391 VM_PKT_DATA_USING_XFER_PAGES
= 0x7,
392 VM_PKT_DATA_USING_GPADL
= 0x8,
393 VM_PKT_DATA_USING_GPA_DIRECT
= 0x9,
394 VM_PKT_CANCEL_REQUEST
= 0xa,
396 VM_PKT_DATA_USING_ADDITIONAL_PKT
= 0xc,
397 VM_PKT_ADDITIONAL_DATA
= 0xd
400 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
403 /* Version 1 messages */
404 enum vmbus_channel_message_type
{
405 CHANNELMSG_INVALID
= 0,
406 CHANNELMSG_OFFERCHANNEL
= 1,
407 CHANNELMSG_RESCIND_CHANNELOFFER
= 2,
408 CHANNELMSG_REQUESTOFFERS
= 3,
409 CHANNELMSG_ALLOFFERS_DELIVERED
= 4,
410 CHANNELMSG_OPENCHANNEL
= 5,
411 CHANNELMSG_OPENCHANNEL_RESULT
= 6,
412 CHANNELMSG_CLOSECHANNEL
= 7,
413 CHANNELMSG_GPADL_HEADER
= 8,
414 CHANNELMSG_GPADL_BODY
= 9,
415 CHANNELMSG_GPADL_CREATED
= 10,
416 CHANNELMSG_GPADL_TEARDOWN
= 11,
417 CHANNELMSG_GPADL_TORNDOWN
= 12,
418 CHANNELMSG_RELID_RELEASED
= 13,
419 CHANNELMSG_INITIATE_CONTACT
= 14,
420 CHANNELMSG_VERSION_RESPONSE
= 15,
421 CHANNELMSG_UNLOAD
= 16,
422 CHANNELMSG_UNLOAD_RESPONSE
= 17,
426 CHANNELMSG_TL_CONNECT_REQUEST
= 21,
430 struct vmbus_channel_message_header
{
431 enum vmbus_channel_message_type msgtype
;
435 /* Query VMBus Version parameters */
436 struct vmbus_channel_query_vmbus_version
{
437 struct vmbus_channel_message_header header
;
441 /* VMBus Version Supported parameters */
442 struct vmbus_channel_version_supported
{
443 struct vmbus_channel_message_header header
;
444 u8 version_supported
;
447 /* Offer Channel parameters */
448 struct vmbus_channel_offer_channel
{
449 struct vmbus_channel_message_header header
;
450 struct vmbus_channel_offer offer
;
454 * win7 and beyond splits this field into a bit field.
456 u8 monitor_allocated
:1;
459 * These are new fields added in win7 and later.
460 * Do not access these fields without checking the
461 * negotiated protocol.
463 * If "is_dedicated_interrupt" is set, we must not set the
464 * associated bit in the channel bitmap while sending the
465 * interrupt to the host.
467 * connection_id is to be used in signaling the host.
469 u16 is_dedicated_interrupt
:1;
474 /* Rescind Offer parameters */
475 struct vmbus_channel_rescind_offer
{
476 struct vmbus_channel_message_header header
;
481 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info
*rbi
)
483 return rbi
->ring_buffer
->pending_send_sz
;
487 * Request Offer -- no parameters, SynIC message contains the partition ID
488 * Set Snoop -- no parameters, SynIC message contains the partition ID
489 * Clear Snoop -- no parameters, SynIC message contains the partition ID
490 * All Offers Delivered -- no parameters, SynIC message contains the partition
492 * Flush Client -- no parameters, SynIC message contains the partition ID
495 /* Open Channel parameters */
496 struct vmbus_channel_open_channel
{
497 struct vmbus_channel_message_header header
;
499 /* Identifies the specific VMBus channel that is being opened. */
502 /* ID making a particular open request at a channel offer unique. */
505 /* GPADL for the channel's ring buffer. */
506 u32 ringbuffer_gpadlhandle
;
509 * Starting with win8, this field will be used to specify
510 * the target virtual processor on which to deliver the interrupt for
511 * the host to guest communication.
512 * Prior to win8, incoming channel interrupts would only
513 * be delivered on cpu 0. Setting this value to 0 would
514 * preserve the earlier behavior.
519 * The upstream ring buffer begins at offset zero in the memory
520 * described by RingBufferGpadlHandle. The downstream ring buffer
521 * follows it at this offset (in pages).
523 u32 downstream_ringbuffer_pageoffset
;
525 /* User-specific data to be passed along to the server endpoint. */
526 unsigned char userdata
[MAX_USER_DEFINED_BYTES
];
529 /* Open Channel Result parameters */
530 struct vmbus_channel_open_result
{
531 struct vmbus_channel_message_header header
;
537 /* Close channel parameters; */
538 struct vmbus_channel_close_channel
{
539 struct vmbus_channel_message_header header
;
543 /* Channel Message GPADL */
544 #define GPADL_TYPE_RING_BUFFER 1
545 #define GPADL_TYPE_SERVER_SAVE_AREA 2
546 #define GPADL_TYPE_TRANSACTION 8
549 * The number of PFNs in a GPADL message is defined by the number of
550 * pages that would be spanned by ByteCount and ByteOffset. If the
551 * implied number of PFNs won't fit in this packet, there will be a
552 * follow-up packet that contains more.
554 struct vmbus_channel_gpadl_header
{
555 struct vmbus_channel_message_header header
;
560 struct gpa_range range
[0];
563 /* This is the followup packet that contains more PFNs. */
564 struct vmbus_channel_gpadl_body
{
565 struct vmbus_channel_message_header header
;
571 struct vmbus_channel_gpadl_created
{
572 struct vmbus_channel_message_header header
;
578 struct vmbus_channel_gpadl_teardown
{
579 struct vmbus_channel_message_header header
;
584 struct vmbus_channel_gpadl_torndown
{
585 struct vmbus_channel_message_header header
;
589 struct vmbus_channel_relid_released
{
590 struct vmbus_channel_message_header header
;
594 struct vmbus_channel_initiate_contact
{
595 struct vmbus_channel_message_header header
;
596 u32 vmbus_version_requested
;
597 u32 target_vcpu
; /* The VCPU the host should respond to */
603 /* Hyper-V socket: guest's connect()-ing to host */
604 struct vmbus_channel_tl_connect_request
{
605 struct vmbus_channel_message_header header
;
606 uuid_le guest_endpoint_id
;
607 uuid_le host_service_id
;
610 struct vmbus_channel_version_response
{
611 struct vmbus_channel_message_header header
;
612 u8 version_supported
;
615 enum vmbus_channel_state
{
617 CHANNEL_OPENING_STATE
,
619 CHANNEL_OPENED_STATE
,
623 * Represents each channel msg on the vmbus connection This is a
624 * variable-size data structure depending on the msg type itself
626 struct vmbus_channel_msginfo
{
627 /* Bookkeeping stuff */
628 struct list_head msglistentry
;
630 /* So far, this is only used to handle gpadl body message */
631 struct list_head submsglist
;
633 /* Synchronize the request/response if needed */
634 struct completion waitevent
;
635 struct vmbus_channel
*waiting_channel
;
637 struct vmbus_channel_version_supported version_supported
;
638 struct vmbus_channel_open_result open_result
;
639 struct vmbus_channel_gpadl_torndown gpadl_torndown
;
640 struct vmbus_channel_gpadl_created gpadl_created
;
641 struct vmbus_channel_version_response version_response
;
646 * The channel message that goes out on the "wire".
647 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
649 unsigned char msg
[0];
652 struct vmbus_close_msg
{
653 struct vmbus_channel_msginfo info
;
654 struct vmbus_channel_close_channel msg
;
657 /* Define connection identifier type. */
658 union hv_connection_id
{
666 /* Definition of the hv_signal_event hypercall input structure. */
667 struct hv_input_signal_event
{
668 union hv_connection_id connectionid
;
673 struct hv_input_signal_event_buffer
{
675 struct hv_input_signal_event event
;
678 enum hv_numa_policy
{
683 enum vmbus_device_type
{
703 struct vmbus_device
{
709 struct vmbus_channel
{
710 struct list_head listentry
;
712 struct hv_device
*device_obj
;
714 enum vmbus_channel_state state
;
716 struct vmbus_channel_offer_channel offermsg
;
718 * These are based on the OfferMsg.MonitorId.
719 * Save it here for easy access.
724 bool rescind
; /* got rescind msg */
726 u32 ringbuffer_gpadlhandle
;
728 /* Allocated memory for ring buffer */
729 void *ringbuffer_pages
;
730 u32 ringbuffer_pagecount
;
731 struct hv_ring_buffer_info outbound
; /* send to parent */
732 struct hv_ring_buffer_info inbound
; /* receive from parent */
734 struct vmbus_close_msg close_msg
;
736 /* Channel callback's invoked in softirq context */
737 struct tasklet_struct callback_event
;
738 void (*onchannel_callback
)(void *context
);
739 void *channel_callback_context
;
742 * A channel can be marked for one of three modes of reading:
743 * BATCHED - callback called from taslket and should read
744 * channel until empty. Interrupts from the host
745 * are masked while read is in process (default).
746 * DIRECT - callback called from tasklet (softirq).
747 * ISR - callback called in interrupt context and must
748 * invoke its own deferred processing.
749 * Host interrupts are disabled and must be re-enabled
750 * when ring is empty.
752 enum hv_callback_mode
{
758 bool is_dedicated_interrupt
;
759 struct hv_input_signal_event_buffer sig_buf
;
760 struct hv_input_signal_event
*sig_event
;
763 * Starting with win8, this field will be used to specify
764 * the target virtual processor on which to deliver the interrupt for
765 * the host to guest communication.
766 * Prior to win8, incoming channel interrupts would only
767 * be delivered on cpu 0. Setting this value to 0 would
768 * preserve the earlier behavior.
771 /* The corresponding CPUID in the guest */
774 * State to manage the CPU affiliation of channels.
776 struct cpumask alloced_cpus_in_node
;
779 * Support for sub-channels. For high performance devices,
780 * it will be useful to have multiple sub-channels to support
781 * a scalable communication infrastructure with the host.
782 * The support for sub-channels is implemented as an extention
783 * to the current infrastructure.
784 * The initial offer is considered the primary channel and this
785 * offer message will indicate if the host supports sub-channels.
786 * The guest is free to ask for sub-channels to be offerred and can
787 * open these sub-channels as a normal "primary" channel. However,
788 * all sub-channels will have the same type and instance guids as the
789 * primary channel. Requests sent on a given channel will result in a
790 * response on the same channel.
794 * Sub-channel creation callback. This callback will be called in
795 * process context when a sub-channel offer is received from the host.
796 * The guest can open the sub-channel in the context of this callback.
798 void (*sc_creation_callback
)(struct vmbus_channel
*new_sc
);
801 * Channel rescind callback. Some channels (the hvsock ones), need to
802 * register a callback which is invoked in vmbus_onoffer_rescind().
804 void (*chn_rescind_callback
)(struct vmbus_channel
*channel
);
807 * The spinlock to protect the structure. It is being used to protect
808 * test-and-set access to various attributes of the structure as well
809 * as all sc_list operations.
813 * All Sub-channels of a primary channel are linked here.
815 struct list_head sc_list
;
817 * Current number of sub-channels.
821 * Number of a sub-channel (position within sc_list) which is supposed
822 * to be used as the next outgoing channel.
826 * The primary channel this sub-channel belongs to.
827 * This will be NULL for the primary channel.
829 struct vmbus_channel
*primary_channel
;
831 * Support per-channel state for use by vmbus drivers.
833 void *per_channel_state
;
835 * To support per-cpu lookup mapping of relid to channel,
836 * link up channels based on their CPU affinity.
838 struct list_head percpu_list
;
841 * Defer freeing channel until after all cpu's have
842 * gone through grace period.
847 * For performance critical channels (storage, networking
848 * etc,), Hyper-V has a mechanism to enhance the throughput
849 * at the expense of latency:
850 * When the host is to be signaled, we just set a bit in a shared page
851 * and this bit will be inspected by the hypervisor within a certain
852 * window and if the bit is set, the host will be signaled. The window
853 * of time is the monitor latency - currently around 100 usecs. This
854 * mechanism improves throughput by:
856 * A) Making the host more efficient - each time it wakes up,
857 * potentially it will process morev number of packets. The
858 * monitor latency allows a batch to build up.
859 * B) By deferring the hypercall to signal, we will also minimize
862 * Clearly, these optimizations improve throughput at the expense of
863 * latency. Furthermore, since the channel is shared for both
864 * control and data messages, control messages currently suffer
865 * unnecessary latency adversley impacting performance and boot
866 * time. To fix this issue, permit tagging the channel as being
867 * in "low latency" mode. In this mode, we will bypass the monitor
873 * NUMA distribution policy:
874 * We support teo policies:
875 * 1) Balanced: Here all performance critical channels are
876 * distributed evenly amongst all the NUMA nodes.
877 * This policy will be the default policy.
878 * 2) Localized: All channels of a given instance of a
879 * performance critical service will be assigned CPUs
880 * within a selected NUMA node.
882 enum hv_numa_policy affinity_policy
;
888 static inline bool is_hvsock_channel(const struct vmbus_channel
*c
)
890 return !!(c
->offermsg
.offer
.chn_flags
&
891 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
);
894 static inline void set_channel_affinity_state(struct vmbus_channel
*c
,
895 enum hv_numa_policy policy
)
897 c
->affinity_policy
= policy
;
900 static inline void set_channel_read_mode(struct vmbus_channel
*c
,
901 enum hv_callback_mode mode
)
903 c
->callback_mode
= mode
;
906 static inline void set_per_channel_state(struct vmbus_channel
*c
, void *s
)
908 c
->per_channel_state
= s
;
911 static inline void *get_per_channel_state(struct vmbus_channel
*c
)
913 return c
->per_channel_state
;
916 static inline void set_channel_pending_send_size(struct vmbus_channel
*c
,
919 c
->outbound
.ring_buffer
->pending_send_sz
= size
;
922 static inline void set_low_latency_mode(struct vmbus_channel
*c
)
924 c
->low_latency
= true;
927 static inline void clear_low_latency_mode(struct vmbus_channel
*c
)
929 c
->low_latency
= false;
932 void vmbus_onmessage(void *context
);
934 int vmbus_request_offers(void);
937 * APIs for managing sub-channels.
940 void vmbus_set_sc_create_callback(struct vmbus_channel
*primary_channel
,
941 void (*sc_cr_cb
)(struct vmbus_channel
*new_sc
));
943 void vmbus_set_chn_rescind_callback(struct vmbus_channel
*channel
,
944 void (*chn_rescind_cb
)(struct vmbus_channel
*));
947 * Retrieve the (sub) channel on which to send an outgoing request.
948 * When a primary channel has multiple sub-channels, we choose a
949 * channel whose VCPU binding is closest to the VCPU on which
950 * this call is being made.
952 struct vmbus_channel
*vmbus_get_outgoing_channel(struct vmbus_channel
*primary
);
955 * Check if sub-channels have already been offerred. This API will be useful
956 * when the driver is unloaded after establishing sub-channels. In this case,
957 * when the driver is re-loaded, the driver would have to check if the
958 * subchannels have already been established before attempting to request
959 * the creation of sub-channels.
960 * This function returns TRUE to indicate that subchannels have already been
962 * This function should be invoked after setting the callback function for
963 * sub-channel creation.
965 bool vmbus_are_subchannels_present(struct vmbus_channel
*primary
);
967 /* The format must be the same as struct vmdata_gpa_direct */
968 struct vmbus_channel_packet_page_buffer
{
976 struct hv_page_buffer range
[MAX_PAGE_BUFFER_COUNT
];
979 /* The format must be the same as struct vmdata_gpa_direct */
980 struct vmbus_channel_packet_multipage_buffer
{
987 u32 rangecount
; /* Always 1 in this case */
988 struct hv_multipage_buffer range
;
991 /* The format must be the same as struct vmdata_gpa_direct */
992 struct vmbus_packet_mpb_array
{
999 u32 rangecount
; /* Always 1 in this case */
1000 struct hv_mpb_array range
;
1004 extern int vmbus_open(struct vmbus_channel
*channel
,
1005 u32 send_ringbuffersize
,
1006 u32 recv_ringbuffersize
,
1009 void (*onchannel_callback
)(void *context
),
1012 extern void vmbus_close(struct vmbus_channel
*channel
);
1014 extern int vmbus_sendpacket(struct vmbus_channel
*channel
,
1018 enum vmbus_packet_type type
,
1021 extern int vmbus_sendpacket_ctl(struct vmbus_channel
*channel
,
1025 enum vmbus_packet_type type
,
1028 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel
*channel
,
1029 struct hv_page_buffer pagebuffers
[],
1035 extern int vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel
*channel
,
1036 struct hv_page_buffer pagebuffers
[],
1043 extern int vmbus_sendpacket_multipagebuffer(struct vmbus_channel
*channel
,
1044 struct hv_multipage_buffer
*mpb
,
1049 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel
*channel
,
1050 struct vmbus_packet_mpb_array
*mpb
,
1056 extern int vmbus_establish_gpadl(struct vmbus_channel
*channel
,
1061 extern int vmbus_teardown_gpadl(struct vmbus_channel
*channel
,
1064 extern int vmbus_recvpacket(struct vmbus_channel
*channel
,
1067 u32
*buffer_actual_len
,
1070 extern int vmbus_recvpacket_raw(struct vmbus_channel
*channel
,
1073 u32
*buffer_actual_len
,
1077 extern void vmbus_ontimer(unsigned long data
);
1079 /* Base driver object */
1084 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1085 * channel flag, actually doesn't mean a synthetic device because the
1086 * offer's if_type/if_instance can change for every new hvsock
1089 * However, to facilitate the notification of new-offer/rescind-offer
1090 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1091 * a special vmbus device, and hence we need the below flag to
1092 * indicate if the driver is the hvsock driver or not: we need to
1093 * specially treat the hvosck offer & driver in vmbus_match().
1097 /* the device type supported by this driver */
1099 const struct hv_vmbus_device_id
*id_table
;
1101 struct device_driver driver
;
1103 /* dynamic device GUID's */
1106 struct list_head list
;
1109 int (*probe
)(struct hv_device
*, const struct hv_vmbus_device_id
*);
1110 int (*remove
)(struct hv_device
*);
1111 void (*shutdown
)(struct hv_device
*);
1115 /* Base device object */
1117 /* the device type id of this device */
1120 /* the device instance id of this device */
1121 uuid_le dev_instance
;
1125 struct device device
;
1127 struct vmbus_channel
*channel
;
1131 static inline struct hv_device
*device_to_hv_device(struct device
*d
)
1133 return container_of(d
, struct hv_device
, device
);
1136 static inline struct hv_driver
*drv_to_hv_drv(struct device_driver
*d
)
1138 return container_of(d
, struct hv_driver
, driver
);
1141 static inline void hv_set_drvdata(struct hv_device
*dev
, void *data
)
1143 dev_set_drvdata(&dev
->device
, data
);
1146 static inline void *hv_get_drvdata(struct hv_device
*dev
)
1148 return dev_get_drvdata(&dev
->device
);
1151 struct hv_ring_buffer_debug_info
{
1152 u32 current_interrupt_mask
;
1153 u32 current_read_index
;
1154 u32 current_write_index
;
1155 u32 bytes_avail_toread
;
1156 u32 bytes_avail_towrite
;
1159 void hv_ringbuffer_get_debuginfo(const struct hv_ring_buffer_info
*ring_info
,
1160 struct hv_ring_buffer_debug_info
*debug_info
);
1162 /* Vmbus interface */
1163 #define vmbus_driver_register(driver) \
1164 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1165 int __must_check
__vmbus_driver_register(struct hv_driver
*hv_driver
,
1166 struct module
*owner
,
1167 const char *mod_name
);
1168 void vmbus_driver_unregister(struct hv_driver
*hv_driver
);
1170 void vmbus_hvsock_device_unregister(struct vmbus_channel
*channel
);
1172 int vmbus_allocate_mmio(struct resource
**new, struct hv_device
*device_obj
,
1173 resource_size_t min
, resource_size_t max
,
1174 resource_size_t size
, resource_size_t align
,
1175 bool fb_overlap_ok
);
1176 void vmbus_free_mmio(resource_size_t start
, resource_size_t size
);
1177 int vmbus_cpu_number_to_vp_number(int cpu_number
);
1178 u64
hv_do_hypercall(u64 control
, void *input
, void *output
);
1181 * GUID definitions of various offer types - services offered to the guest.
1186 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1188 #define HV_NIC_GUID \
1189 .guid = UUID_LE(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1190 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1194 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1196 #define HV_IDE_GUID \
1197 .guid = UUID_LE(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1198 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1202 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1204 #define HV_SCSI_GUID \
1205 .guid = UUID_LE(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1206 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1210 * {0e0b6031-5213-4934-818b-38d90ced39db}
1212 #define HV_SHUTDOWN_GUID \
1213 .guid = UUID_LE(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1214 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1218 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1220 #define HV_TS_GUID \
1221 .guid = UUID_LE(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1222 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1226 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1228 #define HV_HEART_BEAT_GUID \
1229 .guid = UUID_LE(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1230 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1234 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1236 #define HV_KVP_GUID \
1237 .guid = UUID_LE(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1238 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1241 * Dynamic memory GUID
1242 * {525074dc-8985-46e2-8057-a307dc18a502}
1244 #define HV_DM_GUID \
1245 .guid = UUID_LE(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1246 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1250 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1252 #define HV_MOUSE_GUID \
1253 .guid = UUID_LE(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1254 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1258 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1260 #define HV_KBD_GUID \
1261 .guid = UUID_LE(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1262 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1265 * VSS (Backup/Restore) GUID
1267 #define HV_VSS_GUID \
1268 .guid = UUID_LE(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1269 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1271 * Synthetic Video GUID
1272 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1274 #define HV_SYNTHVID_GUID \
1275 .guid = UUID_LE(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1276 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1280 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1282 #define HV_SYNTHFC_GUID \
1283 .guid = UUID_LE(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1284 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1287 * Guest File Copy Service
1288 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1291 #define HV_FCOPY_GUID \
1292 .guid = UUID_LE(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1293 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1296 * NetworkDirect. This is the guest RDMA service.
1297 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1299 #define HV_ND_GUID \
1300 .guid = UUID_LE(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1301 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1304 * PCI Express Pass Through
1305 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1308 #define HV_PCIE_GUID \
1309 .guid = UUID_LE(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1310 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1313 * Linux doesn't support the 3 devices: the first two are for
1314 * Automatic Virtual Machine Activation, and the third is for
1315 * Remote Desktop Virtualization.
1316 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1317 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1318 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1321 #define HV_AVMA1_GUID \
1322 .guid = UUID_LE(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1323 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1325 #define HV_AVMA2_GUID \
1326 .guid = UUID_LE(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1327 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1329 #define HV_RDV_GUID \
1330 .guid = UUID_LE(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1331 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1334 * Common header for Hyper-V ICs
1337 #define ICMSGTYPE_NEGOTIATE 0
1338 #define ICMSGTYPE_HEARTBEAT 1
1339 #define ICMSGTYPE_KVPEXCHANGE 2
1340 #define ICMSGTYPE_SHUTDOWN 3
1341 #define ICMSGTYPE_TIMESYNC 4
1342 #define ICMSGTYPE_VSS 5
1344 #define ICMSGHDRFLAG_TRANSACTION 1
1345 #define ICMSGHDRFLAG_REQUEST 2
1346 #define ICMSGHDRFLAG_RESPONSE 4
1350 * While we want to handle util services as regular devices,
1351 * there is only one instance of each of these services; so
1352 * we statically allocate the service specific state.
1355 struct hv_util_service
{
1358 void (*util_cb
)(void *);
1359 int (*util_init
)(struct hv_util_service
*);
1360 void (*util_deinit
)(void);
1363 struct vmbuspipe_hdr
{
1374 struct ic_version icverframe
;
1376 struct ic_version icvermsg
;
1379 u8 ictransaction_id
;
1384 struct icmsg_negotiate
{
1388 struct ic_version icversion_data
[1]; /* any size array */
1391 struct shutdown_msg_data
{
1393 u32 timeout_seconds
;
1395 u8 display_message
[2048];
1398 struct heartbeat_msg_data
{
1403 /* Time Sync IC defs */
1404 #define ICTIMESYNCFLAG_PROBE 0
1405 #define ICTIMESYNCFLAG_SYNC 1
1406 #define ICTIMESYNCFLAG_SAMPLE 2
1409 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1411 #define WLTIMEDELTA 116444736000000000LL
1414 struct ictimesync_data
{
1421 struct ictimesync_ref_data
{
1423 u64 vmreferencetime
;
1430 struct hyperv_service_callback
{
1434 struct vmbus_channel
*channel
;
1435 void (*callback
)(void *context
);
1438 #define MAX_SRV_VER 0x7ffffff
1439 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr
*icmsghdrp
, u8
*buf
,
1440 const int *fw_version
, int fw_vercnt
,
1441 const int *srv_version
, int srv_vercnt
,
1442 int *nego_fw_version
, int *nego_srv_version
);
1444 void hv_process_channel_removal(u32 relid
);
1446 void vmbus_setevent(struct vmbus_channel
*channel
);
1448 * Negotiated version with the Host.
1451 extern __u32 vmbus_proto_version
;
1453 int vmbus_send_tl_connect_request(const uuid_le
*shv_guest_servie_id
,
1454 const uuid_le
*shv_host_servie_id
);
1455 void vmbus_set_event(struct vmbus_channel
*channel
);
1457 /* Get the start of the ring buffer. */
1458 static inline void *
1459 hv_get_ring_buffer(const struct hv_ring_buffer_info
*ring_info
)
1461 return ring_info
->ring_buffer
->buffer
;
1465 * Mask off host interrupt callback notifications
1467 static inline void hv_begin_read(struct hv_ring_buffer_info
*rbi
)
1469 rbi
->ring_buffer
->interrupt_mask
= 1;
1471 /* make sure mask update is not reordered */
1476 * Re-enable host callback and return number of outstanding bytes
1478 static inline u32
hv_end_read(struct hv_ring_buffer_info
*rbi
)
1481 rbi
->ring_buffer
->interrupt_mask
= 0;
1483 /* make sure mask update is not reordered */
1487 * Now check to see if the ring buffer is still empty.
1488 * If it is not, we raced and we need to process new
1489 * incoming messages.
1491 return hv_get_bytes_to_read(rbi
);
1495 * An API to support in-place processing of incoming VMBUS packets.
1498 /* Get data payload associated with descriptor */
1499 static inline void *hv_pkt_data(const struct vmpacket_descriptor
*desc
)
1501 return (void *)((unsigned long)desc
+ (desc
->offset8
<< 3));
1504 /* Get data size associated with descriptor */
1505 static inline u32
hv_pkt_datalen(const struct vmpacket_descriptor
*desc
)
1507 return (desc
->len8
<< 3) - (desc
->offset8
<< 3);
1511 struct vmpacket_descriptor
*
1512 hv_pkt_iter_first(struct vmbus_channel
*channel
);
1514 struct vmpacket_descriptor
*
1515 __hv_pkt_iter_next(struct vmbus_channel
*channel
,
1516 const struct vmpacket_descriptor
*pkt
);
1518 void hv_pkt_iter_close(struct vmbus_channel
*channel
);
1521 * Get next packet descriptor from iterator
1522 * If at end of list, return NULL and update host.
1524 static inline struct vmpacket_descriptor
*
1525 hv_pkt_iter_next(struct vmbus_channel
*channel
,
1526 const struct vmpacket_descriptor
*pkt
)
1528 struct vmpacket_descriptor
*nxt
;
1530 nxt
= __hv_pkt_iter_next(channel
, pkt
);
1532 hv_pkt_iter_close(channel
);
1537 #define foreach_vmbus_pkt(pkt, channel) \
1538 for (pkt = hv_pkt_iter_first(channel); pkt; \
1539 pkt = hv_pkt_iter_next(channel, pkt))
1541 #endif /* _HYPERV_H */