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>
39 #define MAX_PAGE_BUFFER_COUNT 32
40 #define MAX_MULTIPAGE_BUFFER_COUNT 32 /* 128K */
44 /* Single-page buffer */
45 struct hv_page_buffer {
51 /* Multiple-page buffer */
52 struct hv_multipage_buffer {
53 /* Length and Offset determines the # of pfns in the array */
56 u64 pfn_array[MAX_MULTIPAGE_BUFFER_COUNT];
60 * Multiple-page buffer array; the pfn array is variable size:
61 * The number of entries in the PFN array is determined by
65 /* Length and Offset determines the # of pfns in the array */
71 /* 0x18 includes the proprietary packet header */
72 #define MAX_PAGE_BUFFER_PACKET (0x18 + \
73 (sizeof(struct hv_page_buffer) * \
74 MAX_PAGE_BUFFER_COUNT))
75 #define MAX_MULTIPAGE_BUFFER_PACKET (0x18 + \
76 sizeof(struct hv_multipage_buffer))
81 struct hv_ring_buffer {
82 /* Offset in bytes from the start of ring data below */
85 /* Offset in bytes from the start of ring data below */
91 * Win8 uses some of the reserved bits to implement
92 * interrupt driven flow management. On the send side
93 * we can request that the receiver interrupt the sender
94 * when the ring transitions from being full to being able
95 * to handle a message of size "pending_send_sz".
97 * Add necessary state for this enhancement.
105 u32 feat_pending_send_sz:1;
110 /* Pad it to PAGE_SIZE so that data starts on page boundary */
114 * Ring data starts here + RingDataStartOffset
115 * !!! DO NOT place any fields below this !!!
120 struct hv_ring_buffer_info {
121 struct hv_ring_buffer *ring_buffer;
122 u32 ring_size; /* Include the shared header */
123 spinlock_t ring_lock;
125 u32 ring_datasize; /* < ring_size */
130 static inline u32 hv_get_bytes_to_read(const struct hv_ring_buffer_info *rbi)
132 u32 read_loc, write_loc, dsize, read;
134 dsize = rbi->ring_datasize;
135 read_loc = rbi->ring_buffer->read_index;
136 write_loc = READ_ONCE(rbi->ring_buffer->write_index);
138 read = write_loc >= read_loc ? (write_loc - read_loc) :
139 (dsize - read_loc) + write_loc;
144 static inline u32 hv_get_bytes_to_write(const struct hv_ring_buffer_info *rbi)
146 u32 read_loc, write_loc, dsize, write;
148 dsize = rbi->ring_datasize;
149 read_loc = READ_ONCE(rbi->ring_buffer->read_index);
150 write_loc = rbi->ring_buffer->write_index;
152 write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
153 read_loc - write_loc;
158 * VMBUS version is 32 bit entity broken up into
159 * two 16 bit quantities: major_number. minor_number.
161 * 0 . 13 (Windows Server 2008)
164 * 3 . 0 (Windows 8 R2)
168 #define VERSION_WS2008 ((0 << 16) | (13))
169 #define VERSION_WIN7 ((1 << 16) | (1))
170 #define VERSION_WIN8 ((2 << 16) | (4))
171 #define VERSION_WIN8_1 ((3 << 16) | (0))
172 #define VERSION_WIN10 ((4 << 16) | (0))
174 #define VERSION_INVAL -1
176 #define VERSION_CURRENT VERSION_WIN10
178 /* Make maximum size of pipe payload of 16K */
179 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
181 /* Define PipeMode values. */
182 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
183 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
185 /* The size of the user defined data buffer for non-pipe offers. */
186 #define MAX_USER_DEFINED_BYTES 120
188 /* The size of the user defined data buffer for pipe offers. */
189 #define MAX_PIPE_USER_DEFINED_BYTES 116
192 * At the center of the Channel Management library is the Channel Offer. This
193 * struct contains the fundamental information about an offer.
195 struct vmbus_channel_offer {
200 * These two fields are not currently used.
206 u16 mmio_megabytes; /* in bytes * 1024 * 1024 */
209 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
211 unsigned char user_def[MAX_USER_DEFINED_BYTES];
216 * The following sructure is an integrated pipe protocol, which
217 * is implemented on top of standard user-defined data. Pipe
218 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
223 unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
227 * The sub_channel_index is defined in win8.
229 u16 sub_channel_index;
234 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
235 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
236 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
237 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
238 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
239 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
240 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
241 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
243 struct vmpacket_descriptor {
251 struct vmpacket_header {
252 u32 prev_pkt_start_offset;
253 struct vmpacket_descriptor descriptor;
256 struct vmtransfer_page_range {
261 struct vmtransfer_page_packet_header {
262 struct vmpacket_descriptor d;
267 struct vmtransfer_page_range ranges[1];
270 struct vmgpadl_packet_header {
271 struct vmpacket_descriptor d;
276 struct vmadd_remove_transfer_page_set {
277 struct vmpacket_descriptor d;
284 * This structure defines a range in guest physical space that can be made to
285 * look virtually contiguous.
294 * This is the format for an Establish Gpadl packet, which contains a handle by
295 * which this GPADL will be known and a set of GPA ranges associated with it.
296 * This can be converted to a MDL by the guest OS. If there are multiple GPA
297 * ranges, then the resulting MDL will be "chained," representing multiple VA
300 struct vmestablish_gpadl {
301 struct vmpacket_descriptor d;
304 struct gpa_range range[1];
308 * This is the format for a Teardown Gpadl packet, which indicates that the
309 * GPADL handle in the Establish Gpadl packet will never be referenced again.
311 struct vmteardown_gpadl {
312 struct vmpacket_descriptor d;
314 u32 reserved; /* for alignment to a 8-byte boundary */
318 * This is the format for a GPA-Direct packet, which contains a set of GPA
319 * ranges, in addition to commands and/or data.
321 struct vmdata_gpa_direct {
322 struct vmpacket_descriptor d;
325 struct gpa_range range[1];
328 /* This is the format for a Additional Data Packet. */
329 struct vmadditional_data {
330 struct vmpacket_descriptor d;
334 unsigned char data[1];
337 union vmpacket_largest_possible_header {
338 struct vmpacket_descriptor simple_hdr;
339 struct vmtransfer_page_packet_header xfer_page_hdr;
340 struct vmgpadl_packet_header gpadl_hdr;
341 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr;
342 struct vmestablish_gpadl establish_gpadl_hdr;
343 struct vmteardown_gpadl teardown_gpadl_hdr;
344 struct vmdata_gpa_direct data_gpa_direct_hdr;
347 #define VMPACKET_DATA_START_ADDRESS(__packet) \
348 (void *)(((unsigned char *)__packet) + \
349 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
351 #define VMPACKET_DATA_LENGTH(__packet) \
352 ((((struct vmpacket_descriptor)__packet)->len8 - \
353 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
355 #define VMPACKET_TRANSFER_MODE(__packet) \
356 (((struct IMPACT)__packet)->type)
358 enum vmbus_packet_type {
359 VM_PKT_INVALID = 0x0,
361 VM_PKT_ADD_XFER_PAGESET = 0x2,
362 VM_PKT_RM_XFER_PAGESET = 0x3,
363 VM_PKT_ESTABLISH_GPADL = 0x4,
364 VM_PKT_TEARDOWN_GPADL = 0x5,
365 VM_PKT_DATA_INBAND = 0x6,
366 VM_PKT_DATA_USING_XFER_PAGES = 0x7,
367 VM_PKT_DATA_USING_GPADL = 0x8,
368 VM_PKT_DATA_USING_GPA_DIRECT = 0x9,
369 VM_PKT_CANCEL_REQUEST = 0xa,
371 VM_PKT_DATA_USING_ADDITIONAL_PKT = 0xc,
372 VM_PKT_ADDITIONAL_DATA = 0xd
375 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
378 /* Version 1 messages */
379 enum vmbus_channel_message_type {
380 CHANNELMSG_INVALID = 0,
381 CHANNELMSG_OFFERCHANNEL = 1,
382 CHANNELMSG_RESCIND_CHANNELOFFER = 2,
383 CHANNELMSG_REQUESTOFFERS = 3,
384 CHANNELMSG_ALLOFFERS_DELIVERED = 4,
385 CHANNELMSG_OPENCHANNEL = 5,
386 CHANNELMSG_OPENCHANNEL_RESULT = 6,
387 CHANNELMSG_CLOSECHANNEL = 7,
388 CHANNELMSG_GPADL_HEADER = 8,
389 CHANNELMSG_GPADL_BODY = 9,
390 CHANNELMSG_GPADL_CREATED = 10,
391 CHANNELMSG_GPADL_TEARDOWN = 11,
392 CHANNELMSG_GPADL_TORNDOWN = 12,
393 CHANNELMSG_RELID_RELEASED = 13,
394 CHANNELMSG_INITIATE_CONTACT = 14,
395 CHANNELMSG_VERSION_RESPONSE = 15,
396 CHANNELMSG_UNLOAD = 16,
397 CHANNELMSG_UNLOAD_RESPONSE = 17,
401 CHANNELMSG_TL_CONNECT_REQUEST = 21,
405 struct vmbus_channel_message_header {
406 enum vmbus_channel_message_type msgtype;
410 /* Query VMBus Version parameters */
411 struct vmbus_channel_query_vmbus_version {
412 struct vmbus_channel_message_header header;
416 /* VMBus Version Supported parameters */
417 struct vmbus_channel_version_supported {
418 struct vmbus_channel_message_header header;
419 u8 version_supported;
422 /* Offer Channel parameters */
423 struct vmbus_channel_offer_channel {
424 struct vmbus_channel_message_header header;
425 struct vmbus_channel_offer offer;
429 * win7 and beyond splits this field into a bit field.
431 u8 monitor_allocated:1;
434 * These are new fields added in win7 and later.
435 * Do not access these fields without checking the
436 * negotiated protocol.
438 * If "is_dedicated_interrupt" is set, we must not set the
439 * associated bit in the channel bitmap while sending the
440 * interrupt to the host.
442 * connection_id is to be used in signaling the host.
444 u16 is_dedicated_interrupt:1;
449 /* Rescind Offer parameters */
450 struct vmbus_channel_rescind_offer {
451 struct vmbus_channel_message_header header;
456 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info *rbi)
458 return rbi->ring_buffer->pending_send_sz;
462 * Request Offer -- no parameters, SynIC message contains the partition ID
463 * Set Snoop -- no parameters, SynIC message contains the partition ID
464 * Clear Snoop -- no parameters, SynIC message contains the partition ID
465 * All Offers Delivered -- no parameters, SynIC message contains the partition
467 * Flush Client -- no parameters, SynIC message contains the partition ID
470 /* Open Channel parameters */
471 struct vmbus_channel_open_channel {
472 struct vmbus_channel_message_header header;
474 /* Identifies the specific VMBus channel that is being opened. */
477 /* ID making a particular open request at a channel offer unique. */
480 /* GPADL for the channel's ring buffer. */
481 u32 ringbuffer_gpadlhandle;
484 * Starting with win8, this field will be used to specify
485 * the target virtual processor on which to deliver the interrupt for
486 * the host to guest communication.
487 * Prior to win8, incoming channel interrupts would only
488 * be delivered on cpu 0. Setting this value to 0 would
489 * preserve the earlier behavior.
494 * The upstream ring buffer begins at offset zero in the memory
495 * described by RingBufferGpadlHandle. The downstream ring buffer
496 * follows it at this offset (in pages).
498 u32 downstream_ringbuffer_pageoffset;
500 /* User-specific data to be passed along to the server endpoint. */
501 unsigned char userdata[MAX_USER_DEFINED_BYTES];
504 /* Open Channel Result parameters */
505 struct vmbus_channel_open_result {
506 struct vmbus_channel_message_header header;
512 /* Close channel parameters; */
513 struct vmbus_channel_close_channel {
514 struct vmbus_channel_message_header header;
518 /* Channel Message GPADL */
519 #define GPADL_TYPE_RING_BUFFER 1
520 #define GPADL_TYPE_SERVER_SAVE_AREA 2
521 #define GPADL_TYPE_TRANSACTION 8
524 * The number of PFNs in a GPADL message is defined by the number of
525 * pages that would be spanned by ByteCount and ByteOffset. If the
526 * implied number of PFNs won't fit in this packet, there will be a
527 * follow-up packet that contains more.
529 struct vmbus_channel_gpadl_header {
530 struct vmbus_channel_message_header header;
535 struct gpa_range range[0];
538 /* This is the followup packet that contains more PFNs. */
539 struct vmbus_channel_gpadl_body {
540 struct vmbus_channel_message_header header;
546 struct vmbus_channel_gpadl_created {
547 struct vmbus_channel_message_header header;
553 struct vmbus_channel_gpadl_teardown {
554 struct vmbus_channel_message_header header;
559 struct vmbus_channel_gpadl_torndown {
560 struct vmbus_channel_message_header header;
564 struct vmbus_channel_relid_released {
565 struct vmbus_channel_message_header header;
569 struct vmbus_channel_initiate_contact {
570 struct vmbus_channel_message_header header;
571 u32 vmbus_version_requested;
572 u32 target_vcpu; /* The VCPU the host should respond to */
578 /* Hyper-V socket: guest's connect()-ing to host */
579 struct vmbus_channel_tl_connect_request {
580 struct vmbus_channel_message_header header;
581 uuid_le guest_endpoint_id;
582 uuid_le host_service_id;
585 struct vmbus_channel_version_response {
586 struct vmbus_channel_message_header header;
587 u8 version_supported;
590 enum vmbus_channel_state {
592 CHANNEL_OPENING_STATE,
594 CHANNEL_OPENED_STATE,
598 * Represents each channel msg on the vmbus connection This is a
599 * variable-size data structure depending on the msg type itself
601 struct vmbus_channel_msginfo {
602 /* Bookkeeping stuff */
603 struct list_head msglistentry;
605 /* So far, this is only used to handle gpadl body message */
606 struct list_head submsglist;
608 /* Synchronize the request/response if needed */
609 struct completion waitevent;
610 struct vmbus_channel *waiting_channel;
612 struct vmbus_channel_version_supported version_supported;
613 struct vmbus_channel_open_result open_result;
614 struct vmbus_channel_gpadl_torndown gpadl_torndown;
615 struct vmbus_channel_gpadl_created gpadl_created;
616 struct vmbus_channel_version_response version_response;
621 * The channel message that goes out on the "wire".
622 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
624 unsigned char msg[0];
627 struct vmbus_close_msg {
628 struct vmbus_channel_msginfo info;
629 struct vmbus_channel_close_channel msg;
632 /* Define connection identifier type. */
633 union hv_connection_id {
641 enum hv_numa_policy {
646 enum vmbus_device_type {
666 struct vmbus_device {
672 struct vmbus_channel {
673 struct list_head listentry;
675 struct hv_device *device_obj;
677 enum vmbus_channel_state state;
679 struct vmbus_channel_offer_channel offermsg;
681 * These are based on the OfferMsg.MonitorId.
682 * Save it here for easy access.
687 bool rescind; /* got rescind msg */
688 struct completion rescind_event;
690 u32 ringbuffer_gpadlhandle;
692 /* Allocated memory for ring buffer */
693 void *ringbuffer_pages;
694 u32 ringbuffer_pagecount;
695 struct hv_ring_buffer_info outbound; /* send to parent */
696 struct hv_ring_buffer_info inbound; /* receive from parent */
698 struct vmbus_close_msg close_msg;
701 u64 interrupts; /* Host to Guest interrupts */
702 u64 sig_events; /* Guest to Host events */
704 /* Channel callback's invoked in softirq context */
705 struct tasklet_struct callback_event;
706 void (*onchannel_callback)(void *context);
707 void *channel_callback_context;
710 * A channel can be marked for one of three modes of reading:
711 * BATCHED - callback called from taslket and should read
712 * channel until empty. Interrupts from the host
713 * are masked while read is in process (default).
714 * DIRECT - callback called from tasklet (softirq).
715 * ISR - callback called in interrupt context and must
716 * invoke its own deferred processing.
717 * Host interrupts are disabled and must be re-enabled
718 * when ring is empty.
720 enum hv_callback_mode {
726 bool is_dedicated_interrupt;
730 * Starting with win8, this field will be used to specify
731 * the target virtual processor on which to deliver the interrupt for
732 * the host to guest communication.
733 * Prior to win8, incoming channel interrupts would only
734 * be delivered on cpu 0. Setting this value to 0 would
735 * preserve the earlier behavior.
738 /* The corresponding CPUID in the guest */
741 * State to manage the CPU affiliation of channels.
743 struct cpumask alloced_cpus_in_node;
746 * Support for sub-channels. For high performance devices,
747 * it will be useful to have multiple sub-channels to support
748 * a scalable communication infrastructure with the host.
749 * The support for sub-channels is implemented as an extention
750 * to the current infrastructure.
751 * The initial offer is considered the primary channel and this
752 * offer message will indicate if the host supports sub-channels.
753 * The guest is free to ask for sub-channels to be offerred and can
754 * open these sub-channels as a normal "primary" channel. However,
755 * all sub-channels will have the same type and instance guids as the
756 * primary channel. Requests sent on a given channel will result in a
757 * response on the same channel.
761 * Sub-channel creation callback. This callback will be called in
762 * process context when a sub-channel offer is received from the host.
763 * The guest can open the sub-channel in the context of this callback.
765 void (*sc_creation_callback)(struct vmbus_channel *new_sc);
768 * Channel rescind callback. Some channels (the hvsock ones), need to
769 * register a callback which is invoked in vmbus_onoffer_rescind().
771 void (*chn_rescind_callback)(struct vmbus_channel *channel);
774 * The spinlock to protect the structure. It is being used to protect
775 * test-and-set access to various attributes of the structure as well
776 * as all sc_list operations.
780 * All Sub-channels of a primary channel are linked here.
782 struct list_head sc_list;
784 * Current number of sub-channels.
788 * Number of a sub-channel (position within sc_list) which is supposed
789 * to be used as the next outgoing channel.
793 * The primary channel this sub-channel belongs to.
794 * This will be NULL for the primary channel.
796 struct vmbus_channel *primary_channel;
798 * Support per-channel state for use by vmbus drivers.
800 void *per_channel_state;
802 * To support per-cpu lookup mapping of relid to channel,
803 * link up channels based on their CPU affinity.
805 struct list_head percpu_list;
808 * Defer freeing channel until after all cpu's have
809 * gone through grace period.
814 * For sysfs per-channel properties.
819 * For performance critical channels (storage, networking
820 * etc,), Hyper-V has a mechanism to enhance the throughput
821 * at the expense of latency:
822 * When the host is to be signaled, we just set a bit in a shared page
823 * and this bit will be inspected by the hypervisor within a certain
824 * window and if the bit is set, the host will be signaled. The window
825 * of time is the monitor latency - currently around 100 usecs. This
826 * mechanism improves throughput by:
828 * A) Making the host more efficient - each time it wakes up,
829 * potentially it will process morev number of packets. The
830 * monitor latency allows a batch to build up.
831 * B) By deferring the hypercall to signal, we will also minimize
834 * Clearly, these optimizations improve throughput at the expense of
835 * latency. Furthermore, since the channel is shared for both
836 * control and data messages, control messages currently suffer
837 * unnecessary latency adversley impacting performance and boot
838 * time. To fix this issue, permit tagging the channel as being
839 * in "low latency" mode. In this mode, we will bypass the monitor
845 * NUMA distribution policy:
846 * We support two policies:
847 * 1) Balanced: Here all performance critical channels are
848 * distributed evenly amongst all the NUMA nodes.
849 * This policy will be the default policy.
850 * 2) Localized: All channels of a given instance of a
851 * performance critical service will be assigned CPUs
852 * within a selected NUMA node.
854 enum hv_numa_policy affinity_policy;
860 static inline bool is_hvsock_channel(const struct vmbus_channel *c)
862 return !!(c->offermsg.offer.chn_flags &
863 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER);
866 static inline void set_channel_affinity_state(struct vmbus_channel *c,
867 enum hv_numa_policy policy)
869 c->affinity_policy = policy;
872 static inline void set_channel_read_mode(struct vmbus_channel *c,
873 enum hv_callback_mode mode)
875 c->callback_mode = mode;
878 static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
880 c->per_channel_state = s;
883 static inline void *get_per_channel_state(struct vmbus_channel *c)
885 return c->per_channel_state;
888 static inline void set_channel_pending_send_size(struct vmbus_channel *c,
891 c->outbound.ring_buffer->pending_send_sz = size;
894 static inline void set_low_latency_mode(struct vmbus_channel *c)
896 c->low_latency = true;
899 static inline void clear_low_latency_mode(struct vmbus_channel *c)
901 c->low_latency = false;
904 void vmbus_onmessage(void *context);
906 int vmbus_request_offers(void);
909 * APIs for managing sub-channels.
912 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
913 void (*sc_cr_cb)(struct vmbus_channel *new_sc));
915 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
916 void (*chn_rescind_cb)(struct vmbus_channel *));
919 * Retrieve the (sub) channel on which to send an outgoing request.
920 * When a primary channel has multiple sub-channels, we choose a
921 * channel whose VCPU binding is closest to the VCPU on which
922 * this call is being made.
924 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary);
927 * Check if sub-channels have already been offerred. This API will be useful
928 * when the driver is unloaded after establishing sub-channels. In this case,
929 * when the driver is re-loaded, the driver would have to check if the
930 * subchannels have already been established before attempting to request
931 * the creation of sub-channels.
932 * This function returns TRUE to indicate that subchannels have already been
934 * This function should be invoked after setting the callback function for
935 * sub-channel creation.
937 bool vmbus_are_subchannels_present(struct vmbus_channel *primary);
939 /* The format must be the same as struct vmdata_gpa_direct */
940 struct vmbus_channel_packet_page_buffer {
948 struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
951 /* The format must be the same as struct vmdata_gpa_direct */
952 struct vmbus_channel_packet_multipage_buffer {
959 u32 rangecount; /* Always 1 in this case */
960 struct hv_multipage_buffer range;
963 /* The format must be the same as struct vmdata_gpa_direct */
964 struct vmbus_packet_mpb_array {
971 u32 rangecount; /* Always 1 in this case */
972 struct hv_mpb_array range;
976 extern int vmbus_open(struct vmbus_channel *channel,
977 u32 send_ringbuffersize,
978 u32 recv_ringbuffersize,
981 void (*onchannel_callback)(void *context),
984 extern void vmbus_close(struct vmbus_channel *channel);
986 extern int vmbus_sendpacket(struct vmbus_channel *channel,
990 enum vmbus_packet_type type,
993 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
994 struct hv_page_buffer pagebuffers[],
1000 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1001 struct vmbus_packet_mpb_array *mpb,
1007 extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
1012 extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
1015 extern int vmbus_recvpacket(struct vmbus_channel *channel,
1018 u32 *buffer_actual_len,
1021 extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
1024 u32 *buffer_actual_len,
1028 extern void vmbus_ontimer(unsigned long data);
1030 /* Base driver object */
1035 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1036 * channel flag, actually doesn't mean a synthetic device because the
1037 * offer's if_type/if_instance can change for every new hvsock
1040 * However, to facilitate the notification of new-offer/rescind-offer
1041 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1042 * a special vmbus device, and hence we need the below flag to
1043 * indicate if the driver is the hvsock driver or not: we need to
1044 * specially treat the hvosck offer & driver in vmbus_match().
1048 /* the device type supported by this driver */
1050 const struct hv_vmbus_device_id *id_table;
1052 struct device_driver driver;
1054 /* dynamic device GUID's */
1057 struct list_head list;
1060 int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
1061 int (*remove)(struct hv_device *);
1062 void (*shutdown)(struct hv_device *);
1066 /* Base device object */
1068 /* the device type id of this device */
1071 /* the device instance id of this device */
1072 uuid_le dev_instance;
1076 struct device device;
1078 struct vmbus_channel *channel;
1079 struct kset *channels_kset;
1083 static inline struct hv_device *device_to_hv_device(struct device *d)
1085 return container_of(d, struct hv_device, device);
1088 static inline struct hv_driver *drv_to_hv_drv(struct device_driver *d)
1090 return container_of(d, struct hv_driver, driver);
1093 static inline void hv_set_drvdata(struct hv_device *dev, void *data)
1095 dev_set_drvdata(&dev->device, data);
1098 static inline void *hv_get_drvdata(struct hv_device *dev)
1100 return dev_get_drvdata(&dev->device);
1103 struct hv_ring_buffer_debug_info {
1104 u32 current_interrupt_mask;
1105 u32 current_read_index;
1106 u32 current_write_index;
1107 u32 bytes_avail_toread;
1108 u32 bytes_avail_towrite;
1111 void hv_ringbuffer_get_debuginfo(const struct hv_ring_buffer_info *ring_info,
1112 struct hv_ring_buffer_debug_info *debug_info);
1114 /* Vmbus interface */
1115 #define vmbus_driver_register(driver) \
1116 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1117 int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
1118 struct module *owner,
1119 const char *mod_name);
1120 void vmbus_driver_unregister(struct hv_driver *hv_driver);
1122 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel);
1124 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1125 resource_size_t min, resource_size_t max,
1126 resource_size_t size, resource_size_t align,
1127 bool fb_overlap_ok);
1128 void vmbus_free_mmio(resource_size_t start, resource_size_t size);
1131 * GUID definitions of various offer types - services offered to the guest.
1136 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1138 #define HV_NIC_GUID \
1139 .guid = UUID_LE(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1140 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1144 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1146 #define HV_IDE_GUID \
1147 .guid = UUID_LE(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1148 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1152 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1154 #define HV_SCSI_GUID \
1155 .guid = UUID_LE(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1156 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1160 * {0e0b6031-5213-4934-818b-38d90ced39db}
1162 #define HV_SHUTDOWN_GUID \
1163 .guid = UUID_LE(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1164 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1168 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1170 #define HV_TS_GUID \
1171 .guid = UUID_LE(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1172 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1176 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1178 #define HV_HEART_BEAT_GUID \
1179 .guid = UUID_LE(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1180 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1184 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1186 #define HV_KVP_GUID \
1187 .guid = UUID_LE(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1188 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1191 * Dynamic memory GUID
1192 * {525074dc-8985-46e2-8057-a307dc18a502}
1194 #define HV_DM_GUID \
1195 .guid = UUID_LE(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1196 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1200 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1202 #define HV_MOUSE_GUID \
1203 .guid = UUID_LE(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1204 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1208 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1210 #define HV_KBD_GUID \
1211 .guid = UUID_LE(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1212 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1215 * VSS (Backup/Restore) GUID
1217 #define HV_VSS_GUID \
1218 .guid = UUID_LE(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1219 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1221 * Synthetic Video GUID
1222 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1224 #define HV_SYNTHVID_GUID \
1225 .guid = UUID_LE(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1226 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1230 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1232 #define HV_SYNTHFC_GUID \
1233 .guid = UUID_LE(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1234 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1237 * Guest File Copy Service
1238 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1241 #define HV_FCOPY_GUID \
1242 .guid = UUID_LE(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1243 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1246 * NetworkDirect. This is the guest RDMA service.
1247 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1249 #define HV_ND_GUID \
1250 .guid = UUID_LE(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1251 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1254 * PCI Express Pass Through
1255 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1258 #define HV_PCIE_GUID \
1259 .guid = UUID_LE(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1260 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1263 * Linux doesn't support the 3 devices: the first two are for
1264 * Automatic Virtual Machine Activation, and the third is for
1265 * Remote Desktop Virtualization.
1266 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1267 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1268 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1271 #define HV_AVMA1_GUID \
1272 .guid = UUID_LE(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1273 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1275 #define HV_AVMA2_GUID \
1276 .guid = UUID_LE(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1277 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1279 #define HV_RDV_GUID \
1280 .guid = UUID_LE(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1281 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1284 * Common header for Hyper-V ICs
1287 #define ICMSGTYPE_NEGOTIATE 0
1288 #define ICMSGTYPE_HEARTBEAT 1
1289 #define ICMSGTYPE_KVPEXCHANGE 2
1290 #define ICMSGTYPE_SHUTDOWN 3
1291 #define ICMSGTYPE_TIMESYNC 4
1292 #define ICMSGTYPE_VSS 5
1294 #define ICMSGHDRFLAG_TRANSACTION 1
1295 #define ICMSGHDRFLAG_REQUEST 2
1296 #define ICMSGHDRFLAG_RESPONSE 4
1300 * While we want to handle util services as regular devices,
1301 * there is only one instance of each of these services; so
1302 * we statically allocate the service specific state.
1305 struct hv_util_service {
1308 void (*util_cb)(void *);
1309 int (*util_init)(struct hv_util_service *);
1310 void (*util_deinit)(void);
1313 struct vmbuspipe_hdr {
1324 struct ic_version icverframe;
1326 struct ic_version icvermsg;
1329 u8 ictransaction_id;
1334 struct icmsg_negotiate {
1338 struct ic_version icversion_data[1]; /* any size array */
1341 struct shutdown_msg_data {
1343 u32 timeout_seconds;
1345 u8 display_message[2048];
1348 struct heartbeat_msg_data {
1353 /* Time Sync IC defs */
1354 #define ICTIMESYNCFLAG_PROBE 0
1355 #define ICTIMESYNCFLAG_SYNC 1
1356 #define ICTIMESYNCFLAG_SAMPLE 2
1359 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1361 #define WLTIMEDELTA 116444736000000000LL
1364 struct ictimesync_data {
1371 struct ictimesync_ref_data {
1373 u64 vmreferencetime;
1380 struct hyperv_service_callback {
1384 struct vmbus_channel *channel;
1385 void (*callback)(void *context);
1388 #define MAX_SRV_VER 0x7ffffff
1389 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf,
1390 const int *fw_version, int fw_vercnt,
1391 const int *srv_version, int srv_vercnt,
1392 int *nego_fw_version, int *nego_srv_version);
1394 void hv_process_channel_removal(u32 relid);
1396 void vmbus_setevent(struct vmbus_channel *channel);
1398 * Negotiated version with the Host.
1401 extern __u32 vmbus_proto_version;
1403 int vmbus_send_tl_connect_request(const uuid_le *shv_guest_servie_id,
1404 const uuid_le *shv_host_servie_id);
1405 void vmbus_set_event(struct vmbus_channel *channel);
1407 /* Get the start of the ring buffer. */
1408 static inline void *
1409 hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info)
1411 return ring_info->ring_buffer->buffer;
1415 * Mask off host interrupt callback notifications
1417 static inline void hv_begin_read(struct hv_ring_buffer_info *rbi)
1419 rbi->ring_buffer->interrupt_mask = 1;
1421 /* make sure mask update is not reordered */
1426 * Re-enable host callback and return number of outstanding bytes
1428 static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi)
1431 rbi->ring_buffer->interrupt_mask = 0;
1433 /* make sure mask update is not reordered */
1437 * Now check to see if the ring buffer is still empty.
1438 * If it is not, we raced and we need to process new
1439 * incoming messages.
1441 return hv_get_bytes_to_read(rbi);
1445 * An API to support in-place processing of incoming VMBUS packets.
1448 /* Get data payload associated with descriptor */
1449 static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc)
1451 return (void *)((unsigned long)desc + (desc->offset8 << 3));
1454 /* Get data size associated with descriptor */
1455 static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc)
1457 return (desc->len8 << 3) - (desc->offset8 << 3);
1461 struct vmpacket_descriptor *
1462 hv_pkt_iter_first(struct vmbus_channel *channel);
1464 struct vmpacket_descriptor *
1465 __hv_pkt_iter_next(struct vmbus_channel *channel,
1466 const struct vmpacket_descriptor *pkt);
1468 void hv_pkt_iter_close(struct vmbus_channel *channel);
1471 * Get next packet descriptor from iterator
1472 * If at end of list, return NULL and update host.
1474 static inline struct vmpacket_descriptor *
1475 hv_pkt_iter_next(struct vmbus_channel *channel,
1476 const struct vmpacket_descriptor *pkt)
1478 struct vmpacket_descriptor *nxt;
1480 nxt = __hv_pkt_iter_next(channel, pkt);
1482 hv_pkt_iter_close(channel);
1487 #define foreach_vmbus_pkt(pkt, channel) \
1488 for (pkt = hv_pkt_iter_first(channel); pkt; \
1489 pkt = hv_pkt_iter_next(channel, pkt))
1491 #endif /* _HYPERV_H */