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 */
132 * hv_get_ringbuffer_availbytes()
134 * Get number of bytes available to read and to write to
135 * for the specified ring buffer
138 hv_get_ringbuffer_availbytes(const struct hv_ring_buffer_info
*rbi
,
139 u32
*read
, u32
*write
)
141 u32 read_loc
, write_loc
, dsize
;
143 /* Capture the read/write indices before they changed */
144 read_loc
= rbi
->ring_buffer
->read_index
;
145 write_loc
= rbi
->ring_buffer
->write_index
;
146 dsize
= rbi
->ring_datasize
;
148 *write
= write_loc
>= read_loc
? dsize
- (write_loc
- read_loc
) :
149 read_loc
- write_loc
;
150 *read
= dsize
- *write
;
153 static inline u32
hv_get_bytes_to_read(const struct hv_ring_buffer_info
*rbi
)
155 u32 read_loc
, write_loc
, dsize
, read
;
157 dsize
= rbi
->ring_datasize
;
158 read_loc
= rbi
->ring_buffer
->read_index
;
159 write_loc
= READ_ONCE(rbi
->ring_buffer
->write_index
);
161 read
= write_loc
>= read_loc
? (write_loc
- read_loc
) :
162 (dsize
- read_loc
) + write_loc
;
167 static inline u32
hv_get_bytes_to_write(const struct hv_ring_buffer_info
*rbi
)
169 u32 read_loc
, write_loc
, dsize
, write
;
171 dsize
= rbi
->ring_datasize
;
172 read_loc
= READ_ONCE(rbi
->ring_buffer
->read_index
);
173 write_loc
= rbi
->ring_buffer
->write_index
;
175 write
= write_loc
>= read_loc
? dsize
- (write_loc
- read_loc
) :
176 read_loc
- write_loc
;
181 * VMBUS version is 32 bit entity broken up into
182 * two 16 bit quantities: major_number. minor_number.
184 * 0 . 13 (Windows Server 2008)
187 * 3 . 0 (Windows 8 R2)
191 #define VERSION_WS2008 ((0 << 16) | (13))
192 #define VERSION_WIN7 ((1 << 16) | (1))
193 #define VERSION_WIN8 ((2 << 16) | (4))
194 #define VERSION_WIN8_1 ((3 << 16) | (0))
195 #define VERSION_WIN10 ((4 << 16) | (0))
197 #define VERSION_INVAL -1
199 #define VERSION_CURRENT VERSION_WIN10
201 /* Make maximum size of pipe payload of 16K */
202 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
204 /* Define PipeMode values. */
205 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
206 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
208 /* The size of the user defined data buffer for non-pipe offers. */
209 #define MAX_USER_DEFINED_BYTES 120
211 /* The size of the user defined data buffer for pipe offers. */
212 #define MAX_PIPE_USER_DEFINED_BYTES 116
215 * At the center of the Channel Management library is the Channel Offer. This
216 * struct contains the fundamental information about an offer.
218 struct vmbus_channel_offer
{
223 * These two fields are not currently used.
229 u16 mmio_megabytes
; /* in bytes * 1024 * 1024 */
232 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
234 unsigned char user_def
[MAX_USER_DEFINED_BYTES
];
239 * The following sructure is an integrated pipe protocol, which
240 * is implemented on top of standard user-defined data. Pipe
241 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
246 unsigned char user_def
[MAX_PIPE_USER_DEFINED_BYTES
];
250 * The sub_channel_index is defined in win8.
252 u16 sub_channel_index
;
257 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
258 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
259 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
260 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
261 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
262 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
263 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
264 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
266 struct vmpacket_descriptor
{
274 struct vmpacket_header
{
275 u32 prev_pkt_start_offset
;
276 struct vmpacket_descriptor descriptor
;
279 struct vmtransfer_page_range
{
284 struct vmtransfer_page_packet_header
{
285 struct vmpacket_descriptor d
;
290 struct vmtransfer_page_range ranges
[1];
293 struct vmgpadl_packet_header
{
294 struct vmpacket_descriptor d
;
299 struct vmadd_remove_transfer_page_set
{
300 struct vmpacket_descriptor d
;
307 * This structure defines a range in guest physical space that can be made to
308 * look virtually contiguous.
317 * This is the format for an Establish Gpadl packet, which contains a handle by
318 * which this GPADL will be known and a set of GPA ranges associated with it.
319 * This can be converted to a MDL by the guest OS. If there are multiple GPA
320 * ranges, then the resulting MDL will be "chained," representing multiple VA
323 struct vmestablish_gpadl
{
324 struct vmpacket_descriptor d
;
327 struct gpa_range range
[1];
331 * This is the format for a Teardown Gpadl packet, which indicates that the
332 * GPADL handle in the Establish Gpadl packet will never be referenced again.
334 struct vmteardown_gpadl
{
335 struct vmpacket_descriptor d
;
337 u32 reserved
; /* for alignment to a 8-byte boundary */
341 * This is the format for a GPA-Direct packet, which contains a set of GPA
342 * ranges, in addition to commands and/or data.
344 struct vmdata_gpa_direct
{
345 struct vmpacket_descriptor d
;
348 struct gpa_range range
[1];
351 /* This is the format for a Additional Data Packet. */
352 struct vmadditional_data
{
353 struct vmpacket_descriptor d
;
357 unsigned char data
[1];
360 union vmpacket_largest_possible_header
{
361 struct vmpacket_descriptor simple_hdr
;
362 struct vmtransfer_page_packet_header xfer_page_hdr
;
363 struct vmgpadl_packet_header gpadl_hdr
;
364 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr
;
365 struct vmestablish_gpadl establish_gpadl_hdr
;
366 struct vmteardown_gpadl teardown_gpadl_hdr
;
367 struct vmdata_gpa_direct data_gpa_direct_hdr
;
370 #define VMPACKET_DATA_START_ADDRESS(__packet) \
371 (void *)(((unsigned char *)__packet) + \
372 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
374 #define VMPACKET_DATA_LENGTH(__packet) \
375 ((((struct vmpacket_descriptor)__packet)->len8 - \
376 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
378 #define VMPACKET_TRANSFER_MODE(__packet) \
379 (((struct IMPACT)__packet)->type)
381 enum vmbus_packet_type
{
382 VM_PKT_INVALID
= 0x0,
384 VM_PKT_ADD_XFER_PAGESET
= 0x2,
385 VM_PKT_RM_XFER_PAGESET
= 0x3,
386 VM_PKT_ESTABLISH_GPADL
= 0x4,
387 VM_PKT_TEARDOWN_GPADL
= 0x5,
388 VM_PKT_DATA_INBAND
= 0x6,
389 VM_PKT_DATA_USING_XFER_PAGES
= 0x7,
390 VM_PKT_DATA_USING_GPADL
= 0x8,
391 VM_PKT_DATA_USING_GPA_DIRECT
= 0x9,
392 VM_PKT_CANCEL_REQUEST
= 0xa,
394 VM_PKT_DATA_USING_ADDITIONAL_PKT
= 0xc,
395 VM_PKT_ADDITIONAL_DATA
= 0xd
398 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
401 /* Version 1 messages */
402 enum vmbus_channel_message_type
{
403 CHANNELMSG_INVALID
= 0,
404 CHANNELMSG_OFFERCHANNEL
= 1,
405 CHANNELMSG_RESCIND_CHANNELOFFER
= 2,
406 CHANNELMSG_REQUESTOFFERS
= 3,
407 CHANNELMSG_ALLOFFERS_DELIVERED
= 4,
408 CHANNELMSG_OPENCHANNEL
= 5,
409 CHANNELMSG_OPENCHANNEL_RESULT
= 6,
410 CHANNELMSG_CLOSECHANNEL
= 7,
411 CHANNELMSG_GPADL_HEADER
= 8,
412 CHANNELMSG_GPADL_BODY
= 9,
413 CHANNELMSG_GPADL_CREATED
= 10,
414 CHANNELMSG_GPADL_TEARDOWN
= 11,
415 CHANNELMSG_GPADL_TORNDOWN
= 12,
416 CHANNELMSG_RELID_RELEASED
= 13,
417 CHANNELMSG_INITIATE_CONTACT
= 14,
418 CHANNELMSG_VERSION_RESPONSE
= 15,
419 CHANNELMSG_UNLOAD
= 16,
420 CHANNELMSG_UNLOAD_RESPONSE
= 17,
424 CHANNELMSG_TL_CONNECT_REQUEST
= 21,
428 struct vmbus_channel_message_header
{
429 enum vmbus_channel_message_type msgtype
;
433 /* Query VMBus Version parameters */
434 struct vmbus_channel_query_vmbus_version
{
435 struct vmbus_channel_message_header header
;
439 /* VMBus Version Supported parameters */
440 struct vmbus_channel_version_supported
{
441 struct vmbus_channel_message_header header
;
442 u8 version_supported
;
445 /* Offer Channel parameters */
446 struct vmbus_channel_offer_channel
{
447 struct vmbus_channel_message_header header
;
448 struct vmbus_channel_offer offer
;
452 * win7 and beyond splits this field into a bit field.
454 u8 monitor_allocated
:1;
457 * These are new fields added in win7 and later.
458 * Do not access these fields without checking the
459 * negotiated protocol.
461 * If "is_dedicated_interrupt" is set, we must not set the
462 * associated bit in the channel bitmap while sending the
463 * interrupt to the host.
465 * connection_id is to be used in signaling the host.
467 u16 is_dedicated_interrupt
:1;
472 /* Rescind Offer parameters */
473 struct vmbus_channel_rescind_offer
{
474 struct vmbus_channel_message_header header
;
479 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info
*rbi
)
481 return rbi
->ring_buffer
->pending_send_sz
;
485 * Request Offer -- no parameters, SynIC message contains the partition ID
486 * Set Snoop -- no parameters, SynIC message contains the partition ID
487 * Clear Snoop -- no parameters, SynIC message contains the partition ID
488 * All Offers Delivered -- no parameters, SynIC message contains the partition
490 * Flush Client -- no parameters, SynIC message contains the partition ID
493 /* Open Channel parameters */
494 struct vmbus_channel_open_channel
{
495 struct vmbus_channel_message_header header
;
497 /* Identifies the specific VMBus channel that is being opened. */
500 /* ID making a particular open request at a channel offer unique. */
503 /* GPADL for the channel's ring buffer. */
504 u32 ringbuffer_gpadlhandle
;
507 * Starting with win8, this field will be used to specify
508 * the target virtual processor on which to deliver the interrupt for
509 * the host to guest communication.
510 * Prior to win8, incoming channel interrupts would only
511 * be delivered on cpu 0. Setting this value to 0 would
512 * preserve the earlier behavior.
517 * The upstream ring buffer begins at offset zero in the memory
518 * described by RingBufferGpadlHandle. The downstream ring buffer
519 * follows it at this offset (in pages).
521 u32 downstream_ringbuffer_pageoffset
;
523 /* User-specific data to be passed along to the server endpoint. */
524 unsigned char userdata
[MAX_USER_DEFINED_BYTES
];
527 /* Open Channel Result parameters */
528 struct vmbus_channel_open_result
{
529 struct vmbus_channel_message_header header
;
535 /* Close channel parameters; */
536 struct vmbus_channel_close_channel
{
537 struct vmbus_channel_message_header header
;
541 /* Channel Message GPADL */
542 #define GPADL_TYPE_RING_BUFFER 1
543 #define GPADL_TYPE_SERVER_SAVE_AREA 2
544 #define GPADL_TYPE_TRANSACTION 8
547 * The number of PFNs in a GPADL message is defined by the number of
548 * pages that would be spanned by ByteCount and ByteOffset. If the
549 * implied number of PFNs won't fit in this packet, there will be a
550 * follow-up packet that contains more.
552 struct vmbus_channel_gpadl_header
{
553 struct vmbus_channel_message_header header
;
558 struct gpa_range range
[0];
561 /* This is the followup packet that contains more PFNs. */
562 struct vmbus_channel_gpadl_body
{
563 struct vmbus_channel_message_header header
;
569 struct vmbus_channel_gpadl_created
{
570 struct vmbus_channel_message_header header
;
576 struct vmbus_channel_gpadl_teardown
{
577 struct vmbus_channel_message_header header
;
582 struct vmbus_channel_gpadl_torndown
{
583 struct vmbus_channel_message_header header
;
587 struct vmbus_channel_relid_released
{
588 struct vmbus_channel_message_header header
;
592 struct vmbus_channel_initiate_contact
{
593 struct vmbus_channel_message_header header
;
594 u32 vmbus_version_requested
;
595 u32 target_vcpu
; /* The VCPU the host should respond to */
601 /* Hyper-V socket: guest's connect()-ing to host */
602 struct vmbus_channel_tl_connect_request
{
603 struct vmbus_channel_message_header header
;
604 uuid_le guest_endpoint_id
;
605 uuid_le host_service_id
;
608 struct vmbus_channel_version_response
{
609 struct vmbus_channel_message_header header
;
610 u8 version_supported
;
613 enum vmbus_channel_state
{
615 CHANNEL_OPENING_STATE
,
617 CHANNEL_OPENED_STATE
,
621 * Represents each channel msg on the vmbus connection This is a
622 * variable-size data structure depending on the msg type itself
624 struct vmbus_channel_msginfo
{
625 /* Bookkeeping stuff */
626 struct list_head msglistentry
;
628 /* So far, this is only used to handle gpadl body message */
629 struct list_head submsglist
;
631 /* Synchronize the request/response if needed */
632 struct completion waitevent
;
633 struct vmbus_channel
*waiting_channel
;
635 struct vmbus_channel_version_supported version_supported
;
636 struct vmbus_channel_open_result open_result
;
637 struct vmbus_channel_gpadl_torndown gpadl_torndown
;
638 struct vmbus_channel_gpadl_created gpadl_created
;
639 struct vmbus_channel_version_response version_response
;
644 * The channel message that goes out on the "wire".
645 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
647 unsigned char msg
[0];
650 struct vmbus_close_msg
{
651 struct vmbus_channel_msginfo info
;
652 struct vmbus_channel_close_channel msg
;
655 /* Define connection identifier type. */
656 union hv_connection_id
{
664 enum hv_numa_policy
{
669 enum vmbus_device_type
{
689 struct vmbus_device
{
695 struct vmbus_channel
{
696 struct list_head listentry
;
698 struct hv_device
*device_obj
;
700 enum vmbus_channel_state state
;
702 struct vmbus_channel_offer_channel offermsg
;
704 * These are based on the OfferMsg.MonitorId.
705 * Save it here for easy access.
710 bool rescind
; /* got rescind msg */
711 struct completion rescind_event
;
713 u32 ringbuffer_gpadlhandle
;
715 /* Allocated memory for ring buffer */
716 void *ringbuffer_pages
;
717 u32 ringbuffer_pagecount
;
718 struct hv_ring_buffer_info outbound
; /* send to parent */
719 struct hv_ring_buffer_info inbound
; /* receive from parent */
721 struct vmbus_close_msg close_msg
;
724 u64 interrupts
; /* Host to Guest interrupts */
725 u64 sig_events
; /* Guest to Host events */
727 /* Channel callback's invoked in softirq context */
728 struct tasklet_struct callback_event
;
729 void (*onchannel_callback
)(void *context
);
730 void *channel_callback_context
;
733 * A channel can be marked for one of three modes of reading:
734 * BATCHED - callback called from taslket and should read
735 * channel until empty. Interrupts from the host
736 * are masked while read is in process (default).
737 * DIRECT - callback called from tasklet (softirq).
738 * ISR - callback called in interrupt context and must
739 * invoke its own deferred processing.
740 * Host interrupts are disabled and must be re-enabled
741 * when ring is empty.
743 enum hv_callback_mode
{
749 bool is_dedicated_interrupt
;
753 * Starting with win8, this field will be used to specify
754 * the target virtual processor on which to deliver the interrupt for
755 * the host to guest communication.
756 * Prior to win8, incoming channel interrupts would only
757 * be delivered on cpu 0. Setting this value to 0 would
758 * preserve the earlier behavior.
761 /* The corresponding CPUID in the guest */
764 * State to manage the CPU affiliation of channels.
766 struct cpumask alloced_cpus_in_node
;
769 * Support for sub-channels. For high performance devices,
770 * it will be useful to have multiple sub-channels to support
771 * a scalable communication infrastructure with the host.
772 * The support for sub-channels is implemented as an extention
773 * to the current infrastructure.
774 * The initial offer is considered the primary channel and this
775 * offer message will indicate if the host supports sub-channels.
776 * The guest is free to ask for sub-channels to be offerred and can
777 * open these sub-channels as a normal "primary" channel. However,
778 * all sub-channels will have the same type and instance guids as the
779 * primary channel. Requests sent on a given channel will result in a
780 * response on the same channel.
784 * Sub-channel creation callback. This callback will be called in
785 * process context when a sub-channel offer is received from the host.
786 * The guest can open the sub-channel in the context of this callback.
788 void (*sc_creation_callback
)(struct vmbus_channel
*new_sc
);
791 * Channel rescind callback. Some channels (the hvsock ones), need to
792 * register a callback which is invoked in vmbus_onoffer_rescind().
794 void (*chn_rescind_callback
)(struct vmbus_channel
*channel
);
797 * The spinlock to protect the structure. It is being used to protect
798 * test-and-set access to various attributes of the structure as well
799 * as all sc_list operations.
803 * All Sub-channels of a primary channel are linked here.
805 struct list_head sc_list
;
807 * Current number of sub-channels.
811 * Number of a sub-channel (position within sc_list) which is supposed
812 * to be used as the next outgoing channel.
816 * The primary channel this sub-channel belongs to.
817 * This will be NULL for the primary channel.
819 struct vmbus_channel
*primary_channel
;
821 * Support per-channel state for use by vmbus drivers.
823 void *per_channel_state
;
825 * To support per-cpu lookup mapping of relid to channel,
826 * link up channels based on their CPU affinity.
828 struct list_head percpu_list
;
831 * Defer freeing channel until after all cpu's have
832 * gone through grace period.
837 * For sysfs per-channel properties.
842 * For performance critical channels (storage, networking
843 * etc,), Hyper-V has a mechanism to enhance the throughput
844 * at the expense of latency:
845 * When the host is to be signaled, we just set a bit in a shared page
846 * and this bit will be inspected by the hypervisor within a certain
847 * window and if the bit is set, the host will be signaled. The window
848 * of time is the monitor latency - currently around 100 usecs. This
849 * mechanism improves throughput by:
851 * A) Making the host more efficient - each time it wakes up,
852 * potentially it will process morev number of packets. The
853 * monitor latency allows a batch to build up.
854 * B) By deferring the hypercall to signal, we will also minimize
857 * Clearly, these optimizations improve throughput at the expense of
858 * latency. Furthermore, since the channel is shared for both
859 * control and data messages, control messages currently suffer
860 * unnecessary latency adversley impacting performance and boot
861 * time. To fix this issue, permit tagging the channel as being
862 * in "low latency" mode. In this mode, we will bypass the monitor
868 * NUMA distribution policy:
869 * We support teo policies:
870 * 1) Balanced: Here all performance critical channels are
871 * distributed evenly amongst all the NUMA nodes.
872 * This policy will be the default policy.
873 * 2) Localized: All channels of a given instance of a
874 * performance critical service will be assigned CPUs
875 * within a selected NUMA node.
877 enum hv_numa_policy affinity_policy
;
882 * We must offload the handling of the primary/sub channels
883 * from the single-threaded vmbus_connection.work_queue to
884 * two different workqueue, otherwise we can block
885 * vmbus_connection.work_queue and hang: see vmbus_process_offer().
887 struct work_struct add_channel_work
;
890 static inline bool is_hvsock_channel(const struct vmbus_channel
*c
)
892 return !!(c
->offermsg
.offer
.chn_flags
&
893 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
);
896 static inline void set_channel_affinity_state(struct vmbus_channel
*c
,
897 enum hv_numa_policy policy
)
899 c
->affinity_policy
= policy
;
902 static inline void set_channel_read_mode(struct vmbus_channel
*c
,
903 enum hv_callback_mode mode
)
905 c
->callback_mode
= mode
;
908 static inline void set_per_channel_state(struct vmbus_channel
*c
, void *s
)
910 c
->per_channel_state
= s
;
913 static inline void *get_per_channel_state(struct vmbus_channel
*c
)
915 return c
->per_channel_state
;
918 static inline void set_channel_pending_send_size(struct vmbus_channel
*c
,
921 c
->outbound
.ring_buffer
->pending_send_sz
= size
;
924 static inline void set_low_latency_mode(struct vmbus_channel
*c
)
926 c
->low_latency
= true;
929 static inline void clear_low_latency_mode(struct vmbus_channel
*c
)
931 c
->low_latency
= false;
934 void vmbus_onmessage(void *context
);
936 int vmbus_request_offers(void);
939 * APIs for managing sub-channels.
942 void vmbus_set_sc_create_callback(struct vmbus_channel
*primary_channel
,
943 void (*sc_cr_cb
)(struct vmbus_channel
*new_sc
));
945 void vmbus_set_chn_rescind_callback(struct vmbus_channel
*channel
,
946 void (*chn_rescind_cb
)(struct vmbus_channel
*));
949 * Retrieve the (sub) channel on which to send an outgoing request.
950 * When a primary channel has multiple sub-channels, we choose a
951 * channel whose VCPU binding is closest to the VCPU on which
952 * this call is being made.
954 struct vmbus_channel
*vmbus_get_outgoing_channel(struct vmbus_channel
*primary
);
957 * Check if sub-channels have already been offerred. This API will be useful
958 * when the driver is unloaded after establishing sub-channels. In this case,
959 * when the driver is re-loaded, the driver would have to check if the
960 * subchannels have already been established before attempting to request
961 * the creation of sub-channels.
962 * This function returns TRUE to indicate that subchannels have already been
964 * This function should be invoked after setting the callback function for
965 * sub-channel creation.
967 bool vmbus_are_subchannels_present(struct vmbus_channel
*primary
);
969 /* The format must be the same as struct vmdata_gpa_direct */
970 struct vmbus_channel_packet_page_buffer
{
978 struct hv_page_buffer range
[MAX_PAGE_BUFFER_COUNT
];
981 /* The format must be the same as struct vmdata_gpa_direct */
982 struct vmbus_channel_packet_multipage_buffer
{
989 u32 rangecount
; /* Always 1 in this case */
990 struct hv_multipage_buffer range
;
993 /* The format must be the same as struct vmdata_gpa_direct */
994 struct vmbus_packet_mpb_array
{
1001 u32 rangecount
; /* Always 1 in this case */
1002 struct hv_mpb_array range
;
1006 extern int vmbus_open(struct vmbus_channel
*channel
,
1007 u32 send_ringbuffersize
,
1008 u32 recv_ringbuffersize
,
1011 void (*onchannel_callback
)(void *context
),
1014 extern void vmbus_close(struct vmbus_channel
*channel
);
1016 extern int vmbus_sendpacket(struct vmbus_channel
*channel
,
1020 enum vmbus_packet_type type
,
1023 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel
*channel
,
1024 struct hv_page_buffer pagebuffers
[],
1030 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel
*channel
,
1031 struct vmbus_packet_mpb_array
*mpb
,
1037 extern int vmbus_establish_gpadl(struct vmbus_channel
*channel
,
1042 extern int vmbus_teardown_gpadl(struct vmbus_channel
*channel
,
1045 void vmbus_reset_channel_cb(struct vmbus_channel
*channel
);
1047 extern int vmbus_recvpacket(struct vmbus_channel
*channel
,
1050 u32
*buffer_actual_len
,
1053 extern int vmbus_recvpacket_raw(struct vmbus_channel
*channel
,
1056 u32
*buffer_actual_len
,
1060 extern void vmbus_ontimer(unsigned long data
);
1062 /* Base driver object */
1067 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1068 * channel flag, actually doesn't mean a synthetic device because the
1069 * offer's if_type/if_instance can change for every new hvsock
1072 * However, to facilitate the notification of new-offer/rescind-offer
1073 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1074 * a special vmbus device, and hence we need the below flag to
1075 * indicate if the driver is the hvsock driver or not: we need to
1076 * specially treat the hvosck offer & driver in vmbus_match().
1080 /* the device type supported by this driver */
1082 const struct hv_vmbus_device_id
*id_table
;
1084 struct device_driver driver
;
1086 /* dynamic device GUID's */
1089 struct list_head list
;
1092 int (*probe
)(struct hv_device
*, const struct hv_vmbus_device_id
*);
1093 int (*remove
)(struct hv_device
*);
1094 void (*shutdown
)(struct hv_device
*);
1098 /* Base device object */
1100 /* the device type id of this device */
1103 /* the device instance id of this device */
1104 uuid_le dev_instance
;
1108 struct device device
;
1110 struct vmbus_channel
*channel
;
1111 struct kset
*channels_kset
;
1115 static inline struct hv_device
*device_to_hv_device(struct device
*d
)
1117 return container_of(d
, struct hv_device
, device
);
1120 static inline struct hv_driver
*drv_to_hv_drv(struct device_driver
*d
)
1122 return container_of(d
, struct hv_driver
, driver
);
1125 static inline void hv_set_drvdata(struct hv_device
*dev
, void *data
)
1127 dev_set_drvdata(&dev
->device
, data
);
1130 static inline void *hv_get_drvdata(struct hv_device
*dev
)
1132 return dev_get_drvdata(&dev
->device
);
1135 struct hv_ring_buffer_debug_info
{
1136 u32 current_interrupt_mask
;
1137 u32 current_read_index
;
1138 u32 current_write_index
;
1139 u32 bytes_avail_toread
;
1140 u32 bytes_avail_towrite
;
1144 int hv_ringbuffer_get_debuginfo(const struct hv_ring_buffer_info
*ring_info
,
1145 struct hv_ring_buffer_debug_info
*debug_info
);
1147 /* Vmbus interface */
1148 #define vmbus_driver_register(driver) \
1149 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1150 int __must_check
__vmbus_driver_register(struct hv_driver
*hv_driver
,
1151 struct module
*owner
,
1152 const char *mod_name
);
1153 void vmbus_driver_unregister(struct hv_driver
*hv_driver
);
1155 void vmbus_hvsock_device_unregister(struct vmbus_channel
*channel
);
1157 int vmbus_allocate_mmio(struct resource
**new, struct hv_device
*device_obj
,
1158 resource_size_t min
, resource_size_t max
,
1159 resource_size_t size
, resource_size_t align
,
1160 bool fb_overlap_ok
);
1161 void vmbus_free_mmio(resource_size_t start
, resource_size_t size
);
1164 * GUID definitions of various offer types - services offered to the guest.
1169 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1171 #define HV_NIC_GUID \
1172 .guid = UUID_LE(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1173 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1177 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1179 #define HV_IDE_GUID \
1180 .guid = UUID_LE(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1181 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1185 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1187 #define HV_SCSI_GUID \
1188 .guid = UUID_LE(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1189 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1193 * {0e0b6031-5213-4934-818b-38d90ced39db}
1195 #define HV_SHUTDOWN_GUID \
1196 .guid = UUID_LE(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1197 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1201 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1203 #define HV_TS_GUID \
1204 .guid = UUID_LE(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1205 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1209 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1211 #define HV_HEART_BEAT_GUID \
1212 .guid = UUID_LE(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1213 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1217 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1219 #define HV_KVP_GUID \
1220 .guid = UUID_LE(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1221 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1224 * Dynamic memory GUID
1225 * {525074dc-8985-46e2-8057-a307dc18a502}
1227 #define HV_DM_GUID \
1228 .guid = UUID_LE(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1229 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1233 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1235 #define HV_MOUSE_GUID \
1236 .guid = UUID_LE(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1237 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1241 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1243 #define HV_KBD_GUID \
1244 .guid = UUID_LE(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1245 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1248 * VSS (Backup/Restore) GUID
1250 #define HV_VSS_GUID \
1251 .guid = UUID_LE(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1252 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1254 * Synthetic Video GUID
1255 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1257 #define HV_SYNTHVID_GUID \
1258 .guid = UUID_LE(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1259 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1263 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1265 #define HV_SYNTHFC_GUID \
1266 .guid = UUID_LE(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1267 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1270 * Guest File Copy Service
1271 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1274 #define HV_FCOPY_GUID \
1275 .guid = UUID_LE(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1276 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1279 * NetworkDirect. This is the guest RDMA service.
1280 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1282 #define HV_ND_GUID \
1283 .guid = UUID_LE(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1284 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1287 * PCI Express Pass Through
1288 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1291 #define HV_PCIE_GUID \
1292 .guid = UUID_LE(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1293 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1296 * Linux doesn't support the 3 devices: the first two are for
1297 * Automatic Virtual Machine Activation, and the third is for
1298 * Remote Desktop Virtualization.
1299 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1300 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1301 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1304 #define HV_AVMA1_GUID \
1305 .guid = UUID_LE(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1306 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1308 #define HV_AVMA2_GUID \
1309 .guid = UUID_LE(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1310 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1312 #define HV_RDV_GUID \
1313 .guid = UUID_LE(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1314 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1317 * Common header for Hyper-V ICs
1320 #define ICMSGTYPE_NEGOTIATE 0
1321 #define ICMSGTYPE_HEARTBEAT 1
1322 #define ICMSGTYPE_KVPEXCHANGE 2
1323 #define ICMSGTYPE_SHUTDOWN 3
1324 #define ICMSGTYPE_TIMESYNC 4
1325 #define ICMSGTYPE_VSS 5
1327 #define ICMSGHDRFLAG_TRANSACTION 1
1328 #define ICMSGHDRFLAG_REQUEST 2
1329 #define ICMSGHDRFLAG_RESPONSE 4
1333 * While we want to handle util services as regular devices,
1334 * there is only one instance of each of these services; so
1335 * we statically allocate the service specific state.
1338 struct hv_util_service
{
1341 void (*util_cb
)(void *);
1342 int (*util_init
)(struct hv_util_service
*);
1343 void (*util_deinit
)(void);
1346 struct vmbuspipe_hdr
{
1357 struct ic_version icverframe
;
1359 struct ic_version icvermsg
;
1362 u8 ictransaction_id
;
1367 struct icmsg_negotiate
{
1371 struct ic_version icversion_data
[1]; /* any size array */
1374 struct shutdown_msg_data
{
1376 u32 timeout_seconds
;
1378 u8 display_message
[2048];
1381 struct heartbeat_msg_data
{
1386 /* Time Sync IC defs */
1387 #define ICTIMESYNCFLAG_PROBE 0
1388 #define ICTIMESYNCFLAG_SYNC 1
1389 #define ICTIMESYNCFLAG_SAMPLE 2
1392 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1394 #define WLTIMEDELTA 116444736000000000LL
1397 struct ictimesync_data
{
1404 struct ictimesync_ref_data
{
1406 u64 vmreferencetime
;
1413 struct hyperv_service_callback
{
1417 struct vmbus_channel
*channel
;
1418 void (*callback
)(void *context
);
1421 #define MAX_SRV_VER 0x7ffffff
1422 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr
*icmsghdrp
, u8
*buf
,
1423 const int *fw_version
, int fw_vercnt
,
1424 const int *srv_version
, int srv_vercnt
,
1425 int *nego_fw_version
, int *nego_srv_version
);
1427 void hv_process_channel_removal(u32 relid
);
1429 void vmbus_setevent(struct vmbus_channel
*channel
);
1431 * Negotiated version with the Host.
1434 extern __u32 vmbus_proto_version
;
1436 int vmbus_send_tl_connect_request(const uuid_le
*shv_guest_servie_id
,
1437 const uuid_le
*shv_host_servie_id
);
1438 void vmbus_set_event(struct vmbus_channel
*channel
);
1440 /* Get the start of the ring buffer. */
1441 static inline void *
1442 hv_get_ring_buffer(const struct hv_ring_buffer_info
*ring_info
)
1444 return ring_info
->ring_buffer
->buffer
;
1448 * Mask off host interrupt callback notifications
1450 static inline void hv_begin_read(struct hv_ring_buffer_info
*rbi
)
1452 rbi
->ring_buffer
->interrupt_mask
= 1;
1454 /* make sure mask update is not reordered */
1459 * Re-enable host callback and return number of outstanding bytes
1461 static inline u32
hv_end_read(struct hv_ring_buffer_info
*rbi
)
1464 rbi
->ring_buffer
->interrupt_mask
= 0;
1466 /* make sure mask update is not reordered */
1470 * Now check to see if the ring buffer is still empty.
1471 * If it is not, we raced and we need to process new
1472 * incoming messages.
1474 return hv_get_bytes_to_read(rbi
);
1478 * An API to support in-place processing of incoming VMBUS packets.
1481 /* Get data payload associated with descriptor */
1482 static inline void *hv_pkt_data(const struct vmpacket_descriptor
*desc
)
1484 return (void *)((unsigned long)desc
+ (desc
->offset8
<< 3));
1487 /* Get data size associated with descriptor */
1488 static inline u32
hv_pkt_datalen(const struct vmpacket_descriptor
*desc
)
1490 return (desc
->len8
<< 3) - (desc
->offset8
<< 3);
1494 struct vmpacket_descriptor
*
1495 hv_pkt_iter_first(struct vmbus_channel
*channel
);
1497 struct vmpacket_descriptor
*
1498 __hv_pkt_iter_next(struct vmbus_channel
*channel
,
1499 const struct vmpacket_descriptor
*pkt
);
1501 void hv_pkt_iter_close(struct vmbus_channel
*channel
);
1504 * Get next packet descriptor from iterator
1505 * If at end of list, return NULL and update host.
1507 static inline struct vmpacket_descriptor
*
1508 hv_pkt_iter_next(struct vmbus_channel
*channel
,
1509 const struct vmpacket_descriptor
*pkt
)
1511 struct vmpacket_descriptor
*nxt
;
1513 nxt
= __hv_pkt_iter_next(channel
, pkt
);
1515 hv_pkt_iter_close(channel
);
1520 #define foreach_vmbus_pkt(pkt, channel) \
1521 for (pkt = hv_pkt_iter_first(channel); pkt; \
1522 pkt = hv_pkt_iter_next(channel, pkt))
1524 #endif /* _HYPERV_H */