1 /* SPDX-License-Identifier: GPL-2.0-only */
4 * Copyright (c) 2011, Microsoft Corporation.
7 * Haiyang Zhang <haiyangz@microsoft.com>
8 * Hank Janssen <hjanssen@microsoft.com>
9 * K. Y. Srinivasan <kys@microsoft.com>
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 * 5 . 0 (Newer Windows 10)
188 #define VERSION_WS2008 ((0 << 16) | (13))
189 #define VERSION_WIN7 ((1 << 16) | (1))
190 #define VERSION_WIN8 ((2 << 16) | (4))
191 #define VERSION_WIN8_1 ((3 << 16) | (0))
192 #define VERSION_WIN10 ((4 << 16) | (0))
193 #define VERSION_WIN10_V5 ((5 << 16) | (0))
195 #define VERSION_INVAL -1
197 #define VERSION_CURRENT VERSION_WIN10_V5
199 /* Make maximum size of pipe payload of 16K */
200 #define MAX_PIPE_DATA_PAYLOAD (sizeof(u8) * 16384)
202 /* Define PipeMode values. */
203 #define VMBUS_PIPE_TYPE_BYTE 0x00000000
204 #define VMBUS_PIPE_TYPE_MESSAGE 0x00000004
206 /* The size of the user defined data buffer for non-pipe offers. */
207 #define MAX_USER_DEFINED_BYTES 120
209 /* The size of the user defined data buffer for pipe offers. */
210 #define MAX_PIPE_USER_DEFINED_BYTES 116
213 * At the center of the Channel Management library is the Channel Offer. This
214 * struct contains the fundamental information about an offer.
216 struct vmbus_channel_offer
{
221 * These two fields are not currently used.
227 u16 mmio_megabytes
; /* in bytes * 1024 * 1024 */
230 /* Non-pipes: The user has MAX_USER_DEFINED_BYTES bytes. */
232 unsigned char user_def
[MAX_USER_DEFINED_BYTES
];
237 * The following sructure is an integrated pipe protocol, which
238 * is implemented on top of standard user-defined data. Pipe
239 * clients have MAX_PIPE_USER_DEFINED_BYTES left for their own
244 unsigned char user_def
[MAX_PIPE_USER_DEFINED_BYTES
];
248 * The sub_channel_index is defined in win8.
250 u16 sub_channel_index
;
255 #define VMBUS_CHANNEL_ENUMERATE_DEVICE_INTERFACE 1
256 #define VMBUS_CHANNEL_SERVER_SUPPORTS_TRANSFER_PAGES 2
257 #define VMBUS_CHANNEL_SERVER_SUPPORTS_GPADLS 4
258 #define VMBUS_CHANNEL_NAMED_PIPE_MODE 0x10
259 #define VMBUS_CHANNEL_LOOPBACK_OFFER 0x100
260 #define VMBUS_CHANNEL_PARENT_OFFER 0x200
261 #define VMBUS_CHANNEL_REQUEST_MONITORED_NOTIFICATION 0x400
262 #define VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER 0x2000
264 struct vmpacket_descriptor
{
272 struct vmpacket_header
{
273 u32 prev_pkt_start_offset
;
274 struct vmpacket_descriptor descriptor
;
277 struct vmtransfer_page_range
{
282 struct vmtransfer_page_packet_header
{
283 struct vmpacket_descriptor d
;
288 struct vmtransfer_page_range ranges
[1];
291 struct vmgpadl_packet_header
{
292 struct vmpacket_descriptor d
;
297 struct vmadd_remove_transfer_page_set
{
298 struct vmpacket_descriptor d
;
305 * This structure defines a range in guest physical space that can be made to
306 * look virtually contiguous.
315 * This is the format for an Establish Gpadl packet, which contains a handle by
316 * which this GPADL will be known and a set of GPA ranges associated with it.
317 * This can be converted to a MDL by the guest OS. If there are multiple GPA
318 * ranges, then the resulting MDL will be "chained," representing multiple VA
321 struct vmestablish_gpadl
{
322 struct vmpacket_descriptor d
;
325 struct gpa_range range
[1];
329 * This is the format for a Teardown Gpadl packet, which indicates that the
330 * GPADL handle in the Establish Gpadl packet will never be referenced again.
332 struct vmteardown_gpadl
{
333 struct vmpacket_descriptor d
;
335 u32 reserved
; /* for alignment to a 8-byte boundary */
339 * This is the format for a GPA-Direct packet, which contains a set of GPA
340 * ranges, in addition to commands and/or data.
342 struct vmdata_gpa_direct
{
343 struct vmpacket_descriptor d
;
346 struct gpa_range range
[1];
349 /* This is the format for a Additional Data Packet. */
350 struct vmadditional_data
{
351 struct vmpacket_descriptor d
;
355 unsigned char data
[1];
358 union vmpacket_largest_possible_header
{
359 struct vmpacket_descriptor simple_hdr
;
360 struct vmtransfer_page_packet_header xfer_page_hdr
;
361 struct vmgpadl_packet_header gpadl_hdr
;
362 struct vmadd_remove_transfer_page_set add_rm_xfer_page_hdr
;
363 struct vmestablish_gpadl establish_gpadl_hdr
;
364 struct vmteardown_gpadl teardown_gpadl_hdr
;
365 struct vmdata_gpa_direct data_gpa_direct_hdr
;
368 #define VMPACKET_DATA_START_ADDRESS(__packet) \
369 (void *)(((unsigned char *)__packet) + \
370 ((struct vmpacket_descriptor)__packet)->offset8 * 8)
372 #define VMPACKET_DATA_LENGTH(__packet) \
373 ((((struct vmpacket_descriptor)__packet)->len8 - \
374 ((struct vmpacket_descriptor)__packet)->offset8) * 8)
376 #define VMPACKET_TRANSFER_MODE(__packet) \
377 (((struct IMPACT)__packet)->type)
379 enum vmbus_packet_type
{
380 VM_PKT_INVALID
= 0x0,
382 VM_PKT_ADD_XFER_PAGESET
= 0x2,
383 VM_PKT_RM_XFER_PAGESET
= 0x3,
384 VM_PKT_ESTABLISH_GPADL
= 0x4,
385 VM_PKT_TEARDOWN_GPADL
= 0x5,
386 VM_PKT_DATA_INBAND
= 0x6,
387 VM_PKT_DATA_USING_XFER_PAGES
= 0x7,
388 VM_PKT_DATA_USING_GPADL
= 0x8,
389 VM_PKT_DATA_USING_GPA_DIRECT
= 0x9,
390 VM_PKT_CANCEL_REQUEST
= 0xa,
392 VM_PKT_DATA_USING_ADDITIONAL_PKT
= 0xc,
393 VM_PKT_ADDITIONAL_DATA
= 0xd
396 #define VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED 1
399 /* Version 1 messages */
400 enum vmbus_channel_message_type
{
401 CHANNELMSG_INVALID
= 0,
402 CHANNELMSG_OFFERCHANNEL
= 1,
403 CHANNELMSG_RESCIND_CHANNELOFFER
= 2,
404 CHANNELMSG_REQUESTOFFERS
= 3,
405 CHANNELMSG_ALLOFFERS_DELIVERED
= 4,
406 CHANNELMSG_OPENCHANNEL
= 5,
407 CHANNELMSG_OPENCHANNEL_RESULT
= 6,
408 CHANNELMSG_CLOSECHANNEL
= 7,
409 CHANNELMSG_GPADL_HEADER
= 8,
410 CHANNELMSG_GPADL_BODY
= 9,
411 CHANNELMSG_GPADL_CREATED
= 10,
412 CHANNELMSG_GPADL_TEARDOWN
= 11,
413 CHANNELMSG_GPADL_TORNDOWN
= 12,
414 CHANNELMSG_RELID_RELEASED
= 13,
415 CHANNELMSG_INITIATE_CONTACT
= 14,
416 CHANNELMSG_VERSION_RESPONSE
= 15,
417 CHANNELMSG_UNLOAD
= 16,
418 CHANNELMSG_UNLOAD_RESPONSE
= 17,
422 CHANNELMSG_TL_CONNECT_REQUEST
= 21,
426 struct vmbus_channel_message_header
{
427 enum vmbus_channel_message_type msgtype
;
431 /* Query VMBus Version parameters */
432 struct vmbus_channel_query_vmbus_version
{
433 struct vmbus_channel_message_header header
;
437 /* VMBus Version Supported parameters */
438 struct vmbus_channel_version_supported
{
439 struct vmbus_channel_message_header header
;
440 u8 version_supported
;
443 /* Offer Channel parameters */
444 struct vmbus_channel_offer_channel
{
445 struct vmbus_channel_message_header header
;
446 struct vmbus_channel_offer offer
;
450 * win7 and beyond splits this field into a bit field.
452 u8 monitor_allocated
:1;
455 * These are new fields added in win7 and later.
456 * Do not access these fields without checking the
457 * negotiated protocol.
459 * If "is_dedicated_interrupt" is set, we must not set the
460 * associated bit in the channel bitmap while sending the
461 * interrupt to the host.
463 * connection_id is to be used in signaling the host.
465 u16 is_dedicated_interrupt
:1;
470 /* Rescind Offer parameters */
471 struct vmbus_channel_rescind_offer
{
472 struct vmbus_channel_message_header header
;
477 hv_ringbuffer_pending_size(const struct hv_ring_buffer_info
*rbi
)
479 return rbi
->ring_buffer
->pending_send_sz
;
483 * Request Offer -- no parameters, SynIC message contains the partition ID
484 * Set Snoop -- no parameters, SynIC message contains the partition ID
485 * Clear Snoop -- no parameters, SynIC message contains the partition ID
486 * All Offers Delivered -- no parameters, SynIC message contains the partition
488 * Flush Client -- no parameters, SynIC message contains the partition ID
491 /* Open Channel parameters */
492 struct vmbus_channel_open_channel
{
493 struct vmbus_channel_message_header header
;
495 /* Identifies the specific VMBus channel that is being opened. */
498 /* ID making a particular open request at a channel offer unique. */
501 /* GPADL for the channel's ring buffer. */
502 u32 ringbuffer_gpadlhandle
;
505 * Starting with win8, this field will be used to specify
506 * the target virtual processor on which to deliver the interrupt for
507 * the host to guest communication.
508 * Prior to win8, incoming channel interrupts would only
509 * be delivered on cpu 0. Setting this value to 0 would
510 * preserve the earlier behavior.
515 * The upstream ring buffer begins at offset zero in the memory
516 * described by RingBufferGpadlHandle. The downstream ring buffer
517 * follows it at this offset (in pages).
519 u32 downstream_ringbuffer_pageoffset
;
521 /* User-specific data to be passed along to the server endpoint. */
522 unsigned char userdata
[MAX_USER_DEFINED_BYTES
];
525 /* Open Channel Result parameters */
526 struct vmbus_channel_open_result
{
527 struct vmbus_channel_message_header header
;
533 /* Close channel parameters; */
534 struct vmbus_channel_close_channel
{
535 struct vmbus_channel_message_header header
;
539 /* Channel Message GPADL */
540 #define GPADL_TYPE_RING_BUFFER 1
541 #define GPADL_TYPE_SERVER_SAVE_AREA 2
542 #define GPADL_TYPE_TRANSACTION 8
545 * The number of PFNs in a GPADL message is defined by the number of
546 * pages that would be spanned by ByteCount and ByteOffset. If the
547 * implied number of PFNs won't fit in this packet, there will be a
548 * follow-up packet that contains more.
550 struct vmbus_channel_gpadl_header
{
551 struct vmbus_channel_message_header header
;
556 struct gpa_range range
[0];
559 /* This is the followup packet that contains more PFNs. */
560 struct vmbus_channel_gpadl_body
{
561 struct vmbus_channel_message_header header
;
567 struct vmbus_channel_gpadl_created
{
568 struct vmbus_channel_message_header header
;
574 struct vmbus_channel_gpadl_teardown
{
575 struct vmbus_channel_message_header header
;
580 struct vmbus_channel_gpadl_torndown
{
581 struct vmbus_channel_message_header header
;
585 struct vmbus_channel_relid_released
{
586 struct vmbus_channel_message_header header
;
590 struct vmbus_channel_initiate_contact
{
591 struct vmbus_channel_message_header header
;
592 u32 vmbus_version_requested
;
593 u32 target_vcpu
; /* The VCPU the host should respond to */
606 /* Hyper-V socket: guest's connect()-ing to host */
607 struct vmbus_channel_tl_connect_request
{
608 struct vmbus_channel_message_header header
;
609 guid_t guest_endpoint_id
;
610 guid_t host_service_id
;
613 struct vmbus_channel_version_response
{
614 struct vmbus_channel_message_header header
;
615 u8 version_supported
;
621 * On new hosts that support VMBus protocol 5.0, we must use
622 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
623 * and for subsequent messages, we must use the Message Connection ID
624 * field in the host-returned Version Response Message.
626 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
631 enum vmbus_channel_state
{
633 CHANNEL_OPENING_STATE
,
635 CHANNEL_OPENED_STATE
,
639 * Represents each channel msg on the vmbus connection This is a
640 * variable-size data structure depending on the msg type itself
642 struct vmbus_channel_msginfo
{
643 /* Bookkeeping stuff */
644 struct list_head msglistentry
;
646 /* So far, this is only used to handle gpadl body message */
647 struct list_head submsglist
;
649 /* Synchronize the request/response if needed */
650 struct completion waitevent
;
651 struct vmbus_channel
*waiting_channel
;
653 struct vmbus_channel_version_supported version_supported
;
654 struct vmbus_channel_open_result open_result
;
655 struct vmbus_channel_gpadl_torndown gpadl_torndown
;
656 struct vmbus_channel_gpadl_created gpadl_created
;
657 struct vmbus_channel_version_response version_response
;
662 * The channel message that goes out on the "wire".
663 * It will contain at minimum the VMBUS_CHANNEL_MESSAGE_HEADER header
665 unsigned char msg
[0];
668 struct vmbus_close_msg
{
669 struct vmbus_channel_msginfo info
;
670 struct vmbus_channel_close_channel msg
;
673 /* Define connection identifier type. */
674 union hv_connection_id
{
682 enum hv_numa_policy
{
687 enum vmbus_device_type
{
707 struct vmbus_device
{
713 struct vmbus_channel
{
714 struct list_head listentry
;
716 struct hv_device
*device_obj
;
718 enum vmbus_channel_state state
;
720 struct vmbus_channel_offer_channel offermsg
;
722 * These are based on the OfferMsg.MonitorId.
723 * Save it here for easy access.
728 bool rescind
; /* got rescind msg */
729 struct completion rescind_event
;
731 u32 ringbuffer_gpadlhandle
;
733 /* Allocated memory for ring buffer */
734 struct page
*ringbuffer_page
;
735 u32 ringbuffer_pagecount
;
736 u32 ringbuffer_send_offset
;
737 struct hv_ring_buffer_info outbound
; /* send to parent */
738 struct hv_ring_buffer_info inbound
; /* receive from parent */
740 struct vmbus_close_msg close_msg
;
743 u64 interrupts
; /* Host to Guest interrupts */
744 u64 sig_events
; /* Guest to Host events */
747 * Guest to host interrupts caused by the outbound ring buffer changing
748 * from empty to not empty.
753 * Indicates that a full outbound ring buffer was encountered. The flag
754 * is set to true when a full outbound ring buffer is encountered and
755 * set to false when a write to the outbound ring buffer is completed.
759 /* Channel callback's invoked in softirq context */
760 struct tasklet_struct callback_event
;
761 void (*onchannel_callback
)(void *context
);
762 void *channel_callback_context
;
765 * A channel can be marked for one of three modes of reading:
766 * BATCHED - callback called from taslket and should read
767 * channel until empty. Interrupts from the host
768 * are masked while read is in process (default).
769 * DIRECT - callback called from tasklet (softirq).
770 * ISR - callback called in interrupt context and must
771 * invoke its own deferred processing.
772 * Host interrupts are disabled and must be re-enabled
773 * when ring is empty.
775 enum hv_callback_mode
{
781 bool is_dedicated_interrupt
;
785 * Starting with win8, this field will be used to specify
786 * the target virtual processor on which to deliver the interrupt for
787 * the host to guest communication.
788 * Prior to win8, incoming channel interrupts would only
789 * be delivered on cpu 0. Setting this value to 0 would
790 * preserve the earlier behavior.
793 /* The corresponding CPUID in the guest */
796 * State to manage the CPU affiliation of channels.
798 struct cpumask alloced_cpus_in_node
;
801 * Support for sub-channels. For high performance devices,
802 * it will be useful to have multiple sub-channels to support
803 * a scalable communication infrastructure with the host.
804 * The support for sub-channels is implemented as an extention
805 * to the current infrastructure.
806 * The initial offer is considered the primary channel and this
807 * offer message will indicate if the host supports sub-channels.
808 * The guest is free to ask for sub-channels to be offerred and can
809 * open these sub-channels as a normal "primary" channel. However,
810 * all sub-channels will have the same type and instance guids as the
811 * primary channel. Requests sent on a given channel will result in a
812 * response on the same channel.
816 * Sub-channel creation callback. This callback will be called in
817 * process context when a sub-channel offer is received from the host.
818 * The guest can open the sub-channel in the context of this callback.
820 void (*sc_creation_callback
)(struct vmbus_channel
*new_sc
);
823 * Channel rescind callback. Some channels (the hvsock ones), need to
824 * register a callback which is invoked in vmbus_onoffer_rescind().
826 void (*chn_rescind_callback
)(struct vmbus_channel
*channel
);
829 * The spinlock to protect the structure. It is being used to protect
830 * test-and-set access to various attributes of the structure as well
831 * as all sc_list operations.
835 * All Sub-channels of a primary channel are linked here.
837 struct list_head sc_list
;
839 * The primary channel this sub-channel belongs to.
840 * This will be NULL for the primary channel.
842 struct vmbus_channel
*primary_channel
;
844 * Support per-channel state for use by vmbus drivers.
846 void *per_channel_state
;
848 * To support per-cpu lookup mapping of relid to channel,
849 * link up channels based on their CPU affinity.
851 struct list_head percpu_list
;
854 * Defer freeing channel until after all cpu's have
855 * gone through grace period.
860 * For sysfs per-channel properties.
865 * For performance critical channels (storage, networking
866 * etc,), Hyper-V has a mechanism to enhance the throughput
867 * at the expense of latency:
868 * When the host is to be signaled, we just set a bit in a shared page
869 * and this bit will be inspected by the hypervisor within a certain
870 * window and if the bit is set, the host will be signaled. The window
871 * of time is the monitor latency - currently around 100 usecs. This
872 * mechanism improves throughput by:
874 * A) Making the host more efficient - each time it wakes up,
875 * potentially it will process morev number of packets. The
876 * monitor latency allows a batch to build up.
877 * B) By deferring the hypercall to signal, we will also minimize
880 * Clearly, these optimizations improve throughput at the expense of
881 * latency. Furthermore, since the channel is shared for both
882 * control and data messages, control messages currently suffer
883 * unnecessary latency adversley impacting performance and boot
884 * time. To fix this issue, permit tagging the channel as being
885 * in "low latency" mode. In this mode, we will bypass the monitor
891 * NUMA distribution policy:
892 * We support two policies:
893 * 1) Balanced: Here all performance critical channels are
894 * distributed evenly amongst all the NUMA nodes.
895 * This policy will be the default policy.
896 * 2) Localized: All channels of a given instance of a
897 * performance critical service will be assigned CPUs
898 * within a selected NUMA node.
900 enum hv_numa_policy affinity_policy
;
905 * We must offload the handling of the primary/sub channels
906 * from the single-threaded vmbus_connection.work_queue to
907 * two different workqueue, otherwise we can block
908 * vmbus_connection.work_queue and hang: see vmbus_process_offer().
910 struct work_struct add_channel_work
;
913 * Guest to host interrupts caused by the inbound ring buffer changing
914 * from full to not full while a packet is waiting.
919 * The total number of write operations that encountered a full
920 * outbound ring buffer.
925 * The number of write operations that were the first to encounter a
926 * full outbound ring buffer.
931 static inline bool is_hvsock_channel(const struct vmbus_channel
*c
)
933 return !!(c
->offermsg
.offer
.chn_flags
&
934 VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
);
937 static inline void set_channel_affinity_state(struct vmbus_channel
*c
,
938 enum hv_numa_policy policy
)
940 c
->affinity_policy
= policy
;
943 static inline void set_channel_read_mode(struct vmbus_channel
*c
,
944 enum hv_callback_mode mode
)
946 c
->callback_mode
= mode
;
949 static inline void set_per_channel_state(struct vmbus_channel
*c
, void *s
)
951 c
->per_channel_state
= s
;
954 static inline void *get_per_channel_state(struct vmbus_channel
*c
)
956 return c
->per_channel_state
;
959 static inline void set_channel_pending_send_size(struct vmbus_channel
*c
,
965 spin_lock_irqsave(&c
->outbound
.ring_lock
, flags
);
968 if (!c
->out_full_flag
) {
970 c
->out_full_flag
= true;
972 spin_unlock_irqrestore(&c
->outbound
.ring_lock
, flags
);
974 c
->out_full_flag
= false;
977 c
->outbound
.ring_buffer
->pending_send_sz
= size
;
980 static inline void set_low_latency_mode(struct vmbus_channel
*c
)
982 c
->low_latency
= true;
985 static inline void clear_low_latency_mode(struct vmbus_channel
*c
)
987 c
->low_latency
= false;
990 void vmbus_onmessage(void *context
);
992 int vmbus_request_offers(void);
995 * APIs for managing sub-channels.
998 void vmbus_set_sc_create_callback(struct vmbus_channel
*primary_channel
,
999 void (*sc_cr_cb
)(struct vmbus_channel
*new_sc
));
1001 void vmbus_set_chn_rescind_callback(struct vmbus_channel
*channel
,
1002 void (*chn_rescind_cb
)(struct vmbus_channel
*));
1005 * Check if sub-channels have already been offerred. This API will be useful
1006 * when the driver is unloaded after establishing sub-channels. In this case,
1007 * when the driver is re-loaded, the driver would have to check if the
1008 * subchannels have already been established before attempting to request
1009 * the creation of sub-channels.
1010 * This function returns TRUE to indicate that subchannels have already been
1012 * This function should be invoked after setting the callback function for
1013 * sub-channel creation.
1015 bool vmbus_are_subchannels_present(struct vmbus_channel
*primary
);
1017 /* The format must be the same as struct vmdata_gpa_direct */
1018 struct vmbus_channel_packet_page_buffer
{
1026 struct hv_page_buffer range
[MAX_PAGE_BUFFER_COUNT
];
1029 /* The format must be the same as struct vmdata_gpa_direct */
1030 struct vmbus_channel_packet_multipage_buffer
{
1037 u32 rangecount
; /* Always 1 in this case */
1038 struct hv_multipage_buffer range
;
1041 /* The format must be the same as struct vmdata_gpa_direct */
1042 struct vmbus_packet_mpb_array
{
1049 u32 rangecount
; /* Always 1 in this case */
1050 struct hv_mpb_array range
;
1053 int vmbus_alloc_ring(struct vmbus_channel
*channel
,
1054 u32 send_size
, u32 recv_size
);
1055 void vmbus_free_ring(struct vmbus_channel
*channel
);
1057 int vmbus_connect_ring(struct vmbus_channel
*channel
,
1058 void (*onchannel_callback
)(void *context
),
1060 int vmbus_disconnect_ring(struct vmbus_channel
*channel
);
1062 extern int vmbus_open(struct vmbus_channel
*channel
,
1063 u32 send_ringbuffersize
,
1064 u32 recv_ringbuffersize
,
1067 void (*onchannel_callback
)(void *context
),
1070 extern void vmbus_close(struct vmbus_channel
*channel
);
1072 extern int vmbus_sendpacket(struct vmbus_channel
*channel
,
1076 enum vmbus_packet_type type
,
1079 extern int vmbus_sendpacket_pagebuffer(struct vmbus_channel
*channel
,
1080 struct hv_page_buffer pagebuffers
[],
1086 extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel
*channel
,
1087 struct vmbus_packet_mpb_array
*mpb
,
1093 extern int vmbus_establish_gpadl(struct vmbus_channel
*channel
,
1098 extern int vmbus_teardown_gpadl(struct vmbus_channel
*channel
,
1101 void vmbus_reset_channel_cb(struct vmbus_channel
*channel
);
1103 extern int vmbus_recvpacket(struct vmbus_channel
*channel
,
1106 u32
*buffer_actual_len
,
1109 extern int vmbus_recvpacket_raw(struct vmbus_channel
*channel
,
1112 u32
*buffer_actual_len
,
1116 extern void vmbus_ontimer(unsigned long data
);
1118 /* Base driver object */
1123 * A hvsock offer, which has a VMBUS_CHANNEL_TLNPI_PROVIDER_OFFER
1124 * channel flag, actually doesn't mean a synthetic device because the
1125 * offer's if_type/if_instance can change for every new hvsock
1128 * However, to facilitate the notification of new-offer/rescind-offer
1129 * from vmbus driver to hvsock driver, we can handle hvsock offer as
1130 * a special vmbus device, and hence we need the below flag to
1131 * indicate if the driver is the hvsock driver or not: we need to
1132 * specially treat the hvosck offer & driver in vmbus_match().
1136 /* the device type supported by this driver */
1138 const struct hv_vmbus_device_id
*id_table
;
1140 struct device_driver driver
;
1142 /* dynamic device GUID's */
1145 struct list_head list
;
1148 int (*probe
)(struct hv_device
*, const struct hv_vmbus_device_id
*);
1149 int (*remove
)(struct hv_device
*);
1150 void (*shutdown
)(struct hv_device
*);
1154 /* Base device object */
1156 /* the device type id of this device */
1159 /* the device instance id of this device */
1160 guid_t dev_instance
;
1164 struct device device
;
1165 char *driver_override
; /* Driver name to force a match */
1167 struct vmbus_channel
*channel
;
1168 struct kset
*channels_kset
;
1172 static inline struct hv_device
*device_to_hv_device(struct device
*d
)
1174 return container_of(d
, struct hv_device
, device
);
1177 static inline struct hv_driver
*drv_to_hv_drv(struct device_driver
*d
)
1179 return container_of(d
, struct hv_driver
, driver
);
1182 static inline void hv_set_drvdata(struct hv_device
*dev
, void *data
)
1184 dev_set_drvdata(&dev
->device
, data
);
1187 static inline void *hv_get_drvdata(struct hv_device
*dev
)
1189 return dev_get_drvdata(&dev
->device
);
1192 struct hv_ring_buffer_debug_info
{
1193 u32 current_interrupt_mask
;
1194 u32 current_read_index
;
1195 u32 current_write_index
;
1196 u32 bytes_avail_toread
;
1197 u32 bytes_avail_towrite
;
1201 int hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info
*ring_info
,
1202 struct hv_ring_buffer_debug_info
*debug_info
);
1204 /* Vmbus interface */
1205 #define vmbus_driver_register(driver) \
1206 __vmbus_driver_register(driver, THIS_MODULE, KBUILD_MODNAME)
1207 int __must_check
__vmbus_driver_register(struct hv_driver
*hv_driver
,
1208 struct module
*owner
,
1209 const char *mod_name
);
1210 void vmbus_driver_unregister(struct hv_driver
*hv_driver
);
1212 void vmbus_hvsock_device_unregister(struct vmbus_channel
*channel
);
1214 int vmbus_allocate_mmio(struct resource
**new, struct hv_device
*device_obj
,
1215 resource_size_t min
, resource_size_t max
,
1216 resource_size_t size
, resource_size_t align
,
1217 bool fb_overlap_ok
);
1218 void vmbus_free_mmio(resource_size_t start
, resource_size_t size
);
1221 * GUID definitions of various offer types - services offered to the guest.
1226 * {f8615163-df3e-46c5-913f-f2d2f965ed0e}
1228 #define HV_NIC_GUID \
1229 .guid = GUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x91, 0x3f, \
1230 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e)
1234 * {32412632-86cb-44a2-9b5c-50d1417354f5}
1236 #define HV_IDE_GUID \
1237 .guid = GUID_INIT(0x32412632, 0x86cb, 0x44a2, 0x9b, 0x5c, \
1238 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1242 * {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f}
1244 #define HV_SCSI_GUID \
1245 .guid = GUID_INIT(0xba6163d9, 0x04a1, 0x4d29, 0xb6, 0x05, \
1246 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1250 * {0e0b6031-5213-4934-818b-38d90ced39db}
1252 #define HV_SHUTDOWN_GUID \
1253 .guid = GUID_INIT(0x0e0b6031, 0x5213, 0x4934, 0x81, 0x8b, \
1254 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb)
1258 * {9527E630-D0AE-497b-ADCE-E80AB0175CAF}
1260 #define HV_TS_GUID \
1261 .guid = GUID_INIT(0x9527e630, 0xd0ae, 0x497b, 0xad, 0xce, \
1262 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf)
1266 * {57164f39-9115-4e78-ab55-382f3bd5422d}
1268 #define HV_HEART_BEAT_GUID \
1269 .guid = GUID_INIT(0x57164f39, 0x9115, 0x4e78, 0xab, 0x55, \
1270 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d)
1274 * {a9a0f4e7-5a45-4d96-b827-8a841e8c03e6}
1276 #define HV_KVP_GUID \
1277 .guid = GUID_INIT(0xa9a0f4e7, 0x5a45, 0x4d96, 0xb8, 0x27, \
1278 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6)
1281 * Dynamic memory GUID
1282 * {525074dc-8985-46e2-8057-a307dc18a502}
1284 #define HV_DM_GUID \
1285 .guid = GUID_INIT(0x525074dc, 0x8985, 0x46e2, 0x80, 0x57, \
1286 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02)
1290 * {cfa8b69e-5b4a-4cc0-b98b-8ba1a1f3f95a}
1292 #define HV_MOUSE_GUID \
1293 .guid = GUID_INIT(0xcfa8b69e, 0x5b4a, 0x4cc0, 0xb9, 0x8b, \
1294 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a)
1298 * {f912ad6d-2b17-48ea-bd65-f927a61c7684}
1300 #define HV_KBD_GUID \
1301 .guid = GUID_INIT(0xf912ad6d, 0x2b17, 0x48ea, 0xbd, 0x65, \
1302 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84)
1305 * VSS (Backup/Restore) GUID
1307 #define HV_VSS_GUID \
1308 .guid = GUID_INIT(0x35fa2e29, 0xea23, 0x4236, 0x96, 0xae, \
1309 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40)
1311 * Synthetic Video GUID
1312 * {DA0A7802-E377-4aac-8E77-0558EB1073F8}
1314 #define HV_SYNTHVID_GUID \
1315 .guid = GUID_INIT(0xda0a7802, 0xe377, 0x4aac, 0x8e, 0x77, \
1316 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8)
1320 * {2f9bcc4a-0069-4af3-b76b-6fd0be528cda}
1322 #define HV_SYNTHFC_GUID \
1323 .guid = GUID_INIT(0x2f9bcc4a, 0x0069, 0x4af3, 0xb7, 0x6b, \
1324 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda)
1327 * Guest File Copy Service
1328 * {34D14BE3-DEE4-41c8-9AE7-6B174977C192}
1331 #define HV_FCOPY_GUID \
1332 .guid = GUID_INIT(0x34d14be3, 0xdee4, 0x41c8, 0x9a, 0xe7, \
1333 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92)
1336 * NetworkDirect. This is the guest RDMA service.
1337 * {8c2eaf3d-32a7-4b09-ab99-bd1f1c86b501}
1339 #define HV_ND_GUID \
1340 .guid = GUID_INIT(0x8c2eaf3d, 0x32a7, 0x4b09, 0xab, 0x99, \
1341 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01)
1344 * PCI Express Pass Through
1345 * {44C4F61D-4444-4400-9D52-802E27EDE19F}
1348 #define HV_PCIE_GUID \
1349 .guid = GUID_INIT(0x44c4f61d, 0x4444, 0x4400, 0x9d, 0x52, \
1350 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f)
1353 * Linux doesn't support the 3 devices: the first two are for
1354 * Automatic Virtual Machine Activation, and the third is for
1355 * Remote Desktop Virtualization.
1356 * {f8e65716-3cb3-4a06-9a60-1889c5cccab5}
1357 * {3375baf4-9e15-4b30-b765-67acb10d607b}
1358 * {276aacf4-ac15-426c-98dd-7521ad3f01fe}
1361 #define HV_AVMA1_GUID \
1362 .guid = GUID_INIT(0xf8e65716, 0x3cb3, 0x4a06, 0x9a, 0x60, \
1363 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5)
1365 #define HV_AVMA2_GUID \
1366 .guid = GUID_INIT(0x3375baf4, 0x9e15, 0x4b30, 0xb7, 0x65, \
1367 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b)
1369 #define HV_RDV_GUID \
1370 .guid = GUID_INIT(0x276aacf4, 0xac15, 0x426c, 0x98, 0xdd, \
1371 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe)
1374 * Common header for Hyper-V ICs
1377 #define ICMSGTYPE_NEGOTIATE 0
1378 #define ICMSGTYPE_HEARTBEAT 1
1379 #define ICMSGTYPE_KVPEXCHANGE 2
1380 #define ICMSGTYPE_SHUTDOWN 3
1381 #define ICMSGTYPE_TIMESYNC 4
1382 #define ICMSGTYPE_VSS 5
1384 #define ICMSGHDRFLAG_TRANSACTION 1
1385 #define ICMSGHDRFLAG_REQUEST 2
1386 #define ICMSGHDRFLAG_RESPONSE 4
1390 * While we want to handle util services as regular devices,
1391 * there is only one instance of each of these services; so
1392 * we statically allocate the service specific state.
1395 struct hv_util_service
{
1398 void (*util_cb
)(void *);
1399 int (*util_init
)(struct hv_util_service
*);
1400 void (*util_deinit
)(void);
1403 struct vmbuspipe_hdr
{
1414 struct ic_version icverframe
;
1416 struct ic_version icvermsg
;
1419 u8 ictransaction_id
;
1424 struct icmsg_negotiate
{
1428 struct ic_version icversion_data
[1]; /* any size array */
1431 struct shutdown_msg_data
{
1433 u32 timeout_seconds
;
1435 u8 display_message
[2048];
1438 struct heartbeat_msg_data
{
1443 /* Time Sync IC defs */
1444 #define ICTIMESYNCFLAG_PROBE 0
1445 #define ICTIMESYNCFLAG_SYNC 1
1446 #define ICTIMESYNCFLAG_SAMPLE 2
1449 #define WLTIMEDELTA 116444736000000000L /* in 100ns unit */
1451 #define WLTIMEDELTA 116444736000000000LL
1454 struct ictimesync_data
{
1461 struct ictimesync_ref_data
{
1463 u64 vmreferencetime
;
1470 struct hyperv_service_callback
{
1474 struct vmbus_channel
*channel
;
1475 void (*callback
)(void *context
);
1478 #define MAX_SRV_VER 0x7ffffff
1479 extern bool vmbus_prep_negotiate_resp(struct icmsg_hdr
*icmsghdrp
, u8
*buf
,
1480 const int *fw_version
, int fw_vercnt
,
1481 const int *srv_version
, int srv_vercnt
,
1482 int *nego_fw_version
, int *nego_srv_version
);
1484 void hv_process_channel_removal(struct vmbus_channel
*channel
);
1486 void vmbus_setevent(struct vmbus_channel
*channel
);
1488 * Negotiated version with the Host.
1491 extern __u32 vmbus_proto_version
;
1493 int vmbus_send_tl_connect_request(const guid_t
*shv_guest_servie_id
,
1494 const guid_t
*shv_host_servie_id
);
1495 void vmbus_set_event(struct vmbus_channel
*channel
);
1497 /* Get the start of the ring buffer. */
1498 static inline void *
1499 hv_get_ring_buffer(const struct hv_ring_buffer_info
*ring_info
)
1501 return ring_info
->ring_buffer
->buffer
;
1505 * Mask off host interrupt callback notifications
1507 static inline void hv_begin_read(struct hv_ring_buffer_info
*rbi
)
1509 rbi
->ring_buffer
->interrupt_mask
= 1;
1511 /* make sure mask update is not reordered */
1516 * Re-enable host callback and return number of outstanding bytes
1518 static inline u32
hv_end_read(struct hv_ring_buffer_info
*rbi
)
1521 rbi
->ring_buffer
->interrupt_mask
= 0;
1523 /* make sure mask update is not reordered */
1527 * Now check to see if the ring buffer is still empty.
1528 * If it is not, we raced and we need to process new
1529 * incoming messages.
1531 return hv_get_bytes_to_read(rbi
);
1535 * An API to support in-place processing of incoming VMBUS packets.
1538 /* Get data payload associated with descriptor */
1539 static inline void *hv_pkt_data(const struct vmpacket_descriptor
*desc
)
1541 return (void *)((unsigned long)desc
+ (desc
->offset8
<< 3));
1544 /* Get data size associated with descriptor */
1545 static inline u32
hv_pkt_datalen(const struct vmpacket_descriptor
*desc
)
1547 return (desc
->len8
<< 3) - (desc
->offset8
<< 3);
1551 struct vmpacket_descriptor
*
1552 hv_pkt_iter_first(struct vmbus_channel
*channel
);
1554 struct vmpacket_descriptor
*
1555 __hv_pkt_iter_next(struct vmbus_channel
*channel
,
1556 const struct vmpacket_descriptor
*pkt
);
1558 void hv_pkt_iter_close(struct vmbus_channel
*channel
);
1561 * Get next packet descriptor from iterator
1562 * If at end of list, return NULL and update host.
1564 static inline struct vmpacket_descriptor
*
1565 hv_pkt_iter_next(struct vmbus_channel
*channel
,
1566 const struct vmpacket_descriptor
*pkt
)
1568 struct vmpacket_descriptor
*nxt
;
1570 nxt
= __hv_pkt_iter_next(channel
, pkt
);
1572 hv_pkt_iter_close(channel
);
1577 #define foreach_vmbus_pkt(pkt, channel) \
1578 for (pkt = hv_pkt_iter_first(channel); pkt; \
1579 pkt = hv_pkt_iter_next(channel, pkt))
1581 #endif /* _HYPERV_H */