3 * Copyright (c) 2011, Microsoft Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
16 * Place - Suite 330, Boston, MA 02111-1307 USA.
19 * Haiyang Zhang <haiyangz@microsoft.com>
20 * Hank Janssen <hjanssen@microsoft.com>
21 * K. Y. Srinivasan <kys@microsoft.com>
28 #include <uapi/linux/hyperv.h>
29 #include <uapi/asm/hyperv.h>
31 #include <linux/types.h>
32 #include <linux/scatterlist.h>
33 #include <linux/list.h>
34 #include <linux/timer.h>
35 #include <linux/completion.h>
36 #include <linux/device.h>
37 #include <linux/mod_devicetable.h>
38 #include <linux/interrupt.h>
40 #define MAX_PAGE_BUFFER_COUNT 32
41 #define MAX_MULTIPAGE_BUFFER_COUNT 32 /* 128K */
45 /* Single-page buffer */
46 struct hv_page_buffer {
52 /* Multiple-page buffer */
53 struct hv_multipage_buffer {
54 /* Length and Offset determines the # of pfns in the array */
57 u64 pfn_array[MAX_MULTIPAGE_BUFFER_COUNT];
61 * Multiple-page buffer array; the pfn array is variable size:
62 * The number of entries in the PFN array is determined by
66 /* Length and Offset determines the # of pfns in the array */
72 /* 0x18 includes the proprietary packet header */
73 #define MAX_PAGE_BUFFER_PACKET (0x18 + \
74 (sizeof(struct hv_page_buffer) * \
75 MAX_PAGE_BUFFER_COUNT))
76 #define MAX_MULTIPAGE_BUFFER_PACKET (0x18 + \
77 sizeof(struct hv_multipage_buffer))
82 struct hv_ring_buffer {
83 /* Offset in bytes from the start of ring data below */
86 /* Offset in bytes from the start of ring data below */
92 * Win8 uses some of the reserved bits to implement
93 * interrupt driven flow management. On the send side
94 * we can request that the receiver interrupt the sender
95 * when the ring transitions from being full to being able
96 * to handle a message of size "pending_send_sz".
98 * Add necessary state for this enhancement.
106 u32 feat_pending_send_sz:1;
111 /* Pad it to PAGE_SIZE so that data starts on page boundary */
115 * Ring data starts here + RingDataStartOffset
116 * !!! DO NOT place any fields below this !!!
121 struct hv_ring_buffer_info {
122 struct hv_ring_buffer *ring_buffer;
123 u32 ring_size; /* Include the shared header */
124 spinlock_t ring_lock;
126 u32 ring_datasize; /* < ring_size */
127 u32 ring_data_startoffset;
128 u32 priv_write_index;
130 u32 cached_read_index;
135 * hv_get_ringbuffer_availbytes()
137 * Get number of bytes available to read and to write to
138 * for the specified ring buffer
141 hv_get_ringbuffer_availbytes(const struct hv_ring_buffer_info *rbi,
142 u32 *read, u32 *write)
144 u32 read_loc, write_loc, dsize;
146 /* Capture the read/write indices before they changed */
147 read_loc = rbi->ring_buffer->read_index;
148 write_loc = rbi->ring_buffer->write_index;
149 dsize = rbi->ring_datasize;
151 *write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
152 read_loc - write_loc;
153 *read = dsize - *write;
156 static inline u32 hv_get_bytes_to_read(const struct hv_ring_buffer_info *rbi)
158 u32 read_loc, write_loc, dsize, read;
160 dsize = rbi->ring_datasize;
161 read_loc = rbi->ring_buffer->read_index;
162 write_loc = READ_ONCE(rbi->ring_buffer->write_index);
164 read = write_loc >= read_loc ? (write_loc - read_loc) :
165 (dsize - read_loc) + write_loc;
170 static inline u32 hv_get_bytes_to_write(const struct hv_ring_buffer_info *rbi)
172 u32 read_loc, write_loc, dsize, write;
174 dsize = rbi->ring_datasize;
175 read_loc = READ_ONCE(rbi->ring_buffer->read_index);
176 write_loc = rbi->ring_buffer->write_index;
178 write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
179 read_loc - write_loc;
183 static inline u32 hv_get_cached_bytes_to_write(
184 const struct hv_ring_buffer_info *rbi)
186 u32 read_loc, write_loc, dsize, write;
188 dsize = rbi->ring_datasize;
189 read_loc = rbi->cached_read_index;
190 write_loc = rbi->ring_buffer->write_index;
192 write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
193 read_loc - write_loc;
197 * VMBUS version is 32 bit entity broken up into
198 * two 16 bit quantities: major_number. minor_number.
200 * 0 . 13 (Windows Server 2008)
203 * 3 . 0 (Windows 8 R2)
207 #define VERSION_WS2008 ((0 << 16) | (13))
208 #define VERSION_WIN7 ((1 << 16) | (1))
209 #define VERSION_WIN8 ((2 << 16) | (4))
210 #define VERSION_WIN8_1 ((3 << 16) | (0))
211 #define VERSION_WIN10 ((4 << 16) | (0))
213 #define VERSION_INVAL -1
215 #define VERSION_CURRENT VERSION_WIN10
217 /* Make maximum size of pipe payload of 16K */
218 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
220 /* Define PipeMode values. */
221 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
222 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
224 /* The size of the user defined data buffer for non-pipe offers. */
225 #define MAX_USER_DEFINED_BYTES 120
227 /* The size of the user defined data buffer for pipe offers. */
228 #define MAX_PIPE_USER_DEFINED_BYTES 116
231 * At the center of the Channel Management library is the Channel Offer. This
232 * struct contains the fundamental information about an offer.
234 struct vmbus_channel_offer {
239 * These two fields are not currently used.
245 u16 mmio_megabytes; /* in bytes * 1024 * 1024 */
248 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
250 unsigned char user_def[MAX_USER_DEFINED_BYTES];
255 * The following sructure is an integrated pipe protocol, which
256 * is implemented on top of standard user-defined data. Pipe
257 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
262 unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
266 * The sub_channel_index is defined in win8.
268 u16 sub_channel_index;
273 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
274 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
275 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
276 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
277 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
278 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
279 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
280 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
282 struct vmpacket_descriptor {
290 struct vmpacket_header {
291 u32 prev_pkt_start_offset;
292 struct vmpacket_descriptor descriptor;
295 struct vmtransfer_page_range {
300 struct vmtransfer_page_packet_header {
301 struct vmpacket_descriptor d;
306 struct vmtransfer_page_range ranges[1];
309 struct vmgpadl_packet_header {
310 struct vmpacket_descriptor d;
315 struct vmadd_remove_transfer_page_set {
316 struct vmpacket_descriptor d;
323 * This structure defines a range in guest physical space that can be made to
324 * look virtually contiguous.
333 * This is the format for an Establish Gpadl packet, which contains a handle by
334 * which this GPADL will be known and a set of GPA ranges associated with it.
335 * This can be converted to a MDL by the guest OS. If there are multiple GPA
336 * ranges, then the resulting MDL will be "chained," representing multiple VA
339 struct vmestablish_gpadl {
340 struct vmpacket_descriptor d;
343 struct gpa_range range[1];
347 * This is the format for a Teardown Gpadl packet, which indicates that the
348 * GPADL handle in the Establish Gpadl packet will never be referenced again.
350 struct vmteardown_gpadl {
351 struct vmpacket_descriptor d;
353 u32 reserved; /* for alignment to a 8-byte boundary */
357 * This is the format for a GPA-Direct packet, which contains a set of GPA
358 * ranges, in addition to commands and/or data.
360 struct vmdata_gpa_direct {
361 struct vmpacket_descriptor d;
364 struct gpa_range range[1];
367 /* This is the format for a Additional Data Packet. */
368 struct vmadditional_data {
369 struct vmpacket_descriptor d;
373 unsigned char data[1];
376 union vmpacket_largest_possible_header {
377 struct vmpacket_descriptor simple_hdr;
378 struct vmtransfer_page_packet_header xfer_page_hdr;
379 struct vmgpadl_packet_header gpadl_hdr;
380 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr;
381 struct vmestablish_gpadl establish_gpadl_hdr;
382 struct vmteardown_gpadl teardown_gpadl_hdr;
383 struct vmdata_gpa_direct data_gpa_direct_hdr;
386 #define VMPACKET_DATA_START_ADDRESS(__packet) \
387 (void *)(((unsigned char *)__packet) + \
388 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
390 #define VMPACKET_DATA_LENGTH(__packet) \
391 ((((struct vmpacket_descriptor)__packet)->len8 - \
392 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
394 #define VMPACKET_TRANSFER_MODE(__packet) \
395 (((struct IMPACT)__packet)->type)
397 enum vmbus_packet_type {
398 VM_PKT_INVALID = 0x0,
400 VM_PKT_ADD_XFER_PAGESET = 0x2,
401 VM_PKT_RM_XFER_PAGESET = 0x3,
402 VM_PKT_ESTABLISH_GPADL = 0x4,
403 VM_PKT_TEARDOWN_GPADL = 0x5,
404 VM_PKT_DATA_INBAND = 0x6,
405 VM_PKT_DATA_USING_XFER_PAGES = 0x7,
406 VM_PKT_DATA_USING_GPADL = 0x8,
407 VM_PKT_DATA_USING_GPA_DIRECT = 0x9,
408 VM_PKT_CANCEL_REQUEST = 0xa,
410 VM_PKT_DATA_USING_ADDITIONAL_PKT = 0xc,
411 VM_PKT_ADDITIONAL_DATA = 0xd
414 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
417 /* Version 1 messages */
418 enum vmbus_channel_message_type {
419 CHANNELMSG_INVALID = 0,
420 CHANNELMSG_OFFERCHANNEL = 1,
421 CHANNELMSG_RESCIND_CHANNELOFFER = 2,
422 CHANNELMSG_REQUESTOFFERS = 3,
423 CHANNELMSG_ALLOFFERS_DELIVERED = 4,
424 CHANNELMSG_OPENCHANNEL = 5,
425 CHANNELMSG_OPENCHANNEL_RESULT = 6,
426 CHANNELMSG_CLOSECHANNEL = 7,
427 CHANNELMSG_GPADL_HEADER = 8,
428 CHANNELMSG_GPADL_BODY = 9,
429 CHANNELMSG_GPADL_CREATED = 10,
430 CHANNELMSG_GPADL_TEARDOWN = 11,
431 CHANNELMSG_GPADL_TORNDOWN = 12,
432 CHANNELMSG_RELID_RELEASED = 13,
433 CHANNELMSG_INITIATE_CONTACT = 14,
434 CHANNELMSG_VERSION_RESPONSE = 15,
435 CHANNELMSG_UNLOAD = 16,
436 CHANNELMSG_UNLOAD_RESPONSE = 17,
440 CHANNELMSG_TL_CONNECT_REQUEST = 21,
444 struct vmbus_channel_message_header {
445 enum vmbus_channel_message_type msgtype;
449 /* Query VMBus Version parameters */
450 struct vmbus_channel_query_vmbus_version {
451 struct vmbus_channel_message_header header;
455 /* VMBus Version Supported parameters */
456 struct vmbus_channel_version_supported {
457 struct vmbus_channel_message_header header;
458 u8 version_supported;
461 /* Offer Channel parameters */
462 struct vmbus_channel_offer_channel {
463 struct vmbus_channel_message_header header;
464 struct vmbus_channel_offer offer;
468 * win7 and beyond splits this field into a bit field.
470 u8 monitor_allocated:1;
473 * These are new fields added in win7 and later.
474 * Do not access these fields without checking the
475 * negotiated protocol.
477 * If "is_dedicated_interrupt" is set, we must not set the
478 * associated bit in the channel bitmap while sending the
479 * interrupt to the host.
481 * connection_id is to be used in signaling the host.
483 u16 is_dedicated_interrupt:1;
488 /* Rescind Offer parameters */
489 struct vmbus_channel_rescind_offer {
490 struct vmbus_channel_message_header header;
495 * Request Offer -- no parameters, SynIC message contains the partition ID
496 * Set Snoop -- no parameters, SynIC message contains the partition ID
497 * Clear Snoop -- no parameters, SynIC message contains the partition ID
498 * All Offers Delivered -- no parameters, SynIC message contains the partition
500 * Flush Client -- no parameters, SynIC message contains the partition ID
503 /* Open Channel parameters */
504 struct vmbus_channel_open_channel {
505 struct vmbus_channel_message_header header;
507 /* Identifies the specific VMBus channel that is being opened. */
510 /* ID making a particular open request at a channel offer unique. */
513 /* GPADL for the channel's ring buffer. */
514 u32 ringbuffer_gpadlhandle;
517 * Starting with win8, this field will be used to specify
518 * the target virtual processor on which to deliver the interrupt for
519 * the host to guest communication.
520 * Prior to win8, incoming channel interrupts would only
521 * be delivered on cpu 0. Setting this value to 0 would
522 * preserve the earlier behavior.
527 * The upstream ring buffer begins at offset zero in the memory
528 * described by RingBufferGpadlHandle. The downstream ring buffer
529 * follows it at this offset (in pages).
531 u32 downstream_ringbuffer_pageoffset;
533 /* User-specific data to be passed along to the server endpoint. */
534 unsigned char userdata[MAX_USER_DEFINED_BYTES];
537 /* Open Channel Result parameters */
538 struct vmbus_channel_open_result {
539 struct vmbus_channel_message_header header;
545 /* Close channel parameters; */
546 struct vmbus_channel_close_channel {
547 struct vmbus_channel_message_header header;
551 /* Channel Message GPADL */
552 #define GPADL_TYPE_RING_BUFFER 1
553 #define GPADL_TYPE_SERVER_SAVE_AREA 2
554 #define GPADL_TYPE_TRANSACTION 8
557 * The number of PFNs in a GPADL message is defined by the number of
558 * pages that would be spanned by ByteCount and ByteOffset. If the
559 * implied number of PFNs won't fit in this packet, there will be a
560 * follow-up packet that contains more.
562 struct vmbus_channel_gpadl_header {
563 struct vmbus_channel_message_header header;
568 struct gpa_range range[0];
571 /* This is the followup packet that contains more PFNs. */
572 struct vmbus_channel_gpadl_body {
573 struct vmbus_channel_message_header header;
579 struct vmbus_channel_gpadl_created {
580 struct vmbus_channel_message_header header;
586 struct vmbus_channel_gpadl_teardown {
587 struct vmbus_channel_message_header header;
592 struct vmbus_channel_gpadl_torndown {
593 struct vmbus_channel_message_header header;
597 struct vmbus_channel_relid_released {
598 struct vmbus_channel_message_header header;
602 struct vmbus_channel_initiate_contact {
603 struct vmbus_channel_message_header header;
604 u32 vmbus_version_requested;
605 u32 target_vcpu; /* The VCPU the host should respond to */
611 /* Hyper-V socket: guest's connect()-ing to host */
612 struct vmbus_channel_tl_connect_request {
613 struct vmbus_channel_message_header header;
614 uuid_le guest_endpoint_id;
615 uuid_le host_service_id;
618 struct vmbus_channel_version_response {
619 struct vmbus_channel_message_header header;
620 u8 version_supported;
623 enum vmbus_channel_state {
625 CHANNEL_OPENING_STATE,
627 CHANNEL_OPENED_STATE,
631 * Represents each channel msg on the vmbus connection This is a
632 * variable-size data structure depending on the msg type itself
634 struct vmbus_channel_msginfo {
635 /* Bookkeeping stuff */
636 struct list_head msglistentry;
638 /* So far, this is only used to handle gpadl body message */
639 struct list_head submsglist;
641 /* Synchronize the request/response if needed */
642 struct completion waitevent;
643 struct vmbus_channel *waiting_channel;
645 struct vmbus_channel_version_supported version_supported;
646 struct vmbus_channel_open_result open_result;
647 struct vmbus_channel_gpadl_torndown gpadl_torndown;
648 struct vmbus_channel_gpadl_created gpadl_created;
649 struct vmbus_channel_version_response version_response;
654 * The channel message that goes out on the "wire".
655 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
657 unsigned char msg[0];
660 struct vmbus_close_msg {
661 struct vmbus_channel_msginfo info;
662 struct vmbus_channel_close_channel msg;
665 /* Define connection identifier type. */
666 union hv_connection_id {
674 /* Definition of the hv_signal_event hypercall input structure. */
675 struct hv_input_signal_event {
676 union hv_connection_id connectionid;
681 struct hv_input_signal_event_buffer {
683 struct hv_input_signal_event event;
686 enum hv_numa_policy {
691 enum vmbus_device_type {
711 struct vmbus_device {
717 struct vmbus_channel {
718 struct list_head listentry;
720 struct hv_device *device_obj;
722 enum vmbus_channel_state state;
724 struct vmbus_channel_offer_channel offermsg;
726 * These are based on the OfferMsg.MonitorId.
727 * Save it here for easy access.
732 bool rescind; /* got rescind msg */
734 u32 ringbuffer_gpadlhandle;
736 /* Allocated memory for ring buffer */
737 void *ringbuffer_pages;
738 u32 ringbuffer_pagecount;
739 struct hv_ring_buffer_info outbound; /* send to parent */
740 struct hv_ring_buffer_info inbound; /* receive from parent */
741 spinlock_t inbound_lock;
743 struct vmbus_close_msg close_msg;
745 /* Channel callback's invoked in softirq context */
746 struct tasklet_struct callback_event;
747 void (*onchannel_callback)(void *context);
748 void *channel_callback_context;
751 * A channel can be marked for one of three modes of reading:
752 * BATCHED - callback called from taslket and should read
753 * channel until empty. Interrupts from the host
754 * are masked while read is in process (default).
755 * DIRECT - callback called from tasklet (softirq).
756 * ISR - callback called in interrupt context and must
757 * invoke its own deferred processing.
758 * Host interrupts are disabled and must be re-enabled
759 * when ring is empty.
761 enum hv_callback_mode {
767 bool is_dedicated_interrupt;
768 struct hv_input_signal_event_buffer sig_buf;
769 struct hv_input_signal_event *sig_event;
772 * Starting with win8, this field will be used to specify
773 * the target virtual processor on which to deliver the interrupt for
774 * the host to guest communication.
775 * Prior to win8, incoming channel interrupts would only
776 * be delivered on cpu 0. Setting this value to 0 would
777 * preserve the earlier behavior.
780 /* The corresponding CPUID in the guest */
783 * State to manage the CPU affiliation of channels.
785 struct cpumask alloced_cpus_in_node;
788 * Support for sub-channels. For high performance devices,
789 * it will be useful to have multiple sub-channels to support
790 * a scalable communication infrastructure with the host.
791 * The support for sub-channels is implemented as an extention
792 * to the current infrastructure.
793 * The initial offer is considered the primary channel and this
794 * offer message will indicate if the host supports sub-channels.
795 * The guest is free to ask for sub-channels to be offerred and can
796 * open these sub-channels as a normal "primary" channel. However,
797 * all sub-channels will have the same type and instance guids as the
798 * primary channel. Requests sent on a given channel will result in a
799 * response on the same channel.
803 * Sub-channel creation callback. This callback will be called in
804 * process context when a sub-channel offer is received from the host.
805 * The guest can open the sub-channel in the context of this callback.
807 void (*sc_creation_callback)(struct vmbus_channel *new_sc);
810 * Channel rescind callback. Some channels (the hvsock ones), need to
811 * register a callback which is invoked in vmbus_onoffer_rescind().
813 void (*chn_rescind_callback)(struct vmbus_channel *channel);
816 * The spinlock to protect the structure. It is being used to protect
817 * test-and-set access to various attributes of the structure as well
818 * as all sc_list operations.
822 * All Sub-channels of a primary channel are linked here.
824 struct list_head sc_list;
826 * Current number of sub-channels.
830 * Number of a sub-channel (position within sc_list) which is supposed
831 * to be used as the next outgoing channel.
835 * The primary channel this sub-channel belongs to.
836 * This will be NULL for the primary channel.
838 struct vmbus_channel *primary_channel;
840 * Support per-channel state for use by vmbus drivers.
842 void *per_channel_state;
844 * To support per-cpu lookup mapping of relid to channel,
845 * link up channels based on their CPU affinity.
847 struct list_head percpu_list;
849 * For performance critical channels (storage, networking
850 * etc,), Hyper-V has a mechanism to enhance the throughput
851 * at the expense of latency:
852 * When the host is to be signaled, we just set a bit in a shared page
853 * and this bit will be inspected by the hypervisor within a certain
854 * window and if the bit is set, the host will be signaled. The window
855 * of time is the monitor latency - currently around 100 usecs. This
856 * mechanism improves throughput by:
858 * A) Making the host more efficient - each time it wakes up,
859 * potentially it will process morev number of packets. The
860 * monitor latency allows a batch to build up.
861 * B) By deferring the hypercall to signal, we will also minimize
864 * Clearly, these optimizations improve throughput at the expense of
865 * latency. Furthermore, since the channel is shared for both
866 * control and data messages, control messages currently suffer
867 * unnecessary latency adversley impacting performance and boot
868 * time. To fix this issue, permit tagging the channel as being
869 * in "low latency" mode. In this mode, we will bypass the monitor
875 * NUMA distribution policy:
876 * We support teo policies:
877 * 1) Balanced: Here all performance critical channels are
878 * distributed evenly amongst all the NUMA nodes.
879 * This policy will be the default policy.
880 * 2) Localized: All channels of a given instance of a
881 * performance critical service will be assigned CPUs
882 * within a selected NUMA node.
884 enum hv_numa_policy affinity_policy;
888 static inline bool is_hvsock_channel(const struct vmbus_channel *c)
890 return !!(c->offermsg.offer.chn_flags &
891 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER);
894 static inline void set_channel_affinity_state(struct vmbus_channel *c,
895 enum hv_numa_policy policy)
897 c->affinity_policy = policy;
900 static inline void set_channel_read_mode(struct vmbus_channel *c,
901 enum hv_callback_mode mode)
903 c->callback_mode = mode;
906 static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
908 c->per_channel_state = s;
911 static inline void *get_per_channel_state(struct vmbus_channel *c)
913 return c->per_channel_state;
916 static inline void set_channel_pending_send_size(struct vmbus_channel *c,
919 c->outbound.ring_buffer->pending_send_sz = size;
922 static inline void set_low_latency_mode(struct vmbus_channel *c)
924 c->low_latency = true;
927 static inline void clear_low_latency_mode(struct vmbus_channel *c)
929 c->low_latency = false;
932 void vmbus_onmessage(void *context);
934 int vmbus_request_offers(void);
937 * APIs for managing sub-channels.
940 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
941 void (*sc_cr_cb)(struct vmbus_channel *new_sc));
943 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
944 void (*chn_rescind_cb)(struct vmbus_channel *));
947 * Retrieve the (sub) channel on which to send an outgoing request.
948 * When a primary channel has multiple sub-channels, we choose a
949 * channel whose VCPU binding is closest to the VCPU on which
950 * this call is being made.
952 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary);
955 * Check if sub-channels have already been offerred. This API will be useful
956 * when the driver is unloaded after establishing sub-channels. In this case,
957 * when the driver is re-loaded, the driver would have to check if the
958 * subchannels have already been established before attempting to request
959 * the creation of sub-channels.
960 * This function returns TRUE to indicate that subchannels have already been
962 * This function should be invoked after setting the callback function for
963 * sub-channel creation.
965 bool vmbus_are_subchannels_present(struct vmbus_channel *primary);
967 /* The format must be the same as struct vmdata_gpa_direct */
968 struct vmbus_channel_packet_page_buffer {
976 struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
979 /* The format must be the same as struct vmdata_gpa_direct */
980 struct vmbus_channel_packet_multipage_buffer {
987 u32 rangecount; /* Always 1 in this case */
988 struct hv_multipage_buffer range;
991 /* The format must be the same as struct vmdata_gpa_direct */
992 struct vmbus_packet_mpb_array {
999 u32 rangecount; /* Always 1 in this case */
1000 struct hv_mpb_array range;
1004 extern int vmbus_open(struct vmbus_channel *channel,
1005 u32 send_ringbuffersize,
1006 u32 recv_ringbuffersize,
1009 void(*onchannel_callback)(void *context),
1012 extern void vmbus_close(struct vmbus_channel *channel);
1014 extern int vmbus_sendpacket(struct vmbus_channel *channel,
1018 enum vmbus_packet_type type,
1021 extern int vmbus_sendpacket_ctl(struct vmbus_channel *channel,
1025 enum vmbus_packet_type type,
1028 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1029 struct hv_page_buffer pagebuffers[],
1035 extern int vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel *channel,
1036 struct hv_page_buffer pagebuffers[],
1043 extern int vmbus_sendpacket_multipagebuffer(struct vmbus_channel *channel,
1044 struct hv_multipage_buffer *mpb,
1049 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1050 struct vmbus_packet_mpb_array *mpb,
1056 extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
1061 extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
1064 extern int vmbus_recvpacket(struct vmbus_channel *channel,
1067 u32 *buffer_actual_len,
1070 extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
1073 u32 *buffer_actual_len,
1077 extern void vmbus_ontimer(unsigned long data);
1079 /* Base driver object */
1084 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1085 * channel flag, actually doesn't mean a synthetic device because the
1086 * offer's if_type/if_instance can change for every new hvsock
1089 * However, to facilitate the notification of new-offer/rescind-offer
1090 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1091 * a special vmbus device, and hence we need the below flag to
1092 * indicate if the driver is the hvsock driver or not: we need to
1093 * specially treat the hvosck offer & driver in vmbus_match().
1097 /* the device type supported by this driver */
1099 const struct hv_vmbus_device_id *id_table;
1101 struct device_driver driver;
1103 /* dynamic device GUID's */
1106 struct list_head list;
1109 int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
1110 int (*remove)(struct hv_device *);
1111 void (*shutdown)(struct hv_device *);
1115 /* Base device object */
1117 /* the device type id of this device */
1120 /* the device instance id of this device */
1121 uuid_le dev_instance;
1125 struct device device;
1127 struct vmbus_channel *channel;
1131 static inline struct hv_device *device_to_hv_device(struct device *d)
1133 return container_of(d, struct hv_device, device);
1136 static inline struct hv_driver *drv_to_hv_drv(struct device_driver *d)
1138 return container_of(d, struct hv_driver, driver);
1141 static inline void hv_set_drvdata(struct hv_device *dev, void *data)
1143 dev_set_drvdata(&dev->device, data);
1146 static inline void *hv_get_drvdata(struct hv_device *dev)
1148 return dev_get_drvdata(&dev->device);
1151 /* Vmbus interface */
1152 #define vmbus_driver_register(driver) \
1153 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1154 int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
1155 struct module *owner,
1156 const char *mod_name);
1157 void vmbus_driver_unregister(struct hv_driver *hv_driver);
1159 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel);
1161 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1162 resource_size_t min, resource_size_t max,
1163 resource_size_t size, resource_size_t align,
1164 bool fb_overlap_ok);
1165 void vmbus_free_mmio(resource_size_t start, resource_size_t size);
1166 int vmbus_cpu_number_to_vp_number(int cpu_number);
1167 u64 hv_do_hypercall(u64 control, void *input, void *output);
1170 * GUID definitions of various offer types - services offered to the guest.
1175 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1177 #define HV_NIC_GUID \
1178 .guid = UUID_LE(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1179 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1183 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1185 #define HV_IDE_GUID \
1186 .guid = UUID_LE(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1187 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1191 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1193 #define HV_SCSI_GUID \
1194 .guid = UUID_LE(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1195 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1199 * {0e0b6031-5213-4934-818b-38d90ced39db}
1201 #define HV_SHUTDOWN_GUID \
1202 .guid = UUID_LE(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1203 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1207 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1209 #define HV_TS_GUID \
1210 .guid = UUID_LE(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1211 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1215 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1217 #define HV_HEART_BEAT_GUID \
1218 .guid = UUID_LE(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1219 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1223 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1225 #define HV_KVP_GUID \
1226 .guid = UUID_LE(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1227 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1230 * Dynamic memory GUID
1231 * {525074dc-8985-46e2-8057-a307dc18a502}
1233 #define HV_DM_GUID \
1234 .guid = UUID_LE(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1235 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1239 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1241 #define HV_MOUSE_GUID \
1242 .guid = UUID_LE(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1243 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1247 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1249 #define HV_KBD_GUID \
1250 .guid = UUID_LE(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1251 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1254 * VSS (Backup/Restore) GUID
1256 #define HV_VSS_GUID \
1257 .guid = UUID_LE(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1258 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1260 * Synthetic Video GUID
1261 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1263 #define HV_SYNTHVID_GUID \
1264 .guid = UUID_LE(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1265 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1269 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1271 #define HV_SYNTHFC_GUID \
1272 .guid = UUID_LE(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1273 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1276 * Guest File Copy Service
1277 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1280 #define HV_FCOPY_GUID \
1281 .guid = UUID_LE(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1282 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1285 * NetworkDirect. This is the guest RDMA service.
1286 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1288 #define HV_ND_GUID \
1289 .guid = UUID_LE(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1290 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1293 * PCI Express Pass Through
1294 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1297 #define HV_PCIE_GUID \
1298 .guid = UUID_LE(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1299 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1302 * Linux doesn't support the 3 devices: the first two are for
1303 * Automatic Virtual Machine Activation, and the third is for
1304 * Remote Desktop Virtualization.
1305 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1306 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1307 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1310 #define HV_AVMA1_GUID \
1311 .guid = UUID_LE(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1312 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1314 #define HV_AVMA2_GUID \
1315 .guid = UUID_LE(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1316 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1318 #define HV_RDV_GUID \
1319 .guid = UUID_LE(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1320 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1323 * Common header for Hyper-V ICs
1326 #define ICMSGTYPE_NEGOTIATE 0
1327 #define ICMSGTYPE_HEARTBEAT 1
1328 #define ICMSGTYPE_KVPEXCHANGE 2
1329 #define ICMSGTYPE_SHUTDOWN 3
1330 #define ICMSGTYPE_TIMESYNC 4
1331 #define ICMSGTYPE_VSS 5
1333 #define ICMSGHDRFLAG_TRANSACTION 1
1334 #define ICMSGHDRFLAG_REQUEST 2
1335 #define ICMSGHDRFLAG_RESPONSE 4
1339 * While we want to handle util services as regular devices,
1340 * there is only one instance of each of these services; so
1341 * we statically allocate the service specific state.
1344 struct hv_util_service {
1347 void (*util_cb)(void *);
1348 int (*util_init)(struct hv_util_service *);
1349 void (*util_deinit)(void);
1352 struct vmbuspipe_hdr {
1363 struct ic_version icverframe;
1365 struct ic_version icvermsg;
1368 u8 ictransaction_id;
1373 struct icmsg_negotiate {
1377 struct ic_version icversion_data[1]; /* any size array */
1380 struct shutdown_msg_data {
1382 u32 timeout_seconds;
1384 u8 display_message[2048];
1387 struct heartbeat_msg_data {
1392 /* Time Sync IC defs */
1393 #define ICTIMESYNCFLAG_PROBE 0
1394 #define ICTIMESYNCFLAG_SYNC 1
1395 #define ICTIMESYNCFLAG_SAMPLE 2
1398 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1400 #define WLTIMEDELTA 116444736000000000LL
1403 struct ictimesync_data {
1410 struct ictimesync_ref_data {
1412 u64 vmreferencetime;
1419 struct hyperv_service_callback {
1423 struct vmbus_channel *channel;
1424 void (*callback) (void *context);
1427 #define MAX_SRV_VER 0x7ffffff
1428 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf,
1429 const int *fw_version, int fw_vercnt,
1430 const int *srv_version, int srv_vercnt,
1431 int *nego_fw_version, int *nego_srv_version);
1433 void hv_event_tasklet_disable(struct vmbus_channel *channel);
1434 void hv_event_tasklet_enable(struct vmbus_channel *channel);
1436 void hv_process_channel_removal(struct vmbus_channel *channel, u32 relid);
1438 void vmbus_setevent(struct vmbus_channel *channel);
1440 * Negotiated version with the Host.
1443 extern __u32 vmbus_proto_version;
1445 int vmbus_send_tl_connect_request(const uuid_le *shv_guest_servie_id,
1446 const uuid_le *shv_host_servie_id);
1447 void vmbus_set_event(struct vmbus_channel *channel);
1449 /* Get the start of the ring buffer. */
1450 static inline void *
1451 hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info)
1453 return ring_info->ring_buffer->buffer;
1457 * To optimize the flow management on the send-side,
1458 * when the sender is blocked because of lack of
1459 * sufficient space in the ring buffer, potential the
1460 * consumer of the ring buffer can signal the producer.
1461 * This is controlled by the following parameters:
1463 * 1. pending_send_sz: This is the size in bytes that the
1464 * producer is trying to send.
1465 * 2. The feature bit feat_pending_send_sz set to indicate if
1466 * the consumer of the ring will signal when the ring
1467 * state transitions from being full to a state where
1468 * there is room for the producer to send the pending packet.
1471 static inline void hv_signal_on_read(struct vmbus_channel *channel)
1473 u32 cur_write_sz, cached_write_sz;
1475 struct hv_ring_buffer_info *rbi = &channel->inbound;
1478 * Issue a full memory barrier before making the signaling decision.
1479 * Here is the reason for having this barrier:
1480 * If the reading of the pend_sz (in this function)
1481 * were to be reordered and read before we commit the new read
1482 * index (in the calling function) we could
1483 * have a problem. If the host were to set the pending_sz after we
1484 * have sampled pending_sz and go to sleep before we commit the
1485 * read index, we could miss sending the interrupt. Issue a full
1486 * memory barrier to address this.
1490 pending_sz = READ_ONCE(rbi->ring_buffer->pending_send_sz);
1491 /* If the other end is not blocked on write don't bother. */
1492 if (pending_sz == 0)
1495 cur_write_sz = hv_get_bytes_to_write(rbi);
1497 if (cur_write_sz < pending_sz)
1500 cached_write_sz = hv_get_cached_bytes_to_write(rbi);
1501 if (cached_write_sz < pending_sz)
1502 vmbus_setevent(channel);
1508 init_cached_read_index(struct vmbus_channel *channel)
1510 struct hv_ring_buffer_info *rbi = &channel->inbound;
1512 rbi->cached_read_index = rbi->ring_buffer->read_index;
1516 * Mask off host interrupt callback notifications
1518 static inline void hv_begin_read(struct hv_ring_buffer_info *rbi)
1520 rbi->ring_buffer->interrupt_mask = 1;
1522 /* make sure mask update is not reordered */
1527 * Re-enable host callback and return number of outstanding bytes
1529 static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi)
1532 rbi->ring_buffer->interrupt_mask = 0;
1534 /* make sure mask update is not reordered */
1538 * Now check to see if the ring buffer is still empty.
1539 * If it is not, we raced and we need to process new
1540 * incoming messages.
1542 return hv_get_bytes_to_read(rbi);
1546 * An API to support in-place processing of incoming VMBUS packets.
1548 #define VMBUS_PKT_TRAILER 8
1550 static inline struct vmpacket_descriptor *
1551 get_next_pkt_raw(struct vmbus_channel *channel)
1553 struct hv_ring_buffer_info *ring_info = &channel->inbound;
1554 u32 priv_read_loc = ring_info->priv_read_index;
1555 void *ring_buffer = hv_get_ring_buffer(ring_info);
1556 u32 dsize = ring_info->ring_datasize;
1558 * delta is the difference between what is available to read and
1559 * what was already consumed in place. We commit read index after
1560 * the whole batch is processed.
1562 u32 delta = priv_read_loc >= ring_info->ring_buffer->read_index ?
1563 priv_read_loc - ring_info->ring_buffer->read_index :
1564 (dsize - ring_info->ring_buffer->read_index) + priv_read_loc;
1565 u32 bytes_avail_toread = (hv_get_bytes_to_read(ring_info) - delta);
1567 if (bytes_avail_toread < sizeof(struct vmpacket_descriptor))
1570 return ring_buffer + priv_read_loc;
1574 * A helper function to step through packets "in-place"
1575 * This API is to be called after each successful call
1576 * get_next_pkt_raw().
1578 static inline void put_pkt_raw(struct vmbus_channel *channel,
1579 struct vmpacket_descriptor *desc)
1581 struct hv_ring_buffer_info *ring_info = &channel->inbound;
1582 u32 packetlen = desc->len8 << 3;
1583 u32 dsize = ring_info->ring_datasize;
1586 * Include the packet trailer.
1588 ring_info->priv_read_index += packetlen + VMBUS_PKT_TRAILER;
1589 ring_info->priv_read_index %= dsize;
1593 * This call commits the read index and potentially signals the host.
1594 * Here is the pattern for using the "in-place" consumption APIs:
1596 * init_cached_read_index();
1598 * while (get_next_pkt_raw() {
1599 * process the packet "in-place";
1602 * if (packets processed in place)
1603 * commit_rd_index();
1605 static inline void commit_rd_index(struct vmbus_channel *channel)
1607 struct hv_ring_buffer_info *ring_info = &channel->inbound;
1609 * Make sure all reads are done before we update the read index since
1610 * the writer may start writing to the read area once the read index
1614 ring_info->ring_buffer->read_index = ring_info->priv_read_index;
1616 hv_signal_on_read(channel);
1620 #endif /* _HYPERV_H */