2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
18 * Haiyang Zhang <haiyangz@microsoft.com>
19 * Hank Janssen <hjanssen@microsoft.com>
21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
23 #include <linux/kernel.h>
24 #include <linux/interrupt.h>
25 #include <linux/sched.h>
26 #include <linux/wait.h>
28 #include <linux/slab.h>
29 #include <linux/list.h>
30 #include <linux/module.h>
31 #include <linux/completion.h>
32 #include <linux/delay.h>
33 #include <linux/hyperv.h>
34 #include <asm/mshyperv.h>
36 #include "hyperv_vmbus.h"
38 static void init_vp_index(struct vmbus_channel
*channel
, u16 dev_type
);
40 static const struct vmbus_device vmbus_devs
[] = {
48 { .dev_type
= HV_SCSI
,
72 { .dev_type
= HV_PCIE
,
77 /* Synthetic Frame Buffer */
83 /* Synthetic Keyboard */
90 { .dev_type
= HV_MOUSE
,
104 .perf_device
= false,
110 .perf_device
= false,
114 { .dev_type
= HV_SHUTDOWN
,
116 .perf_device
= false,
120 { .dev_type
= HV_FCOPY
,
122 .perf_device
= false,
126 { .dev_type
= HV_BACKUP
,
128 .perf_device
= false,
134 .perf_device
= false,
138 { .dev_type
= HV_UNKNOWN
,
139 .perf_device
= false,
143 static const struct {
145 } vmbus_unsupported_devs
[] = {
152 * The rescinded channel may be blocked waiting for a response from the host;
155 static void vmbus_rescind_cleanup(struct vmbus_channel
*channel
)
157 struct vmbus_channel_msginfo
*msginfo
;
161 spin_lock_irqsave(&vmbus_connection
.channelmsg_lock
, flags
);
163 list_for_each_entry(msginfo
, &vmbus_connection
.chn_msg_list
,
166 if (msginfo
->waiting_channel
== channel
) {
167 complete(&msginfo
->waitevent
);
171 spin_unlock_irqrestore(&vmbus_connection
.channelmsg_lock
, flags
);
174 static bool is_unsupported_vmbus_devs(const uuid_le
*guid
)
178 for (i
= 0; i
< ARRAY_SIZE(vmbus_unsupported_devs
); i
++)
179 if (!uuid_le_cmp(*guid
, vmbus_unsupported_devs
[i
].guid
))
184 static u16
hv_get_dev_type(const struct vmbus_channel
*channel
)
186 const uuid_le
*guid
= &channel
->offermsg
.offer
.if_type
;
189 if (is_hvsock_channel(channel
) || is_unsupported_vmbus_devs(guid
))
192 for (i
= HV_IDE
; i
< HV_UNKNOWN
; i
++) {
193 if (!uuid_le_cmp(*guid
, vmbus_devs
[i
].guid
))
196 pr_info("Unknown GUID: %pUl\n", guid
);
201 * vmbus_prep_negotiate_resp() - Create default response for Hyper-V Negotiate message
202 * @icmsghdrp: Pointer to msg header structure
203 * @icmsg_negotiate: Pointer to negotiate message structure
204 * @buf: Raw buffer channel data
206 * @icmsghdrp is of type &struct icmsg_hdr.
207 * Set up and fill in default negotiate response message.
209 * The fw_version and fw_vercnt specifies the framework version that
212 * The srv_version and srv_vercnt specifies the service
213 * versions we can support.
215 * Versions are given in decreasing order.
217 * nego_fw_version and nego_srv_version store the selected protocol versions.
219 * Mainly used by Hyper-V drivers.
221 bool vmbus_prep_negotiate_resp(struct icmsg_hdr
*icmsghdrp
,
222 u8
*buf
, const int *fw_version
, int fw_vercnt
,
223 const int *srv_version
, int srv_vercnt
,
224 int *nego_fw_version
, int *nego_srv_version
)
226 int icframe_major
, icframe_minor
;
227 int icmsg_major
, icmsg_minor
;
228 int fw_major
, fw_minor
;
229 int srv_major
, srv_minor
;
231 bool found_match
= false;
232 struct icmsg_negotiate
*negop
;
234 icmsghdrp
->icmsgsize
= 0x10;
235 negop
= (struct icmsg_negotiate
*)&buf
[
236 sizeof(struct vmbuspipe_hdr
) +
237 sizeof(struct icmsg_hdr
)];
239 icframe_major
= negop
->icframe_vercnt
;
242 icmsg_major
= negop
->icmsg_vercnt
;
246 * Select the framework version number we will
250 for (i
= 0; i
< fw_vercnt
; i
++) {
251 fw_major
= (fw_version
[i
] >> 16);
252 fw_minor
= (fw_version
[i
] & 0xFFFF);
254 for (j
= 0; j
< negop
->icframe_vercnt
; j
++) {
255 if ((negop
->icversion_data
[j
].major
== fw_major
) &&
256 (negop
->icversion_data
[j
].minor
== fw_minor
)) {
257 icframe_major
= negop
->icversion_data
[j
].major
;
258 icframe_minor
= negop
->icversion_data
[j
].minor
;
273 for (i
= 0; i
< srv_vercnt
; i
++) {
274 srv_major
= (srv_version
[i
] >> 16);
275 srv_minor
= (srv_version
[i
] & 0xFFFF);
277 for (j
= negop
->icframe_vercnt
;
278 (j
< negop
->icframe_vercnt
+ negop
->icmsg_vercnt
);
281 if ((negop
->icversion_data
[j
].major
== srv_major
) &&
282 (negop
->icversion_data
[j
].minor
== srv_minor
)) {
284 icmsg_major
= negop
->icversion_data
[j
].major
;
285 icmsg_minor
= negop
->icversion_data
[j
].minor
;
296 * Respond with the framework and service
297 * version numbers we can support.
302 negop
->icframe_vercnt
= 0;
303 negop
->icmsg_vercnt
= 0;
305 negop
->icframe_vercnt
= 1;
306 negop
->icmsg_vercnt
= 1;
310 *nego_fw_version
= (icframe_major
<< 16) | icframe_minor
;
312 if (nego_srv_version
)
313 *nego_srv_version
= (icmsg_major
<< 16) | icmsg_minor
;
315 negop
->icversion_data
[0].major
= icframe_major
;
316 negop
->icversion_data
[0].minor
= icframe_minor
;
317 negop
->icversion_data
[1].major
= icmsg_major
;
318 negop
->icversion_data
[1].minor
= icmsg_minor
;
322 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp
);
325 * alloc_channel - Allocate and initialize a vmbus channel object
327 static struct vmbus_channel
*alloc_channel(void)
329 struct vmbus_channel
*channel
;
331 channel
= kzalloc(sizeof(*channel
), GFP_ATOMIC
);
335 spin_lock_init(&channel
->inbound_lock
);
336 spin_lock_init(&channel
->lock
);
338 INIT_LIST_HEAD(&channel
->sc_list
);
339 INIT_LIST_HEAD(&channel
->percpu_list
);
341 tasklet_init(&channel
->callback_event
,
342 vmbus_on_event
, (unsigned long)channel
);
348 * free_channel - Release the resources used by the vmbus channel object
350 static void free_channel(struct vmbus_channel
*channel
)
352 tasklet_kill(&channel
->callback_event
);
354 kfree_rcu(channel
, rcu
);
357 static void percpu_channel_enq(void *arg
)
359 struct vmbus_channel
*channel
= arg
;
360 struct hv_per_cpu_context
*hv_cpu
361 = this_cpu_ptr(hv_context
.cpu_context
);
363 list_add_tail_rcu(&channel
->percpu_list
, &hv_cpu
->chan_list
);
366 static void percpu_channel_deq(void *arg
)
368 struct vmbus_channel
*channel
= arg
;
370 list_del_rcu(&channel
->percpu_list
);
374 static void vmbus_release_relid(u32 relid
)
376 struct vmbus_channel_relid_released msg
;
378 memset(&msg
, 0, sizeof(struct vmbus_channel_relid_released
));
379 msg
.child_relid
= relid
;
380 msg
.header
.msgtype
= CHANNELMSG_RELID_RELEASED
;
381 vmbus_post_msg(&msg
, sizeof(struct vmbus_channel_relid_released
),
385 void hv_process_channel_removal(struct vmbus_channel
*channel
, u32 relid
)
388 struct vmbus_channel
*primary_channel
;
390 BUG_ON(!channel
->rescind
);
391 BUG_ON(!mutex_is_locked(&vmbus_connection
.channel_mutex
));
393 if (channel
->target_cpu
!= get_cpu()) {
395 smp_call_function_single(channel
->target_cpu
,
396 percpu_channel_deq
, channel
, true);
398 percpu_channel_deq(channel
);
402 if (channel
->primary_channel
== NULL
) {
403 list_del(&channel
->listentry
);
405 primary_channel
= channel
;
407 primary_channel
= channel
->primary_channel
;
408 spin_lock_irqsave(&primary_channel
->lock
, flags
);
409 list_del(&channel
->sc_list
);
410 primary_channel
->num_sc
--;
411 spin_unlock_irqrestore(&primary_channel
->lock
, flags
);
415 * We need to free the bit for init_vp_index() to work in the case
416 * of sub-channel, when we reload drivers like hv_netvsc.
418 if (channel
->affinity_policy
== HV_LOCALIZED
)
419 cpumask_clear_cpu(channel
->target_cpu
,
420 &primary_channel
->alloced_cpus_in_node
);
422 vmbus_release_relid(relid
);
424 free_channel(channel
);
427 void vmbus_free_channels(void)
429 struct vmbus_channel
*channel
, *tmp
;
431 mutex_lock(&vmbus_connection
.channel_mutex
);
432 list_for_each_entry_safe(channel
, tmp
, &vmbus_connection
.chn_list
,
434 /* hv_process_channel_removal() needs this */
435 channel
->rescind
= true;
437 vmbus_device_unregister(channel
->device_obj
);
439 mutex_unlock(&vmbus_connection
.channel_mutex
);
443 * vmbus_process_offer - Process the offer by creating a channel/device
444 * associated with this offer
446 static void vmbus_process_offer(struct vmbus_channel
*newchannel
)
448 struct vmbus_channel
*channel
;
454 /* Make sure this is a new offer */
455 mutex_lock(&vmbus_connection
.channel_mutex
);
457 list_for_each_entry(channel
, &vmbus_connection
.chn_list
, listentry
) {
458 if (!uuid_le_cmp(channel
->offermsg
.offer
.if_type
,
459 newchannel
->offermsg
.offer
.if_type
) &&
460 !uuid_le_cmp(channel
->offermsg
.offer
.if_instance
,
461 newchannel
->offermsg
.offer
.if_instance
)) {
468 list_add_tail(&newchannel
->listentry
,
469 &vmbus_connection
.chn_list
);
471 mutex_unlock(&vmbus_connection
.channel_mutex
);
475 * Check to see if this is a sub-channel.
477 if (newchannel
->offermsg
.offer
.sub_channel_index
!= 0) {
479 * Process the sub-channel.
481 newchannel
->primary_channel
= channel
;
482 spin_lock_irqsave(&channel
->lock
, flags
);
483 list_add_tail(&newchannel
->sc_list
, &channel
->sc_list
);
485 spin_unlock_irqrestore(&channel
->lock
, flags
);
490 dev_type
= hv_get_dev_type(newchannel
);
492 init_vp_index(newchannel
, dev_type
);
494 if (newchannel
->target_cpu
!= get_cpu()) {
496 smp_call_function_single(newchannel
->target_cpu
,
500 percpu_channel_enq(newchannel
);
505 * This state is used to indicate a successful open
506 * so that when we do close the channel normally, we
507 * can cleanup properly
509 newchannel
->state
= CHANNEL_OPEN_STATE
;
512 if (channel
->sc_creation_callback
!= NULL
)
513 channel
->sc_creation_callback(newchannel
);
518 * Start the process of binding this offer to the driver
519 * We need to set the DeviceObject field before calling
520 * vmbus_child_dev_add()
522 newchannel
->device_obj
= vmbus_device_create(
523 &newchannel
->offermsg
.offer
.if_type
,
524 &newchannel
->offermsg
.offer
.if_instance
,
526 if (!newchannel
->device_obj
)
529 newchannel
->device_obj
->device_id
= dev_type
;
531 * Add the new device to the bus. This will kick off device-driver
532 * binding which eventually invokes the device driver's AddDevice()
535 mutex_lock(&vmbus_connection
.channel_mutex
);
536 ret
= vmbus_device_register(newchannel
->device_obj
);
537 mutex_unlock(&vmbus_connection
.channel_mutex
);
540 pr_err("unable to add child device object (relid %d)\n",
541 newchannel
->offermsg
.child_relid
);
542 kfree(newchannel
->device_obj
);
548 mutex_lock(&vmbus_connection
.channel_mutex
);
549 list_del(&newchannel
->listentry
);
550 mutex_unlock(&vmbus_connection
.channel_mutex
);
552 if (newchannel
->target_cpu
!= get_cpu()) {
554 smp_call_function_single(newchannel
->target_cpu
,
555 percpu_channel_deq
, newchannel
, true);
557 percpu_channel_deq(newchannel
);
561 vmbus_release_relid(newchannel
->offermsg
.child_relid
);
564 free_channel(newchannel
);
568 * We use this state to statically distribute the channel interrupt load.
570 static int next_numa_node_id
;
573 * Starting with Win8, we can statically distribute the incoming
574 * channel interrupt load by binding a channel to VCPU.
575 * We do this in a hierarchical fashion:
576 * First distribute the primary channels across available NUMA nodes
577 * and then distribute the subchannels amongst the CPUs in the NUMA
578 * node assigned to the primary channel.
580 * For pre-win8 hosts or non-performance critical channels we assign the
581 * first CPU in the first NUMA node.
583 static void init_vp_index(struct vmbus_channel
*channel
, u16 dev_type
)
586 bool perf_chn
= vmbus_devs
[dev_type
].perf_device
;
587 struct vmbus_channel
*primary
= channel
->primary_channel
;
589 struct cpumask available_mask
;
590 struct cpumask
*alloced_mask
;
592 if ((vmbus_proto_version
== VERSION_WS2008
) ||
593 (vmbus_proto_version
== VERSION_WIN7
) || (!perf_chn
)) {
595 * Prior to win8, all channel interrupts are
596 * delivered on cpu 0.
597 * Also if the channel is not a performance critical
598 * channel, bind it to cpu 0.
600 channel
->numa_node
= 0;
601 channel
->target_cpu
= 0;
602 channel
->target_vp
= hv_context
.vp_index
[0];
607 * Based on the channel affinity policy, we will assign the NUMA
611 if ((channel
->affinity_policy
== HV_BALANCED
) || (!primary
)) {
613 next_node
= next_numa_node_id
++;
614 if (next_node
== nr_node_ids
) {
615 next_node
= next_numa_node_id
= 0;
618 if (cpumask_empty(cpumask_of_node(next_node
)))
622 channel
->numa_node
= next_node
;
625 alloced_mask
= &hv_context
.hv_numa_map
[primary
->numa_node
];
627 if (cpumask_weight(alloced_mask
) ==
628 cpumask_weight(cpumask_of_node(primary
->numa_node
))) {
630 * We have cycled through all the CPUs in the node;
631 * reset the alloced map.
633 cpumask_clear(alloced_mask
);
636 cpumask_xor(&available_mask
, alloced_mask
,
637 cpumask_of_node(primary
->numa_node
));
641 if (primary
->affinity_policy
== HV_LOCALIZED
) {
643 * Normally Hyper-V host doesn't create more subchannels
644 * than there are VCPUs on the node but it is possible when not
645 * all present VCPUs on the node are initialized by guest.
646 * Clear the alloced_cpus_in_node to start over.
648 if (cpumask_equal(&primary
->alloced_cpus_in_node
,
649 cpumask_of_node(primary
->numa_node
)))
650 cpumask_clear(&primary
->alloced_cpus_in_node
);
654 cur_cpu
= cpumask_next(cur_cpu
, &available_mask
);
655 if (cur_cpu
>= nr_cpu_ids
) {
657 cpumask_copy(&available_mask
,
658 cpumask_of_node(primary
->numa_node
));
662 if (primary
->affinity_policy
== HV_LOCALIZED
) {
664 * NOTE: in the case of sub-channel, we clear the
665 * sub-channel related bit(s) in
666 * primary->alloced_cpus_in_node in
667 * hv_process_channel_removal(), so when we
668 * reload drivers like hv_netvsc in SMP guest, here
669 * we're able to re-allocate
670 * bit from primary->alloced_cpus_in_node.
672 if (!cpumask_test_cpu(cur_cpu
,
673 &primary
->alloced_cpus_in_node
)) {
674 cpumask_set_cpu(cur_cpu
,
675 &primary
->alloced_cpus_in_node
);
676 cpumask_set_cpu(cur_cpu
, alloced_mask
);
680 cpumask_set_cpu(cur_cpu
, alloced_mask
);
685 channel
->target_cpu
= cur_cpu
;
686 channel
->target_vp
= hv_context
.vp_index
[cur_cpu
];
689 static void vmbus_wait_for_unload(void)
693 struct hv_message
*msg
;
694 struct vmbus_channel_message_header
*hdr
;
698 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
699 * used for initial contact or to CPU0 depending on host version. When
700 * we're crashing on a different CPU let's hope that IRQ handler on
701 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
702 * functional and vmbus_unload_response() will complete
703 * vmbus_connection.unload_event. If not, the last thing we can do is
704 * read message pages for all CPUs directly.
707 if (completion_done(&vmbus_connection
.unload_event
))
710 for_each_online_cpu(cpu
) {
711 struct hv_per_cpu_context
*hv_cpu
712 = per_cpu_ptr(hv_context
.cpu_context
, cpu
);
714 page_addr
= hv_cpu
->synic_message_page
;
715 msg
= (struct hv_message
*)page_addr
716 + VMBUS_MESSAGE_SINT
;
718 message_type
= READ_ONCE(msg
->header
.message_type
);
719 if (message_type
== HVMSG_NONE
)
722 hdr
= (struct vmbus_channel_message_header
*)
725 if (hdr
->msgtype
== CHANNELMSG_UNLOAD_RESPONSE
)
726 complete(&vmbus_connection
.unload_event
);
728 vmbus_signal_eom(msg
, message_type
);
735 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
736 * maybe-pending messages on all CPUs to be able to receive new
737 * messages after we reconnect.
739 for_each_online_cpu(cpu
) {
740 struct hv_per_cpu_context
*hv_cpu
741 = per_cpu_ptr(hv_context
.cpu_context
, cpu
);
743 page_addr
= hv_cpu
->synic_message_page
;
744 msg
= (struct hv_message
*)page_addr
+ VMBUS_MESSAGE_SINT
;
745 msg
->header
.message_type
= HVMSG_NONE
;
750 * vmbus_unload_response - Handler for the unload response.
752 static void vmbus_unload_response(struct vmbus_channel_message_header
*hdr
)
755 * This is a global event; just wakeup the waiting thread.
756 * Once we successfully unload, we can cleanup the monitor state.
758 complete(&vmbus_connection
.unload_event
);
761 void vmbus_initiate_unload(bool crash
)
763 struct vmbus_channel_message_header hdr
;
765 /* Pre-Win2012R2 hosts don't support reconnect */
766 if (vmbus_proto_version
< VERSION_WIN8_1
)
769 init_completion(&vmbus_connection
.unload_event
);
770 memset(&hdr
, 0, sizeof(struct vmbus_channel_message_header
));
771 hdr
.msgtype
= CHANNELMSG_UNLOAD
;
772 vmbus_post_msg(&hdr
, sizeof(struct vmbus_channel_message_header
),
776 * vmbus_initiate_unload() is also called on crash and the crash can be
777 * happening in an interrupt context, where scheduling is impossible.
780 wait_for_completion(&vmbus_connection
.unload_event
);
782 vmbus_wait_for_unload();
786 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
789 static void vmbus_onoffer(struct vmbus_channel_message_header
*hdr
)
791 struct vmbus_channel_offer_channel
*offer
;
792 struct vmbus_channel
*newchannel
;
794 offer
= (struct vmbus_channel_offer_channel
*)hdr
;
796 /* Allocate the channel object and save this offer. */
797 newchannel
= alloc_channel();
799 vmbus_release_relid(offer
->child_relid
);
800 pr_err("Unable to allocate channel object\n");
805 * Setup state for signalling the host.
807 newchannel
->sig_event
= (struct hv_input_signal_event
*)
808 (ALIGN((unsigned long)
809 &newchannel
->sig_buf
,
810 HV_HYPERCALL_PARAM_ALIGN
));
812 newchannel
->sig_event
->connectionid
.asu32
= 0;
813 newchannel
->sig_event
->connectionid
.u
.id
= VMBUS_EVENT_CONNECTION_ID
;
814 newchannel
->sig_event
->flag_number
= 0;
815 newchannel
->sig_event
->rsvdz
= 0;
817 if (vmbus_proto_version
!= VERSION_WS2008
) {
818 newchannel
->is_dedicated_interrupt
=
819 (offer
->is_dedicated_interrupt
!= 0);
820 newchannel
->sig_event
->connectionid
.u
.id
=
821 offer
->connection_id
;
824 memcpy(&newchannel
->offermsg
, offer
,
825 sizeof(struct vmbus_channel_offer_channel
));
826 newchannel
->monitor_grp
= (u8
)offer
->monitorid
/ 32;
827 newchannel
->monitor_bit
= (u8
)offer
->monitorid
% 32;
829 vmbus_process_offer(newchannel
);
833 * vmbus_onoffer_rescind - Rescind offer handler.
835 * We queue a work item to process this offer synchronously
837 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header
*hdr
)
839 struct vmbus_channel_rescind_offer
*rescind
;
840 struct vmbus_channel
*channel
;
844 rescind
= (struct vmbus_channel_rescind_offer
*)hdr
;
846 mutex_lock(&vmbus_connection
.channel_mutex
);
847 channel
= relid2channel(rescind
->child_relid
);
849 if (channel
== NULL
) {
851 * This is very impossible, because in
852 * vmbus_process_offer(), we have already invoked
853 * vmbus_release_relid() on error.
858 spin_lock_irqsave(&channel
->lock
, flags
);
859 channel
->rescind
= true;
860 spin_unlock_irqrestore(&channel
->lock
, flags
);
862 vmbus_rescind_cleanup(channel
);
864 if (channel
->device_obj
) {
865 if (channel
->chn_rescind_callback
) {
866 channel
->chn_rescind_callback(channel
);
870 * We will have to unregister this device from the
873 dev
= get_device(&channel
->device_obj
->device
);
875 vmbus_device_unregister(channel
->device_obj
);
879 hv_process_channel_removal(channel
,
880 channel
->offermsg
.child_relid
);
884 mutex_unlock(&vmbus_connection
.channel_mutex
);
887 void vmbus_hvsock_device_unregister(struct vmbus_channel
*channel
)
889 mutex_lock(&vmbus_connection
.channel_mutex
);
891 BUG_ON(!is_hvsock_channel(channel
));
893 channel
->rescind
= true;
894 vmbus_device_unregister(channel
->device_obj
);
896 mutex_unlock(&vmbus_connection
.channel_mutex
);
898 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister
);
902 * vmbus_onoffers_delivered -
903 * This is invoked when all offers have been delivered.
905 * Nothing to do here.
907 static void vmbus_onoffers_delivered(
908 struct vmbus_channel_message_header
*hdr
)
913 * vmbus_onopen_result - Open result handler.
915 * This is invoked when we received a response to our channel open request.
916 * Find the matching request, copy the response and signal the requesting
919 static void vmbus_onopen_result(struct vmbus_channel_message_header
*hdr
)
921 struct vmbus_channel_open_result
*result
;
922 struct vmbus_channel_msginfo
*msginfo
;
923 struct vmbus_channel_message_header
*requestheader
;
924 struct vmbus_channel_open_channel
*openmsg
;
927 result
= (struct vmbus_channel_open_result
*)hdr
;
930 * Find the open msg, copy the result and signal/unblock the wait event
932 spin_lock_irqsave(&vmbus_connection
.channelmsg_lock
, flags
);
934 list_for_each_entry(msginfo
, &vmbus_connection
.chn_msg_list
,
937 (struct vmbus_channel_message_header
*)msginfo
->msg
;
939 if (requestheader
->msgtype
== CHANNELMSG_OPENCHANNEL
) {
941 (struct vmbus_channel_open_channel
*)msginfo
->msg
;
942 if (openmsg
->child_relid
== result
->child_relid
&&
943 openmsg
->openid
== result
->openid
) {
944 memcpy(&msginfo
->response
.open_result
,
947 struct vmbus_channel_open_result
));
948 complete(&msginfo
->waitevent
);
953 spin_unlock_irqrestore(&vmbus_connection
.channelmsg_lock
, flags
);
957 * vmbus_ongpadl_created - GPADL created handler.
959 * This is invoked when we received a response to our gpadl create request.
960 * Find the matching request, copy the response and signal the requesting
963 static void vmbus_ongpadl_created(struct vmbus_channel_message_header
*hdr
)
965 struct vmbus_channel_gpadl_created
*gpadlcreated
;
966 struct vmbus_channel_msginfo
*msginfo
;
967 struct vmbus_channel_message_header
*requestheader
;
968 struct vmbus_channel_gpadl_header
*gpadlheader
;
971 gpadlcreated
= (struct vmbus_channel_gpadl_created
*)hdr
;
974 * Find the establish msg, copy the result and signal/unblock the wait
977 spin_lock_irqsave(&vmbus_connection
.channelmsg_lock
, flags
);
979 list_for_each_entry(msginfo
, &vmbus_connection
.chn_msg_list
,
982 (struct vmbus_channel_message_header
*)msginfo
->msg
;
984 if (requestheader
->msgtype
== CHANNELMSG_GPADL_HEADER
) {
986 (struct vmbus_channel_gpadl_header
*)requestheader
;
988 if ((gpadlcreated
->child_relid
==
989 gpadlheader
->child_relid
) &&
990 (gpadlcreated
->gpadl
== gpadlheader
->gpadl
)) {
991 memcpy(&msginfo
->response
.gpadl_created
,
994 struct vmbus_channel_gpadl_created
));
995 complete(&msginfo
->waitevent
);
1000 spin_unlock_irqrestore(&vmbus_connection
.channelmsg_lock
, flags
);
1004 * vmbus_ongpadl_torndown - GPADL torndown handler.
1006 * This is invoked when we received a response to our gpadl teardown request.
1007 * Find the matching request, copy the response and signal the requesting
1010 static void vmbus_ongpadl_torndown(
1011 struct vmbus_channel_message_header
*hdr
)
1013 struct vmbus_channel_gpadl_torndown
*gpadl_torndown
;
1014 struct vmbus_channel_msginfo
*msginfo
;
1015 struct vmbus_channel_message_header
*requestheader
;
1016 struct vmbus_channel_gpadl_teardown
*gpadl_teardown
;
1017 unsigned long flags
;
1019 gpadl_torndown
= (struct vmbus_channel_gpadl_torndown
*)hdr
;
1022 * Find the open msg, copy the result and signal/unblock the wait event
1024 spin_lock_irqsave(&vmbus_connection
.channelmsg_lock
, flags
);
1026 list_for_each_entry(msginfo
, &vmbus_connection
.chn_msg_list
,
1029 (struct vmbus_channel_message_header
*)msginfo
->msg
;
1031 if (requestheader
->msgtype
== CHANNELMSG_GPADL_TEARDOWN
) {
1033 (struct vmbus_channel_gpadl_teardown
*)requestheader
;
1035 if (gpadl_torndown
->gpadl
== gpadl_teardown
->gpadl
) {
1036 memcpy(&msginfo
->response
.gpadl_torndown
,
1039 struct vmbus_channel_gpadl_torndown
));
1040 complete(&msginfo
->waitevent
);
1045 spin_unlock_irqrestore(&vmbus_connection
.channelmsg_lock
, flags
);
1049 * vmbus_onversion_response - Version response handler
1051 * This is invoked when we received a response to our initiate contact request.
1052 * Find the matching request, copy the response and signal the requesting
1055 static void vmbus_onversion_response(
1056 struct vmbus_channel_message_header
*hdr
)
1058 struct vmbus_channel_msginfo
*msginfo
;
1059 struct vmbus_channel_message_header
*requestheader
;
1060 struct vmbus_channel_version_response
*version_response
;
1061 unsigned long flags
;
1063 version_response
= (struct vmbus_channel_version_response
*)hdr
;
1064 spin_lock_irqsave(&vmbus_connection
.channelmsg_lock
, flags
);
1066 list_for_each_entry(msginfo
, &vmbus_connection
.chn_msg_list
,
1069 (struct vmbus_channel_message_header
*)msginfo
->msg
;
1071 if (requestheader
->msgtype
==
1072 CHANNELMSG_INITIATE_CONTACT
) {
1073 memcpy(&msginfo
->response
.version_response
,
1075 sizeof(struct vmbus_channel_version_response
));
1076 complete(&msginfo
->waitevent
);
1079 spin_unlock_irqrestore(&vmbus_connection
.channelmsg_lock
, flags
);
1082 /* Channel message dispatch table */
1083 struct vmbus_channel_message_table_entry
1084 channel_message_table
[CHANNELMSG_COUNT
] = {
1085 {CHANNELMSG_INVALID
, 0, NULL
},
1086 {CHANNELMSG_OFFERCHANNEL
, 0, vmbus_onoffer
},
1087 {CHANNELMSG_RESCIND_CHANNELOFFER
, 0, vmbus_onoffer_rescind
},
1088 {CHANNELMSG_REQUESTOFFERS
, 0, NULL
},
1089 {CHANNELMSG_ALLOFFERS_DELIVERED
, 1, vmbus_onoffers_delivered
},
1090 {CHANNELMSG_OPENCHANNEL
, 0, NULL
},
1091 {CHANNELMSG_OPENCHANNEL_RESULT
, 1, vmbus_onopen_result
},
1092 {CHANNELMSG_CLOSECHANNEL
, 0, NULL
},
1093 {CHANNELMSG_GPADL_HEADER
, 0, NULL
},
1094 {CHANNELMSG_GPADL_BODY
, 0, NULL
},
1095 {CHANNELMSG_GPADL_CREATED
, 1, vmbus_ongpadl_created
},
1096 {CHANNELMSG_GPADL_TEARDOWN
, 0, NULL
},
1097 {CHANNELMSG_GPADL_TORNDOWN
, 1, vmbus_ongpadl_torndown
},
1098 {CHANNELMSG_RELID_RELEASED
, 0, NULL
},
1099 {CHANNELMSG_INITIATE_CONTACT
, 0, NULL
},
1100 {CHANNELMSG_VERSION_RESPONSE
, 1, vmbus_onversion_response
},
1101 {CHANNELMSG_UNLOAD
, 0, NULL
},
1102 {CHANNELMSG_UNLOAD_RESPONSE
, 1, vmbus_unload_response
},
1103 {CHANNELMSG_18
, 0, NULL
},
1104 {CHANNELMSG_19
, 0, NULL
},
1105 {CHANNELMSG_20
, 0, NULL
},
1106 {CHANNELMSG_TL_CONNECT_REQUEST
, 0, NULL
},
1110 * vmbus_onmessage - Handler for channel protocol messages.
1112 * This is invoked in the vmbus worker thread context.
1114 void vmbus_onmessage(void *context
)
1116 struct hv_message
*msg
= context
;
1117 struct vmbus_channel_message_header
*hdr
;
1120 hdr
= (struct vmbus_channel_message_header
*)msg
->u
.payload
;
1121 size
= msg
->header
.payload_size
;
1123 if (hdr
->msgtype
>= CHANNELMSG_COUNT
) {
1124 pr_err("Received invalid channel message type %d size %d\n",
1125 hdr
->msgtype
, size
);
1126 print_hex_dump_bytes("", DUMP_PREFIX_NONE
,
1127 (unsigned char *)msg
->u
.payload
, size
);
1131 if (channel_message_table
[hdr
->msgtype
].message_handler
)
1132 channel_message_table
[hdr
->msgtype
].message_handler(hdr
);
1134 pr_err("Unhandled channel message type %d\n", hdr
->msgtype
);
1138 * vmbus_request_offers - Send a request to get all our pending offers.
1140 int vmbus_request_offers(void)
1142 struct vmbus_channel_message_header
*msg
;
1143 struct vmbus_channel_msginfo
*msginfo
;
1146 msginfo
= kmalloc(sizeof(*msginfo
) +
1147 sizeof(struct vmbus_channel_message_header
),
1152 msg
= (struct vmbus_channel_message_header
*)msginfo
->msg
;
1154 msg
->msgtype
= CHANNELMSG_REQUESTOFFERS
;
1157 ret
= vmbus_post_msg(msg
, sizeof(struct vmbus_channel_message_header
),
1160 pr_err("Unable to request offers - %d\n", ret
);
1172 * Retrieve the (sub) channel on which to send an outgoing request.
1173 * When a primary channel has multiple sub-channels, we try to
1174 * distribute the load equally amongst all available channels.
1176 struct vmbus_channel
*vmbus_get_outgoing_channel(struct vmbus_channel
*primary
)
1178 struct list_head
*cur
, *tmp
;
1180 struct vmbus_channel
*cur_channel
;
1181 struct vmbus_channel
*outgoing_channel
= primary
;
1185 if (list_empty(&primary
->sc_list
))
1186 return outgoing_channel
;
1188 next_channel
= primary
->next_oc
++;
1190 if (next_channel
> (primary
->num_sc
)) {
1191 primary
->next_oc
= 0;
1192 return outgoing_channel
;
1195 cur_cpu
= hv_context
.vp_index
[get_cpu()];
1197 list_for_each_safe(cur
, tmp
, &primary
->sc_list
) {
1198 cur_channel
= list_entry(cur
, struct vmbus_channel
, sc_list
);
1199 if (cur_channel
->state
!= CHANNEL_OPENED_STATE
)
1202 if (cur_channel
->target_vp
== cur_cpu
)
1205 if (i
== next_channel
)
1211 return outgoing_channel
;
1213 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel
);
1215 static void invoke_sc_cb(struct vmbus_channel
*primary_channel
)
1217 struct list_head
*cur
, *tmp
;
1218 struct vmbus_channel
*cur_channel
;
1220 if (primary_channel
->sc_creation_callback
== NULL
)
1223 list_for_each_safe(cur
, tmp
, &primary_channel
->sc_list
) {
1224 cur_channel
= list_entry(cur
, struct vmbus_channel
, sc_list
);
1226 primary_channel
->sc_creation_callback(cur_channel
);
1230 void vmbus_set_sc_create_callback(struct vmbus_channel
*primary_channel
,
1231 void (*sc_cr_cb
)(struct vmbus_channel
*new_sc
))
1233 primary_channel
->sc_creation_callback
= sc_cr_cb
;
1235 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback
);
1237 bool vmbus_are_subchannels_present(struct vmbus_channel
*primary
)
1241 ret
= !list_empty(&primary
->sc_list
);
1245 * Invoke the callback on sub-channel creation.
1246 * This will present a uniform interface to the
1249 invoke_sc_cb(primary
);
1254 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present
);
1256 void vmbus_set_chn_rescind_callback(struct vmbus_channel
*channel
,
1257 void (*chn_rescind_cb
)(struct vmbus_channel
*))
1259 channel
->chn_rescind_callback
= chn_rescind_cb
;
1261 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback
);