1 /* SPDX-License-Identifier: GPL-2.0-only */
4 * Copyright (c) 2011, Microsoft Corporation.
15 #include <uapi/linux/hyperv.h>
17 #include <linux/types.h>
18 #include <linux/scatterlist.h>
19 #include <linux/list.h>
20 #include <linux/timer.h>
21 #include <linux/completion.h>
22 #include <linux/device.h>
23 #include <linux/mod_devicetable.h>
24 #include <linux/interrupt.h>
25 #include <linux/reciprocal_div.h>
27 #define MAX_PAGE_BUFFER_COUNT 32
28 #define MAX_MULTIPAGE_BUFFER_COUNT 32 /* 128K */
32 /* Single-page buffer */
33 struct hv_page_buffer {
39 /* Multiple-page buffer */
40 struct hv_multipage_buffer {
41 /* Length and Offset determines the # of pfns in the array */
44 u64 pfn_array[MAX_MULTIPAGE_BUFFER_COUNT];
48 * Multiple-page buffer array; the pfn array is variable size:
49 * The number of entries in the PFN array is determined by
53 /* Length and Offset determines the # of pfns in the array */
59 /* 0x18 includes the proprietary packet header */
60 #define MAX_PAGE_BUFFER_PACKET (0x18 + \
61 (sizeof(struct hv_page_buffer) * \
62 MAX_PAGE_BUFFER_COUNT))
63 #define MAX_MULTIPAGE_BUFFER_PACKET (0x18 + \
64 sizeof(struct hv_multipage_buffer))
69 struct hv_ring_buffer {
70 /* Offset in bytes from the start of ring data below */
73 /* Offset in bytes from the start of ring data below */
79 * WS2012/Win8 and later versions of Hyper-V implement interrupt
80 * driven flow management. The feature bit feat_pending_send_sz
81 * is set by the host on the host->guest ring buffer, and by the
82 * guest on the guest->host ring buffer.
84 * The meaning of the feature bit is a bit complex in that it has
85 * semantics that apply to both ring buffers. If the guest sets
86 * the feature bit in the guest->host ring buffer, the guest is
87 * telling the host that:
88 * 1) It will set the pending_send_sz field in the guest->host ring
89 * buffer when it is waiting for space to become available, and
90 * 2) It will read the pending_send_sz field in the host->guest
91 * ring buffer and interrupt the host when it frees enough space
93 * Similarly, if the host sets the feature bit in the host->guest
94 * ring buffer, the host is telling the guest that:
95 * 1) It will set the pending_send_sz field in the host->guest ring
96 * buffer when it is waiting for space to become available, and
97 * 2) It will read the pending_send_sz field in the guest->host
98 * ring buffer and interrupt the guest when it frees enough space
100 * If either the guest or host does not set the feature bit that it
101 * owns, that guest or host must do polling if it encounters a full
102 * ring buffer, and not signal the other end with an interrupt.
108 u32 feat_pending_send_sz:1;
113 /* Pad it to PAGE_SIZE so that data starts on page boundary */
117 * Ring data starts here + RingDataStartOffset
118 * !!! DO NOT place any fields below this !!!
123 struct hv_ring_buffer_info {
124 struct hv_ring_buffer *ring_buffer;
125 u32 ring_size; /* Include the shared header */
126 struct reciprocal_value ring_size_div10_reciprocal;
127 spinlock_t ring_lock;
129 u32 ring_datasize; /* < ring_size */
132 * The ring buffer mutex lock. This lock prevents the ring buffer from
133 * being freed while the ring buffer is being accessed.
135 struct mutex ring_buffer_mutex;
139 static inline u32 hv_get_bytes_to_read(const struct hv_ring_buffer_info *rbi)
141 u32 read_loc, write_loc, dsize, read;
143 dsize = rbi->ring_datasize;
144 read_loc = rbi->ring_buffer->read_index;
145 write_loc = READ_ONCE(rbi->ring_buffer->write_index);
147 read = write_loc >= read_loc ? (write_loc - read_loc) :
148 (dsize - read_loc) + write_loc;
153 static inline u32 hv_get_bytes_to_write(const struct hv_ring_buffer_info *rbi)
155 u32 read_loc, write_loc, dsize, write;
157 dsize = rbi->ring_datasize;
158 read_loc = READ_ONCE(rbi->ring_buffer->read_index);
159 write_loc = rbi->ring_buffer->write_index;
161 write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
162 read_loc - write_loc;
166 static inline u32 hv_get_avail_to_write_percent(
167 const struct hv_ring_buffer_info *rbi)
169 u32 avail_write = hv_get_bytes_to_write(rbi);
171 return reciprocal_divide(
172 (avail_write << 3) + (avail_write << 1),
173 rbi->ring_size_div10_reciprocal);
177 * VMBUS version is 32 bit entity broken up into
178 * two 16 bit quantities: major_number. minor_number.
180 * 0 . 13 (Windows Server 2008)
183 * 3 . 0 (Windows 8 R2)
185 * 4 . 1 (Windows 10 RS3)
186 * 5 . 0 (Newer Windows 10)
187 * 5 . 1 (Windows 10 RS4)
188 * 5 . 2 (Windows Server 2019, RS5)
191 #define VERSION_WS2008 ((0 << 16) | (13))
192 #define VERSION_WIN7 ((1 << 16) | (1))
193 #define VERSION_WIN8 ((2 << 16) | (4))
194 #define VERSION_WIN8_1 ((3 << 16) | (0))
195 #define VERSION_WIN10 ((4 << 16) | (0))
196 #define VERSION_WIN10_V4_1 ((4 << 16) | (1))
197 #define VERSION_WIN10_V5 ((5 << 16) | (0))
198 #define VERSION_WIN10_V5_1 ((5 << 16) | (1))
199 #define VERSION_WIN10_V5_2 ((5 << 16) | (2))
201 /* Make maximum size of pipe payload of 16K */
202 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
204 /* Define PipeMode values. */
205 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
206 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
208 /* The size of the user defined data buffer for non-pipe offers. */
209 #define MAX_USER_DEFINED_BYTES 120
211 /* The size of the user defined data buffer for pipe offers. */
212 #define MAX_PIPE_USER_DEFINED_BYTES 116
215 * At the center of the Channel Management library is the Channel Offer. This
216 * struct contains the fundamental information about an offer.
218 struct vmbus_channel_offer {
223 * These two fields are not currently used.
229 u16 mmio_megabytes; /* in bytes * 1024 * 1024 */
232 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
234 unsigned char user_def[MAX_USER_DEFINED_BYTES];
239 * The following sructure is an integrated pipe protocol, which
240 * is implemented on top of standard user-defined data. Pipe
241 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
246 unsigned char user_def[MAX_PIPE_USER_DEFINED_BYTES];
250 * The sub_channel_index is defined in Win8: a value of zero means a
251 * primary channel and a value of non-zero means a sub-channel.
253 * Before Win8, the field is reserved, meaning it's always zero.
255 u16 sub_channel_index;
260 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
261 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
262 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
263 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
264 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
265 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
266 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
267 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
269 struct vmpacket_descriptor {
277 struct vmpacket_header {
278 u32 prev_pkt_start_offset;
279 struct vmpacket_descriptor descriptor;
282 struct vmtransfer_page_range {
287 struct vmtransfer_page_packet_header {
288 struct vmpacket_descriptor d;
293 struct vmtransfer_page_range ranges[1];
296 struct vmgpadl_packet_header {
297 struct vmpacket_descriptor d;
302 struct vmadd_remove_transfer_page_set {
303 struct vmpacket_descriptor d;
310 * This structure defines a range in guest physical space that can be made to
311 * look virtually contiguous.
320 * This is the format for an Establish Gpadl packet, which contains a handle by
321 * which this GPADL will be known and a set of GPA ranges associated with it.
322 * This can be converted to a MDL by the guest OS. If there are multiple GPA
323 * ranges, then the resulting MDL will be "chained," representing multiple VA
326 struct vmestablish_gpadl {
327 struct vmpacket_descriptor d;
330 struct gpa_range range[1];
334 * This is the format for a Teardown Gpadl packet, which indicates that the
335 * GPADL handle in the Establish Gpadl packet will never be referenced again.
337 struct vmteardown_gpadl {
338 struct vmpacket_descriptor d;
340 u32 reserved; /* for alignment to a 8-byte boundary */
344 * This is the format for a GPA-Direct packet, which contains a set of GPA
345 * ranges, in addition to commands and/or data.
347 struct vmdata_gpa_direct {
348 struct vmpacket_descriptor d;
351 struct gpa_range range[1];
354 /* This is the format for a Additional Data Packet. */
355 struct vmadditional_data {
356 struct vmpacket_descriptor d;
360 unsigned char data[1];
363 union vmpacket_largest_possible_header {
364 struct vmpacket_descriptor simple_hdr;
365 struct vmtransfer_page_packet_header xfer_page_hdr;
366 struct vmgpadl_packet_header gpadl_hdr;
367 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr;
368 struct vmestablish_gpadl establish_gpadl_hdr;
369 struct vmteardown_gpadl teardown_gpadl_hdr;
370 struct vmdata_gpa_direct data_gpa_direct_hdr;
373 #define VMPACKET_DATA_START_ADDRESS(__packet) \
374 (void *)(((unsigned char *)__packet) + \
375 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
377 #define VMPACKET_DATA_LENGTH(__packet) \
378 ((((struct vmpacket_descriptor)__packet)->len8 - \
379 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
381 #define VMPACKET_TRANSFER_MODE(__packet) \
382 (((struct IMPACT)__packet)->type)
384 enum vmbus_packet_type {
385 VM_PKT_INVALID = 0x0,
387 VM_PKT_ADD_XFER_PAGESET = 0x2,
388 VM_PKT_RM_XFER_PAGESET = 0x3,
389 VM_PKT_ESTABLISH_GPADL = 0x4,
390 VM_PKT_TEARDOWN_GPADL = 0x5,
391 VM_PKT_DATA_INBAND = 0x6,
392 VM_PKT_DATA_USING_XFER_PAGES = 0x7,
393 VM_PKT_DATA_USING_GPADL = 0x8,
394 VM_PKT_DATA_USING_GPA_DIRECT = 0x9,
395 VM_PKT_CANCEL_REQUEST = 0xa,
397 VM_PKT_DATA_USING_ADDITIONAL_PKT = 0xc,
398 VM_PKT_ADDITIONAL_DATA = 0xd
401 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
404 /* Version 1 messages */
405 enum vmbus_channel_message_type {
406 CHANNELMSG_INVALID = 0,
407 CHANNELMSG_OFFERCHANNEL = 1,
408 CHANNELMSG_RESCIND_CHANNELOFFER = 2,
409 CHANNELMSG_REQUESTOFFERS = 3,
410 CHANNELMSG_ALLOFFERS_DELIVERED = 4,
411 CHANNELMSG_OPENCHANNEL = 5,
412 CHANNELMSG_OPENCHANNEL_RESULT = 6,
413 CHANNELMSG_CLOSECHANNEL = 7,
414 CHANNELMSG_GPADL_HEADER = 8,
415 CHANNELMSG_GPADL_BODY = 9,
416 CHANNELMSG_GPADL_CREATED = 10,
417 CHANNELMSG_GPADL_TEARDOWN = 11,
418 CHANNELMSG_GPADL_TORNDOWN = 12,
419 CHANNELMSG_RELID_RELEASED = 13,
420 CHANNELMSG_INITIATE_CONTACT = 14,
421 CHANNELMSG_VERSION_RESPONSE = 15,
422 CHANNELMSG_UNLOAD = 16,
423 CHANNELMSG_UNLOAD_RESPONSE = 17,
427 CHANNELMSG_TL_CONNECT_REQUEST = 21,
431 /* Hyper-V supports about 2048 channels, and the RELIDs start with 1. */
432 #define INVALID_RELID U32_MAX
434 struct vmbus_channel_message_header {
435 enum vmbus_channel_message_type msgtype;
439 /* Query VMBus Version parameters */
440 struct vmbus_channel_query_vmbus_version {
441 struct vmbus_channel_message_header header;
445 /* VMBus Version Supported parameters */
446 struct vmbus_channel_version_supported {
447 struct vmbus_channel_message_header header;
448 u8 version_supported;
451 /* Offer Channel parameters */
452 struct vmbus_channel_offer_channel {
453 struct vmbus_channel_message_header header;
454 struct vmbus_channel_offer offer;
458 * win7 and beyond splits this field into a bit field.
460 u8 monitor_allocated:1;
463 * These are new fields added in win7 and later.
464 * Do not access these fields without checking the
465 * negotiated protocol.
467 * If "is_dedicated_interrupt" is set, we must not set the
468 * associated bit in the channel bitmap while sending the
469 * interrupt to the host.
471 * connection_id is to be used in signaling the host.
473 u16 is_dedicated_interrupt:1;
478 /* Rescind Offer parameters */
479 struct vmbus_channel_rescind_offer {
480 struct vmbus_channel_message_header header;
485 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info *rbi)
487 return rbi->ring_buffer->pending_send_sz;
491 * Request Offer -- no parameters, SynIC message contains the partition ID
492 * Set Snoop -- no parameters, SynIC message contains the partition ID
493 * Clear Snoop -- no parameters, SynIC message contains the partition ID
494 * All Offers Delivered -- no parameters, SynIC message contains the partition
496 * Flush Client -- no parameters, SynIC message contains the partition ID
499 /* Open Channel parameters */
500 struct vmbus_channel_open_channel {
501 struct vmbus_channel_message_header header;
503 /* Identifies the specific VMBus channel that is being opened. */
506 /* ID making a particular open request at a channel offer unique. */
509 /* GPADL for the channel's ring buffer. */
510 u32 ringbuffer_gpadlhandle;
513 * Starting with win8, this field will be used to specify
514 * the target virtual processor on which to deliver the interrupt for
515 * the host to guest communication.
516 * Prior to win8, incoming channel interrupts would only
517 * be delivered on cpu 0. Setting this value to 0 would
518 * preserve the earlier behavior.
523 * The upstream ring buffer begins at offset zero in the memory
524 * described by RingBufferGpadlHandle. The downstream ring buffer
525 * follows it at this offset (in pages).
527 u32 downstream_ringbuffer_pageoffset;
529 /* User-specific data to be passed along to the server endpoint. */
530 unsigned char userdata[MAX_USER_DEFINED_BYTES];
533 /* Open Channel Result parameters */
534 struct vmbus_channel_open_result {
535 struct vmbus_channel_message_header header;
541 /* Close channel parameters; */
542 struct vmbus_channel_close_channel {
543 struct vmbus_channel_message_header header;
547 /* Channel Message GPADL */
548 #define GPADL_TYPE_RING_BUFFER 1
549 #define GPADL_TYPE_SERVER_SAVE_AREA 2
550 #define GPADL_TYPE_TRANSACTION 8
553 * The number of PFNs in a GPADL message is defined by the number of
554 * pages that would be spanned by ByteCount and ByteOffset. If the
555 * implied number of PFNs won't fit in this packet, there will be a
556 * follow-up packet that contains more.
558 struct vmbus_channel_gpadl_header {
559 struct vmbus_channel_message_header header;
564 struct gpa_range range[0];
567 /* This is the followup packet that contains more PFNs. */
568 struct vmbus_channel_gpadl_body {
569 struct vmbus_channel_message_header header;
575 struct vmbus_channel_gpadl_created {
576 struct vmbus_channel_message_header header;
582 struct vmbus_channel_gpadl_teardown {
583 struct vmbus_channel_message_header header;
588 struct vmbus_channel_gpadl_torndown {
589 struct vmbus_channel_message_header header;
593 struct vmbus_channel_relid_released {
594 struct vmbus_channel_message_header header;
598 struct vmbus_channel_initiate_contact {
599 struct vmbus_channel_message_header header;
600 u32 vmbus_version_requested;
601 u32 target_vcpu; /* The VCPU the host should respond to */
614 /* Hyper-V socket: guest's connect()-ing to host */
615 struct vmbus_channel_tl_connect_request {
616 struct vmbus_channel_message_header header;
617 guid_t guest_endpoint_id;
618 guid_t host_service_id;
621 struct vmbus_channel_version_response {
622 struct vmbus_channel_message_header header;
623 u8 version_supported;
629 * On new hosts that support VMBus protocol 5.0, we must use
630 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
631 * and for subsequent messages, we must use the Message Connection ID
632 * field in the host-returned Version Response Message.
634 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
639 enum vmbus_channel_state {
641 CHANNEL_OPENING_STATE,
643 CHANNEL_OPENED_STATE,
647 * Represents each channel msg on the vmbus connection This is a
648 * variable-size data structure depending on the msg type itself
650 struct vmbus_channel_msginfo {
651 /* Bookkeeping stuff */
652 struct list_head msglistentry;
654 /* So far, this is only used to handle gpadl body message */
655 struct list_head submsglist;
657 /* Synchronize the request/response if needed */
658 struct completion waitevent;
659 struct vmbus_channel *waiting_channel;
661 struct vmbus_channel_version_supported version_supported;
662 struct vmbus_channel_open_result open_result;
663 struct vmbus_channel_gpadl_torndown gpadl_torndown;
664 struct vmbus_channel_gpadl_created gpadl_created;
665 struct vmbus_channel_version_response version_response;
670 * The channel message that goes out on the "wire".
671 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
673 unsigned char msg[0];
676 struct vmbus_close_msg {
677 struct vmbus_channel_msginfo info;
678 struct vmbus_channel_close_channel msg;
681 /* Define connection identifier type. */
682 union hv_connection_id {
690 enum hv_numa_policy {
695 enum vmbus_device_type {
715 struct vmbus_device {
721 struct vmbus_channel {
722 struct list_head listentry;
724 struct hv_device *device_obj;
726 enum vmbus_channel_state state;
728 struct vmbus_channel_offer_channel offermsg;
730 * These are based on the OfferMsg.MonitorId.
731 * Save it here for easy access.
736 bool rescind; /* got rescind msg */
737 struct completion rescind_event;
739 u32 ringbuffer_gpadlhandle;
741 /* Allocated memory for ring buffer */
742 struct page *ringbuffer_page;
743 u32 ringbuffer_pagecount;
744 u32 ringbuffer_send_offset;
745 struct hv_ring_buffer_info outbound; /* send to parent */
746 struct hv_ring_buffer_info inbound; /* receive from parent */
748 struct vmbus_close_msg close_msg;
751 u64 interrupts; /* Host to Guest interrupts */
752 u64 sig_events; /* Guest to Host events */
755 * Guest to host interrupts caused by the outbound ring buffer changing
756 * from empty to not empty.
761 * Indicates that a full outbound ring buffer was encountered. The flag
762 * is set to true when a full outbound ring buffer is encountered and
763 * set to false when a write to the outbound ring buffer is completed.
767 /* Channel callback's invoked in softirq context */
768 struct tasklet_struct callback_event;
769 void (*onchannel_callback)(void *context);
770 void *channel_callback_context;
773 * A channel can be marked for one of three modes of reading:
774 * BATCHED - callback called from taslket and should read
775 * channel until empty. Interrupts from the host
776 * are masked while read is in process (default).
777 * DIRECT - callback called from tasklet (softirq).
778 * ISR - callback called in interrupt context and must
779 * invoke its own deferred processing.
780 * Host interrupts are disabled and must be re-enabled
781 * when ring is empty.
783 enum hv_callback_mode {
789 bool is_dedicated_interrupt;
793 * Starting with win8, this field will be used to specify
794 * the target virtual processor on which to deliver the interrupt for
795 * the host to guest communication.
796 * Prior to win8, incoming channel interrupts would only
797 * be delivered on cpu 0. Setting this value to 0 would
798 * preserve the earlier behavior.
801 /* The corresponding CPUID in the guest */
804 * State to manage the CPU affiliation of channels.
806 struct cpumask alloced_cpus_in_node;
809 * Support for sub-channels. For high performance devices,
810 * it will be useful to have multiple sub-channels to support
811 * a scalable communication infrastructure with the host.
812 * The support for sub-channels is implemented as an extention
813 * to the current infrastructure.
814 * The initial offer is considered the primary channel and this
815 * offer message will indicate if the host supports sub-channels.
816 * The guest is free to ask for sub-channels to be offerred and can
817 * open these sub-channels as a normal "primary" channel. However,
818 * all sub-channels will have the same type and instance guids as the
819 * primary channel. Requests sent on a given channel will result in a
820 * response on the same channel.
824 * Sub-channel creation callback. This callback will be called in
825 * process context when a sub-channel offer is received from the host.
826 * The guest can open the sub-channel in the context of this callback.
828 void (*sc_creation_callback)(struct vmbus_channel *new_sc);
831 * Channel rescind callback. Some channels (the hvsock ones), need to
832 * register a callback which is invoked in vmbus_onoffer_rescind().
834 void (*chn_rescind_callback)(struct vmbus_channel *channel);
837 * The spinlock to protect the structure. It is being used to protect
838 * test-and-set access to various attributes of the structure as well
839 * as all sc_list operations.
843 * All Sub-channels of a primary channel are linked here.
845 struct list_head sc_list;
847 * The primary channel this sub-channel belongs to.
848 * This will be NULL for the primary channel.
850 struct vmbus_channel *primary_channel;
852 * Support per-channel state for use by vmbus drivers.
854 void *per_channel_state;
856 * To support per-cpu lookup mapping of relid to channel,
857 * link up channels based on their CPU affinity.
859 struct list_head percpu_list;
862 * Defer freeing channel until after all cpu's have
863 * gone through grace period.
868 * For sysfs per-channel properties.
873 * For performance critical channels (storage, networking
874 * etc,), Hyper-V has a mechanism to enhance the throughput
875 * at the expense of latency:
876 * When the host is to be signaled, we just set a bit in a shared page
877 * and this bit will be inspected by the hypervisor within a certain
878 * window and if the bit is set, the host will be signaled. The window
879 * of time is the monitor latency - currently around 100 usecs. This
880 * mechanism improves throughput by:
882 * A) Making the host more efficient - each time it wakes up,
883 * potentially it will process morev number of packets. The
884 * monitor latency allows a batch to build up.
885 * B) By deferring the hypercall to signal, we will also minimize
888 * Clearly, these optimizations improve throughput at the expense of
889 * latency. Furthermore, since the channel is shared for both
890 * control and data messages, control messages currently suffer
891 * unnecessary latency adversley impacting performance and boot
892 * time. To fix this issue, permit tagging the channel as being
893 * in "low latency" mode. In this mode, we will bypass the monitor
899 * NUMA distribution policy:
900 * We support two policies:
901 * 1) Balanced: Here all performance critical channels are
902 * distributed evenly amongst all the NUMA nodes.
903 * This policy will be the default policy.
904 * 2) Localized: All channels of a given instance of a
905 * performance critical service will be assigned CPUs
906 * within a selected NUMA node.
908 enum hv_numa_policy affinity_policy;
913 * We must offload the handling of the primary/sub channels
914 * from the single-threaded vmbus_connection.work_queue to
915 * two different workqueue, otherwise we can block
916 * vmbus_connection.work_queue and hang: see vmbus_process_offer().
918 struct work_struct add_channel_work;
921 * Guest to host interrupts caused by the inbound ring buffer changing
922 * from full to not full while a packet is waiting.
927 * The total number of write operations that encountered a full
928 * outbound ring buffer.
933 * The number of write operations that were the first to encounter a
934 * full outbound ring buffer.
938 /* enabling/disabling fuzz testing on the channel (default is false)*/
939 bool fuzz_testing_state;
942 * Interrupt delay will delay the guest from emptying the ring buffer
943 * for a specific amount of time. The delay is in microseconds and will
944 * be between 1 to a maximum of 1000, its default is 0 (no delay).
945 * The Message delay will delay guest reading on a per message basis
946 * in microseconds between 1 to 1000 with the default being 0
949 u32 fuzz_testing_interrupt_delay;
950 u32 fuzz_testing_message_delay;
954 static inline bool is_hvsock_channel(const struct vmbus_channel *c)
956 return !!(c->offermsg.offer.chn_flags &
957 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER);
960 static inline bool is_sub_channel(const struct vmbus_channel *c)
962 return c->offermsg.offer.sub_channel_index != 0;
965 static inline void set_channel_affinity_state(struct vmbus_channel *c,
966 enum hv_numa_policy policy)
968 c->affinity_policy = policy;
971 static inline void set_channel_read_mode(struct vmbus_channel *c,
972 enum hv_callback_mode mode)
974 c->callback_mode = mode;
977 static inline void set_per_channel_state(struct vmbus_channel *c, void *s)
979 c->per_channel_state = s;
982 static inline void *get_per_channel_state(struct vmbus_channel *c)
984 return c->per_channel_state;
987 static inline void set_channel_pending_send_size(struct vmbus_channel *c,
993 spin_lock_irqsave(&c->outbound.ring_lock, flags);
996 if (!c->out_full_flag) {
998 c->out_full_flag = true;
1000 spin_unlock_irqrestore(&c->outbound.ring_lock, flags);
1002 c->out_full_flag = false;
1005 c->outbound.ring_buffer->pending_send_sz = size;
1008 static inline void set_low_latency_mode(struct vmbus_channel *c)
1010 c->low_latency = true;
1013 static inline void clear_low_latency_mode(struct vmbus_channel *c)
1015 c->low_latency = false;
1018 void vmbus_onmessage(void *context);
1020 int vmbus_request_offers(void);
1023 * APIs for managing sub-channels.
1026 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1027 void (*sc_cr_cb)(struct vmbus_channel *new_sc));
1029 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1030 void (*chn_rescind_cb)(struct vmbus_channel *));
1033 * Check if sub-channels have already been offerred. This API will be useful
1034 * when the driver is unloaded after establishing sub-channels. In this case,
1035 * when the driver is re-loaded, the driver would have to check if the
1036 * subchannels have already been established before attempting to request
1037 * the creation of sub-channels.
1038 * This function returns TRUE to indicate that subchannels have already been
1040 * This function should be invoked after setting the callback function for
1041 * sub-channel creation.
1043 bool vmbus_are_subchannels_present(struct vmbus_channel *primary);
1045 /* The format must be the same as struct vmdata_gpa_direct */
1046 struct vmbus_channel_packet_page_buffer {
1054 struct hv_page_buffer range[MAX_PAGE_BUFFER_COUNT];
1057 /* The format must be the same as struct vmdata_gpa_direct */
1058 struct vmbus_channel_packet_multipage_buffer {
1065 u32 rangecount; /* Always 1 in this case */
1066 struct hv_multipage_buffer range;
1069 /* The format must be the same as struct vmdata_gpa_direct */
1070 struct vmbus_packet_mpb_array {
1077 u32 rangecount; /* Always 1 in this case */
1078 struct hv_mpb_array range;
1081 int vmbus_alloc_ring(struct vmbus_channel *channel,
1082 u32 send_size, u32 recv_size);
1083 void vmbus_free_ring(struct vmbus_channel *channel);
1085 int vmbus_connect_ring(struct vmbus_channel *channel,
1086 void (*onchannel_callback)(void *context),
1088 int vmbus_disconnect_ring(struct vmbus_channel *channel);
1090 extern int vmbus_open(struct vmbus_channel *channel,
1091 u32 send_ringbuffersize,
1092 u32 recv_ringbuffersize,
1095 void (*onchannel_callback)(void *context),
1098 extern void vmbus_close(struct vmbus_channel *channel);
1100 extern int vmbus_sendpacket(struct vmbus_channel *channel,
1104 enum vmbus_packet_type type,
1107 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1108 struct hv_page_buffer pagebuffers[],
1114 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1115 struct vmbus_packet_mpb_array *mpb,
1121 extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
1126 extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
1129 void vmbus_reset_channel_cb(struct vmbus_channel *channel);
1131 extern int vmbus_recvpacket(struct vmbus_channel *channel,
1134 u32 *buffer_actual_len,
1137 extern int vmbus_recvpacket_raw(struct vmbus_channel *channel,
1140 u32 *buffer_actual_len,
1144 extern void vmbus_ontimer(unsigned long data);
1146 /* Base driver object */
1151 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1152 * channel flag, actually doesn't mean a synthetic device because the
1153 * offer's if_type/if_instance can change for every new hvsock
1156 * However, to facilitate the notification of new-offer/rescind-offer
1157 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1158 * a special vmbus device, and hence we need the below flag to
1159 * indicate if the driver is the hvsock driver or not: we need to
1160 * specially treat the hvosck offer & driver in vmbus_match().
1164 /* the device type supported by this driver */
1166 const struct hv_vmbus_device_id *id_table;
1168 struct device_driver driver;
1170 /* dynamic device GUID's */
1173 struct list_head list;
1176 int (*probe)(struct hv_device *, const struct hv_vmbus_device_id *);
1177 int (*remove)(struct hv_device *);
1178 void (*shutdown)(struct hv_device *);
1180 int (*suspend)(struct hv_device *);
1181 int (*resume)(struct hv_device *);
1185 /* Base device object */
1187 /* the device type id of this device */
1190 /* the device instance id of this device */
1191 guid_t dev_instance;
1195 struct device device;
1196 char *driver_override; /* Driver name to force a match */
1198 struct vmbus_channel *channel;
1199 struct kset *channels_kset;
1201 /* place holder to keep track of the dir for hv device in debugfs */
1202 struct dentry *debug_dir;
1207 static inline struct hv_device *device_to_hv_device(struct device *d)
1209 return container_of(d, struct hv_device, device);
1212 static inline struct hv_driver *drv_to_hv_drv(struct device_driver *d)
1214 return container_of(d, struct hv_driver, driver);
1217 static inline void hv_set_drvdata(struct hv_device *dev, void *data)
1219 dev_set_drvdata(&dev->device, data);
1222 static inline void *hv_get_drvdata(struct hv_device *dev)
1224 return dev_get_drvdata(&dev->device);
1227 struct hv_ring_buffer_debug_info {
1228 u32 current_interrupt_mask;
1229 u32 current_read_index;
1230 u32 current_write_index;
1231 u32 bytes_avail_toread;
1232 u32 bytes_avail_towrite;
1236 int hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
1237 struct hv_ring_buffer_debug_info *debug_info);
1239 /* Vmbus interface */
1240 #define vmbus_driver_register(driver) \
1241 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1242 int __must_check __vmbus_driver_register(struct hv_driver *hv_driver,
1243 struct module *owner,
1244 const char *mod_name);
1245 void vmbus_driver_unregister(struct hv_driver *hv_driver);
1247 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel);
1249 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1250 resource_size_t min, resource_size_t max,
1251 resource_size_t size, resource_size_t align,
1252 bool fb_overlap_ok);
1253 void vmbus_free_mmio(resource_size_t start, resource_size_t size);
1256 * GUID definitions of various offer types - services offered to the guest.
1261 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1263 #define HV_NIC_GUID \
1264 .guid = GUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1265 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1269 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1271 #define HV_IDE_GUID \
1272 .guid = GUID_INIT(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1273 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1277 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1279 #define HV_SCSI_GUID \
1280 .guid = GUID_INIT(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1281 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1285 * {0e0b6031-5213-4934-818b-38d90ced39db}
1287 #define HV_SHUTDOWN_GUID \
1288 .guid = GUID_INIT(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1289 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1293 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1295 #define HV_TS_GUID \
1296 .guid = GUID_INIT(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1297 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1301 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1303 #define HV_HEART_BEAT_GUID \
1304 .guid = GUID_INIT(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1305 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1309 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1311 #define HV_KVP_GUID \
1312 .guid = GUID_INIT(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1313 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1316 * Dynamic memory GUID
1317 * {525074dc-8985-46e2-8057-a307dc18a502}
1319 #define HV_DM_GUID \
1320 .guid = GUID_INIT(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1321 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1325 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1327 #define HV_MOUSE_GUID \
1328 .guid = GUID_INIT(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1329 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1333 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1335 #define HV_KBD_GUID \
1336 .guid = GUID_INIT(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1337 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1340 * VSS (Backup/Restore) GUID
1342 #define HV_VSS_GUID \
1343 .guid = GUID_INIT(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1344 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1346 * Synthetic Video GUID
1347 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1349 #define HV_SYNTHVID_GUID \
1350 .guid = GUID_INIT(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1351 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1355 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1357 #define HV_SYNTHFC_GUID \
1358 .guid = GUID_INIT(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1359 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1362 * Guest File Copy Service
1363 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1366 #define HV_FCOPY_GUID \
1367 .guid = GUID_INIT(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1368 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1371 * NetworkDirect. This is the guest RDMA service.
1372 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1374 #define HV_ND_GUID \
1375 .guid = GUID_INIT(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1376 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1379 * PCI Express Pass Through
1380 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1383 #define HV_PCIE_GUID \
1384 .guid = GUID_INIT(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1385 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1388 * Linux doesn't support the 3 devices: the first two are for
1389 * Automatic Virtual Machine Activation, and the third is for
1390 * Remote Desktop Virtualization.
1391 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1392 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1393 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1396 #define HV_AVMA1_GUID \
1397 .guid = GUID_INIT(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1398 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1400 #define HV_AVMA2_GUID \
1401 .guid = GUID_INIT(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1402 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1404 #define HV_RDV_GUID \
1405 .guid = GUID_INIT(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1406 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1409 * Common header for Hyper-V ICs
1412 #define ICMSGTYPE_NEGOTIATE 0
1413 #define ICMSGTYPE_HEARTBEAT 1
1414 #define ICMSGTYPE_KVPEXCHANGE 2
1415 #define ICMSGTYPE_SHUTDOWN 3
1416 #define ICMSGTYPE_TIMESYNC 4
1417 #define ICMSGTYPE_VSS 5
1419 #define ICMSGHDRFLAG_TRANSACTION 1
1420 #define ICMSGHDRFLAG_REQUEST 2
1421 #define ICMSGHDRFLAG_RESPONSE 4
1425 * While we want to handle util services as regular devices,
1426 * there is only one instance of each of these services; so
1427 * we statically allocate the service specific state.
1430 struct hv_util_service {
1433 void (*util_cb)(void *);
1434 int (*util_init)(struct hv_util_service *);
1435 void (*util_deinit)(void);
1438 struct vmbuspipe_hdr {
1449 struct ic_version icverframe;
1451 struct ic_version icvermsg;
1454 u8 ictransaction_id;
1459 struct icmsg_negotiate {
1463 struct ic_version icversion_data[1]; /* any size array */
1466 struct shutdown_msg_data {
1468 u32 timeout_seconds;
1470 u8 display_message[2048];
1473 struct heartbeat_msg_data {
1478 /* Time Sync IC defs */
1479 #define ICTIMESYNCFLAG_PROBE 0
1480 #define ICTIMESYNCFLAG_SYNC 1
1481 #define ICTIMESYNCFLAG_SAMPLE 2
1484 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1486 #define WLTIMEDELTA 116444736000000000LL
1489 struct ictimesync_data {
1496 struct ictimesync_ref_data {
1498 u64 vmreferencetime;
1505 struct hyperv_service_callback {
1509 struct vmbus_channel *channel;
1510 void (*callback)(void *context);
1513 #define MAX_SRV_VER 0x7ffffff
1514 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf,
1515 const int *fw_version, int fw_vercnt,
1516 const int *srv_version, int srv_vercnt,
1517 int *nego_fw_version, int *nego_srv_version);
1519 void hv_process_channel_removal(struct vmbus_channel *channel);
1521 void vmbus_setevent(struct vmbus_channel *channel);
1523 * Negotiated version with the Host.
1526 extern __u32 vmbus_proto_version;
1528 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
1529 const guid_t *shv_host_servie_id);
1530 void vmbus_set_event(struct vmbus_channel *channel);
1532 /* Get the start of the ring buffer. */
1533 static inline void *
1534 hv_get_ring_buffer(const struct hv_ring_buffer_info *ring_info)
1536 return ring_info->ring_buffer->buffer;
1540 * Mask off host interrupt callback notifications
1542 static inline void hv_begin_read(struct hv_ring_buffer_info *rbi)
1544 rbi->ring_buffer->interrupt_mask = 1;
1546 /* make sure mask update is not reordered */
1551 * Re-enable host callback and return number of outstanding bytes
1553 static inline u32 hv_end_read(struct hv_ring_buffer_info *rbi)
1556 rbi->ring_buffer->interrupt_mask = 0;
1558 /* make sure mask update is not reordered */
1562 * Now check to see if the ring buffer is still empty.
1563 * If it is not, we raced and we need to process new
1564 * incoming messages.
1566 return hv_get_bytes_to_read(rbi);
1570 * An API to support in-place processing of incoming VMBUS packets.
1573 /* Get data payload associated with descriptor */
1574 static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc)
1576 return (void *)((unsigned long)desc + (desc->offset8 << 3));
1579 /* Get data size associated with descriptor */
1580 static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc)
1582 return (desc->len8 << 3) - (desc->offset8 << 3);
1586 struct vmpacket_descriptor *
1587 hv_pkt_iter_first(struct vmbus_channel *channel);
1589 struct vmpacket_descriptor *
1590 __hv_pkt_iter_next(struct vmbus_channel *channel,
1591 const struct vmpacket_descriptor *pkt);
1593 void hv_pkt_iter_close(struct vmbus_channel *channel);
1596 * Get next packet descriptor from iterator
1597 * If at end of list, return NULL and update host.
1599 static inline struct vmpacket_descriptor *
1600 hv_pkt_iter_next(struct vmbus_channel *channel,
1601 const struct vmpacket_descriptor *pkt)
1603 struct vmpacket_descriptor *nxt;
1605 nxt = __hv_pkt_iter_next(channel, pkt);
1607 hv_pkt_iter_close(channel);
1612 #define foreach_vmbus_pkt(pkt, channel) \
1613 for (pkt = hv_pkt_iter_first(channel); pkt; \
1614 pkt = hv_pkt_iter_next(channel, pkt))
1617 * Interface for passing data between SR-IOV PF and VF drivers. The VF driver
1618 * sends requests to read and write blocks. Each block must be 128 bytes or
1619 * smaller. Optionally, the VF driver can register a callback function which
1620 * will be invoked when the host says that one or more of the first 64 block
1621 * IDs is "invalid" which means that the VF driver should reread them.
1623 #define HV_CONFIG_BLOCK_SIZE_MAX 128
1625 int hyperv_read_cfg_blk(struct pci_dev *dev, void *buf, unsigned int buf_len,
1626 unsigned int block_id, unsigned int *bytes_returned);
1627 int hyperv_write_cfg_blk(struct pci_dev *dev, void *buf, unsigned int len,
1628 unsigned int block_id);
1629 int hyperv_reg_block_invalidate(struct pci_dev *dev, void *context,
1630 void (*block_invalidate)(void *context,
1633 struct hyperv_pci_block_ops {
1634 int (*read_block)(struct pci_dev *dev, void *buf, unsigned int buf_len,
1635 unsigned int block_id, unsigned int *bytes_returned);
1636 int (*write_block)(struct pci_dev *dev, void *buf, unsigned int len,
1637 unsigned int block_id);
1638 int (*reg_blk_invalidate)(struct pci_dev *dev, void *context,
1639 void (*block_invalidate)(void *context,
1643 extern struct hyperv_pci_block_ops hvpci_block_ops;
1645 #endif /* _HYPERV_H */