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.
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 */
145 * The ring buffer mutex lock. This lock prevents the ring buffer from
146 * being freed while the ring buffer is being accessed.
148 struct mutex ring_buffer_mutex;
152 static inline u32 hv_get_bytes_to_read(const struct hv_ring_buffer_info *rbi)
154 u32 read_loc, write_loc, dsize, read;
156 dsize = rbi->ring_datasize;
157 read_loc = rbi->ring_buffer->read_index;
158 write_loc = READ_ONCE(rbi->ring_buffer->write_index);
160 read = write_loc >= read_loc ? (write_loc - read_loc) :
161 (dsize - read_loc) + write_loc;
166 static inline u32 hv_get_bytes_to_write(const struct hv_ring_buffer_info *rbi)
168 u32 read_loc, write_loc, dsize, write;
170 dsize = rbi->ring_datasize;
171 read_loc = READ_ONCE(rbi->ring_buffer->read_index);
172 write_loc = rbi->ring_buffer->write_index;
174 write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
175 read_loc - write_loc;
179 static inline u32 hv_get_avail_to_write_percent(
180 const struct hv_ring_buffer_info *rbi)
182 u32 avail_write = hv_get_bytes_to_write(rbi);
184 return reciprocal_divide(
185 (avail_write << 3) + (avail_write << 1),
186 rbi->ring_size_div10_reciprocal);
190 * VMBUS version is 32 bit entity broken up into
191 * two 16 bit quantities: major_number. minor_number.
193 * 0 . 13 (Windows Server 2008)
196 * 3 . 0 (Windows 8 R2)
198 * 5 . 0 (Newer Windows 10)
201 #define VERSION_WS2008 ((0 << 16) | (13))
202 #define VERSION_WIN7 ((1 << 16) | (1))
203 #define VERSION_WIN8 ((2 << 16) | (4))
204 #define VERSION_WIN8_1 ((3 << 16) | (0))
205 #define VERSION_WIN10 ((4 << 16) | (0))
206 #define VERSION_WIN10_V5 ((5 << 16) | (0))
208 #define VERSION_INVAL -1
210 #define VERSION_CURRENT VERSION_WIN10_V5
212 /* Make maximum size of pipe payload of 16K */
213 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
215 /* Define PipeMode values. */
216 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
217 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
219 /* The size of the user defined data buffer for non-pipe offers. */
220 #define MAX_USER_DEFINED_BYTES 120
222 /* The size of the user defined data buffer for pipe offers. */
223 #define MAX_PIPE_USER_DEFINED_BYTES 116
226 * At the center of the Channel Management library is the Channel Offer. This
227 * struct contains the fundamental information about an offer.
229 struct vmbus_channel_offer {
234 * These two fields are not currently used.
240 u16 mmio_megabytes; /* in bytes * 1024 * 1024 */
243 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
245 unsigned char user_def[MAX_USER_DEFINED_BYTES];
250 * The following sructure is an integrated pipe protocol, which
251 * is implemented on top of standard user-defined data. Pipe
252 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
257 unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
261 * The sub_channel_index is defined in win8.
263 u16 sub_channel_index;
268 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
269 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
270 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
271 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
272 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
273 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
274 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
275 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
277 struct vmpacket_descriptor {
285 struct vmpacket_header {
286 u32 prev_pkt_start_offset;
287 struct vmpacket_descriptor descriptor;
290 struct vmtransfer_page_range {
295 struct vmtransfer_page_packet_header {
296 struct vmpacket_descriptor d;
301 struct vmtransfer_page_range ranges[1];
304 struct vmgpadl_packet_header {
305 struct vmpacket_descriptor d;
310 struct vmadd_remove_transfer_page_set {
311 struct vmpacket_descriptor d;
318 * This structure defines a range in guest physical space that can be made to
319 * look virtually contiguous.
328 * This is the format for an Establish Gpadl packet, which contains a handle by
329 * which this GPADL will be known and a set of GPA ranges associated with it.
330 * This can be converted to a MDL by the guest OS. If there are multiple GPA
331 * ranges, then the resulting MDL will be "chained," representing multiple VA
334 struct vmestablish_gpadl {
335 struct vmpacket_descriptor d;
338 struct gpa_range range[1];
342 * This is the format for a Teardown Gpadl packet, which indicates that the
343 * GPADL handle in the Establish Gpadl packet will never be referenced again.
345 struct vmteardown_gpadl {
346 struct vmpacket_descriptor d;
348 u32 reserved; /* for alignment to a 8-byte boundary */
352 * This is the format for a GPA-Direct packet, which contains a set of GPA
353 * ranges, in addition to commands and/or data.
355 struct vmdata_gpa_direct {
356 struct vmpacket_descriptor d;
359 struct gpa_range range[1];
362 /* This is the format for a Additional Data Packet. */
363 struct vmadditional_data {
364 struct vmpacket_descriptor d;
368 unsigned char data[1];
371 union vmpacket_largest_possible_header {
372 struct vmpacket_descriptor simple_hdr;
373 struct vmtransfer_page_packet_header xfer_page_hdr;
374 struct vmgpadl_packet_header gpadl_hdr;
375 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr;
376 struct vmestablish_gpadl establish_gpadl_hdr;
377 struct vmteardown_gpadl teardown_gpadl_hdr;
378 struct vmdata_gpa_direct data_gpa_direct_hdr;
381 #define VMPACKET_DATA_START_ADDRESS(__packet) \
382 (void *)(((unsigned char *)__packet) + \
383 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
385 #define VMPACKET_DATA_LENGTH(__packet) \
386 ((((struct vmpacket_descriptor)__packet)->len8 - \
387 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
389 #define VMPACKET_TRANSFER_MODE(__packet) \
390 (((struct IMPACT)__packet)->type)
392 enum vmbus_packet_type {
393 VM_PKT_INVALID = 0x0,
395 VM_PKT_ADD_XFER_PAGESET = 0x2,
396 VM_PKT_RM_XFER_PAGESET = 0x3,
397 VM_PKT_ESTABLISH_GPADL = 0x4,
398 VM_PKT_TEARDOWN_GPADL = 0x5,
399 VM_PKT_DATA_INBAND = 0x6,
400 VM_PKT_DATA_USING_XFER_PAGES = 0x7,
401 VM_PKT_DATA_USING_GPADL = 0x8,
402 VM_PKT_DATA_USING_GPA_DIRECT = 0x9,
403 VM_PKT_CANCEL_REQUEST = 0xa,
405 VM_PKT_DATA_USING_ADDITIONAL_PKT = 0xc,
406 VM_PKT_ADDITIONAL_DATA = 0xd
409 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
412 /* Version 1 messages */
413 enum vmbus_channel_message_type {
414 CHANNELMSG_INVALID = 0,
415 CHANNELMSG_OFFERCHANNEL = 1,
416 CHANNELMSG_RESCIND_CHANNELOFFER = 2,
417 CHANNELMSG_REQUESTOFFERS = 3,
418 CHANNELMSG_ALLOFFERS_DELIVERED = 4,
419 CHANNELMSG_OPENCHANNEL = 5,
420 CHANNELMSG_OPENCHANNEL_RESULT = 6,
421 CHANNELMSG_CLOSECHANNEL = 7,
422 CHANNELMSG_GPADL_HEADER = 8,
423 CHANNELMSG_GPADL_BODY = 9,
424 CHANNELMSG_GPADL_CREATED = 10,
425 CHANNELMSG_GPADL_TEARDOWN = 11,
426 CHANNELMSG_GPADL_TORNDOWN = 12,
427 CHANNELMSG_RELID_RELEASED = 13,
428 CHANNELMSG_INITIATE_CONTACT = 14,
429 CHANNELMSG_VERSION_RESPONSE = 15,
430 CHANNELMSG_UNLOAD = 16,
431 CHANNELMSG_UNLOAD_RESPONSE = 17,
435 CHANNELMSG_TL_CONNECT_REQUEST = 21,
439 struct vmbus_channel_message_header {
440 enum vmbus_channel_message_type msgtype;
444 /* Query VMBus Version parameters */
445 struct vmbus_channel_query_vmbus_version {
446 struct vmbus_channel_message_header header;
450 /* VMBus Version Supported parameters */
451 struct vmbus_channel_version_supported {
452 struct vmbus_channel_message_header header;
453 u8 version_supported;
456 /* Offer Channel parameters */
457 struct vmbus_channel_offer_channel {
458 struct vmbus_channel_message_header header;
459 struct vmbus_channel_offer offer;
463 * win7 and beyond splits this field into a bit field.
465 u8 monitor_allocated:1;
468 * These are new fields added in win7 and later.
469 * Do not access these fields without checking the
470 * negotiated protocol.
472 * If "is_dedicated_interrupt" is set, we must not set the
473 * associated bit in the channel bitmap while sending the
474 * interrupt to the host.
476 * connection_id is to be used in signaling the host.
478 u16 is_dedicated_interrupt:1;
483 /* Rescind Offer parameters */
484 struct vmbus_channel_rescind_offer {
485 struct vmbus_channel_message_header header;
490 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info *rbi)
492 return rbi->ring_buffer->pending_send_sz;
496 * Request Offer -- no parameters, SynIC message contains the partition ID
497 * Set Snoop -- no parameters, SynIC message contains the partition ID
498 * Clear Snoop -- no parameters, SynIC message contains the partition ID
499 * All Offers Delivered -- no parameters, SynIC message contains the partition
501 * Flush Client -- no parameters, SynIC message contains the partition ID
504 /* Open Channel parameters */
505 struct vmbus_channel_open_channel {
506 struct vmbus_channel_message_header header;
508 /* Identifies the specific VMBus channel that is being opened. */
511 /* ID making a particular open request at a channel offer unique. */
514 /* GPADL for the channel's ring buffer. */
515 u32 ringbuffer_gpadlhandle;
518 * Starting with win8, this field will be used to specify
519 * the target virtual processor on which to deliver the interrupt for
520 * the host to guest communication.
521 * Prior to win8, incoming channel interrupts would only
522 * be delivered on cpu 0. Setting this value to 0 would
523 * preserve the earlier behavior.
528 * The upstream ring buffer begins at offset zero in the memory
529 * described by RingBufferGpadlHandle. The downstream ring buffer
530 * follows it at this offset (in pages).
532 u32 downstream_ringbuffer_pageoffset;
534 /* User-specific data to be passed along to the server endpoint. */
535 unsigned char userdata[MAX_USER_DEFINED_BYTES];
538 /* Open Channel Result parameters */
539 struct vmbus_channel_open_result {
540 struct vmbus_channel_message_header header;
546 /* Close channel parameters; */
547 struct vmbus_channel_close_channel {
548 struct vmbus_channel_message_header header;
552 /* Channel Message GPADL */
553 #define GPADL_TYPE_RING_BUFFER 1
554 #define GPADL_TYPE_SERVER_SAVE_AREA 2
555 #define GPADL_TYPE_TRANSACTION 8
558 * The number of PFNs in a GPADL message is defined by the number of
559 * pages that would be spanned by ByteCount and ByteOffset. If the
560 * implied number of PFNs won't fit in this packet, there will be a
561 * follow-up packet that contains more.
563 struct vmbus_channel_gpadl_header {
564 struct vmbus_channel_message_header header;
569 struct gpa_range range[0];
572 /* This is the followup packet that contains more PFNs. */
573 struct vmbus_channel_gpadl_body {
574 struct vmbus_channel_message_header header;
580 struct vmbus_channel_gpadl_created {
581 struct vmbus_channel_message_header header;
587 struct vmbus_channel_gpadl_teardown {
588 struct vmbus_channel_message_header header;
593 struct vmbus_channel_gpadl_torndown {
594 struct vmbus_channel_message_header header;
598 struct vmbus_channel_relid_released {
599 struct vmbus_channel_message_header header;
603 struct vmbus_channel_initiate_contact {
604 struct vmbus_channel_message_header header;
605 u32 vmbus_version_requested;
606 u32 target_vcpu; /* The VCPU the host should respond to */
619 /* Hyper-V socket: guest's connect()-ing to host */
620 struct vmbus_channel_tl_connect_request {
621 struct vmbus_channel_message_header header;
622 guid_t guest_endpoint_id;
623 guid_t host_service_id;
626 struct vmbus_channel_version_response {
627 struct vmbus_channel_message_header header;
628 u8 version_supported;
634 * On new hosts that support VMBus protocol 5.0, we must use
635 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
636 * and for subsequent messages, we must use the Message Connection ID
637 * field in the host-returned Version Response Message.
639 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
644 enum vmbus_channel_state {
646 CHANNEL_OPENING_STATE,
648 CHANNEL_OPENED_STATE,
652 * Represents each channel msg on the vmbus connection This is a
653 * variable-size data structure depending on the msg type itself
655 struct vmbus_channel_msginfo {
656 /* Bookkeeping stuff */
657 struct list_head msglistentry;
659 /* So far, this is only used to handle gpadl body message */
660 struct list_head submsglist;
662 /* Synchronize the request/response if needed */
663 struct completion waitevent;
664 struct vmbus_channel *waiting_channel;
666 struct vmbus_channel_version_supported version_supported;
667 struct vmbus_channel_open_result open_result;
668 struct vmbus_channel_gpadl_torndown gpadl_torndown;
669 struct vmbus_channel_gpadl_created gpadl_created;
670 struct vmbus_channel_version_response version_response;
675 * The channel message that goes out on the "wire".
676 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
678 unsigned char msg[0];
681 struct vmbus_close_msg {
682 struct vmbus_channel_msginfo info;
683 struct vmbus_channel_close_channel msg;
686 /* Define connection identifier type. */
687 union hv_connection_id {
695 enum hv_numa_policy {
700 enum vmbus_device_type {
720 struct vmbus_device {
726 struct vmbus_channel {
727 struct list_head listentry;
729 struct hv_device *device_obj;
731 enum vmbus_channel_state state;
733 struct vmbus_channel_offer_channel offermsg;
735 * These are based on the OfferMsg.MonitorId.
736 * Save it here for easy access.
741 bool rescind; /* got rescind msg */
742 struct completion rescind_event;
744 u32 ringbuffer_gpadlhandle;
746 /* Allocated memory for ring buffer */
747 struct page *ringbuffer_page;
748 u32 ringbuffer_pagecount;
749 u32 ringbuffer_send_offset;
750 struct hv_ring_buffer_info outbound; /* send to parent */
751 struct hv_ring_buffer_info inbound; /* receive from parent */
753 struct vmbus_close_msg close_msg;
756 u64 interrupts; /* Host to Guest interrupts */
757 u64 sig_events; /* Guest to Host events */
760 * Guest to host interrupts caused by the outbound ring buffer changing
761 * from empty to not empty.
766 * Indicates that a full outbound ring buffer was encountered. The flag
767 * is set to true when a full outbound ring buffer is encountered and
768 * set to false when a write to the outbound ring buffer is completed.
772 /* Channel callback's invoked in softirq context */
773 struct tasklet_struct callback_event;
774 void (*onchannel_callback)(void *context);
775 void *channel_callback_context;
778 * A channel can be marked for one of three modes of reading:
779 * BATCHED - callback called from taslket and should read
780 * channel until empty. Interrupts from the host
781 * are masked while read is in process (default).
782 * DIRECT - callback called from tasklet (softirq).
783 * ISR - callback called in interrupt context and must
784 * invoke its own deferred processing.
785 * Host interrupts are disabled and must be re-enabled
786 * when ring is empty.
788 enum hv_callback_mode {
794 bool is_dedicated_interrupt;
798 * Starting with win8, this field will be used to specify
799 * the target virtual processor on which to deliver the interrupt for
800 * the host to guest communication.
801 * Prior to win8, incoming channel interrupts would only
802 * be delivered on cpu 0. Setting this value to 0 would
803 * preserve the earlier behavior.
806 /* The corresponding CPUID in the guest */
809 * State to manage the CPU affiliation of channels.
811 struct cpumask alloced_cpus_in_node;
814 * Support for sub-channels. For high performance devices,
815 * it will be useful to have multiple sub-channels to support
816 * a scalable communication infrastructure with the host.
817 * The support for sub-channels is implemented as an extention
818 * to the current infrastructure.
819 * The initial offer is considered the primary channel and this
820 * offer message will indicate if the host supports sub-channels.
821 * The guest is free to ask for sub-channels to be offerred and can
822 * open these sub-channels as a normal "primary" channel. However,
823 * all sub-channels will have the same type and instance guids as the
824 * primary channel. Requests sent on a given channel will result in a
825 * response on the same channel.
829 * Sub-channel creation callback. This callback will be called in
830 * process context when a sub-channel offer is received from the host.
831 * The guest can open the sub-channel in the context of this callback.
833 void (*sc_creation_callback)(struct vmbus_channel *new_sc);
836 * Channel rescind callback. Some channels (the hvsock ones), need to
837 * register a callback which is invoked in vmbus_onoffer_rescind().
839 void (*chn_rescind_callback)(struct vmbus_channel *channel);
842 * The spinlock to protect the structure. It is being used to protect
843 * test-and-set access to various attributes of the structure as well
844 * as all sc_list operations.
848 * All Sub-channels of a primary channel are linked here.
850 struct list_head sc_list;
852 * The primary channel this sub-channel belongs to.
853 * This will be NULL for the primary channel.
855 struct vmbus_channel *primary_channel;
857 * Support per-channel state for use by vmbus drivers.
859 void *per_channel_state;
861 * To support per-cpu lookup mapping of relid to channel,
862 * link up channels based on their CPU affinity.
864 struct list_head percpu_list;
867 * Defer freeing channel until after all cpu's have
868 * gone through grace period.
873 * For sysfs per-channel properties.
878 * For performance critical channels (storage, networking
879 * etc,), Hyper-V has a mechanism to enhance the throughput
880 * at the expense of latency:
881 * When the host is to be signaled, we just set a bit in a shared page
882 * and this bit will be inspected by the hypervisor within a certain
883 * window and if the bit is set, the host will be signaled. The window
884 * of time is the monitor latency - currently around 100 usecs. This
885 * mechanism improves throughput by:
887 * A) Making the host more efficient - each time it wakes up,
888 * potentially it will process morev number of packets. The
889 * monitor latency allows a batch to build up.
890 * B) By deferring the hypercall to signal, we will also minimize
893 * Clearly, these optimizations improve throughput at the expense of
894 * latency. Furthermore, since the channel is shared for both
895 * control and data messages, control messages currently suffer
896 * unnecessary latency adversley impacting performance and boot
897 * time. To fix this issue, permit tagging the channel as being
898 * in "low latency" mode. In this mode, we will bypass the monitor
904 * NUMA distribution policy:
905 * We support two policies:
906 * 1) Balanced: Here all performance critical channels are
907 * distributed evenly amongst all the NUMA nodes.
908 * This policy will be the default policy.
909 * 2) Localized: All channels of a given instance of a
910 * performance critical service will be assigned CPUs
911 * within a selected NUMA node.
913 enum hv_numa_policy affinity_policy;
918 * We must offload the handling of the primary/sub channels
919 * from the single-threaded vmbus_connection.work_queue to
920 * two different workqueue, otherwise we can block
921 * vmbus_connection.work_queue and hang: see vmbus_process_offer().
923 struct work_struct add_channel_work;
926 * Guest to host interrupts caused by the inbound ring buffer changing
927 * from full to not full while a packet is waiting.
932 * The total number of write operations that encountered a full
933 * outbound ring buffer.
938 * The number of write operations that were the first to encounter a
939 * full outbound ring buffer.
944 static inline bool is_hvsock_channel(const struct vmbus_channel *c)
946 return !!(c->offermsg.offer.chn_flags &
947 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER);
950 static inline void set_channel_affinity_state(struct vmbus_channel *c,
951 enum hv_numa_policy policy)
953 c->affinity_policy = policy;
956 static inline void set_channel_read_mode(struct vmbus_channel *c,
957 enum hv_callback_mode mode)
959 c->callback_mode = mode;
962 static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
964 c->per_channel_state = s;
967 static inline void *get_per_channel_state(struct vmbus_channel *c)
969 return c->per_channel_state;
972 static inline void set_channel_pending_send_size(struct vmbus_channel *c,
978 spin_lock_irqsave(&c->outbound.ring_lock, flags);
981 if (!c->out_full_flag) {
983 c->out_full_flag = true;
985 spin_unlock_irqrestore(&c->outbound.ring_lock, flags);
987 c->out_full_flag = false;
990 c->outbound.ring_buffer->pending_send_sz = size;
993 static inline void set_low_latency_mode(struct vmbus_channel *c)
995 c->low_latency = true;
998 static inline void clear_low_latency_mode(struct vmbus_channel *c)
1000 c->low_latency = false;
1003 void vmbus_onmessage(void *context);
1005 int vmbus_request_offers(void);
1008 * APIs for managing sub-channels.
1011 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1012 void (*sc_cr_cb)(struct vmbus_channel *new_sc));
1014 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1015 void (*chn_rescind_cb)(struct vmbus_channel *));
1018 * Check if sub-channels have already been offerred. This API will be useful
1019 * when the driver is unloaded after establishing sub-channels. In this case,
1020 * when the driver is re-loaded, the driver would have to check if the
1021 * subchannels have already been established before attempting to request
1022 * the creation of sub-channels.
1023 * This function returns TRUE to indicate that subchannels have already been
1025 * This function should be invoked after setting the callback function for
1026 * sub-channel creation.
1028 bool vmbus_are_subchannels_present(struct vmbus_channel *primary);
1030 /* The format must be the same as struct vmdata_gpa_direct */
1031 struct vmbus_channel_packet_page_buffer {
1039 struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
1042 /* The format must be the same as struct vmdata_gpa_direct */
1043 struct vmbus_channel_packet_multipage_buffer {
1050 u32 rangecount; /* Always 1 in this case */
1051 struct hv_multipage_buffer range;
1054 /* The format must be the same as struct vmdata_gpa_direct */
1055 struct vmbus_packet_mpb_array {
1062 u32 rangecount; /* Always 1 in this case */
1063 struct hv_mpb_array range;
1066 int vmbus_alloc_ring(struct vmbus_channel *channel,
1067 u32 send_size, u32 recv_size);
1068 void vmbus_free_ring(struct vmbus_channel *channel);
1070 int vmbus_connect_ring(struct vmbus_channel *channel,
1071 void (*onchannel_callback)(void *context),
1073 int vmbus_disconnect_ring(struct vmbus_channel *channel);
1075 extern int vmbus_open(struct vmbus_channel *channel,
1076 u32 send_ringbuffersize,
1077 u32 recv_ringbuffersize,
1080 void (*onchannel_callback)(void *context),
1083 extern void vmbus_close(struct vmbus_channel *channel);
1085 extern int vmbus_sendpacket(struct vmbus_channel *channel,
1089 enum vmbus_packet_type type,
1092 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1093 struct hv_page_buffer pagebuffers[],
1099 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1100 struct vmbus_packet_mpb_array *mpb,
1106 extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
1111 extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
1114 void vmbus_reset_channel_cb(struct vmbus_channel *channel);
1116 extern int vmbus_recvpacket(struct vmbus_channel *channel,
1119 u32 *buffer_actual_len,
1122 extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
1125 u32 *buffer_actual_len,
1129 extern void vmbus_ontimer(unsigned long data);
1131 /* Base driver object */
1136 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1137 * channel flag, actually doesn't mean a synthetic device because the
1138 * offer's if_type/if_instance can change for every new hvsock
1141 * However, to facilitate the notification of new-offer/rescind-offer
1142 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1143 * a special vmbus device, and hence we need the below flag to
1144 * indicate if the driver is the hvsock driver or not: we need to
1145 * specially treat the hvosck offer & driver in vmbus_match().
1149 /* the device type supported by this driver */
1151 const struct hv_vmbus_device_id *id_table;
1153 struct device_driver driver;
1155 /* dynamic device GUID's */
1158 struct list_head list;
1161 int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
1162 int (*remove)(struct hv_device *);
1163 void (*shutdown)(struct hv_device *);
1167 /* Base device object */
1169 /* the device type id of this device */
1172 /* the device instance id of this device */
1173 guid_t dev_instance;
1177 struct device device;
1178 char *driver_override; /* Driver name to force a match */
1180 struct vmbus_channel *channel;
1181 struct kset *channels_kset;
1185 static inline struct hv_device *device_to_hv_device(struct device *d)
1187 return container_of(d, struct hv_device, device);
1190 static inline struct hv_driver *drv_to_hv_drv(struct device_driver *d)
1192 return container_of(d, struct hv_driver, driver);
1195 static inline void hv_set_drvdata(struct hv_device *dev, void *data)
1197 dev_set_drvdata(&dev->device, data);
1200 static inline void *hv_get_drvdata(struct hv_device *dev)
1202 return dev_get_drvdata(&dev->device);
1205 struct hv_ring_buffer_debug_info {
1206 u32 current_interrupt_mask;
1207 u32 current_read_index;
1208 u32 current_write_index;
1209 u32 bytes_avail_toread;
1210 u32 bytes_avail_towrite;
1214 int hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
1215 struct hv_ring_buffer_debug_info *debug_info);
1217 /* Vmbus interface */
1218 #define vmbus_driver_register(driver) \
1219 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1220 int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
1221 struct module *owner,
1222 const char *mod_name);
1223 void vmbus_driver_unregister(struct hv_driver *hv_driver);
1225 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel);
1227 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1228 resource_size_t min, resource_size_t max,
1229 resource_size_t size, resource_size_t align,
1230 bool fb_overlap_ok);
1231 void vmbus_free_mmio(resource_size_t start, resource_size_t size);
1234 * GUID definitions of various offer types - services offered to the guest.
1239 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1241 #define HV_NIC_GUID \
1242 .guid = GUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1243 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1247 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1249 #define HV_IDE_GUID \
1250 .guid = GUID_INIT(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1251 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1255 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1257 #define HV_SCSI_GUID \
1258 .guid = GUID_INIT(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1259 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1263 * {0e0b6031-5213-4934-818b-38d90ced39db}
1265 #define HV_SHUTDOWN_GUID \
1266 .guid = GUID_INIT(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1267 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1271 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1273 #define HV_TS_GUID \
1274 .guid = GUID_INIT(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1275 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1279 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1281 #define HV_HEART_BEAT_GUID \
1282 .guid = GUID_INIT(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1283 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1287 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1289 #define HV_KVP_GUID \
1290 .guid = GUID_INIT(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1291 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1294 * Dynamic memory GUID
1295 * {525074dc-8985-46e2-8057-a307dc18a502}
1297 #define HV_DM_GUID \
1298 .guid = GUID_INIT(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1299 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1303 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1305 #define HV_MOUSE_GUID \
1306 .guid = GUID_INIT(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1307 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1311 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1313 #define HV_KBD_GUID \
1314 .guid = GUID_INIT(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1315 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1318 * VSS (Backup/Restore) GUID
1320 #define HV_VSS_GUID \
1321 .guid = GUID_INIT(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1322 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1324 * Synthetic Video GUID
1325 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1327 #define HV_SYNTHVID_GUID \
1328 .guid = GUID_INIT(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1329 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1333 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1335 #define HV_SYNTHFC_GUID \
1336 .guid = GUID_INIT(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1337 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1340 * Guest File Copy Service
1341 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1344 #define HV_FCOPY_GUID \
1345 .guid = GUID_INIT(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1346 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1349 * NetworkDirect. This is the guest RDMA service.
1350 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1352 #define HV_ND_GUID \
1353 .guid = GUID_INIT(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1354 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1357 * PCI Express Pass Through
1358 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1361 #define HV_PCIE_GUID \
1362 .guid = GUID_INIT(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1363 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1366 * Linux doesn't support the 3 devices: the first two are for
1367 * Automatic Virtual Machine Activation, and the third is for
1368 * Remote Desktop Virtualization.
1369 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1370 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1371 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1374 #define HV_AVMA1_GUID \
1375 .guid = GUID_INIT(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1376 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1378 #define HV_AVMA2_GUID \
1379 .guid = GUID_INIT(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1380 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1382 #define HV_RDV_GUID \
1383 .guid = GUID_INIT(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1384 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1387 * Common header for Hyper-V ICs
1390 #define ICMSGTYPE_NEGOTIATE 0
1391 #define ICMSGTYPE_HEARTBEAT 1
1392 #define ICMSGTYPE_KVPEXCHANGE 2
1393 #define ICMSGTYPE_SHUTDOWN 3
1394 #define ICMSGTYPE_TIMESYNC 4
1395 #define ICMSGTYPE_VSS 5
1397 #define ICMSGHDRFLAG_TRANSACTION 1
1398 #define ICMSGHDRFLAG_REQUEST 2
1399 #define ICMSGHDRFLAG_RESPONSE 4
1403 * While we want to handle util services as regular devices,
1404 * there is only one instance of each of these services; so
1405 * we statically allocate the service specific state.
1408 struct hv_util_service {
1411 void (*util_cb)(void *);
1412 int (*util_init)(struct hv_util_service *);
1413 void (*util_deinit)(void);
1416 struct vmbuspipe_hdr {
1427 struct ic_version icverframe;
1429 struct ic_version icvermsg;
1432 u8 ictransaction_id;
1437 struct icmsg_negotiate {
1441 struct ic_version icversion_data[1]; /* any size array */
1444 struct shutdown_msg_data {
1446 u32 timeout_seconds;
1448 u8 display_message[2048];
1451 struct heartbeat_msg_data {
1456 /* Time Sync IC defs */
1457 #define ICTIMESYNCFLAG_PROBE 0
1458 #define ICTIMESYNCFLAG_SYNC 1
1459 #define ICTIMESYNCFLAG_SAMPLE 2
1462 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1464 #define WLTIMEDELTA 116444736000000000LL
1467 struct ictimesync_data {
1474 struct ictimesync_ref_data {
1476 u64 vmreferencetime;
1483 struct hyperv_service_callback {
1487 struct vmbus_channel *channel;
1488 void (*callback)(void *context);
1491 #define MAX_SRV_VER 0x7ffffff
1492 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf,
1493 const int *fw_version, int fw_vercnt,
1494 const int *srv_version, int srv_vercnt,
1495 int *nego_fw_version, int *nego_srv_version);
1497 void hv_process_channel_removal(struct vmbus_channel *channel);
1499 void vmbus_setevent(struct vmbus_channel *channel);
1501 * Negotiated version with the Host.
1504 extern __u32 vmbus_proto_version;
1506 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
1507 const guid_t *shv_host_servie_id);
1508 void vmbus_set_event(struct vmbus_channel *channel);
1510 /* Get the start of the ring buffer. */
1511 static inline void *
1512 hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info)
1514 return ring_info->ring_buffer->buffer;
1518 * Mask off host interrupt callback notifications
1520 static inline void hv_begin_read(struct hv_ring_buffer_info *rbi)
1522 rbi->ring_buffer->interrupt_mask = 1;
1524 /* make sure mask update is not reordered */
1529 * Re-enable host callback and return number of outstanding bytes
1531 static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi)
1534 rbi->ring_buffer->interrupt_mask = 0;
1536 /* make sure mask update is not reordered */
1540 * Now check to see if the ring buffer is still empty.
1541 * If it is not, we raced and we need to process new
1542 * incoming messages.
1544 return hv_get_bytes_to_read(rbi);
1548 * An API to support in-place processing of incoming VMBUS packets.
1551 /* Get data payload associated with descriptor */
1552 static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc)
1554 return (void *)((unsigned long)desc + (desc->offset8 << 3));
1557 /* Get data size associated with descriptor */
1558 static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc)
1560 return (desc->len8 << 3) - (desc->offset8 << 3);
1564 struct vmpacket_descriptor *
1565 hv_pkt_iter_first(struct vmbus_channel *channel);
1567 struct vmpacket_descriptor *
1568 __hv_pkt_iter_next(struct vmbus_channel *channel,
1569 const struct vmpacket_descriptor *pkt);
1571 void hv_pkt_iter_close(struct vmbus_channel *channel);
1574 * Get next packet descriptor from iterator
1575 * If at end of list, return NULL and update host.
1577 static inline struct vmpacket_descriptor *
1578 hv_pkt_iter_next(struct vmbus_channel *channel,
1579 const struct vmpacket_descriptor *pkt)
1581 struct vmpacket_descriptor *nxt;
1583 nxt = __hv_pkt_iter_next(channel, pkt);
1585 hv_pkt_iter_close(channel);
1590 #define foreach_vmbus_pkt(pkt, channel) \
1591 for (pkt = hv_pkt_iter_first(channel); pkt; \
1592 pkt = hv_pkt_iter_next(channel, pkt))
1594 #endif /* _HYPERV_H */