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>
30 #include <linux/types.h>
31 #include <linux/scatterlist.h>
32 #include <linux/list.h>
33 #include <linux/timer.h>
34 #include <linux/completion.h>
35 #include <linux/device.h>
36 #include <linux/mod_devicetable.h>
37 #include <linux/interrupt.h>
38 #include <linux/reciprocal_div.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 * WS2012/Win8 and later versions of Hyper-V implement interrupt
93 * driven flow management. The feature bit feat_pending_send_sz
94 * is set by the host on the host->guest ring buffer, and by the
95 * guest on the guest->host ring buffer.
97 * The meaning of the feature bit is a bit complex in that it has
98 * semantics that apply to both ring buffers. If the guest sets
99 * the feature bit in the guest->host ring buffer, the guest is
100 * telling the host that:
101 * 1) It will set the pending_send_sz field in the guest->host ring
102 * buffer when it is waiting for space to become available, and
103 * 2) It will read the pending_send_sz field in the host->guest
104 * ring buffer and interrupt the host when it frees enough space
106 * Similarly, if the host sets the feature bit in the host->guest
107 * ring buffer, the host is telling the guest that:
108 * 1) It will set the pending_send_sz field in the host->guest ring
109 * buffer when it is waiting for space to become available, and
110 * 2) It will read the pending_send_sz field in the guest->host
111 * ring buffer and interrupt the guest when it frees enough space
113 * If either the guest or host does not set the feature bit that it
114 * owns, that guest or host must do polling if it encounters a full
115 * ring buffer, and not signal the other end with an interrupt.
121 u32 feat_pending_send_sz
:1;
126 /* Pad it to PAGE_SIZE so that data starts on page boundary */
130 * Ring data starts here + RingDataStartOffset
131 * !!! DO NOT place any fields below this !!!
136 struct hv_ring_buffer_info
{
137 struct hv_ring_buffer
*ring_buffer
;
138 u32 ring_size
; /* Include the shared header */
139 struct reciprocal_value ring_size_div10_reciprocal
;
140 spinlock_t ring_lock
;
142 u32 ring_datasize
; /* < ring_size */
147 static inline u32
hv_get_bytes_to_read(const struct hv_ring_buffer_info
*rbi
)
149 u32 read_loc
, write_loc
, dsize
, read
;
151 dsize
= rbi
->ring_datasize
;
152 read_loc
= rbi
->ring_buffer
->read_index
;
153 write_loc
= READ_ONCE(rbi
->ring_buffer
->write_index
);
155 read
= write_loc
>= read_loc
? (write_loc
- read_loc
) :
156 (dsize
- read_loc
) + write_loc
;
161 static inline u32
hv_get_bytes_to_write(const struct hv_ring_buffer_info
*rbi
)
163 u32 read_loc
, write_loc
, dsize
, write
;
165 dsize
= rbi
->ring_datasize
;
166 read_loc
= READ_ONCE(rbi
->ring_buffer
->read_index
);
167 write_loc
= rbi
->ring_buffer
->write_index
;
169 write
= write_loc
>= read_loc
? dsize
- (write_loc
- read_loc
) :
170 read_loc
- write_loc
;
174 static inline u32
hv_get_avail_to_write_percent(
175 const struct hv_ring_buffer_info
*rbi
)
177 u32 avail_write
= hv_get_bytes_to_write(rbi
);
179 return reciprocal_divide(
180 (avail_write
<< 3) + (avail_write
<< 1),
181 rbi
->ring_size_div10_reciprocal
);
185 * VMBUS version is 32 bit entity broken up into
186 * two 16 bit quantities: major_number. minor_number.
188 * 0 . 13 (Windows Server 2008)
191 * 3 . 0 (Windows 8 R2)
193 * 5 . 0 (Newer Windows 10)
196 #define VERSION_WS2008 ((0 << 16) | (13))
197 #define VERSION_WIN7 ((1 << 16) | (1))
198 #define VERSION_WIN8 ((2 << 16) | (4))
199 #define VERSION_WIN8_1 ((3 << 16) | (0))
200 #define VERSION_WIN10 ((4 << 16) | (0))
201 #define VERSION_WIN10_V5 ((5 << 16) | (0))
203 #define VERSION_INVAL -1
205 #define VERSION_CURRENT VERSION_WIN10_V5
207 /* Make maximum size of pipe payload of 16K */
208 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
210 /* Define PipeMode values. */
211 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
212 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
214 /* The size of the user defined data buffer for non-pipe offers. */
215 #define MAX_USER_DEFINED_BYTES 120
217 /* The size of the user defined data buffer for pipe offers. */
218 #define MAX_PIPE_USER_DEFINED_BYTES 116
221 * At the center of the Channel Management library is the Channel Offer. This
222 * struct contains the fundamental information about an offer.
224 struct vmbus_channel_offer
{
229 * These two fields are not currently used.
235 u16 mmio_megabytes
; /* in bytes * 1024 * 1024 */
238 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
240 unsigned char user_def
[MAX_USER_DEFINED_BYTES
];
245 * The following sructure is an integrated pipe protocol, which
246 * is implemented on top of standard user-defined data. Pipe
247 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
252 unsigned char user_def
[MAX_PIPE_USER_DEFINED_BYTES
];
256 * The sub_channel_index is defined in win8.
258 u16 sub_channel_index
;
263 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
264 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
265 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
266 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
267 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
268 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
269 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
270 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
272 struct vmpacket_descriptor
{
280 struct vmpacket_header
{
281 u32 prev_pkt_start_offset
;
282 struct vmpacket_descriptor descriptor
;
285 struct vmtransfer_page_range
{
290 struct vmtransfer_page_packet_header
{
291 struct vmpacket_descriptor d
;
296 struct vmtransfer_page_range ranges
[1];
299 struct vmgpadl_packet_header
{
300 struct vmpacket_descriptor d
;
305 struct vmadd_remove_transfer_page_set
{
306 struct vmpacket_descriptor d
;
313 * This structure defines a range in guest physical space that can be made to
314 * look virtually contiguous.
323 * This is the format for an Establish Gpadl packet, which contains a handle by
324 * which this GPADL will be known and a set of GPA ranges associated with it.
325 * This can be converted to a MDL by the guest OS. If there are multiple GPA
326 * ranges, then the resulting MDL will be "chained," representing multiple VA
329 struct vmestablish_gpadl
{
330 struct vmpacket_descriptor d
;
333 struct gpa_range range
[1];
337 * This is the format for a Teardown Gpadl packet, which indicates that the
338 * GPADL handle in the Establish Gpadl packet will never be referenced again.
340 struct vmteardown_gpadl
{
341 struct vmpacket_descriptor d
;
343 u32 reserved
; /* for alignment to a 8-byte boundary */
347 * This is the format for a GPA-Direct packet, which contains a set of GPA
348 * ranges, in addition to commands and/or data.
350 struct vmdata_gpa_direct
{
351 struct vmpacket_descriptor d
;
354 struct gpa_range range
[1];
357 /* This is the format for a Additional Data Packet. */
358 struct vmadditional_data
{
359 struct vmpacket_descriptor d
;
363 unsigned char data
[1];
366 union vmpacket_largest_possible_header
{
367 struct vmpacket_descriptor simple_hdr
;
368 struct vmtransfer_page_packet_header xfer_page_hdr
;
369 struct vmgpadl_packet_header gpadl_hdr
;
370 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr
;
371 struct vmestablish_gpadl establish_gpadl_hdr
;
372 struct vmteardown_gpadl teardown_gpadl_hdr
;
373 struct vmdata_gpa_direct data_gpa_direct_hdr
;
376 #define VMPACKET_DATA_START_ADDRESS(__packet) \
377 (void *)(((unsigned char *)__packet) + \
378 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
380 #define VMPACKET_DATA_LENGTH(__packet) \
381 ((((struct vmpacket_descriptor)__packet)->len8 - \
382 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
384 #define VMPACKET_TRANSFER_MODE(__packet) \
385 (((struct IMPACT)__packet)->type)
387 enum vmbus_packet_type
{
388 VM_PKT_INVALID
= 0x0,
390 VM_PKT_ADD_XFER_PAGESET
= 0x2,
391 VM_PKT_RM_XFER_PAGESET
= 0x3,
392 VM_PKT_ESTABLISH_GPADL
= 0x4,
393 VM_PKT_TEARDOWN_GPADL
= 0x5,
394 VM_PKT_DATA_INBAND
= 0x6,
395 VM_PKT_DATA_USING_XFER_PAGES
= 0x7,
396 VM_PKT_DATA_USING_GPADL
= 0x8,
397 VM_PKT_DATA_USING_GPA_DIRECT
= 0x9,
398 VM_PKT_CANCEL_REQUEST
= 0xa,
400 VM_PKT_DATA_USING_ADDITIONAL_PKT
= 0xc,
401 VM_PKT_ADDITIONAL_DATA
= 0xd
404 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
407 /* Version 1 messages */
408 enum vmbus_channel_message_type
{
409 CHANNELMSG_INVALID
= 0,
410 CHANNELMSG_OFFERCHANNEL
= 1,
411 CHANNELMSG_RESCIND_CHANNELOFFER
= 2,
412 CHANNELMSG_REQUESTOFFERS
= 3,
413 CHANNELMSG_ALLOFFERS_DELIVERED
= 4,
414 CHANNELMSG_OPENCHANNEL
= 5,
415 CHANNELMSG_OPENCHANNEL_RESULT
= 6,
416 CHANNELMSG_CLOSECHANNEL
= 7,
417 CHANNELMSG_GPADL_HEADER
= 8,
418 CHANNELMSG_GPADL_BODY
= 9,
419 CHANNELMSG_GPADL_CREATED
= 10,
420 CHANNELMSG_GPADL_TEARDOWN
= 11,
421 CHANNELMSG_GPADL_TORNDOWN
= 12,
422 CHANNELMSG_RELID_RELEASED
= 13,
423 CHANNELMSG_INITIATE_CONTACT
= 14,
424 CHANNELMSG_VERSION_RESPONSE
= 15,
425 CHANNELMSG_UNLOAD
= 16,
426 CHANNELMSG_UNLOAD_RESPONSE
= 17,
430 CHANNELMSG_TL_CONNECT_REQUEST
= 21,
434 struct vmbus_channel_message_header
{
435 enum vmbus_channel_message_type msgtype
;
439 /* Query VMBus Version parameters */
440 struct vmbus_channel_query_vmbus_version
{
441 struct vmbus_channel_message_header header
;
445 /* VMBus Version Supported parameters */
446 struct vmbus_channel_version_supported
{
447 struct vmbus_channel_message_header header
;
448 u8 version_supported
;
451 /* Offer Channel parameters */
452 struct vmbus_channel_offer_channel
{
453 struct vmbus_channel_message_header header
;
454 struct vmbus_channel_offer offer
;
458 * win7 and beyond splits this field into a bit field.
460 u8 monitor_allocated
:1;
463 * These are new fields added in win7 and later.
464 * Do not access these fields without checking the
465 * negotiated protocol.
467 * If "is_dedicated_interrupt" is set, we must not set the
468 * associated bit in the channel bitmap while sending the
469 * interrupt to the host.
471 * connection_id is to be used in signaling the host.
473 u16 is_dedicated_interrupt
:1;
478 /* Rescind Offer parameters */
479 struct vmbus_channel_rescind_offer
{
480 struct vmbus_channel_message_header header
;
485 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info
*rbi
)
487 return rbi
->ring_buffer
->pending_send_sz
;
491 * Request Offer -- no parameters, SynIC message contains the partition ID
492 * Set Snoop -- no parameters, SynIC message contains the partition ID
493 * Clear Snoop -- no parameters, SynIC message contains the partition ID
494 * All Offers Delivered -- no parameters, SynIC message contains the partition
496 * Flush Client -- no parameters, SynIC message contains the partition ID
499 /* Open Channel parameters */
500 struct vmbus_channel_open_channel
{
501 struct vmbus_channel_message_header header
;
503 /* Identifies the specific VMBus channel that is being opened. */
506 /* ID making a particular open request at a channel offer unique. */
509 /* GPADL for the channel's ring buffer. */
510 u32 ringbuffer_gpadlhandle
;
513 * Starting with win8, this field will be used to specify
514 * the target virtual processor on which to deliver the interrupt for
515 * the host to guest communication.
516 * Prior to win8, incoming channel interrupts would only
517 * be delivered on cpu 0. Setting this value to 0 would
518 * preserve the earlier behavior.
523 * The upstream ring buffer begins at offset zero in the memory
524 * described by RingBufferGpadlHandle. The downstream ring buffer
525 * follows it at this offset (in pages).
527 u32 downstream_ringbuffer_pageoffset
;
529 /* User-specific data to be passed along to the server endpoint. */
530 unsigned char userdata
[MAX_USER_DEFINED_BYTES
];
533 /* Open Channel Result parameters */
534 struct vmbus_channel_open_result
{
535 struct vmbus_channel_message_header header
;
541 /* Close channel parameters; */
542 struct vmbus_channel_close_channel
{
543 struct vmbus_channel_message_header header
;
547 /* Channel Message GPADL */
548 #define GPADL_TYPE_RING_BUFFER 1
549 #define GPADL_TYPE_SERVER_SAVE_AREA 2
550 #define GPADL_TYPE_TRANSACTION 8
553 * The number of PFNs in a GPADL message is defined by the number of
554 * pages that would be spanned by ByteCount and ByteOffset. If the
555 * implied number of PFNs won't fit in this packet, there will be a
556 * follow-up packet that contains more.
558 struct vmbus_channel_gpadl_header
{
559 struct vmbus_channel_message_header header
;
564 struct gpa_range range
[0];
567 /* This is the followup packet that contains more PFNs. */
568 struct vmbus_channel_gpadl_body
{
569 struct vmbus_channel_message_header header
;
575 struct vmbus_channel_gpadl_created
{
576 struct vmbus_channel_message_header header
;
582 struct vmbus_channel_gpadl_teardown
{
583 struct vmbus_channel_message_header header
;
588 struct vmbus_channel_gpadl_torndown
{
589 struct vmbus_channel_message_header header
;
593 struct vmbus_channel_relid_released
{
594 struct vmbus_channel_message_header header
;
598 struct vmbus_channel_initiate_contact
{
599 struct vmbus_channel_message_header header
;
600 u32 vmbus_version_requested
;
601 u32 target_vcpu
; /* The VCPU the host should respond to */
614 /* Hyper-V socket: guest's connect()-ing to host */
615 struct vmbus_channel_tl_connect_request
{
616 struct vmbus_channel_message_header header
;
617 uuid_le guest_endpoint_id
;
618 uuid_le host_service_id
;
621 struct vmbus_channel_version_response
{
622 struct vmbus_channel_message_header header
;
623 u8 version_supported
;
629 * On new hosts that support VMBus protocol 5.0, we must use
630 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
631 * and for subsequent messages, we must use the Message Connection ID
632 * field in the host-returned Version Response Message.
634 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
639 enum vmbus_channel_state
{
641 CHANNEL_OPENING_STATE
,
643 CHANNEL_OPENED_STATE
,
647 * Represents each channel msg on the vmbus connection This is a
648 * variable-size data structure depending on the msg type itself
650 struct vmbus_channel_msginfo
{
651 /* Bookkeeping stuff */
652 struct list_head msglistentry
;
654 /* So far, this is only used to handle gpadl body message */
655 struct list_head submsglist
;
657 /* Synchronize the request/response if needed */
658 struct completion waitevent
;
659 struct vmbus_channel
*waiting_channel
;
661 struct vmbus_channel_version_supported version_supported
;
662 struct vmbus_channel_open_result open_result
;
663 struct vmbus_channel_gpadl_torndown gpadl_torndown
;
664 struct vmbus_channel_gpadl_created gpadl_created
;
665 struct vmbus_channel_version_response version_response
;
670 * The channel message that goes out on the "wire".
671 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
673 unsigned char msg
[0];
676 struct vmbus_close_msg
{
677 struct vmbus_channel_msginfo info
;
678 struct vmbus_channel_close_channel msg
;
681 /* Define connection identifier type. */
682 union hv_connection_id
{
690 enum hv_numa_policy
{
695 enum vmbus_device_type
{
715 struct vmbus_device
{
721 struct vmbus_channel
{
722 struct list_head listentry
;
724 struct hv_device
*device_obj
;
726 enum vmbus_channel_state state
;
728 struct vmbus_channel_offer_channel offermsg
;
730 * These are based on the OfferMsg.MonitorId.
731 * Save it here for easy access.
736 bool rescind
; /* got rescind msg */
737 struct completion rescind_event
;
739 u32 ringbuffer_gpadlhandle
;
741 /* Allocated memory for ring buffer */
742 struct page
*ringbuffer_page
;
743 u32 ringbuffer_pagecount
;
744 u32 ringbuffer_send_offset
;
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
;
751 u64 interrupts
; /* Host to Guest interrupts */
752 u64 sig_events
; /* Guest to Host events */
754 /* Channel callback's invoked in softirq context */
755 struct tasklet_struct callback_event
;
756 void (*onchannel_callback
)(void *context
);
757 void *channel_callback_context
;
760 * A channel can be marked for one of three modes of reading:
761 * BATCHED - callback called from taslket and should read
762 * channel until empty. Interrupts from the host
763 * are masked while read is in process (default).
764 * DIRECT - callback called from tasklet (softirq).
765 * ISR - callback called in interrupt context and must
766 * invoke its own deferred processing.
767 * Host interrupts are disabled and must be re-enabled
768 * when ring is empty.
770 enum hv_callback_mode
{
776 bool is_dedicated_interrupt
;
780 * Starting with win8, this field will be used to specify
781 * the target virtual processor on which to deliver the interrupt for
782 * the host to guest communication.
783 * Prior to win8, incoming channel interrupts would only
784 * be delivered on cpu 0. Setting this value to 0 would
785 * preserve the earlier behavior.
788 /* The corresponding CPUID in the guest */
791 * State to manage the CPU affiliation of channels.
793 struct cpumask alloced_cpus_in_node
;
796 * Support for sub-channels. For high performance devices,
797 * it will be useful to have multiple sub-channels to support
798 * a scalable communication infrastructure with the host.
799 * The support for sub-channels is implemented as an extention
800 * to the current infrastructure.
801 * The initial offer is considered the primary channel and this
802 * offer message will indicate if the host supports sub-channels.
803 * The guest is free to ask for sub-channels to be offerred and can
804 * open these sub-channels as a normal "primary" channel. However,
805 * all sub-channels will have the same type and instance guids as the
806 * primary channel. Requests sent on a given channel will result in a
807 * response on the same channel.
811 * Sub-channel creation callback. This callback will be called in
812 * process context when a sub-channel offer is received from the host.
813 * The guest can open the sub-channel in the context of this callback.
815 void (*sc_creation_callback
)(struct vmbus_channel
*new_sc
);
818 * Channel rescind callback. Some channels (the hvsock ones), need to
819 * register a callback which is invoked in vmbus_onoffer_rescind().
821 void (*chn_rescind_callback
)(struct vmbus_channel
*channel
);
824 * The spinlock to protect the structure. It is being used to protect
825 * test-and-set access to various attributes of the structure as well
826 * as all sc_list operations.
830 * All Sub-channels of a primary channel are linked here.
832 struct list_head sc_list
;
834 * Current number of sub-channels.
838 * Number of a sub-channel (position within sc_list) which is supposed
839 * to be used as the next outgoing channel.
843 * The primary channel this sub-channel belongs to.
844 * This will be NULL for the primary channel.
846 struct vmbus_channel
*primary_channel
;
848 * Support per-channel state for use by vmbus drivers.
850 void *per_channel_state
;
852 * To support per-cpu lookup mapping of relid to channel,
853 * link up channels based on their CPU affinity.
855 struct list_head percpu_list
;
858 * Defer freeing channel until after all cpu's have
859 * gone through grace period.
864 * For sysfs per-channel properties.
869 * For performance critical channels (storage, networking
870 * etc,), Hyper-V has a mechanism to enhance the throughput
871 * at the expense of latency:
872 * When the host is to be signaled, we just set a bit in a shared page
873 * and this bit will be inspected by the hypervisor within a certain
874 * window and if the bit is set, the host will be signaled. The window
875 * of time is the monitor latency - currently around 100 usecs. This
876 * mechanism improves throughput by:
878 * A) Making the host more efficient - each time it wakes up,
879 * potentially it will process morev number of packets. The
880 * monitor latency allows a batch to build up.
881 * B) By deferring the hypercall to signal, we will also minimize
884 * Clearly, these optimizations improve throughput at the expense of
885 * latency. Furthermore, since the channel is shared for both
886 * control and data messages, control messages currently suffer
887 * unnecessary latency adversley impacting performance and boot
888 * time. To fix this issue, permit tagging the channel as being
889 * in "low latency" mode. In this mode, we will bypass the monitor
895 * NUMA distribution policy:
896 * We support two policies:
897 * 1) Balanced: Here all performance critical channels are
898 * distributed evenly amongst all the NUMA nodes.
899 * This policy will be the default policy.
900 * 2) Localized: All channels of a given instance of a
901 * performance critical service will be assigned CPUs
902 * within a selected NUMA node.
904 enum hv_numa_policy affinity_policy
;
910 static inline bool is_hvsock_channel(const struct vmbus_channel
*c
)
912 return !!(c
->offermsg
.offer
.chn_flags
&
913 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
);
916 static inline void set_channel_affinity_state(struct vmbus_channel
*c
,
917 enum hv_numa_policy policy
)
919 c
->affinity_policy
= policy
;
922 static inline void set_channel_read_mode(struct vmbus_channel
*c
,
923 enum hv_callback_mode mode
)
925 c
->callback_mode
= mode
;
928 static inline void set_per_channel_state(struct vmbus_channel
*c
, void *s
)
930 c
->per_channel_state
= s
;
933 static inline void *get_per_channel_state(struct vmbus_channel
*c
)
935 return c
->per_channel_state
;
938 static inline void set_channel_pending_send_size(struct vmbus_channel
*c
,
941 c
->outbound
.ring_buffer
->pending_send_sz
= size
;
944 static inline void set_low_latency_mode(struct vmbus_channel
*c
)
946 c
->low_latency
= true;
949 static inline void clear_low_latency_mode(struct vmbus_channel
*c
)
951 c
->low_latency
= false;
954 void vmbus_onmessage(void *context
);
956 int vmbus_request_offers(void);
959 * APIs for managing sub-channels.
962 void vmbus_set_sc_create_callback(struct vmbus_channel
*primary_channel
,
963 void (*sc_cr_cb
)(struct vmbus_channel
*new_sc
));
965 void vmbus_set_chn_rescind_callback(struct vmbus_channel
*channel
,
966 void (*chn_rescind_cb
)(struct vmbus_channel
*));
969 * Retrieve the (sub) channel on which to send an outgoing request.
970 * When a primary channel has multiple sub-channels, we choose a
971 * channel whose VCPU binding is closest to the VCPU on which
972 * this call is being made.
974 struct vmbus_channel
*vmbus_get_outgoing_channel(struct vmbus_channel
*primary
);
977 * Check if sub-channels have already been offerred. This API will be useful
978 * when the driver is unloaded after establishing sub-channels. In this case,
979 * when the driver is re-loaded, the driver would have to check if the
980 * subchannels have already been established before attempting to request
981 * the creation of sub-channels.
982 * This function returns TRUE to indicate that subchannels have already been
984 * This function should be invoked after setting the callback function for
985 * sub-channel creation.
987 bool vmbus_are_subchannels_present(struct vmbus_channel
*primary
);
989 /* The format must be the same as struct vmdata_gpa_direct */
990 struct vmbus_channel_packet_page_buffer
{
998 struct hv_page_buffer range
[MAX_PAGE_BUFFER_COUNT
];
1001 /* The format must be the same as struct vmdata_gpa_direct */
1002 struct vmbus_channel_packet_multipage_buffer
{
1009 u32 rangecount
; /* Always 1 in this case */
1010 struct hv_multipage_buffer range
;
1013 /* The format must be the same as struct vmdata_gpa_direct */
1014 struct vmbus_packet_mpb_array
{
1021 u32 rangecount
; /* Always 1 in this case */
1022 struct hv_mpb_array range
;
1025 int vmbus_alloc_ring(struct vmbus_channel
*channel
,
1026 u32 send_size
, u32 recv_size
);
1027 void vmbus_free_ring(struct vmbus_channel
*channel
);
1029 int vmbus_connect_ring(struct vmbus_channel
*channel
,
1030 void (*onchannel_callback
)(void *context
),
1032 int vmbus_disconnect_ring(struct vmbus_channel
*channel
);
1034 extern int vmbus_open(struct vmbus_channel
*channel
,
1035 u32 send_ringbuffersize
,
1036 u32 recv_ringbuffersize
,
1039 void (*onchannel_callback
)(void *context
),
1042 extern void vmbus_close(struct vmbus_channel
*channel
);
1044 extern int vmbus_sendpacket(struct vmbus_channel
*channel
,
1048 enum vmbus_packet_type type
,
1051 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel
*channel
,
1052 struct hv_page_buffer pagebuffers
[],
1058 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel
*channel
,
1059 struct vmbus_packet_mpb_array
*mpb
,
1065 extern int vmbus_establish_gpadl(struct vmbus_channel
*channel
,
1070 extern int vmbus_teardown_gpadl(struct vmbus_channel
*channel
,
1073 void vmbus_reset_channel_cb(struct vmbus_channel
*channel
);
1075 extern int vmbus_recvpacket(struct vmbus_channel
*channel
,
1078 u32
*buffer_actual_len
,
1081 extern int vmbus_recvpacket_raw(struct vmbus_channel
*channel
,
1084 u32
*buffer_actual_len
,
1088 extern void vmbus_ontimer(unsigned long data
);
1090 /* Base driver object */
1095 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1096 * channel flag, actually doesn't mean a synthetic device because the
1097 * offer's if_type/if_instance can change for every new hvsock
1100 * However, to facilitate the notification of new-offer/rescind-offer
1101 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1102 * a special vmbus device, and hence we need the below flag to
1103 * indicate if the driver is the hvsock driver or not: we need to
1104 * specially treat the hvosck offer & driver in vmbus_match().
1108 /* the device type supported by this driver */
1110 const struct hv_vmbus_device_id
*id_table
;
1112 struct device_driver driver
;
1114 /* dynamic device GUID's */
1117 struct list_head list
;
1120 int (*probe
)(struct hv_device
*, const struct hv_vmbus_device_id
*);
1121 int (*remove
)(struct hv_device
*);
1122 void (*shutdown
)(struct hv_device
*);
1126 /* Base device object */
1128 /* the device type id of this device */
1131 /* the device instance id of this device */
1132 uuid_le dev_instance
;
1136 struct device device
;
1137 char *driver_override
; /* Driver name to force a match */
1139 struct vmbus_channel
*channel
;
1140 struct kset
*channels_kset
;
1144 static inline struct hv_device
*device_to_hv_device(struct device
*d
)
1146 return container_of(d
, struct hv_device
, device
);
1149 static inline struct hv_driver
*drv_to_hv_drv(struct device_driver
*d
)
1151 return container_of(d
, struct hv_driver
, driver
);
1154 static inline void hv_set_drvdata(struct hv_device
*dev
, void *data
)
1156 dev_set_drvdata(&dev
->device
, data
);
1159 static inline void *hv_get_drvdata(struct hv_device
*dev
)
1161 return dev_get_drvdata(&dev
->device
);
1164 struct hv_ring_buffer_debug_info
{
1165 u32 current_interrupt_mask
;
1166 u32 current_read_index
;
1167 u32 current_write_index
;
1168 u32 bytes_avail_toread
;
1169 u32 bytes_avail_towrite
;
1172 void hv_ringbuffer_get_debuginfo(const struct hv_ring_buffer_info
*ring_info
,
1173 struct hv_ring_buffer_debug_info
*debug_info
);
1175 /* Vmbus interface */
1176 #define vmbus_driver_register(driver) \
1177 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1178 int __must_check
__vmbus_driver_register(struct hv_driver
*hv_driver
,
1179 struct module
*owner
,
1180 const char *mod_name
);
1181 void vmbus_driver_unregister(struct hv_driver
*hv_driver
);
1183 void vmbus_hvsock_device_unregister(struct vmbus_channel
*channel
);
1185 int vmbus_allocate_mmio(struct resource
**new, struct hv_device
*device_obj
,
1186 resource_size_t min
, resource_size_t max
,
1187 resource_size_t size
, resource_size_t align
,
1188 bool fb_overlap_ok
);
1189 void vmbus_free_mmio(resource_size_t start
, resource_size_t size
);
1192 * GUID definitions of various offer types - services offered to the guest.
1197 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1199 #define HV_NIC_GUID \
1200 .guid = UUID_LE(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1201 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1205 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1207 #define HV_IDE_GUID \
1208 .guid = UUID_LE(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1209 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1213 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1215 #define HV_SCSI_GUID \
1216 .guid = UUID_LE(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1217 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1221 * {0e0b6031-5213-4934-818b-38d90ced39db}
1223 #define HV_SHUTDOWN_GUID \
1224 .guid = UUID_LE(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1225 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1229 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1231 #define HV_TS_GUID \
1232 .guid = UUID_LE(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1233 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1237 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1239 #define HV_HEART_BEAT_GUID \
1240 .guid = UUID_LE(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1241 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1245 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1247 #define HV_KVP_GUID \
1248 .guid = UUID_LE(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1249 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1252 * Dynamic memory GUID
1253 * {525074dc-8985-46e2-8057-a307dc18a502}
1255 #define HV_DM_GUID \
1256 .guid = UUID_LE(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1257 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1261 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1263 #define HV_MOUSE_GUID \
1264 .guid = UUID_LE(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1265 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1269 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1271 #define HV_KBD_GUID \
1272 .guid = UUID_LE(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1273 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1276 * VSS (Backup/Restore) GUID
1278 #define HV_VSS_GUID \
1279 .guid = UUID_LE(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1280 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1282 * Synthetic Video GUID
1283 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1285 #define HV_SYNTHVID_GUID \
1286 .guid = UUID_LE(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1287 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1291 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1293 #define HV_SYNTHFC_GUID \
1294 .guid = UUID_LE(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1295 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1298 * Guest File Copy Service
1299 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1302 #define HV_FCOPY_GUID \
1303 .guid = UUID_LE(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1304 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1307 * NetworkDirect. This is the guest RDMA service.
1308 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1310 #define HV_ND_GUID \
1311 .guid = UUID_LE(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1312 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1315 * PCI Express Pass Through
1316 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1319 #define HV_PCIE_GUID \
1320 .guid = UUID_LE(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1321 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1324 * Linux doesn't support the 3 devices: the first two are for
1325 * Automatic Virtual Machine Activation, and the third is for
1326 * Remote Desktop Virtualization.
1327 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1328 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1329 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1332 #define HV_AVMA1_GUID \
1333 .guid = UUID_LE(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1334 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1336 #define HV_AVMA2_GUID \
1337 .guid = UUID_LE(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1338 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1340 #define HV_RDV_GUID \
1341 .guid = UUID_LE(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1342 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1345 * Common header for Hyper-V ICs
1348 #define ICMSGTYPE_NEGOTIATE 0
1349 #define ICMSGTYPE_HEARTBEAT 1
1350 #define ICMSGTYPE_KVPEXCHANGE 2
1351 #define ICMSGTYPE_SHUTDOWN 3
1352 #define ICMSGTYPE_TIMESYNC 4
1353 #define ICMSGTYPE_VSS 5
1355 #define ICMSGHDRFLAG_TRANSACTION 1
1356 #define ICMSGHDRFLAG_REQUEST 2
1357 #define ICMSGHDRFLAG_RESPONSE 4
1361 * While we want to handle util services as regular devices,
1362 * there is only one instance of each of these services; so
1363 * we statically allocate the service specific state.
1366 struct hv_util_service
{
1369 void (*util_cb
)(void *);
1370 int (*util_init
)(struct hv_util_service
*);
1371 void (*util_deinit
)(void);
1374 struct vmbuspipe_hdr
{
1385 struct ic_version icverframe
;
1387 struct ic_version icvermsg
;
1390 u8 ictransaction_id
;
1395 struct icmsg_negotiate
{
1399 struct ic_version icversion_data
[1]; /* any size array */
1402 struct shutdown_msg_data
{
1404 u32 timeout_seconds
;
1406 u8 display_message
[2048];
1409 struct heartbeat_msg_data
{
1414 /* Time Sync IC defs */
1415 #define ICTIMESYNCFLAG_PROBE 0
1416 #define ICTIMESYNCFLAG_SYNC 1
1417 #define ICTIMESYNCFLAG_SAMPLE 2
1420 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1422 #define WLTIMEDELTA 116444736000000000LL
1425 struct ictimesync_data
{
1432 struct ictimesync_ref_data
{
1434 u64 vmreferencetime
;
1441 struct hyperv_service_callback
{
1445 struct vmbus_channel
*channel
;
1446 void (*callback
)(void *context
);
1449 #define MAX_SRV_VER 0x7ffffff
1450 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr
*icmsghdrp
, u8
*buf
,
1451 const int *fw_version
, int fw_vercnt
,
1452 const int *srv_version
, int srv_vercnt
,
1453 int *nego_fw_version
, int *nego_srv_version
);
1455 void hv_process_channel_removal(struct vmbus_channel
*channel
);
1457 void vmbus_setevent(struct vmbus_channel
*channel
);
1459 * Negotiated version with the Host.
1462 extern __u32 vmbus_proto_version
;
1464 int vmbus_send_tl_connect_request(const uuid_le
*shv_guest_servie_id
,
1465 const uuid_le
*shv_host_servie_id
);
1466 void vmbus_set_event(struct vmbus_channel
*channel
);
1468 /* Get the start of the ring buffer. */
1469 static inline void *
1470 hv_get_ring_buffer(const struct hv_ring_buffer_info
*ring_info
)
1472 return ring_info
->ring_buffer
->buffer
;
1476 * Mask off host interrupt callback notifications
1478 static inline void hv_begin_read(struct hv_ring_buffer_info
*rbi
)
1480 rbi
->ring_buffer
->interrupt_mask
= 1;
1482 /* make sure mask update is not reordered */
1487 * Re-enable host callback and return number of outstanding bytes
1489 static inline u32
hv_end_read(struct hv_ring_buffer_info
*rbi
)
1492 rbi
->ring_buffer
->interrupt_mask
= 0;
1494 /* make sure mask update is not reordered */
1498 * Now check to see if the ring buffer is still empty.
1499 * If it is not, we raced and we need to process new
1500 * incoming messages.
1502 return hv_get_bytes_to_read(rbi
);
1506 * An API to support in-place processing of incoming VMBUS packets.
1509 /* Get data payload associated with descriptor */
1510 static inline void *hv_pkt_data(const struct vmpacket_descriptor
*desc
)
1512 return (void *)((unsigned long)desc
+ (desc
->offset8
<< 3));
1515 /* Get data size associated with descriptor */
1516 static inline u32
hv_pkt_datalen(const struct vmpacket_descriptor
*desc
)
1518 return (desc
->len8
<< 3) - (desc
->offset8
<< 3);
1522 struct vmpacket_descriptor
*
1523 hv_pkt_iter_first(struct vmbus_channel
*channel
);
1525 struct vmpacket_descriptor
*
1526 __hv_pkt_iter_next(struct vmbus_channel
*channel
,
1527 const struct vmpacket_descriptor
*pkt
);
1529 void hv_pkt_iter_close(struct vmbus_channel
*channel
);
1532 * Get next packet descriptor from iterator
1533 * If at end of list, return NULL and update host.
1535 static inline struct vmpacket_descriptor
*
1536 hv_pkt_iter_next(struct vmbus_channel
*channel
,
1537 const struct vmpacket_descriptor
*pkt
)
1539 struct vmpacket_descriptor
*nxt
;
1541 nxt
= __hv_pkt_iter_next(channel
, pkt
);
1543 hv_pkt_iter_close(channel
);
1548 #define foreach_vmbus_pkt(pkt, channel) \
1549 for (pkt = hv_pkt_iter_first(channel); pkt; \
1550 pkt = hv_pkt_iter_next(channel, pkt))
1552 #endif /* _HYPERV_H */