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/workqueue.h>
36 #include <linux/completion.h>
37 #include <linux/device.h>
38 #include <linux/mod_devicetable.h>
41 #define MAX_PAGE_BUFFER_COUNT 32
42 #define MAX_MULTIPAGE_BUFFER_COUNT 32 /* 128K */
46 /* Single-page buffer */
47 struct hv_page_buffer
{
53 /* Multiple-page buffer */
54 struct hv_multipage_buffer
{
55 /* Length and Offset determines the # of pfns in the array */
58 u64 pfn_array
[MAX_MULTIPAGE_BUFFER_COUNT
];
62 * Multiple-page buffer array; the pfn array is variable size:
63 * The number of entries in the PFN array is determined by
67 /* Length and Offset determines the # of pfns in the array */
73 /* 0x18 includes the proprietary packet header */
74 #define MAX_PAGE_BUFFER_PACKET (0x18 + \
75 (sizeof(struct hv_page_buffer) * \
76 MAX_PAGE_BUFFER_COUNT))
77 #define MAX_MULTIPAGE_BUFFER_PACKET (0x18 + \
78 sizeof(struct hv_multipage_buffer))
83 struct hv_ring_buffer
{
84 /* Offset in bytes from the start of ring data below */
87 /* Offset in bytes from the start of ring data below */
93 * Win8 uses some of the reserved bits to implement
94 * interrupt driven flow management. On the send side
95 * we can request that the receiver interrupt the sender
96 * when the ring transitions from being full to being able
97 * to handle a message of size "pending_send_sz".
99 * Add necessary state for this enhancement.
107 u32 feat_pending_send_sz
:1;
112 /* Pad it to PAGE_SIZE so that data starts on page boundary */
116 * Ring data starts here + RingDataStartOffset
117 * !!! DO NOT place any fields below this !!!
122 struct hv_ring_buffer_info
{
123 struct hv_ring_buffer
*ring_buffer
;
124 u32 ring_size
; /* Include the shared header */
125 spinlock_t ring_lock
;
127 u32 ring_datasize
; /* < ring_size */
128 u32 ring_data_startoffset
;
129 u32 priv_write_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(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(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(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
;
184 * VMBUS version is 32 bit entity broken up into
185 * two 16 bit quantities: major_number. minor_number.
187 * 0 . 13 (Windows Server 2008)
190 * 3 . 0 (Windows 8 R2)
194 #define VERSION_WS2008 ((0 << 16) | (13))
195 #define VERSION_WIN7 ((1 << 16) | (1))
196 #define VERSION_WIN8 ((2 << 16) | (4))
197 #define VERSION_WIN8_1 ((3 << 16) | (0))
198 #define VERSION_WIN10 ((4 << 16) | (0))
200 #define VERSION_INVAL -1
202 #define VERSION_CURRENT VERSION_WIN10
204 /* Make maximum size of pipe payload of 16K */
205 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
207 /* Define PipeMode values. */
208 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
209 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
211 /* The size of the user defined data buffer for non-pipe offers. */
212 #define MAX_USER_DEFINED_BYTES 120
214 /* The size of the user defined data buffer for pipe offers. */
215 #define MAX_PIPE_USER_DEFINED_BYTES 116
218 * At the center of the Channel Management library is the Channel Offer. This
219 * struct contains the fundamental information about an offer.
221 struct vmbus_channel_offer
{
226 * These two fields are not currently used.
232 u16 mmio_megabytes
; /* in bytes * 1024 * 1024 */
235 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
237 unsigned char user_def
[MAX_USER_DEFINED_BYTES
];
242 * The following sructure is an integrated pipe protocol, which
243 * is implemented on top of standard user-defined data. Pipe
244 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
249 unsigned char user_def
[MAX_PIPE_USER_DEFINED_BYTES
];
253 * The sub_channel_index is defined in win8.
255 u16 sub_channel_index
;
260 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
261 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
262 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
263 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
264 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
265 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
266 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
267 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
269 struct vmpacket_descriptor
{
277 struct vmpacket_header
{
278 u32 prev_pkt_start_offset
;
279 struct vmpacket_descriptor descriptor
;
282 struct vmtransfer_page_range
{
287 struct vmtransfer_page_packet_header
{
288 struct vmpacket_descriptor d
;
293 struct vmtransfer_page_range ranges
[1];
296 struct vmgpadl_packet_header
{
297 struct vmpacket_descriptor d
;
302 struct vmadd_remove_transfer_page_set
{
303 struct vmpacket_descriptor d
;
310 * This structure defines a range in guest physical space that can be made to
311 * look virtually contiguous.
320 * This is the format for an Establish Gpadl packet, which contains a handle by
321 * which this GPADL will be known and a set of GPA ranges associated with it.
322 * This can be converted to a MDL by the guest OS. If there are multiple GPA
323 * ranges, then the resulting MDL will be "chained," representing multiple VA
326 struct vmestablish_gpadl
{
327 struct vmpacket_descriptor d
;
330 struct gpa_range range
[1];
334 * This is the format for a Teardown Gpadl packet, which indicates that the
335 * GPADL handle in the Establish Gpadl packet will never be referenced again.
337 struct vmteardown_gpadl
{
338 struct vmpacket_descriptor d
;
340 u32 reserved
; /* for alignment to a 8-byte boundary */
344 * This is the format for a GPA-Direct packet, which contains a set of GPA
345 * ranges, in addition to commands and/or data.
347 struct vmdata_gpa_direct
{
348 struct vmpacket_descriptor d
;
351 struct gpa_range range
[1];
354 /* This is the format for a Additional Data Packet. */
355 struct vmadditional_data
{
356 struct vmpacket_descriptor d
;
360 unsigned char data
[1];
363 union vmpacket_largest_possible_header
{
364 struct vmpacket_descriptor simple_hdr
;
365 struct vmtransfer_page_packet_header xfer_page_hdr
;
366 struct vmgpadl_packet_header gpadl_hdr
;
367 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr
;
368 struct vmestablish_gpadl establish_gpadl_hdr
;
369 struct vmteardown_gpadl teardown_gpadl_hdr
;
370 struct vmdata_gpa_direct data_gpa_direct_hdr
;
373 #define VMPACKET_DATA_START_ADDRESS(__packet) \
374 (void *)(((unsigned char *)__packet) + \
375 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
377 #define VMPACKET_DATA_LENGTH(__packet) \
378 ((((struct vmpacket_descriptor)__packet)->len8 - \
379 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
381 #define VMPACKET_TRANSFER_MODE(__packet) \
382 (((struct IMPACT)__packet)->type)
384 enum vmbus_packet_type
{
385 VM_PKT_INVALID
= 0x0,
387 VM_PKT_ADD_XFER_PAGESET
= 0x2,
388 VM_PKT_RM_XFER_PAGESET
= 0x3,
389 VM_PKT_ESTABLISH_GPADL
= 0x4,
390 VM_PKT_TEARDOWN_GPADL
= 0x5,
391 VM_PKT_DATA_INBAND
= 0x6,
392 VM_PKT_DATA_USING_XFER_PAGES
= 0x7,
393 VM_PKT_DATA_USING_GPADL
= 0x8,
394 VM_PKT_DATA_USING_GPA_DIRECT
= 0x9,
395 VM_PKT_CANCEL_REQUEST
= 0xa,
397 VM_PKT_DATA_USING_ADDITIONAL_PKT
= 0xc,
398 VM_PKT_ADDITIONAL_DATA
= 0xd
401 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
404 /* Version 1 messages */
405 enum vmbus_channel_message_type
{
406 CHANNELMSG_INVALID
= 0,
407 CHANNELMSG_OFFERCHANNEL
= 1,
408 CHANNELMSG_RESCIND_CHANNELOFFER
= 2,
409 CHANNELMSG_REQUESTOFFERS
= 3,
410 CHANNELMSG_ALLOFFERS_DELIVERED
= 4,
411 CHANNELMSG_OPENCHANNEL
= 5,
412 CHANNELMSG_OPENCHANNEL_RESULT
= 6,
413 CHANNELMSG_CLOSECHANNEL
= 7,
414 CHANNELMSG_GPADL_HEADER
= 8,
415 CHANNELMSG_GPADL_BODY
= 9,
416 CHANNELMSG_GPADL_CREATED
= 10,
417 CHANNELMSG_GPADL_TEARDOWN
= 11,
418 CHANNELMSG_GPADL_TORNDOWN
= 12,
419 CHANNELMSG_RELID_RELEASED
= 13,
420 CHANNELMSG_INITIATE_CONTACT
= 14,
421 CHANNELMSG_VERSION_RESPONSE
= 15,
422 CHANNELMSG_UNLOAD
= 16,
423 CHANNELMSG_UNLOAD_RESPONSE
= 17,
427 CHANNELMSG_TL_CONNECT_REQUEST
= 21,
431 struct vmbus_channel_message_header
{
432 enum vmbus_channel_message_type msgtype
;
436 /* Query VMBus Version parameters */
437 struct vmbus_channel_query_vmbus_version
{
438 struct vmbus_channel_message_header header
;
442 /* VMBus Version Supported parameters */
443 struct vmbus_channel_version_supported
{
444 struct vmbus_channel_message_header header
;
445 u8 version_supported
;
448 /* Offer Channel parameters */
449 struct vmbus_channel_offer_channel
{
450 struct vmbus_channel_message_header header
;
451 struct vmbus_channel_offer offer
;
455 * win7 and beyond splits this field into a bit field.
457 u8 monitor_allocated
:1;
460 * These are new fields added in win7 and later.
461 * Do not access these fields without checking the
462 * negotiated protocol.
464 * If "is_dedicated_interrupt" is set, we must not set the
465 * associated bit in the channel bitmap while sending the
466 * interrupt to the host.
468 * connection_id is to be used in signaling the host.
470 u16 is_dedicated_interrupt
:1;
475 /* Rescind Offer parameters */
476 struct vmbus_channel_rescind_offer
{
477 struct vmbus_channel_message_header header
;
482 * Request Offer -- no parameters, SynIC message contains the partition ID
483 * Set Snoop -- no parameters, SynIC message contains the partition ID
484 * Clear Snoop -- no parameters, SynIC message contains the partition ID
485 * All Offers Delivered -- no parameters, SynIC message contains the partition
487 * Flush Client -- no parameters, SynIC message contains the partition ID
490 /* Open Channel parameters */
491 struct vmbus_channel_open_channel
{
492 struct vmbus_channel_message_header header
;
494 /* Identifies the specific VMBus channel that is being opened. */
497 /* ID making a particular open request at a channel offer unique. */
500 /* GPADL for the channel's ring buffer. */
501 u32 ringbuffer_gpadlhandle
;
504 * Starting with win8, this field will be used to specify
505 * the target virtual processor on which to deliver the interrupt for
506 * the host to guest communication.
507 * Prior to win8, incoming channel interrupts would only
508 * be delivered on cpu 0. Setting this value to 0 would
509 * preserve the earlier behavior.
514 * The upstream ring buffer begins at offset zero in the memory
515 * described by RingBufferGpadlHandle. The downstream ring buffer
516 * follows it at this offset (in pages).
518 u32 downstream_ringbuffer_pageoffset
;
520 /* User-specific data to be passed along to the server endpoint. */
521 unsigned char userdata
[MAX_USER_DEFINED_BYTES
];
524 /* Open Channel Result parameters */
525 struct vmbus_channel_open_result
{
526 struct vmbus_channel_message_header header
;
532 /* Close channel parameters; */
533 struct vmbus_channel_close_channel
{
534 struct vmbus_channel_message_header header
;
538 /* Channel Message GPADL */
539 #define GPADL_TYPE_RING_BUFFER 1
540 #define GPADL_TYPE_SERVER_SAVE_AREA 2
541 #define GPADL_TYPE_TRANSACTION 8
544 * The number of PFNs in a GPADL message is defined by the number of
545 * pages that would be spanned by ByteCount and ByteOffset. If the
546 * implied number of PFNs won't fit in this packet, there will be a
547 * follow-up packet that contains more.
549 struct vmbus_channel_gpadl_header
{
550 struct vmbus_channel_message_header header
;
555 struct gpa_range range
[0];
558 /* This is the followup packet that contains more PFNs. */
559 struct vmbus_channel_gpadl_body
{
560 struct vmbus_channel_message_header header
;
566 struct vmbus_channel_gpadl_created
{
567 struct vmbus_channel_message_header header
;
573 struct vmbus_channel_gpadl_teardown
{
574 struct vmbus_channel_message_header header
;
579 struct vmbus_channel_gpadl_torndown
{
580 struct vmbus_channel_message_header header
;
584 struct vmbus_channel_relid_released
{
585 struct vmbus_channel_message_header header
;
589 struct vmbus_channel_initiate_contact
{
590 struct vmbus_channel_message_header header
;
591 u32 vmbus_version_requested
;
592 u32 target_vcpu
; /* The VCPU the host should respond to */
598 /* Hyper-V socket: guest's connect()-ing to host */
599 struct vmbus_channel_tl_connect_request
{
600 struct vmbus_channel_message_header header
;
601 uuid_le guest_endpoint_id
;
602 uuid_le host_service_id
;
605 struct vmbus_channel_version_response
{
606 struct vmbus_channel_message_header header
;
607 u8 version_supported
;
610 enum vmbus_channel_state
{
612 CHANNEL_OPENING_STATE
,
614 CHANNEL_OPENED_STATE
,
618 * Represents each channel msg on the vmbus connection This is a
619 * variable-size data structure depending on the msg type itself
621 struct vmbus_channel_msginfo
{
622 /* Bookkeeping stuff */
623 struct list_head msglistentry
;
625 /* So far, this is only used to handle gpadl body message */
626 struct list_head submsglist
;
628 /* Synchronize the request/response if needed */
629 struct completion waitevent
;
631 struct vmbus_channel_version_supported version_supported
;
632 struct vmbus_channel_open_result open_result
;
633 struct vmbus_channel_gpadl_torndown gpadl_torndown
;
634 struct vmbus_channel_gpadl_created gpadl_created
;
635 struct vmbus_channel_version_response version_response
;
640 * The channel message that goes out on the "wire".
641 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
643 unsigned char msg
[0];
646 struct vmbus_close_msg
{
647 struct vmbus_channel_msginfo info
;
648 struct vmbus_channel_close_channel msg
;
651 /* Define connection identifier type. */
652 union hv_connection_id
{
660 /* Definition of the hv_signal_event hypercall input structure. */
661 struct hv_input_signal_event
{
662 union hv_connection_id connectionid
;
667 struct hv_input_signal_event_buffer
{
669 struct hv_input_signal_event event
;
672 enum hv_signal_policy
{
673 HV_SIGNAL_POLICY_DEFAULT
= 0,
674 HV_SIGNAL_POLICY_EXPLICIT
,
677 enum hv_numa_policy
{
682 enum vmbus_device_type
{
702 struct vmbus_device
{
708 struct vmbus_channel
{
709 /* Unique channel id */
712 struct list_head listentry
;
714 struct hv_device
*device_obj
;
716 enum vmbus_channel_state state
;
718 struct vmbus_channel_offer_channel offermsg
;
720 * These are based on the OfferMsg.MonitorId.
721 * Save it here for easy access.
726 bool rescind
; /* got rescind msg */
728 u32 ringbuffer_gpadlhandle
;
730 /* Allocated memory for ring buffer */
731 void *ringbuffer_pages
;
732 u32 ringbuffer_pagecount
;
733 struct hv_ring_buffer_info outbound
; /* send to parent */
734 struct hv_ring_buffer_info inbound
; /* receive from parent */
735 spinlock_t inbound_lock
;
737 struct vmbus_close_msg close_msg
;
739 /* Channel callback are invoked in this workqueue context */
740 /* HANDLE dataWorkQueue; */
742 void (*onchannel_callback
)(void *context
);
743 void *channel_callback_context
;
746 * A channel can be marked for efficient (batched)
748 * If batched_reading is set to "true", we read until the
749 * channel is empty and hold off interrupts from the host
750 * during the entire read process.
751 * If batched_reading is set to "false", the client is not
752 * going to perform batched reading.
754 * By default we will enable batched reading; specific
755 * drivers that don't want this behavior can turn it off.
758 bool batched_reading
;
760 bool is_dedicated_interrupt
;
761 struct hv_input_signal_event_buffer sig_buf
;
762 struct hv_input_signal_event
*sig_event
;
765 * Starting with win8, this field will be used to specify
766 * the target virtual processor on which to deliver the interrupt for
767 * the host to guest communication.
768 * Prior to win8, incoming channel interrupts would only
769 * be delivered on cpu 0. Setting this value to 0 would
770 * preserve the earlier behavior.
773 /* The corresponding CPUID in the guest */
776 * State to manage the CPU affiliation of channels.
778 struct cpumask alloced_cpus_in_node
;
781 * Support for sub-channels. For high performance devices,
782 * it will be useful to have multiple sub-channels to support
783 * a scalable communication infrastructure with the host.
784 * The support for sub-channels is implemented as an extention
785 * to the current infrastructure.
786 * The initial offer is considered the primary channel and this
787 * offer message will indicate if the host supports sub-channels.
788 * The guest is free to ask for sub-channels to be offerred and can
789 * open these sub-channels as a normal "primary" channel. However,
790 * all sub-channels will have the same type and instance guids as the
791 * primary channel. Requests sent on a given channel will result in a
792 * response on the same channel.
796 * Sub-channel creation callback. This callback will be called in
797 * process context when a sub-channel offer is received from the host.
798 * The guest can open the sub-channel in the context of this callback.
800 void (*sc_creation_callback
)(struct vmbus_channel
*new_sc
);
803 * Channel rescind callback. Some channels (the hvsock ones), need to
804 * register a callback which is invoked in vmbus_onoffer_rescind().
806 void (*chn_rescind_callback
)(struct vmbus_channel
*channel
);
809 * The spinlock to protect the structure. It is being used to protect
810 * test-and-set access to various attributes of the structure as well
811 * as all sc_list operations.
815 * All Sub-channels of a primary channel are linked here.
817 struct list_head sc_list
;
819 * Current number of sub-channels.
823 * Number of a sub-channel (position within sc_list) which is supposed
824 * to be used as the next outgoing channel.
828 * The primary channel this sub-channel belongs to.
829 * This will be NULL for the primary channel.
831 struct vmbus_channel
*primary_channel
;
833 * Support per-channel state for use by vmbus drivers.
835 void *per_channel_state
;
837 * To support per-cpu lookup mapping of relid to channel,
838 * link up channels based on their CPU affinity.
840 struct list_head percpu_list
;
842 * Host signaling policy: The default policy will be
843 * based on the ring buffer state. We will also support
844 * a policy where the client driver can have explicit
847 enum hv_signal_policy signal_policy
;
849 * On the channel send side, many of the VMBUS
850 * device drivers explicity serialize access to the
851 * outgoing ring buffer. Give more control to the
852 * VMBUS device drivers in terms how to serialize
853 * accesss to the outgoing ring buffer.
854 * The default behavior will be to aquire the
855 * ring lock to preserve the current behavior.
857 bool acquire_ring_lock
;
859 * For performance critical channels (storage, networking
860 * etc,), Hyper-V has a mechanism to enhance the throughput
861 * at the expense of latency:
862 * When the host is to be signaled, we just set a bit in a shared page
863 * and this bit will be inspected by the hypervisor within a certain
864 * window and if the bit is set, the host will be signaled. The window
865 * of time is the monitor latency - currently around 100 usecs. This
866 * mechanism improves throughput by:
868 * A) Making the host more efficient - each time it wakes up,
869 * potentially it will process morev number of packets. The
870 * monitor latency allows a batch to build up.
871 * B) By deferring the hypercall to signal, we will also minimize
874 * Clearly, these optimizations improve throughput at the expense of
875 * latency. Furthermore, since the channel is shared for both
876 * control and data messages, control messages currently suffer
877 * unnecessary latency adversley impacting performance and boot
878 * time. To fix this issue, permit tagging the channel as being
879 * in "low latency" mode. In this mode, we will bypass the monitor
885 * NUMA distribution policy:
886 * We support teo policies:
887 * 1) Balanced: Here all performance critical channels are
888 * distributed evenly amongst all the NUMA nodes.
889 * This policy will be the default policy.
890 * 2) Localized: All channels of a given instance of a
891 * performance critical service will be assigned CPUs
892 * within a selected NUMA node.
894 enum hv_numa_policy affinity_policy
;
898 static inline void set_channel_lock_state(struct vmbus_channel
*c
, bool state
)
900 c
->acquire_ring_lock
= state
;
903 static inline bool is_hvsock_channel(const struct vmbus_channel
*c
)
905 return !!(c
->offermsg
.offer
.chn_flags
&
906 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
);
909 static inline void set_channel_signal_state(struct vmbus_channel
*c
,
910 enum hv_signal_policy policy
)
912 c
->signal_policy
= policy
;
915 static inline void set_channel_affinity_state(struct vmbus_channel
*c
,
916 enum hv_numa_policy policy
)
918 c
->affinity_policy
= policy
;
921 static inline void set_channel_read_state(struct vmbus_channel
*c
, bool state
)
923 c
->batched_reading
= state
;
926 static inline void set_per_channel_state(struct vmbus_channel
*c
, void *s
)
928 c
->per_channel_state
= s
;
931 static inline void *get_per_channel_state(struct vmbus_channel
*c
)
933 return c
->per_channel_state
;
936 static inline void set_channel_pending_send_size(struct vmbus_channel
*c
,
939 c
->outbound
.ring_buffer
->pending_send_sz
= size
;
942 static inline void set_low_latency_mode(struct vmbus_channel
*c
)
944 c
->low_latency
= true;
947 static inline void clear_low_latency_mode(struct vmbus_channel
*c
)
949 c
->low_latency
= false;
952 void vmbus_onmessage(void *context
);
954 int vmbus_request_offers(void);
957 * APIs for managing sub-channels.
960 void vmbus_set_sc_create_callback(struct vmbus_channel
*primary_channel
,
961 void (*sc_cr_cb
)(struct vmbus_channel
*new_sc
));
963 void vmbus_set_chn_rescind_callback(struct vmbus_channel
*channel
,
964 void (*chn_rescind_cb
)(struct vmbus_channel
*));
967 * Retrieve the (sub) channel on which to send an outgoing request.
968 * When a primary channel has multiple sub-channels, we choose a
969 * channel whose VCPU binding is closest to the VCPU on which
970 * this call is being made.
972 struct vmbus_channel
*vmbus_get_outgoing_channel(struct vmbus_channel
*primary
);
975 * Check if sub-channels have already been offerred. This API will be useful
976 * when the driver is unloaded after establishing sub-channels. In this case,
977 * when the driver is re-loaded, the driver would have to check if the
978 * subchannels have already been established before attempting to request
979 * the creation of sub-channels.
980 * This function returns TRUE to indicate that subchannels have already been
982 * This function should be invoked after setting the callback function for
983 * sub-channel creation.
985 bool vmbus_are_subchannels_present(struct vmbus_channel
*primary
);
987 /* The format must be the same as struct vmdata_gpa_direct */
988 struct vmbus_channel_packet_page_buffer
{
996 struct hv_page_buffer range
[MAX_PAGE_BUFFER_COUNT
];
999 /* The format must be the same as struct vmdata_gpa_direct */
1000 struct vmbus_channel_packet_multipage_buffer
{
1007 u32 rangecount
; /* Always 1 in this case */
1008 struct hv_multipage_buffer range
;
1011 /* The format must be the same as struct vmdata_gpa_direct */
1012 struct vmbus_packet_mpb_array
{
1019 u32 rangecount
; /* Always 1 in this case */
1020 struct hv_mpb_array range
;
1024 extern int vmbus_open(struct vmbus_channel
*channel
,
1025 u32 send_ringbuffersize
,
1026 u32 recv_ringbuffersize
,
1029 void(*onchannel_callback
)(void *context
),
1032 extern void vmbus_close(struct vmbus_channel
*channel
);
1034 extern int vmbus_sendpacket(struct vmbus_channel
*channel
,
1038 enum vmbus_packet_type type
,
1041 extern int vmbus_sendpacket_ctl(struct vmbus_channel
*channel
,
1045 enum vmbus_packet_type type
,
1049 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel
*channel
,
1050 struct hv_page_buffer pagebuffers
[],
1056 extern int vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel
*channel
,
1057 struct hv_page_buffer pagebuffers
[],
1065 extern int vmbus_sendpacket_multipagebuffer(struct vmbus_channel
*channel
,
1066 struct hv_multipage_buffer
*mpb
,
1071 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel
*channel
,
1072 struct vmbus_packet_mpb_array
*mpb
,
1078 extern int vmbus_establish_gpadl(struct vmbus_channel
*channel
,
1083 extern int vmbus_teardown_gpadl(struct vmbus_channel
*channel
,
1086 extern int vmbus_recvpacket(struct vmbus_channel
*channel
,
1089 u32
*buffer_actual_len
,
1092 extern int vmbus_recvpacket_raw(struct vmbus_channel
*channel
,
1095 u32
*buffer_actual_len
,
1099 extern void vmbus_ontimer(unsigned long data
);
1101 /* Base driver object */
1106 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1107 * channel flag, actually doesn't mean a synthetic device because the
1108 * offer's if_type/if_instance can change for every new hvsock
1111 * However, to facilitate the notification of new-offer/rescind-offer
1112 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1113 * a special vmbus device, and hence we need the below flag to
1114 * indicate if the driver is the hvsock driver or not: we need to
1115 * specially treat the hvosck offer & driver in vmbus_match().
1119 /* the device type supported by this driver */
1121 const struct hv_vmbus_device_id
*id_table
;
1123 struct device_driver driver
;
1125 int (*probe
)(struct hv_device
*, const struct hv_vmbus_device_id
*);
1126 int (*remove
)(struct hv_device
*);
1127 void (*shutdown
)(struct hv_device
*);
1131 /* Base device object */
1133 /* the device type id of this device */
1136 /* the device instance id of this device */
1137 uuid_le dev_instance
;
1141 struct device device
;
1143 struct vmbus_channel
*channel
;
1147 static inline struct hv_device
*device_to_hv_device(struct device
*d
)
1149 return container_of(d
, struct hv_device
, device
);
1152 static inline struct hv_driver
*drv_to_hv_drv(struct device_driver
*d
)
1154 return container_of(d
, struct hv_driver
, driver
);
1157 static inline void hv_set_drvdata(struct hv_device
*dev
, void *data
)
1159 dev_set_drvdata(&dev
->device
, data
);
1162 static inline void *hv_get_drvdata(struct hv_device
*dev
)
1164 return dev_get_drvdata(&dev
->device
);
1167 /* Vmbus interface */
1168 #define vmbus_driver_register(driver) \
1169 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1170 int __must_check
__vmbus_driver_register(struct hv_driver
*hv_driver
,
1171 struct module
*owner
,
1172 const char *mod_name
);
1173 void vmbus_driver_unregister(struct hv_driver
*hv_driver
);
1175 static inline const char *vmbus_dev_name(const struct hv_device
*device_obj
)
1177 const struct kobject
*kobj
= &device_obj
->device
.kobj
;
1182 void vmbus_hvsock_device_unregister(struct vmbus_channel
*channel
);
1184 int vmbus_allocate_mmio(struct resource
**new, struct hv_device
*device_obj
,
1185 resource_size_t min
, resource_size_t max
,
1186 resource_size_t size
, resource_size_t align
,
1187 bool fb_overlap_ok
);
1188 void vmbus_free_mmio(resource_size_t start
, resource_size_t size
);
1189 int vmbus_cpu_number_to_vp_number(int cpu_number
);
1190 u64
hv_do_hypercall(u64 control
, void *input
, void *output
);
1193 * GUID definitions of various offer types - services offered to the guest.
1198 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1200 #define HV_NIC_GUID \
1201 .guid = UUID_LE(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1202 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1206 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1208 #define HV_IDE_GUID \
1209 .guid = UUID_LE(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1210 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1214 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1216 #define HV_SCSI_GUID \
1217 .guid = UUID_LE(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1218 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1222 * {0e0b6031-5213-4934-818b-38d90ced39db}
1224 #define HV_SHUTDOWN_GUID \
1225 .guid = UUID_LE(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1226 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1230 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1232 #define HV_TS_GUID \
1233 .guid = UUID_LE(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1234 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1238 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1240 #define HV_HEART_BEAT_GUID \
1241 .guid = UUID_LE(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1242 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1246 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1248 #define HV_KVP_GUID \
1249 .guid = UUID_LE(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1250 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1253 * Dynamic memory GUID
1254 * {525074dc-8985-46e2-8057-a307dc18a502}
1256 #define HV_DM_GUID \
1257 .guid = UUID_LE(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1258 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1262 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1264 #define HV_MOUSE_GUID \
1265 .guid = UUID_LE(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1266 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1270 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1272 #define HV_KBD_GUID \
1273 .guid = UUID_LE(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1274 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1277 * VSS (Backup/Restore) GUID
1279 #define HV_VSS_GUID \
1280 .guid = UUID_LE(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1281 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1283 * Synthetic Video GUID
1284 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1286 #define HV_SYNTHVID_GUID \
1287 .guid = UUID_LE(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1288 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1292 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1294 #define HV_SYNTHFC_GUID \
1295 .guid = UUID_LE(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1296 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1299 * Guest File Copy Service
1300 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1303 #define HV_FCOPY_GUID \
1304 .guid = UUID_LE(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1305 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1308 * NetworkDirect. This is the guest RDMA service.
1309 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1311 #define HV_ND_GUID \
1312 .guid = UUID_LE(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1313 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1316 * PCI Express Pass Through
1317 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1320 #define HV_PCIE_GUID \
1321 .guid = UUID_LE(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1322 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1325 * Common header for Hyper-V ICs
1328 #define ICMSGTYPE_NEGOTIATE 0
1329 #define ICMSGTYPE_HEARTBEAT 1
1330 #define ICMSGTYPE_KVPEXCHANGE 2
1331 #define ICMSGTYPE_SHUTDOWN 3
1332 #define ICMSGTYPE_TIMESYNC 4
1333 #define ICMSGTYPE_VSS 5
1335 #define ICMSGHDRFLAG_TRANSACTION 1
1336 #define ICMSGHDRFLAG_REQUEST 2
1337 #define ICMSGHDRFLAG_RESPONSE 4
1341 * While we want to handle util services as regular devices,
1342 * there is only one instance of each of these services; so
1343 * we statically allocate the service specific state.
1346 struct hv_util_service
{
1349 void (*util_cb
)(void *);
1350 int (*util_init
)(struct hv_util_service
*);
1351 void (*util_deinit
)(void);
1354 struct vmbuspipe_hdr
{
1365 struct ic_version icverframe
;
1367 struct ic_version icvermsg
;
1370 u8 ictransaction_id
;
1375 struct icmsg_negotiate
{
1379 struct ic_version icversion_data
[1]; /* any size array */
1382 struct shutdown_msg_data
{
1384 u32 timeout_seconds
;
1386 u8 display_message
[2048];
1389 struct heartbeat_msg_data
{
1394 /* Time Sync IC defs */
1395 #define ICTIMESYNCFLAG_PROBE 0
1396 #define ICTIMESYNCFLAG_SYNC 1
1397 #define ICTIMESYNCFLAG_SAMPLE 2
1400 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1402 #define WLTIMEDELTA 116444736000000000LL
1405 struct ictimesync_data
{
1412 struct hyperv_service_callback
{
1416 struct vmbus_channel
*channel
;
1417 void (*callback
) (void *context
);
1420 #define MAX_SRV_VER 0x7ffffff
1421 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr
*,
1422 struct icmsg_negotiate
*, u8
*, int,
1425 void hv_event_tasklet_disable(struct vmbus_channel
*channel
);
1426 void hv_event_tasklet_enable(struct vmbus_channel
*channel
);
1428 void hv_process_channel_removal(struct vmbus_channel
*channel
, u32 relid
);
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(struct hv_ring_buffer_info
*ring_info
)
1444 return (void *)ring_info
->ring_buffer
->buffer
;
1448 * To optimize the flow management on the send-side,
1449 * when the sender is blocked because of lack of
1450 * sufficient space in the ring buffer, potential the
1451 * consumer of the ring buffer can signal the producer.
1452 * This is controlled by the following parameters:
1454 * 1. pending_send_sz: This is the size in bytes that the
1455 * producer is trying to send.
1456 * 2. The feature bit feat_pending_send_sz set to indicate if
1457 * the consumer of the ring will signal when the ring
1458 * state transitions from being full to a state where
1459 * there is room for the producer to send the pending packet.
1462 static inline bool hv_need_to_signal_on_read(struct hv_ring_buffer_info
*rbi
)
1468 * Issue a full memory barrier before making the signaling decision.
1469 * Here is the reason for having this barrier:
1470 * If the reading of the pend_sz (in this function)
1471 * were to be reordered and read before we commit the new read
1472 * index (in the calling function) we could
1473 * have a problem. If the host were to set the pending_sz after we
1474 * have sampled pending_sz and go to sleep before we commit the
1475 * read index, we could miss sending the interrupt. Issue a full
1476 * memory barrier to address this.
1480 pending_sz
= READ_ONCE(rbi
->ring_buffer
->pending_send_sz
);
1481 /* If the other end is not blocked on write don't bother. */
1482 if (pending_sz
== 0)
1485 cur_write_sz
= hv_get_bytes_to_write(rbi
);
1487 if (cur_write_sz
>= pending_sz
)
1494 * An API to support in-place processing of incoming VMBUS packets.
1496 #define VMBUS_PKT_TRAILER 8
1498 static inline struct vmpacket_descriptor
*
1499 get_next_pkt_raw(struct vmbus_channel
*channel
)
1501 struct hv_ring_buffer_info
*ring_info
= &channel
->inbound
;
1502 u32 priv_read_loc
= ring_info
->priv_read_index
;
1503 void *ring_buffer
= hv_get_ring_buffer(ring_info
);
1504 u32 dsize
= ring_info
->ring_datasize
;
1506 * delta is the difference between what is available to read and
1507 * what was already consumed in place. We commit read index after
1508 * the whole batch is processed.
1510 u32 delta
= priv_read_loc
>= ring_info
->ring_buffer
->read_index
?
1511 priv_read_loc
- ring_info
->ring_buffer
->read_index
:
1512 (dsize
- ring_info
->ring_buffer
->read_index
) + priv_read_loc
;
1513 u32 bytes_avail_toread
= (hv_get_bytes_to_read(ring_info
) - delta
);
1515 if (bytes_avail_toread
< sizeof(struct vmpacket_descriptor
))
1518 return ring_buffer
+ priv_read_loc
;
1522 * A helper function to step through packets "in-place"
1523 * This API is to be called after each successful call
1524 * get_next_pkt_raw().
1526 static inline void put_pkt_raw(struct vmbus_channel
*channel
,
1527 struct vmpacket_descriptor
*desc
)
1529 struct hv_ring_buffer_info
*ring_info
= &channel
->inbound
;
1530 u32 packetlen
= desc
->len8
<< 3;
1531 u32 dsize
= ring_info
->ring_datasize
;
1534 * Include the packet trailer.
1536 ring_info
->priv_read_index
+= packetlen
+ VMBUS_PKT_TRAILER
;
1537 ring_info
->priv_read_index
%= dsize
;
1541 * This call commits the read index and potentially signals the host.
1542 * Here is the pattern for using the "in-place" consumption APIs:
1544 * while (get_next_pkt_raw() {
1545 * process the packet "in-place";
1548 * if (packets processed in place)
1549 * commit_rd_index();
1551 static inline void commit_rd_index(struct vmbus_channel
*channel
)
1553 struct hv_ring_buffer_info
*ring_info
= &channel
->inbound
;
1555 * Make sure all reads are done before we update the read index since
1556 * the writer may start writing to the read area once the read index
1560 ring_info
->ring_buffer
->read_index
= ring_info
->priv_read_index
;
1562 if (hv_need_to_signal_on_read(ring_info
))
1563 vmbus_set_event(channel
);
1567 #endif /* _HYPERV_H */