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152 * The rescinded channel may be blocked waiting for a response from the host;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
);
162 channel
->rescind
= true;
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
->lock
);
337 INIT_LIST_HEAD(&channel
->sc_list
);
338 INIT_LIST_HEAD(&channel
->percpu_list
);
340 tasklet_init(&channel
->callback_event
,
341 vmbus_on_event
, (unsigned long)channel
);
347 * free_channel - Release the resources used by the vmbus channel object
349 static void free_channel(struct vmbus_channel
*channel
)
351 tasklet_kill(&channel
->callback_event
);
353 kfree_rcu(channel
, rcu
);
356 static void percpu_channel_enq(void *arg
)
358 struct vmbus_channel
*channel
= arg
;
359 struct hv_per_cpu_context
*hv_cpu
360 = this_cpu_ptr(hv_context
.cpu_context
);
362 list_add_tail_rcu(&channel
->percpu_list
, &hv_cpu
->chan_list
);
365 static void percpu_channel_deq(void *arg
)
367 struct vmbus_channel
*channel
= arg
;
369 list_del_rcu(&channel
->percpu_list
);
373 static void vmbus_release_relid(u32 relid
)
375 struct vmbus_channel_relid_released msg
;
377 memset(&msg
, 0, sizeof(struct vmbus_channel_relid_released
));
378 msg
.child_relid
= relid
;
379 msg
.header
.msgtype
= CHANNELMSG_RELID_RELEASED
;
380 vmbus_post_msg(&msg
, sizeof(struct vmbus_channel_relid_released
),
384 void hv_process_channel_removal(u32 relid
)
387 struct vmbus_channel
*primary_channel
, *channel
;
389 BUG_ON(!mutex_is_locked(&vmbus_connection
.channel_mutex
));
392 * Make sure channel is valid as we may have raced.
394 channel
= relid2channel(relid
);
398 BUG_ON(!channel
->rescind
);
399 if (channel
->target_cpu
!= get_cpu()) {
401 smp_call_function_single(channel
->target_cpu
,
402 percpu_channel_deq
, channel
, true);
404 percpu_channel_deq(channel
);
408 if (channel
->primary_channel
== NULL
) {
409 list_del(&channel
->listentry
);
411 primary_channel
= channel
;
413 primary_channel
= channel
->primary_channel
;
414 spin_lock_irqsave(&primary_channel
->lock
, flags
);
415 list_del(&channel
->sc_list
);
416 primary_channel
->num_sc
--;
417 spin_unlock_irqrestore(&primary_channel
->lock
, flags
);
421 * We need to free the bit for init_vp_index() to work in the case
422 * of sub-channel, when we reload drivers like hv_netvsc.
424 if (channel
->affinity_policy
== HV_LOCALIZED
)
425 cpumask_clear_cpu(channel
->target_cpu
,
426 &primary_channel
->alloced_cpus_in_node
);
428 vmbus_release_relid(relid
);
430 free_channel(channel
);
433 void vmbus_free_channels(void)
435 struct vmbus_channel
*channel
, *tmp
;
437 list_for_each_entry_safe(channel
, tmp
, &vmbus_connection
.chn_list
,
439 /* hv_process_channel_removal() needs this */
440 channel
->rescind
= true;
442 vmbus_device_unregister(channel
->device_obj
);
447 * vmbus_process_offer - Process the offer by creating a channel/device
448 * associated with this offer
450 static void vmbus_process_offer(struct vmbus_channel
*newchannel
)
452 struct vmbus_channel
*channel
;
458 /* Make sure this is a new offer */
459 mutex_lock(&vmbus_connection
.channel_mutex
);
462 * Now that we have acquired the channel_mutex,
463 * we can release the potentially racing rescind thread.
465 atomic_dec(&vmbus_connection
.offer_in_progress
);
467 list_for_each_entry(channel
, &vmbus_connection
.chn_list
, listentry
) {
468 if (!uuid_le_cmp(channel
->offermsg
.offer
.if_type
,
469 newchannel
->offermsg
.offer
.if_type
) &&
470 !uuid_le_cmp(channel
->offermsg
.offer
.if_instance
,
471 newchannel
->offermsg
.offer
.if_instance
)) {
478 list_add_tail(&newchannel
->listentry
,
479 &vmbus_connection
.chn_list
);
481 mutex_unlock(&vmbus_connection
.channel_mutex
);
485 * Check to see if this is a sub-channel.
487 if (newchannel
->offermsg
.offer
.sub_channel_index
!= 0) {
489 * Process the sub-channel.
491 newchannel
->primary_channel
= channel
;
492 spin_lock_irqsave(&channel
->lock
, flags
);
493 list_add_tail(&newchannel
->sc_list
, &channel
->sc_list
);
495 spin_unlock_irqrestore(&channel
->lock
, flags
);
501 dev_type
= hv_get_dev_type(newchannel
);
503 init_vp_index(newchannel
, dev_type
);
505 if (newchannel
->target_cpu
!= get_cpu()) {
507 smp_call_function_single(newchannel
->target_cpu
,
511 percpu_channel_enq(newchannel
);
516 * This state is used to indicate a successful open
517 * so that when we do close the channel normally, we
518 * can cleanup properly
520 newchannel
->state
= CHANNEL_OPEN_STATE
;
523 if (channel
->sc_creation_callback
!= NULL
)
524 channel
->sc_creation_callback(newchannel
);
525 newchannel
->probe_done
= true;
530 * Start the process of binding this offer to the driver
531 * We need to set the DeviceObject field before calling
532 * vmbus_child_dev_add()
534 newchannel
->device_obj
= vmbus_device_create(
535 &newchannel
->offermsg
.offer
.if_type
,
536 &newchannel
->offermsg
.offer
.if_instance
,
538 if (!newchannel
->device_obj
)
541 newchannel
->device_obj
->device_id
= dev_type
;
543 * Add the new device to the bus. This will kick off device-driver
544 * binding which eventually invokes the device driver's AddDevice()
547 ret
= vmbus_device_register(newchannel
->device_obj
);
550 pr_err("unable to add child device object (relid %d)\n",
551 newchannel
->offermsg
.child_relid
);
552 kfree(newchannel
->device_obj
);
556 newchannel
->probe_done
= true;
560 mutex_lock(&vmbus_connection
.channel_mutex
);
561 list_del(&newchannel
->listentry
);
562 mutex_unlock(&vmbus_connection
.channel_mutex
);
564 if (newchannel
->target_cpu
!= get_cpu()) {
566 smp_call_function_single(newchannel
->target_cpu
,
567 percpu_channel_deq
, newchannel
, true);
569 percpu_channel_deq(newchannel
);
573 vmbus_release_relid(newchannel
->offermsg
.child_relid
);
576 free_channel(newchannel
);
580 * We use this state to statically distribute the channel interrupt load.
582 static int next_numa_node_id
;
585 * Starting with Win8, we can statically distribute the incoming
586 * channel interrupt load by binding a channel to VCPU.
587 * We do this in a hierarchical fashion:
588 * First distribute the primary channels across available NUMA nodes
589 * and then distribute the subchannels amongst the CPUs in the NUMA
590 * node assigned to the primary channel.
592 * For pre-win8 hosts or non-performance critical channels we assign the
593 * first CPU in the first NUMA node.
595 static void init_vp_index(struct vmbus_channel
*channel
, u16 dev_type
)
598 bool perf_chn
= vmbus_devs
[dev_type
].perf_device
;
599 struct vmbus_channel
*primary
= channel
->primary_channel
;
601 struct cpumask available_mask
;
602 struct cpumask
*alloced_mask
;
604 if ((vmbus_proto_version
== VERSION_WS2008
) ||
605 (vmbus_proto_version
== VERSION_WIN7
) || (!perf_chn
)) {
607 * Prior to win8, all channel interrupts are
608 * delivered on cpu 0.
609 * Also if the channel is not a performance critical
610 * channel, bind it to cpu 0.
612 channel
->numa_node
= 0;
613 channel
->target_cpu
= 0;
614 channel
->target_vp
= hv_context
.vp_index
[0];
619 * Based on the channel affinity policy, we will assign the NUMA
623 if ((channel
->affinity_policy
== HV_BALANCED
) || (!primary
)) {
625 next_node
= next_numa_node_id
++;
626 if (next_node
== nr_node_ids
) {
627 next_node
= next_numa_node_id
= 0;
630 if (cpumask_empty(cpumask_of_node(next_node
)))
634 channel
->numa_node
= next_node
;
637 alloced_mask
= &hv_context
.hv_numa_map
[primary
->numa_node
];
639 if (cpumask_weight(alloced_mask
) ==
640 cpumask_weight(cpumask_of_node(primary
->numa_node
))) {
642 * We have cycled through all the CPUs in the node;
643 * reset the alloced map.
645 cpumask_clear(alloced_mask
);
648 cpumask_xor(&available_mask
, alloced_mask
,
649 cpumask_of_node(primary
->numa_node
));
653 if (primary
->affinity_policy
== HV_LOCALIZED
) {
655 * Normally Hyper-V host doesn't create more subchannels
656 * than there are VCPUs on the node but it is possible when not
657 * all present VCPUs on the node are initialized by guest.
658 * Clear the alloced_cpus_in_node to start over.
660 if (cpumask_equal(&primary
->alloced_cpus_in_node
,
661 cpumask_of_node(primary
->numa_node
)))
662 cpumask_clear(&primary
->alloced_cpus_in_node
);
666 cur_cpu
= cpumask_next(cur_cpu
, &available_mask
);
667 if (cur_cpu
>= nr_cpu_ids
) {
669 cpumask_copy(&available_mask
,
670 cpumask_of_node(primary
->numa_node
));
674 if (primary
->affinity_policy
== HV_LOCALIZED
) {
676 * NOTE: in the case of sub-channel, we clear the
677 * sub-channel related bit(s) in
678 * primary->alloced_cpus_in_node in
679 * hv_process_channel_removal(), so when we
680 * reload drivers like hv_netvsc in SMP guest, here
681 * we're able to re-allocate
682 * bit from primary->alloced_cpus_in_node.
684 if (!cpumask_test_cpu(cur_cpu
,
685 &primary
->alloced_cpus_in_node
)) {
686 cpumask_set_cpu(cur_cpu
,
687 &primary
->alloced_cpus_in_node
);
688 cpumask_set_cpu(cur_cpu
, alloced_mask
);
692 cpumask_set_cpu(cur_cpu
, alloced_mask
);
697 channel
->target_cpu
= cur_cpu
;
698 channel
->target_vp
= hv_context
.vp_index
[cur_cpu
];
701 static void vmbus_wait_for_unload(void)
705 struct hv_message
*msg
;
706 struct vmbus_channel_message_header
*hdr
;
710 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
711 * used for initial contact or to CPU0 depending on host version. When
712 * we're crashing on a different CPU let's hope that IRQ handler on
713 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
714 * functional and vmbus_unload_response() will complete
715 * vmbus_connection.unload_event. If not, the last thing we can do is
716 * read message pages for all CPUs directly.
719 if (completion_done(&vmbus_connection
.unload_event
))
722 for_each_online_cpu(cpu
) {
723 struct hv_per_cpu_context
*hv_cpu
724 = per_cpu_ptr(hv_context
.cpu_context
, cpu
);
726 page_addr
= hv_cpu
->synic_message_page
;
727 msg
= (struct hv_message
*)page_addr
728 + VMBUS_MESSAGE_SINT
;
730 message_type
= READ_ONCE(msg
->header
.message_type
);
731 if (message_type
== HVMSG_NONE
)
734 hdr
= (struct vmbus_channel_message_header
*)
737 if (hdr
->msgtype
== CHANNELMSG_UNLOAD_RESPONSE
)
738 complete(&vmbus_connection
.unload_event
);
740 vmbus_signal_eom(msg
, message_type
);
747 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
748 * maybe-pending messages on all CPUs to be able to receive new
749 * messages after we reconnect.
751 for_each_online_cpu(cpu
) {
752 struct hv_per_cpu_context
*hv_cpu
753 = per_cpu_ptr(hv_context
.cpu_context
, cpu
);
755 page_addr
= hv_cpu
->synic_message_page
;
756 msg
= (struct hv_message
*)page_addr
+ VMBUS_MESSAGE_SINT
;
757 msg
->header
.message_type
= HVMSG_NONE
;
762 * vmbus_unload_response - Handler for the unload response.
764 static void vmbus_unload_response(struct vmbus_channel_message_header
*hdr
)
767 * This is a global event; just wakeup the waiting thread.
768 * Once we successfully unload, we can cleanup the monitor state.
770 complete(&vmbus_connection
.unload_event
);
773 void vmbus_initiate_unload(bool crash
)
775 struct vmbus_channel_message_header hdr
;
777 /* Pre-Win2012R2 hosts don't support reconnect */
778 if (vmbus_proto_version
< VERSION_WIN8_1
)
781 init_completion(&vmbus_connection
.unload_event
);
782 memset(&hdr
, 0, sizeof(struct vmbus_channel_message_header
));
783 hdr
.msgtype
= CHANNELMSG_UNLOAD
;
784 vmbus_post_msg(&hdr
, sizeof(struct vmbus_channel_message_header
),
788 * vmbus_initiate_unload() is also called on crash and the crash can be
789 * happening in an interrupt context, where scheduling is impossible.
792 wait_for_completion(&vmbus_connection
.unload_event
);
794 vmbus_wait_for_unload();
798 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
801 static void vmbus_onoffer(struct vmbus_channel_message_header
*hdr
)
803 struct vmbus_channel_offer_channel
*offer
;
804 struct vmbus_channel
*newchannel
;
806 offer
= (struct vmbus_channel_offer_channel
*)hdr
;
808 /* Allocate the channel object and save this offer. */
809 newchannel
= alloc_channel();
811 vmbus_release_relid(offer
->child_relid
);
812 atomic_dec(&vmbus_connection
.offer_in_progress
);
813 pr_err("Unable to allocate channel object\n");
818 * Setup state for signalling the host.
820 newchannel
->sig_event
= (struct hv_input_signal_event
*)
821 (ALIGN((unsigned long)
822 &newchannel
->sig_buf
,
823 HV_HYPERCALL_PARAM_ALIGN
));
825 newchannel
->sig_event
->connectionid
.asu32
= 0;
826 newchannel
->sig_event
->connectionid
.u
.id
= VMBUS_EVENT_CONNECTION_ID
;
827 newchannel
->sig_event
->flag_number
= 0;
828 newchannel
->sig_event
->rsvdz
= 0;
830 if (vmbus_proto_version
!= VERSION_WS2008
) {
831 newchannel
->is_dedicated_interrupt
=
832 (offer
->is_dedicated_interrupt
!= 0);
833 newchannel
->sig_event
->connectionid
.u
.id
=
834 offer
->connection_id
;
837 memcpy(&newchannel
->offermsg
, offer
,
838 sizeof(struct vmbus_channel_offer_channel
));
839 newchannel
->monitor_grp
= (u8
)offer
->monitorid
/ 32;
840 newchannel
->monitor_bit
= (u8
)offer
->monitorid
% 32;
842 vmbus_process_offer(newchannel
);
846 * vmbus_onoffer_rescind - Rescind offer handler.
848 * We queue a work item to process this offer synchronously
850 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header
*hdr
)
852 struct vmbus_channel_rescind_offer
*rescind
;
853 struct vmbus_channel
*channel
;
856 rescind
= (struct vmbus_channel_rescind_offer
*)hdr
;
859 * The offer msg and the corresponding rescind msg
860 * from the host are guranteed to be ordered -
861 * offer comes in first and then the rescind.
862 * Since we process these events in work elements,
863 * and with preemption, we may end up processing
864 * the events out of order. Given that we handle these
865 * work elements on the same CPU, this is possible only
866 * in the case of preemption. In any case wait here
867 * until the offer processing has moved beyond the
868 * point where the channel is discoverable.
871 while (atomic_read(&vmbus_connection
.offer_in_progress
) != 0) {
873 * We wait here until any channel offer is currently
879 mutex_lock(&vmbus_connection
.channel_mutex
);
880 channel
= relid2channel(rescind
->child_relid
);
881 mutex_unlock(&vmbus_connection
.channel_mutex
);
883 if (channel
== NULL
) {
885 * We failed in processing the offer message;
886 * we would have cleaned up the relid in that
893 * Now wait for offer handling to complete.
895 while (READ_ONCE(channel
->probe_done
) == false) {
897 * We wait here until any channel offer is currently
904 * At this point, the rescind handling can proceed safely.
907 if (channel
->device_obj
) {
908 if (channel
->chn_rescind_callback
) {
909 channel
->chn_rescind_callback(channel
);
910 vmbus_rescind_cleanup(channel
);
914 * We will have to unregister this device from the
917 dev
= get_device(&channel
->device_obj
->device
);
919 vmbus_rescind_cleanup(channel
);
920 vmbus_device_unregister(channel
->device_obj
);
924 if (channel
->primary_channel
!= NULL
) {
926 * Sub-channel is being rescinded. Following is the channel
927 * close sequence when initiated from the driveri (refer to
928 * vmbus_close() for details):
929 * 1. Close all sub-channels first
930 * 2. Then close the primary channel.
932 mutex_lock(&vmbus_connection
.channel_mutex
);
933 vmbus_rescind_cleanup(channel
);
934 if (channel
->state
== CHANNEL_OPEN_STATE
) {
936 * The channel is currently not open;
937 * it is safe for us to cleanup the channel.
939 hv_process_channel_removal(rescind
->child_relid
);
941 mutex_unlock(&vmbus_connection
.channel_mutex
);
945 void vmbus_hvsock_device_unregister(struct vmbus_channel
*channel
)
947 BUG_ON(!is_hvsock_channel(channel
));
949 channel
->rescind
= true;
950 vmbus_device_unregister(channel
->device_obj
);
952 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister
);
956 * vmbus_onoffers_delivered -
957 * This is invoked when all offers have been delivered.
959 * Nothing to do here.
961 static void vmbus_onoffers_delivered(
962 struct vmbus_channel_message_header
*hdr
)
967 * vmbus_onopen_result - Open result handler.
969 * This is invoked when we received a response to our channel open request.
970 * Find the matching request, copy the response and signal the requesting
973 static void vmbus_onopen_result(struct vmbus_channel_message_header
*hdr
)
975 struct vmbus_channel_open_result
*result
;
976 struct vmbus_channel_msginfo
*msginfo
;
977 struct vmbus_channel_message_header
*requestheader
;
978 struct vmbus_channel_open_channel
*openmsg
;
981 result
= (struct vmbus_channel_open_result
*)hdr
;
984 * Find the open msg, copy the result and signal/unblock the wait event
986 spin_lock_irqsave(&vmbus_connection
.channelmsg_lock
, flags
);
988 list_for_each_entry(msginfo
, &vmbus_connection
.chn_msg_list
,
991 (struct vmbus_channel_message_header
*)msginfo
->msg
;
993 if (requestheader
->msgtype
== CHANNELMSG_OPENCHANNEL
) {
995 (struct vmbus_channel_open_channel
*)msginfo
->msg
;
996 if (openmsg
->child_relid
== result
->child_relid
&&
997 openmsg
->openid
== result
->openid
) {
998 memcpy(&msginfo
->response
.open_result
,
1001 struct vmbus_channel_open_result
));
1002 complete(&msginfo
->waitevent
);
1007 spin_unlock_irqrestore(&vmbus_connection
.channelmsg_lock
, flags
);
1011 * vmbus_ongpadl_created - GPADL created handler.
1013 * This is invoked when we received a response to our gpadl create request.
1014 * Find the matching request, copy the response and signal the requesting
1017 static void vmbus_ongpadl_created(struct vmbus_channel_message_header
*hdr
)
1019 struct vmbus_channel_gpadl_created
*gpadlcreated
;
1020 struct vmbus_channel_msginfo
*msginfo
;
1021 struct vmbus_channel_message_header
*requestheader
;
1022 struct vmbus_channel_gpadl_header
*gpadlheader
;
1023 unsigned long flags
;
1025 gpadlcreated
= (struct vmbus_channel_gpadl_created
*)hdr
;
1028 * Find the establish msg, copy the result and signal/unblock the wait
1031 spin_lock_irqsave(&vmbus_connection
.channelmsg_lock
, flags
);
1033 list_for_each_entry(msginfo
, &vmbus_connection
.chn_msg_list
,
1036 (struct vmbus_channel_message_header
*)msginfo
->msg
;
1038 if (requestheader
->msgtype
== CHANNELMSG_GPADL_HEADER
) {
1040 (struct vmbus_channel_gpadl_header
*)requestheader
;
1042 if ((gpadlcreated
->child_relid
==
1043 gpadlheader
->child_relid
) &&
1044 (gpadlcreated
->gpadl
== gpadlheader
->gpadl
)) {
1045 memcpy(&msginfo
->response
.gpadl_created
,
1048 struct vmbus_channel_gpadl_created
));
1049 complete(&msginfo
->waitevent
);
1054 spin_unlock_irqrestore(&vmbus_connection
.channelmsg_lock
, flags
);
1058 * vmbus_ongpadl_torndown - GPADL torndown handler.
1060 * This is invoked when we received a response to our gpadl teardown request.
1061 * Find the matching request, copy the response and signal the requesting
1064 static void vmbus_ongpadl_torndown(
1065 struct vmbus_channel_message_header
*hdr
)
1067 struct vmbus_channel_gpadl_torndown
*gpadl_torndown
;
1068 struct vmbus_channel_msginfo
*msginfo
;
1069 struct vmbus_channel_message_header
*requestheader
;
1070 struct vmbus_channel_gpadl_teardown
*gpadl_teardown
;
1071 unsigned long flags
;
1073 gpadl_torndown
= (struct vmbus_channel_gpadl_torndown
*)hdr
;
1076 * Find the open msg, copy the result and signal/unblock the wait event
1078 spin_lock_irqsave(&vmbus_connection
.channelmsg_lock
, flags
);
1080 list_for_each_entry(msginfo
, &vmbus_connection
.chn_msg_list
,
1083 (struct vmbus_channel_message_header
*)msginfo
->msg
;
1085 if (requestheader
->msgtype
== CHANNELMSG_GPADL_TEARDOWN
) {
1087 (struct vmbus_channel_gpadl_teardown
*)requestheader
;
1089 if (gpadl_torndown
->gpadl
== gpadl_teardown
->gpadl
) {
1090 memcpy(&msginfo
->response
.gpadl_torndown
,
1093 struct vmbus_channel_gpadl_torndown
));
1094 complete(&msginfo
->waitevent
);
1099 spin_unlock_irqrestore(&vmbus_connection
.channelmsg_lock
, flags
);
1103 * vmbus_onversion_response - Version response handler
1105 * This is invoked when we received a response to our initiate contact request.
1106 * Find the matching request, copy the response and signal the requesting
1109 static void vmbus_onversion_response(
1110 struct vmbus_channel_message_header
*hdr
)
1112 struct vmbus_channel_msginfo
*msginfo
;
1113 struct vmbus_channel_message_header
*requestheader
;
1114 struct vmbus_channel_version_response
*version_response
;
1115 unsigned long flags
;
1117 version_response
= (struct vmbus_channel_version_response
*)hdr
;
1118 spin_lock_irqsave(&vmbus_connection
.channelmsg_lock
, flags
);
1120 list_for_each_entry(msginfo
, &vmbus_connection
.chn_msg_list
,
1123 (struct vmbus_channel_message_header
*)msginfo
->msg
;
1125 if (requestheader
->msgtype
==
1126 CHANNELMSG_INITIATE_CONTACT
) {
1127 memcpy(&msginfo
->response
.version_response
,
1129 sizeof(struct vmbus_channel_version_response
));
1130 complete(&msginfo
->waitevent
);
1133 spin_unlock_irqrestore(&vmbus_connection
.channelmsg_lock
, flags
);
1136 /* Channel message dispatch table */
1137 const struct vmbus_channel_message_table_entry
1138 channel_message_table
[CHANNELMSG_COUNT
] = {
1139 { CHANNELMSG_INVALID
, 0, NULL
},
1140 { CHANNELMSG_OFFERCHANNEL
, 0, vmbus_onoffer
},
1141 { CHANNELMSG_RESCIND_CHANNELOFFER
, 0, vmbus_onoffer_rescind
},
1142 { CHANNELMSG_REQUESTOFFERS
, 0, NULL
},
1143 { CHANNELMSG_ALLOFFERS_DELIVERED
, 1, vmbus_onoffers_delivered
},
1144 { CHANNELMSG_OPENCHANNEL
, 0, NULL
},
1145 { CHANNELMSG_OPENCHANNEL_RESULT
, 1, vmbus_onopen_result
},
1146 { CHANNELMSG_CLOSECHANNEL
, 0, NULL
},
1147 { CHANNELMSG_GPADL_HEADER
, 0, NULL
},
1148 { CHANNELMSG_GPADL_BODY
, 0, NULL
},
1149 { CHANNELMSG_GPADL_CREATED
, 1, vmbus_ongpadl_created
},
1150 { CHANNELMSG_GPADL_TEARDOWN
, 0, NULL
},
1151 { CHANNELMSG_GPADL_TORNDOWN
, 1, vmbus_ongpadl_torndown
},
1152 { CHANNELMSG_RELID_RELEASED
, 0, NULL
},
1153 { CHANNELMSG_INITIATE_CONTACT
, 0, NULL
},
1154 { CHANNELMSG_VERSION_RESPONSE
, 1, vmbus_onversion_response
},
1155 { CHANNELMSG_UNLOAD
, 0, NULL
},
1156 { CHANNELMSG_UNLOAD_RESPONSE
, 1, vmbus_unload_response
},
1157 { CHANNELMSG_18
, 0, NULL
},
1158 { CHANNELMSG_19
, 0, NULL
},
1159 { CHANNELMSG_20
, 0, NULL
},
1160 { CHANNELMSG_TL_CONNECT_REQUEST
, 0, NULL
},
1164 * vmbus_onmessage - Handler for channel protocol messages.
1166 * This is invoked in the vmbus worker thread context.
1168 void vmbus_onmessage(void *context
)
1170 struct hv_message
*msg
= context
;
1171 struct vmbus_channel_message_header
*hdr
;
1174 hdr
= (struct vmbus_channel_message_header
*)msg
->u
.payload
;
1175 size
= msg
->header
.payload_size
;
1177 if (hdr
->msgtype
>= CHANNELMSG_COUNT
) {
1178 pr_err("Received invalid channel message type %d size %d\n",
1179 hdr
->msgtype
, size
);
1180 print_hex_dump_bytes("", DUMP_PREFIX_NONE
,
1181 (unsigned char *)msg
->u
.payload
, size
);
1185 if (channel_message_table
[hdr
->msgtype
].message_handler
)
1186 channel_message_table
[hdr
->msgtype
].message_handler(hdr
);
1188 pr_err("Unhandled channel message type %d\n", hdr
->msgtype
);
1192 * vmbus_request_offers - Send a request to get all our pending offers.
1194 int vmbus_request_offers(void)
1196 struct vmbus_channel_message_header
*msg
;
1197 struct vmbus_channel_msginfo
*msginfo
;
1200 msginfo
= kmalloc(sizeof(*msginfo
) +
1201 sizeof(struct vmbus_channel_message_header
),
1206 msg
= (struct vmbus_channel_message_header
*)msginfo
->msg
;
1208 msg
->msgtype
= CHANNELMSG_REQUESTOFFERS
;
1211 ret
= vmbus_post_msg(msg
, sizeof(struct vmbus_channel_message_header
),
1214 pr_err("Unable to request offers - %d\n", ret
);
1226 * Retrieve the (sub) channel on which to send an outgoing request.
1227 * When a primary channel has multiple sub-channels, we try to
1228 * distribute the load equally amongst all available channels.
1230 struct vmbus_channel
*vmbus_get_outgoing_channel(struct vmbus_channel
*primary
)
1232 struct list_head
*cur
, *tmp
;
1234 struct vmbus_channel
*cur_channel
;
1235 struct vmbus_channel
*outgoing_channel
= primary
;
1239 if (list_empty(&primary
->sc_list
))
1240 return outgoing_channel
;
1242 next_channel
= primary
->next_oc
++;
1244 if (next_channel
> (primary
->num_sc
)) {
1245 primary
->next_oc
= 0;
1246 return outgoing_channel
;
1249 cur_cpu
= hv_context
.vp_index
[get_cpu()];
1251 list_for_each_safe(cur
, tmp
, &primary
->sc_list
) {
1252 cur_channel
= list_entry(cur
, struct vmbus_channel
, sc_list
);
1253 if (cur_channel
->state
!= CHANNEL_OPENED_STATE
)
1256 if (cur_channel
->target_vp
== cur_cpu
)
1259 if (i
== next_channel
)
1265 return outgoing_channel
;
1267 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel
);
1269 static void invoke_sc_cb(struct vmbus_channel
*primary_channel
)
1271 struct list_head
*cur
, *tmp
;
1272 struct vmbus_channel
*cur_channel
;
1274 if (primary_channel
->sc_creation_callback
== NULL
)
1277 list_for_each_safe(cur
, tmp
, &primary_channel
->sc_list
) {
1278 cur_channel
= list_entry(cur
, struct vmbus_channel
, sc_list
);
1280 primary_channel
->sc_creation_callback(cur_channel
);
1284 void vmbus_set_sc_create_callback(struct vmbus_channel
*primary_channel
,
1285 void (*sc_cr_cb
)(struct vmbus_channel
*new_sc
))
1287 primary_channel
->sc_creation_callback
= sc_cr_cb
;
1289 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback
);
1291 bool vmbus_are_subchannels_present(struct vmbus_channel
*primary
)
1295 ret
= !list_empty(&primary
->sc_list
);
1299 * Invoke the callback on sub-channel creation.
1300 * This will present a uniform interface to the
1303 invoke_sc_cb(primary
);
1308 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present
);
1310 void vmbus_set_chn_rescind_callback(struct vmbus_channel
*channel
,
1311 void (*chn_rescind_cb
)(struct vmbus_channel
*))
1313 channel
->chn_rescind_callback
= chn_rescind_cb
;
1315 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback
);