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
;
130 u32 cached_read_index
;
135 * hv_get_ringbuffer_availbytes()
137 * Get number of bytes available to read and to write to
138 * for the specified ring buffer
141 hv_get_ringbuffer_availbytes(const struct hv_ring_buffer_info
*rbi
,
142 u32
*read
, u32
*write
)
144 u32 read_loc
, write_loc
, dsize
;
146 /* Capture the read/write indices before they changed */
147 read_loc
= rbi
->ring_buffer
->read_index
;
148 write_loc
= rbi
->ring_buffer
->write_index
;
149 dsize
= rbi
->ring_datasize
;
151 *write
= write_loc
>= read_loc
? dsize
- (write_loc
- read_loc
) :
152 read_loc
- write_loc
;
153 *read
= dsize
- *write
;
156 static inline u32
hv_get_bytes_to_read(const struct hv_ring_buffer_info
*rbi
)
158 u32 read_loc
, write_loc
, dsize
, read
;
160 dsize
= rbi
->ring_datasize
;
161 read_loc
= rbi
->ring_buffer
->read_index
;
162 write_loc
= READ_ONCE(rbi
->ring_buffer
->write_index
);
164 read
= write_loc
>= read_loc
? (write_loc
- read_loc
) :
165 (dsize
- read_loc
) + write_loc
;
170 static inline u32
hv_get_bytes_to_write(const struct hv_ring_buffer_info
*rbi
)
172 u32 read_loc
, write_loc
, dsize
, write
;
174 dsize
= rbi
->ring_datasize
;
175 read_loc
= READ_ONCE(rbi
->ring_buffer
->read_index
);
176 write_loc
= rbi
->ring_buffer
->write_index
;
178 write
= write_loc
>= read_loc
? dsize
- (write_loc
- read_loc
) :
179 read_loc
- write_loc
;
183 static inline u32
hv_get_cached_bytes_to_write(
184 const struct hv_ring_buffer_info
*rbi
)
186 u32 read_loc
, write_loc
, dsize
, write
;
188 dsize
= rbi
->ring_datasize
;
189 read_loc
= rbi
->cached_read_index
;
190 write_loc
= rbi
->ring_buffer
->write_index
;
192 write
= write_loc
>= read_loc
? dsize
- (write_loc
- read_loc
) :
193 read_loc
- write_loc
;
197 * VMBUS version is 32 bit entity broken up into
198 * two 16 bit quantities: major_number. minor_number.
200 * 0 . 13 (Windows Server 2008)
203 * 3 . 0 (Windows 8 R2)
207 #define VERSION_WS2008 ((0 << 16) | (13))
208 #define VERSION_WIN7 ((1 << 16) | (1))
209 #define VERSION_WIN8 ((2 << 16) | (4))
210 #define VERSION_WIN8_1 ((3 << 16) | (0))
211 #define VERSION_WIN10 ((4 << 16) | (0))
213 #define VERSION_INVAL -1
215 #define VERSION_CURRENT VERSION_WIN10
217 /* Make maximum size of pipe payload of 16K */
218 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
220 /* Define PipeMode values. */
221 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
222 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
224 /* The size of the user defined data buffer for non-pipe offers. */
225 #define MAX_USER_DEFINED_BYTES 120
227 /* The size of the user defined data buffer for pipe offers. */
228 #define MAX_PIPE_USER_DEFINED_BYTES 116
231 * At the center of the Channel Management library is the Channel Offer. This
232 * struct contains the fundamental information about an offer.
234 struct vmbus_channel_offer
{
239 * These two fields are not currently used.
245 u16 mmio_megabytes
; /* in bytes * 1024 * 1024 */
248 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
250 unsigned char user_def
[MAX_USER_DEFINED_BYTES
];
255 * The following sructure is an integrated pipe protocol, which
256 * is implemented on top of standard user-defined data. Pipe
257 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
262 unsigned char user_def
[MAX_PIPE_USER_DEFINED_BYTES
];
266 * The sub_channel_index is defined in win8.
268 u16 sub_channel_index
;
273 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
274 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
275 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
276 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
277 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
278 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
279 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
280 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
282 struct vmpacket_descriptor
{
290 struct vmpacket_header
{
291 u32 prev_pkt_start_offset
;
292 struct vmpacket_descriptor descriptor
;
295 struct vmtransfer_page_range
{
300 struct vmtransfer_page_packet_header
{
301 struct vmpacket_descriptor d
;
306 struct vmtransfer_page_range ranges
[1];
309 struct vmgpadl_packet_header
{
310 struct vmpacket_descriptor d
;
315 struct vmadd_remove_transfer_page_set
{
316 struct vmpacket_descriptor d
;
323 * This structure defines a range in guest physical space that can be made to
324 * look virtually contiguous.
333 * This is the format for an Establish Gpadl packet, which contains a handle by
334 * which this GPADL will be known and a set of GPA ranges associated with it.
335 * This can be converted to a MDL by the guest OS. If there are multiple GPA
336 * ranges, then the resulting MDL will be "chained," representing multiple VA
339 struct vmestablish_gpadl
{
340 struct vmpacket_descriptor d
;
343 struct gpa_range range
[1];
347 * This is the format for a Teardown Gpadl packet, which indicates that the
348 * GPADL handle in the Establish Gpadl packet will never be referenced again.
350 struct vmteardown_gpadl
{
351 struct vmpacket_descriptor d
;
353 u32 reserved
; /* for alignment to a 8-byte boundary */
357 * This is the format for a GPA-Direct packet, which contains a set of GPA
358 * ranges, in addition to commands and/or data.
360 struct vmdata_gpa_direct
{
361 struct vmpacket_descriptor d
;
364 struct gpa_range range
[1];
367 /* This is the format for a Additional Data Packet. */
368 struct vmadditional_data
{
369 struct vmpacket_descriptor d
;
373 unsigned char data
[1];
376 union vmpacket_largest_possible_header
{
377 struct vmpacket_descriptor simple_hdr
;
378 struct vmtransfer_page_packet_header xfer_page_hdr
;
379 struct vmgpadl_packet_header gpadl_hdr
;
380 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr
;
381 struct vmestablish_gpadl establish_gpadl_hdr
;
382 struct vmteardown_gpadl teardown_gpadl_hdr
;
383 struct vmdata_gpa_direct data_gpa_direct_hdr
;
386 #define VMPACKET_DATA_START_ADDRESS(__packet) \
387 (void *)(((unsigned char *)__packet) + \
388 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
390 #define VMPACKET_DATA_LENGTH(__packet) \
391 ((((struct vmpacket_descriptor)__packet)->len8 - \
392 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
394 #define VMPACKET_TRANSFER_MODE(__packet) \
395 (((struct IMPACT)__packet)->type)
397 enum vmbus_packet_type
{
398 VM_PKT_INVALID
= 0x0,
400 VM_PKT_ADD_XFER_PAGESET
= 0x2,
401 VM_PKT_RM_XFER_PAGESET
= 0x3,
402 VM_PKT_ESTABLISH_GPADL
= 0x4,
403 VM_PKT_TEARDOWN_GPADL
= 0x5,
404 VM_PKT_DATA_INBAND
= 0x6,
405 VM_PKT_DATA_USING_XFER_PAGES
= 0x7,
406 VM_PKT_DATA_USING_GPADL
= 0x8,
407 VM_PKT_DATA_USING_GPA_DIRECT
= 0x9,
408 VM_PKT_CANCEL_REQUEST
= 0xa,
410 VM_PKT_DATA_USING_ADDITIONAL_PKT
= 0xc,
411 VM_PKT_ADDITIONAL_DATA
= 0xd
414 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
417 /* Version 1 messages */
418 enum vmbus_channel_message_type
{
419 CHANNELMSG_INVALID
= 0,
420 CHANNELMSG_OFFERCHANNEL
= 1,
421 CHANNELMSG_RESCIND_CHANNELOFFER
= 2,
422 CHANNELMSG_REQUESTOFFERS
= 3,
423 CHANNELMSG_ALLOFFERS_DELIVERED
= 4,
424 CHANNELMSG_OPENCHANNEL
= 5,
425 CHANNELMSG_OPENCHANNEL_RESULT
= 6,
426 CHANNELMSG_CLOSECHANNEL
= 7,
427 CHANNELMSG_GPADL_HEADER
= 8,
428 CHANNELMSG_GPADL_BODY
= 9,
429 CHANNELMSG_GPADL_CREATED
= 10,
430 CHANNELMSG_GPADL_TEARDOWN
= 11,
431 CHANNELMSG_GPADL_TORNDOWN
= 12,
432 CHANNELMSG_RELID_RELEASED
= 13,
433 CHANNELMSG_INITIATE_CONTACT
= 14,
434 CHANNELMSG_VERSION_RESPONSE
= 15,
435 CHANNELMSG_UNLOAD
= 16,
436 CHANNELMSG_UNLOAD_RESPONSE
= 17,
440 CHANNELMSG_TL_CONNECT_REQUEST
= 21,
444 struct vmbus_channel_message_header
{
445 enum vmbus_channel_message_type msgtype
;
449 /* Query VMBus Version parameters */
450 struct vmbus_channel_query_vmbus_version
{
451 struct vmbus_channel_message_header header
;
455 /* VMBus Version Supported parameters */
456 struct vmbus_channel_version_supported
{
457 struct vmbus_channel_message_header header
;
458 u8 version_supported
;
461 /* Offer Channel parameters */
462 struct vmbus_channel_offer_channel
{
463 struct vmbus_channel_message_header header
;
464 struct vmbus_channel_offer offer
;
468 * win7 and beyond splits this field into a bit field.
470 u8 monitor_allocated
:1;
473 * These are new fields added in win7 and later.
474 * Do not access these fields without checking the
475 * negotiated protocol.
477 * If "is_dedicated_interrupt" is set, we must not set the
478 * associated bit in the channel bitmap while sending the
479 * interrupt to the host.
481 * connection_id is to be used in signaling the host.
483 u16 is_dedicated_interrupt
:1;
488 /* Rescind Offer parameters */
489 struct vmbus_channel_rescind_offer
{
490 struct vmbus_channel_message_header header
;
495 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info
*rbi
)
497 return rbi
->ring_buffer
->pending_send_sz
;
501 * Request Offer -- no parameters, SynIC message contains the partition ID
502 * Set Snoop -- no parameters, SynIC message contains the partition ID
503 * Clear Snoop -- no parameters, SynIC message contains the partition ID
504 * All Offers Delivered -- no parameters, SynIC message contains the partition
506 * Flush Client -- no parameters, SynIC message contains the partition ID
509 /* Open Channel parameters */
510 struct vmbus_channel_open_channel
{
511 struct vmbus_channel_message_header header
;
513 /* Identifies the specific VMBus channel that is being opened. */
516 /* ID making a particular open request at a channel offer unique. */
519 /* GPADL for the channel's ring buffer. */
520 u32 ringbuffer_gpadlhandle
;
523 * Starting with win8, this field will be used to specify
524 * the target virtual processor on which to deliver the interrupt for
525 * the host to guest communication.
526 * Prior to win8, incoming channel interrupts would only
527 * be delivered on cpu 0. Setting this value to 0 would
528 * preserve the earlier behavior.
533 * The upstream ring buffer begins at offset zero in the memory
534 * described by RingBufferGpadlHandle. The downstream ring buffer
535 * follows it at this offset (in pages).
537 u32 downstream_ringbuffer_pageoffset
;
539 /* User-specific data to be passed along to the server endpoint. */
540 unsigned char userdata
[MAX_USER_DEFINED_BYTES
];
543 /* Open Channel Result parameters */
544 struct vmbus_channel_open_result
{
545 struct vmbus_channel_message_header header
;
551 /* Close channel parameters; */
552 struct vmbus_channel_close_channel
{
553 struct vmbus_channel_message_header header
;
557 /* Channel Message GPADL */
558 #define GPADL_TYPE_RING_BUFFER 1
559 #define GPADL_TYPE_SERVER_SAVE_AREA 2
560 #define GPADL_TYPE_TRANSACTION 8
563 * The number of PFNs in a GPADL message is defined by the number of
564 * pages that would be spanned by ByteCount and ByteOffset. If the
565 * implied number of PFNs won't fit in this packet, there will be a
566 * follow-up packet that contains more.
568 struct vmbus_channel_gpadl_header
{
569 struct vmbus_channel_message_header header
;
574 struct gpa_range range
[0];
577 /* This is the followup packet that contains more PFNs. */
578 struct vmbus_channel_gpadl_body
{
579 struct vmbus_channel_message_header header
;
585 struct vmbus_channel_gpadl_created
{
586 struct vmbus_channel_message_header header
;
592 struct vmbus_channel_gpadl_teardown
{
593 struct vmbus_channel_message_header header
;
598 struct vmbus_channel_gpadl_torndown
{
599 struct vmbus_channel_message_header header
;
603 struct vmbus_channel_relid_released
{
604 struct vmbus_channel_message_header header
;
608 struct vmbus_channel_initiate_contact
{
609 struct vmbus_channel_message_header header
;
610 u32 vmbus_version_requested
;
611 u32 target_vcpu
; /* The VCPU the host should respond to */
617 /* Hyper-V socket: guest's connect()-ing to host */
618 struct vmbus_channel_tl_connect_request
{
619 struct vmbus_channel_message_header header
;
620 uuid_le guest_endpoint_id
;
621 uuid_le host_service_id
;
624 struct vmbus_channel_version_response
{
625 struct vmbus_channel_message_header header
;
626 u8 version_supported
;
629 enum vmbus_channel_state
{
631 CHANNEL_OPENING_STATE
,
633 CHANNEL_OPENED_STATE
,
637 * Represents each channel msg on the vmbus connection This is a
638 * variable-size data structure depending on the msg type itself
640 struct vmbus_channel_msginfo
{
641 /* Bookkeeping stuff */
642 struct list_head msglistentry
;
644 /* So far, this is only used to handle gpadl body message */
645 struct list_head submsglist
;
647 /* Synchronize the request/response if needed */
648 struct completion waitevent
;
649 struct vmbus_channel
*waiting_channel
;
651 struct vmbus_channel_version_supported version_supported
;
652 struct vmbus_channel_open_result open_result
;
653 struct vmbus_channel_gpadl_torndown gpadl_torndown
;
654 struct vmbus_channel_gpadl_created gpadl_created
;
655 struct vmbus_channel_version_response version_response
;
660 * The channel message that goes out on the "wire".
661 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
663 unsigned char msg
[0];
666 struct vmbus_close_msg
{
667 struct vmbus_channel_msginfo info
;
668 struct vmbus_channel_close_channel msg
;
671 /* Define connection identifier type. */
672 union hv_connection_id
{
680 /* Definition of the hv_signal_event hypercall input structure. */
681 struct hv_input_signal_event
{
682 union hv_connection_id connectionid
;
687 struct hv_input_signal_event_buffer
{
689 struct hv_input_signal_event event
;
692 enum hv_numa_policy
{
697 enum vmbus_device_type
{
717 struct vmbus_device
{
723 struct vmbus_channel
{
724 struct list_head listentry
;
726 struct hv_device
*device_obj
;
728 enum vmbus_channel_state state
;
730 struct vmbus_channel_offer_channel offermsg
;
732 * These are based on the OfferMsg.MonitorId.
733 * Save it here for easy access.
738 bool rescind
; /* got rescind msg */
740 u32 ringbuffer_gpadlhandle
;
742 /* Allocated memory for ring buffer */
743 void *ringbuffer_pages
;
744 u32 ringbuffer_pagecount
;
745 struct hv_ring_buffer_info outbound
; /* send to parent */
746 struct hv_ring_buffer_info inbound
; /* receive from parent */
748 struct vmbus_close_msg close_msg
;
750 /* Channel callback's invoked in softirq context */
751 struct tasklet_struct callback_event
;
752 void (*onchannel_callback
)(void *context
);
753 void *channel_callback_context
;
756 * A channel can be marked for one of three modes of reading:
757 * BATCHED - callback called from taslket and should read
758 * channel until empty. Interrupts from the host
759 * are masked while read is in process (default).
760 * DIRECT - callback called from tasklet (softirq).
761 * ISR - callback called in interrupt context and must
762 * invoke its own deferred processing.
763 * Host interrupts are disabled and must be re-enabled
764 * when ring is empty.
766 enum hv_callback_mode
{
772 bool is_dedicated_interrupt
;
773 struct hv_input_signal_event_buffer sig_buf
;
774 struct hv_input_signal_event
*sig_event
;
777 * Starting with win8, this field will be used to specify
778 * the target virtual processor on which to deliver the interrupt for
779 * the host to guest communication.
780 * Prior to win8, incoming channel interrupts would only
781 * be delivered on cpu 0. Setting this value to 0 would
782 * preserve the earlier behavior.
785 /* The corresponding CPUID in the guest */
788 * State to manage the CPU affiliation of channels.
790 struct cpumask alloced_cpus_in_node
;
793 * Support for sub-channels. For high performance devices,
794 * it will be useful to have multiple sub-channels to support
795 * a scalable communication infrastructure with the host.
796 * The support for sub-channels is implemented as an extention
797 * to the current infrastructure.
798 * The initial offer is considered the primary channel and this
799 * offer message will indicate if the host supports sub-channels.
800 * The guest is free to ask for sub-channels to be offerred and can
801 * open these sub-channels as a normal "primary" channel. However,
802 * all sub-channels will have the same type and instance guids as the
803 * primary channel. Requests sent on a given channel will result in a
804 * response on the same channel.
808 * Sub-channel creation callback. This callback will be called in
809 * process context when a sub-channel offer is received from the host.
810 * The guest can open the sub-channel in the context of this callback.
812 void (*sc_creation_callback
)(struct vmbus_channel
*new_sc
);
815 * Channel rescind callback. Some channels (the hvsock ones), need to
816 * register a callback which is invoked in vmbus_onoffer_rescind().
818 void (*chn_rescind_callback
)(struct vmbus_channel
*channel
);
821 * The spinlock to protect the structure. It is being used to protect
822 * test-and-set access to various attributes of the structure as well
823 * as all sc_list operations.
827 * All Sub-channels of a primary channel are linked here.
829 struct list_head sc_list
;
831 * Current number of sub-channels.
835 * Number of a sub-channel (position within sc_list) which is supposed
836 * to be used as the next outgoing channel.
840 * The primary channel this sub-channel belongs to.
841 * This will be NULL for the primary channel.
843 struct vmbus_channel
*primary_channel
;
845 * Support per-channel state for use by vmbus drivers.
847 void *per_channel_state
;
849 * To support per-cpu lookup mapping of relid to channel,
850 * link up channels based on their CPU affinity.
852 struct list_head percpu_list
;
855 * Defer freeing channel until after all cpu's have
856 * gone through grace period.
861 * For performance critical channels (storage, networking
862 * etc,), Hyper-V has a mechanism to enhance the throughput
863 * at the expense of latency:
864 * When the host is to be signaled, we just set a bit in a shared page
865 * and this bit will be inspected by the hypervisor within a certain
866 * window and if the bit is set, the host will be signaled. The window
867 * of time is the monitor latency - currently around 100 usecs. This
868 * mechanism improves throughput by:
870 * A) Making the host more efficient - each time it wakes up,
871 * potentially it will process morev number of packets. The
872 * monitor latency allows a batch to build up.
873 * B) By deferring the hypercall to signal, we will also minimize
876 * Clearly, these optimizations improve throughput at the expense of
877 * latency. Furthermore, since the channel is shared for both
878 * control and data messages, control messages currently suffer
879 * unnecessary latency adversley impacting performance and boot
880 * time. To fix this issue, permit tagging the channel as being
881 * in "low latency" mode. In this mode, we will bypass the monitor
887 * NUMA distribution policy:
888 * We support teo policies:
889 * 1) Balanced: Here all performance critical channels are
890 * distributed evenly amongst all the NUMA nodes.
891 * This policy will be the default policy.
892 * 2) Localized: All channels of a given instance of a
893 * performance critical service will be assigned CPUs
894 * within a selected NUMA node.
896 enum hv_numa_policy affinity_policy
;
902 static inline bool is_hvsock_channel(const struct vmbus_channel
*c
)
904 return !!(c
->offermsg
.offer
.chn_flags
&
905 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
);
908 static inline void set_channel_affinity_state(struct vmbus_channel
*c
,
909 enum hv_numa_policy policy
)
911 c
->affinity_policy
= policy
;
914 static inline void set_channel_read_mode(struct vmbus_channel
*c
,
915 enum hv_callback_mode mode
)
917 c
->callback_mode
= mode
;
920 static inline void set_per_channel_state(struct vmbus_channel
*c
, void *s
)
922 c
->per_channel_state
= s
;
925 static inline void *get_per_channel_state(struct vmbus_channel
*c
)
927 return c
->per_channel_state
;
930 static inline void set_channel_pending_send_size(struct vmbus_channel
*c
,
933 c
->outbound
.ring_buffer
->pending_send_sz
= size
;
936 static inline void set_low_latency_mode(struct vmbus_channel
*c
)
938 c
->low_latency
= true;
941 static inline void clear_low_latency_mode(struct vmbus_channel
*c
)
943 c
->low_latency
= false;
946 void vmbus_onmessage(void *context
);
948 int vmbus_request_offers(void);
951 * APIs for managing sub-channels.
954 void vmbus_set_sc_create_callback(struct vmbus_channel
*primary_channel
,
955 void (*sc_cr_cb
)(struct vmbus_channel
*new_sc
));
957 void vmbus_set_chn_rescind_callback(struct vmbus_channel
*channel
,
958 void (*chn_rescind_cb
)(struct vmbus_channel
*));
961 * Retrieve the (sub) channel on which to send an outgoing request.
962 * When a primary channel has multiple sub-channels, we choose a
963 * channel whose VCPU binding is closest to the VCPU on which
964 * this call is being made.
966 struct vmbus_channel
*vmbus_get_outgoing_channel(struct vmbus_channel
*primary
);
969 * Check if sub-channels have already been offerred. This API will be useful
970 * when the driver is unloaded after establishing sub-channels. In this case,
971 * when the driver is re-loaded, the driver would have to check if the
972 * subchannels have already been established before attempting to request
973 * the creation of sub-channels.
974 * This function returns TRUE to indicate that subchannels have already been
976 * This function should be invoked after setting the callback function for
977 * sub-channel creation.
979 bool vmbus_are_subchannels_present(struct vmbus_channel
*primary
);
981 /* The format must be the same as struct vmdata_gpa_direct */
982 struct vmbus_channel_packet_page_buffer
{
990 struct hv_page_buffer range
[MAX_PAGE_BUFFER_COUNT
];
993 /* The format must be the same as struct vmdata_gpa_direct */
994 struct vmbus_channel_packet_multipage_buffer
{
1001 u32 rangecount
; /* Always 1 in this case */
1002 struct hv_multipage_buffer range
;
1005 /* The format must be the same as struct vmdata_gpa_direct */
1006 struct vmbus_packet_mpb_array
{
1013 u32 rangecount
; /* Always 1 in this case */
1014 struct hv_mpb_array range
;
1018 extern int vmbus_open(struct vmbus_channel
*channel
,
1019 u32 send_ringbuffersize
,
1020 u32 recv_ringbuffersize
,
1023 void (*onchannel_callback
)(void *context
),
1026 extern void vmbus_close(struct vmbus_channel
*channel
);
1028 extern int vmbus_sendpacket(struct vmbus_channel
*channel
,
1032 enum vmbus_packet_type type
,
1035 extern int vmbus_sendpacket_ctl(struct vmbus_channel
*channel
,
1039 enum vmbus_packet_type type
,
1042 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel
*channel
,
1043 struct hv_page_buffer pagebuffers
[],
1049 extern int vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel
*channel
,
1050 struct hv_page_buffer pagebuffers
[],
1057 extern int vmbus_sendpacket_multipagebuffer(struct vmbus_channel
*channel
,
1058 struct hv_multipage_buffer
*mpb
,
1063 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel
*channel
,
1064 struct vmbus_packet_mpb_array
*mpb
,
1070 extern int vmbus_establish_gpadl(struct vmbus_channel
*channel
,
1075 extern int vmbus_teardown_gpadl(struct vmbus_channel
*channel
,
1078 extern int vmbus_recvpacket(struct vmbus_channel
*channel
,
1081 u32
*buffer_actual_len
,
1084 extern int vmbus_recvpacket_raw(struct vmbus_channel
*channel
,
1087 u32
*buffer_actual_len
,
1091 extern void vmbus_ontimer(unsigned long data
);
1093 /* Base driver object */
1098 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1099 * channel flag, actually doesn't mean a synthetic device because the
1100 * offer's if_type/if_instance can change for every new hvsock
1103 * However, to facilitate the notification of new-offer/rescind-offer
1104 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1105 * a special vmbus device, and hence we need the below flag to
1106 * indicate if the driver is the hvsock driver or not: we need to
1107 * specially treat the hvosck offer & driver in vmbus_match().
1111 /* the device type supported by this driver */
1113 const struct hv_vmbus_device_id
*id_table
;
1115 struct device_driver driver
;
1117 /* dynamic device GUID's */
1120 struct list_head list
;
1123 int (*probe
)(struct hv_device
*, const struct hv_vmbus_device_id
*);
1124 int (*remove
)(struct hv_device
*);
1125 void (*shutdown
)(struct hv_device
*);
1129 /* Base device object */
1131 /* the device type id of this device */
1134 /* the device instance id of this device */
1135 uuid_le dev_instance
;
1139 struct device device
;
1141 struct vmbus_channel
*channel
;
1145 static inline struct hv_device
*device_to_hv_device(struct device
*d
)
1147 return container_of(d
, struct hv_device
, device
);
1150 static inline struct hv_driver
*drv_to_hv_drv(struct device_driver
*d
)
1152 return container_of(d
, struct hv_driver
, driver
);
1155 static inline void hv_set_drvdata(struct hv_device
*dev
, void *data
)
1157 dev_set_drvdata(&dev
->device
, data
);
1160 static inline void *hv_get_drvdata(struct hv_device
*dev
)
1162 return dev_get_drvdata(&dev
->device
);
1165 struct hv_ring_buffer_debug_info
{
1166 u32 current_interrupt_mask
;
1167 u32 current_read_index
;
1168 u32 current_write_index
;
1169 u32 bytes_avail_toread
;
1170 u32 bytes_avail_towrite
;
1173 void hv_ringbuffer_get_debuginfo(const struct hv_ring_buffer_info
*ring_info
,
1174 struct hv_ring_buffer_debug_info
*debug_info
);
1176 /* Vmbus interface */
1177 #define vmbus_driver_register(driver) \
1178 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1179 int __must_check
__vmbus_driver_register(struct hv_driver
*hv_driver
,
1180 struct module
*owner
,
1181 const char *mod_name
);
1182 void vmbus_driver_unregister(struct hv_driver
*hv_driver
);
1184 void vmbus_hvsock_device_unregister(struct vmbus_channel
*channel
);
1186 int vmbus_allocate_mmio(struct resource
**new, struct hv_device
*device_obj
,
1187 resource_size_t min
, resource_size_t max
,
1188 resource_size_t size
, resource_size_t align
,
1189 bool fb_overlap_ok
);
1190 void vmbus_free_mmio(resource_size_t start
, resource_size_t size
);
1191 int vmbus_cpu_number_to_vp_number(int cpu_number
);
1192 u64
hv_do_hypercall(u64 control
, void *input
, void *output
);
1195 * GUID definitions of various offer types - services offered to the guest.
1200 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1202 #define HV_NIC_GUID \
1203 .guid = UUID_LE(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1204 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1208 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1210 #define HV_IDE_GUID \
1211 .guid = UUID_LE(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1212 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1216 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1218 #define HV_SCSI_GUID \
1219 .guid = UUID_LE(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1220 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1224 * {0e0b6031-5213-4934-818b-38d90ced39db}
1226 #define HV_SHUTDOWN_GUID \
1227 .guid = UUID_LE(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1228 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1232 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1234 #define HV_TS_GUID \
1235 .guid = UUID_LE(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1236 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1240 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1242 #define HV_HEART_BEAT_GUID \
1243 .guid = UUID_LE(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1244 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1248 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1250 #define HV_KVP_GUID \
1251 .guid = UUID_LE(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1252 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1255 * Dynamic memory GUID
1256 * {525074dc-8985-46e2-8057-a307dc18a502}
1258 #define HV_DM_GUID \
1259 .guid = UUID_LE(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1260 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1264 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1266 #define HV_MOUSE_GUID \
1267 .guid = UUID_LE(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1268 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1272 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1274 #define HV_KBD_GUID \
1275 .guid = UUID_LE(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1276 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1279 * VSS (Backup/Restore) GUID
1281 #define HV_VSS_GUID \
1282 .guid = UUID_LE(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1283 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1285 * Synthetic Video GUID
1286 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1288 #define HV_SYNTHVID_GUID \
1289 .guid = UUID_LE(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1290 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1294 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1296 #define HV_SYNTHFC_GUID \
1297 .guid = UUID_LE(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1298 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1301 * Guest File Copy Service
1302 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1305 #define HV_FCOPY_GUID \
1306 .guid = UUID_LE(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1307 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1310 * NetworkDirect. This is the guest RDMA service.
1311 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1313 #define HV_ND_GUID \
1314 .guid = UUID_LE(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1315 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1318 * PCI Express Pass Through
1319 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1322 #define HV_PCIE_GUID \
1323 .guid = UUID_LE(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1324 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1327 * Linux doesn't support the 3 devices: the first two are for
1328 * Automatic Virtual Machine Activation, and the third is for
1329 * Remote Desktop Virtualization.
1330 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1331 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1332 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1335 #define HV_AVMA1_GUID \
1336 .guid = UUID_LE(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1337 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1339 #define HV_AVMA2_GUID \
1340 .guid = UUID_LE(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1341 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1343 #define HV_RDV_GUID \
1344 .guid = UUID_LE(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1345 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1348 * Common header for Hyper-V ICs
1351 #define ICMSGTYPE_NEGOTIATE 0
1352 #define ICMSGTYPE_HEARTBEAT 1
1353 #define ICMSGTYPE_KVPEXCHANGE 2
1354 #define ICMSGTYPE_SHUTDOWN 3
1355 #define ICMSGTYPE_TIMESYNC 4
1356 #define ICMSGTYPE_VSS 5
1358 #define ICMSGHDRFLAG_TRANSACTION 1
1359 #define ICMSGHDRFLAG_REQUEST 2
1360 #define ICMSGHDRFLAG_RESPONSE 4
1364 * While we want to handle util services as regular devices,
1365 * there is only one instance of each of these services; so
1366 * we statically allocate the service specific state.
1369 struct hv_util_service
{
1372 void (*util_cb
)(void *);
1373 int (*util_init
)(struct hv_util_service
*);
1374 void (*util_deinit
)(void);
1377 struct vmbuspipe_hdr
{
1388 struct ic_version icverframe
;
1390 struct ic_version icvermsg
;
1393 u8 ictransaction_id
;
1398 struct icmsg_negotiate
{
1402 struct ic_version icversion_data
[1]; /* any size array */
1405 struct shutdown_msg_data
{
1407 u32 timeout_seconds
;
1409 u8 display_message
[2048];
1412 struct heartbeat_msg_data
{
1417 /* Time Sync IC defs */
1418 #define ICTIMESYNCFLAG_PROBE 0
1419 #define ICTIMESYNCFLAG_SYNC 1
1420 #define ICTIMESYNCFLAG_SAMPLE 2
1423 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1425 #define WLTIMEDELTA 116444736000000000LL
1428 struct ictimesync_data
{
1435 struct ictimesync_ref_data
{
1437 u64 vmreferencetime
;
1444 struct hyperv_service_callback
{
1448 struct vmbus_channel
*channel
;
1449 void (*callback
)(void *context
);
1452 #define MAX_SRV_VER 0x7ffffff
1453 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr
*icmsghdrp
, u8
*buf
,
1454 const int *fw_version
, int fw_vercnt
,
1455 const int *srv_version
, int srv_vercnt
,
1456 int *nego_fw_version
, int *nego_srv_version
);
1458 void hv_process_channel_removal(u32 relid
);
1460 void vmbus_setevent(struct vmbus_channel
*channel
);
1462 * Negotiated version with the Host.
1465 extern __u32 vmbus_proto_version
;
1467 int vmbus_send_tl_connect_request(const uuid_le
*shv_guest_servie_id
,
1468 const uuid_le
*shv_host_servie_id
);
1469 void vmbus_set_event(struct vmbus_channel
*channel
);
1471 /* Get the start of the ring buffer. */
1472 static inline void *
1473 hv_get_ring_buffer(const struct hv_ring_buffer_info
*ring_info
)
1475 return ring_info
->ring_buffer
->buffer
;
1479 * To optimize the flow management on the send-side,
1480 * when the sender is blocked because of lack of
1481 * sufficient space in the ring buffer, potential the
1482 * consumer of the ring buffer can signal the producer.
1483 * This is controlled by the following parameters:
1485 * 1. pending_send_sz: This is the size in bytes that the
1486 * producer is trying to send.
1487 * 2. The feature bit feat_pending_send_sz set to indicate if
1488 * the consumer of the ring will signal when the ring
1489 * state transitions from being full to a state where
1490 * there is room for the producer to send the pending packet.
1493 static inline void hv_signal_on_read(struct vmbus_channel
*channel
)
1495 u32 cur_write_sz
, cached_write_sz
;
1497 struct hv_ring_buffer_info
*rbi
= &channel
->inbound
;
1500 * Issue a full memory barrier before making the signaling decision.
1501 * Here is the reason for having this barrier:
1502 * If the reading of the pend_sz (in this function)
1503 * were to be reordered and read before we commit the new read
1504 * index (in the calling function) we could
1505 * have a problem. If the host were to set the pending_sz after we
1506 * have sampled pending_sz and go to sleep before we commit the
1507 * read index, we could miss sending the interrupt. Issue a full
1508 * memory barrier to address this.
1512 pending_sz
= READ_ONCE(rbi
->ring_buffer
->pending_send_sz
);
1513 /* If the other end is not blocked on write don't bother. */
1514 if (pending_sz
== 0)
1517 cur_write_sz
= hv_get_bytes_to_write(rbi
);
1519 if (cur_write_sz
< pending_sz
)
1522 cached_write_sz
= hv_get_cached_bytes_to_write(rbi
);
1523 if (cached_write_sz
< pending_sz
)
1524 vmbus_setevent(channel
);
1528 * Mask off host interrupt callback notifications
1530 static inline void hv_begin_read(struct hv_ring_buffer_info
*rbi
)
1532 rbi
->ring_buffer
->interrupt_mask
= 1;
1534 /* make sure mask update is not reordered */
1539 * Re-enable host callback and return number of outstanding bytes
1541 static inline u32
hv_end_read(struct hv_ring_buffer_info
*rbi
)
1544 rbi
->ring_buffer
->interrupt_mask
= 0;
1546 /* make sure mask update is not reordered */
1550 * Now check to see if the ring buffer is still empty.
1551 * If it is not, we raced and we need to process new
1552 * incoming messages.
1554 return hv_get_bytes_to_read(rbi
);
1558 * An API to support in-place processing of incoming VMBUS packets.
1561 /* Get data payload associated with descriptor */
1562 static inline void *hv_pkt_data(const struct vmpacket_descriptor
*desc
)
1564 return (void *)((unsigned long)desc
+ (desc
->offset8
<< 3));
1567 /* Get data size associated with descriptor */
1568 static inline u32
hv_pkt_datalen(const struct vmpacket_descriptor
*desc
)
1570 return (desc
->len8
<< 3) - (desc
->offset8
<< 3);
1574 struct vmpacket_descriptor
*
1575 hv_pkt_iter_first(struct vmbus_channel
*channel
);
1577 struct vmpacket_descriptor
*
1578 __hv_pkt_iter_next(struct vmbus_channel
*channel
,
1579 const struct vmpacket_descriptor
*pkt
);
1581 void hv_pkt_iter_close(struct vmbus_channel
*channel
);
1584 * Get next packet descriptor from iterator
1585 * If at end of list, return NULL and update host.
1587 static inline struct vmpacket_descriptor
*
1588 hv_pkt_iter_next(struct vmbus_channel
*channel
,
1589 const struct vmpacket_descriptor
*pkt
)
1591 struct vmpacket_descriptor
*nxt
;
1593 nxt
= __hv_pkt_iter_next(channel
, pkt
);
1595 hv_pkt_iter_close(channel
);
1600 #define foreach_vmbus_pkt(pkt, channel) \
1601 for (pkt = hv_pkt_iter_first(channel); pkt; \
1602 pkt = hv_pkt_iter_next(channel, pkt))
1604 #endif /* _HYPERV_H */