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152 * The rescinded channel may be blocked waiting for a response from the host;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 kobject_put(&channel->kobj);
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;
378 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
379 msg.child_relid = relid;
380 msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
381 ret = vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released),
384 trace_vmbus_release_relid(&msg, ret);
387 void hv_process_channel_removal(u32 relid)
390 struct vmbus_channel *primary_channel, *channel;
392 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
395 * Make sure channel is valid as we may have raced.
397 channel = relid2channel(relid);
401 BUG_ON(!channel->rescind);
402 if (channel->target_cpu != get_cpu()) {
404 smp_call_function_single(channel->target_cpu,
405 percpu_channel_deq, channel, true);
407 percpu_channel_deq(channel);
411 if (channel->primary_channel == NULL) {
412 list_del(&channel->listentry);
414 primary_channel = channel;
416 primary_channel = channel->primary_channel;
417 spin_lock_irqsave(&primary_channel->lock, flags);
418 list_del(&channel->sc_list);
419 primary_channel->num_sc--;
420 spin_unlock_irqrestore(&primary_channel->lock, flags);
424 * We need to free the bit for init_vp_index() to work in the case
425 * of sub-channel, when we reload drivers like hv_netvsc.
427 if (channel->affinity_policy == HV_LOCALIZED)
428 cpumask_clear_cpu(channel->target_cpu,
429 &primary_channel->alloced_cpus_in_node);
431 vmbus_release_relid(relid);
433 free_channel(channel);
436 void vmbus_free_channels(void)
438 struct vmbus_channel *channel, *tmp;
440 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
442 /* hv_process_channel_removal() needs this */
443 channel->rescind = true;
445 vmbus_device_unregister(channel->device_obj);
450 * vmbus_process_offer - Process the offer by creating a channel/device
451 * associated with this offer
453 static void vmbus_process_offer(struct vmbus_channel *newchannel)
455 struct vmbus_channel *channel;
461 /* Make sure this is a new offer */
462 mutex_lock(&vmbus_connection.channel_mutex);
465 * Now that we have acquired the channel_mutex,
466 * we can release the potentially racing rescind thread.
468 atomic_dec(&vmbus_connection.offer_in_progress);
470 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
471 if (!uuid_le_cmp(channel->offermsg.offer.if_type,
472 newchannel->offermsg.offer.if_type) &&
473 !uuid_le_cmp(channel->offermsg.offer.if_instance,
474 newchannel->offermsg.offer.if_instance)) {
481 list_add_tail(&newchannel->listentry,
482 &vmbus_connection.chn_list);
484 mutex_unlock(&vmbus_connection.channel_mutex);
488 * Check to see if this is a sub-channel.
490 if (newchannel->offermsg.offer.sub_channel_index != 0) {
492 * Process the sub-channel.
494 newchannel->primary_channel = channel;
495 spin_lock_irqsave(&channel->lock, flags);
496 list_add_tail(&newchannel->sc_list, &channel->sc_list);
498 spin_unlock_irqrestore(&channel->lock, flags);
504 dev_type = hv_get_dev_type(newchannel);
506 init_vp_index(newchannel, dev_type);
508 if (newchannel->target_cpu != get_cpu()) {
510 smp_call_function_single(newchannel->target_cpu,
514 percpu_channel_enq(newchannel);
519 * This state is used to indicate a successful open
520 * so that when we do close the channel normally, we
521 * can cleanup properly
523 newchannel->state = CHANNEL_OPEN_STATE;
526 struct hv_device *dev
527 = newchannel->primary_channel->device_obj;
529 if (vmbus_add_channel_kobj(dev, newchannel)) {
530 atomic_dec(&vmbus_connection.offer_in_progress);
534 if (channel->sc_creation_callback != NULL)
535 channel->sc_creation_callback(newchannel);
536 newchannel->probe_done = true;
541 * Start the process of binding this offer to the driver
542 * We need to set the DeviceObject field before calling
543 * vmbus_child_dev_add()
545 newchannel->device_obj = vmbus_device_create(
546 &newchannel->offermsg.offer.if_type,
547 &newchannel->offermsg.offer.if_instance,
549 if (!newchannel->device_obj)
552 newchannel->device_obj->device_id = dev_type;
554 * Add the new device to the bus. This will kick off device-driver
555 * binding which eventually invokes the device driver's AddDevice()
558 ret = vmbus_device_register(newchannel->device_obj);
561 pr_err("unable to add child device object (relid %d)\n",
562 newchannel->offermsg.child_relid);
563 kfree(newchannel->device_obj);
567 newchannel->probe_done = true;
571 mutex_lock(&vmbus_connection.channel_mutex);
572 list_del(&newchannel->listentry);
573 mutex_unlock(&vmbus_connection.channel_mutex);
575 if (newchannel->target_cpu != get_cpu()) {
577 smp_call_function_single(newchannel->target_cpu,
578 percpu_channel_deq, newchannel, true);
580 percpu_channel_deq(newchannel);
584 vmbus_release_relid(newchannel->offermsg.child_relid);
587 free_channel(newchannel);
591 * We use this state to statically distribute the channel interrupt load.
593 static int next_numa_node_id;
596 * Starting with Win8, we can statically distribute the incoming
597 * channel interrupt load by binding a channel to VCPU.
598 * We do this in a hierarchical fashion:
599 * First distribute the primary channels across available NUMA nodes
600 * and then distribute the subchannels amongst the CPUs in the NUMA
601 * node assigned to the primary channel.
603 * For pre-win8 hosts or non-performance critical channels we assign the
604 * first CPU in the first NUMA node.
606 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
609 bool perf_chn = vmbus_devs[dev_type].perf_device;
610 struct vmbus_channel *primary = channel->primary_channel;
612 struct cpumask available_mask;
613 struct cpumask *alloced_mask;
615 if ((vmbus_proto_version == VERSION_WS2008) ||
616 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn)) {
618 * Prior to win8, all channel interrupts are
619 * delivered on cpu 0.
620 * Also if the channel is not a performance critical
621 * channel, bind it to cpu 0.
623 channel->numa_node = 0;
624 channel->target_cpu = 0;
625 channel->target_vp = hv_cpu_number_to_vp_number(0);
630 * Based on the channel affinity policy, we will assign the NUMA
634 if ((channel->affinity_policy == HV_BALANCED) || (!primary)) {
636 next_node = next_numa_node_id++;
637 if (next_node == nr_node_ids) {
638 next_node = next_numa_node_id = 0;
641 if (cpumask_empty(cpumask_of_node(next_node)))
645 channel->numa_node = next_node;
648 alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
650 if (cpumask_weight(alloced_mask) ==
651 cpumask_weight(cpumask_of_node(primary->numa_node))) {
653 * We have cycled through all the CPUs in the node;
654 * reset the alloced map.
656 cpumask_clear(alloced_mask);
659 cpumask_xor(&available_mask, alloced_mask,
660 cpumask_of_node(primary->numa_node));
664 if (primary->affinity_policy == HV_LOCALIZED) {
666 * Normally Hyper-V host doesn't create more subchannels
667 * than there are VCPUs on the node but it is possible when not
668 * all present VCPUs on the node are initialized by guest.
669 * Clear the alloced_cpus_in_node to start over.
671 if (cpumask_equal(&primary->alloced_cpus_in_node,
672 cpumask_of_node(primary->numa_node)))
673 cpumask_clear(&primary->alloced_cpus_in_node);
677 cur_cpu = cpumask_next(cur_cpu, &available_mask);
678 if (cur_cpu >= nr_cpu_ids) {
680 cpumask_copy(&available_mask,
681 cpumask_of_node(primary->numa_node));
685 if (primary->affinity_policy == HV_LOCALIZED) {
687 * NOTE: in the case of sub-channel, we clear the
688 * sub-channel related bit(s) in
689 * primary->alloced_cpus_in_node in
690 * hv_process_channel_removal(), so when we
691 * reload drivers like hv_netvsc in SMP guest, here
692 * we're able to re-allocate
693 * bit from primary->alloced_cpus_in_node.
695 if (!cpumask_test_cpu(cur_cpu,
696 &primary->alloced_cpus_in_node)) {
697 cpumask_set_cpu(cur_cpu,
698 &primary->alloced_cpus_in_node);
699 cpumask_set_cpu(cur_cpu, alloced_mask);
703 cpumask_set_cpu(cur_cpu, alloced_mask);
708 channel->target_cpu = cur_cpu;
709 channel->target_vp = hv_cpu_number_to_vp_number(cur_cpu);
712 static void vmbus_wait_for_unload(void)
716 struct hv_message *msg;
717 struct vmbus_channel_message_header *hdr;
721 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
722 * used for initial contact or to CPU0 depending on host version. When
723 * we're crashing on a different CPU let's hope that IRQ handler on
724 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
725 * functional and vmbus_unload_response() will complete
726 * vmbus_connection.unload_event. If not, the last thing we can do is
727 * read message pages for all CPUs directly.
730 if (completion_done(&vmbus_connection.unload_event))
733 for_each_online_cpu(cpu) {
734 struct hv_per_cpu_context *hv_cpu
735 = per_cpu_ptr(hv_context.cpu_context, cpu);
737 page_addr = hv_cpu->synic_message_page;
738 msg = (struct hv_message *)page_addr
739 + VMBUS_MESSAGE_SINT;
741 message_type = READ_ONCE(msg->header.message_type);
742 if (message_type == HVMSG_NONE)
745 hdr = (struct vmbus_channel_message_header *)
748 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
749 complete(&vmbus_connection.unload_event);
751 vmbus_signal_eom(msg, message_type);
758 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
759 * maybe-pending messages on all CPUs to be able to receive new
760 * messages after we reconnect.
762 for_each_online_cpu(cpu) {
763 struct hv_per_cpu_context *hv_cpu
764 = per_cpu_ptr(hv_context.cpu_context, cpu);
766 page_addr = hv_cpu->synic_message_page;
767 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
768 msg->header.message_type = HVMSG_NONE;
773 * vmbus_unload_response - Handler for the unload response.
775 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
778 * This is a global event; just wakeup the waiting thread.
779 * Once we successfully unload, we can cleanup the monitor state.
781 complete(&vmbus_connection.unload_event);
784 void vmbus_initiate_unload(bool crash)
786 struct vmbus_channel_message_header hdr;
788 /* Pre-Win2012R2 hosts don't support reconnect */
789 if (vmbus_proto_version < VERSION_WIN8_1)
792 init_completion(&vmbus_connection.unload_event);
793 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
794 hdr.msgtype = CHANNELMSG_UNLOAD;
795 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
799 * vmbus_initiate_unload() is also called on crash and the crash can be
800 * happening in an interrupt context, where scheduling is impossible.
803 wait_for_completion(&vmbus_connection.unload_event);
805 vmbus_wait_for_unload();
809 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
812 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
814 struct vmbus_channel_offer_channel *offer;
815 struct vmbus_channel *newchannel;
817 offer = (struct vmbus_channel_offer_channel *)hdr;
819 trace_vmbus_onoffer(offer);
821 /* Allocate the channel object and save this offer. */
822 newchannel = alloc_channel();
824 vmbus_release_relid(offer->child_relid);
825 atomic_dec(&vmbus_connection.offer_in_progress);
826 pr_err("Unable to allocate channel object\n");
831 * Setup state for signalling the host.
833 newchannel->sig_event = VMBUS_EVENT_CONNECTION_ID;
835 if (vmbus_proto_version != VERSION_WS2008) {
836 newchannel->is_dedicated_interrupt =
837 (offer->is_dedicated_interrupt != 0);
838 newchannel->sig_event = offer->connection_id;
841 memcpy(&newchannel->offermsg, offer,
842 sizeof(struct vmbus_channel_offer_channel));
843 newchannel->monitor_grp = (u8)offer->monitorid / 32;
844 newchannel->monitor_bit = (u8)offer->monitorid % 32;
846 vmbus_process_offer(newchannel);
850 * vmbus_onoffer_rescind - Rescind offer handler.
852 * We queue a work item to process this offer synchronously
854 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
856 struct vmbus_channel_rescind_offer *rescind;
857 struct vmbus_channel *channel;
860 rescind = (struct vmbus_channel_rescind_offer *)hdr;
862 trace_vmbus_onoffer_rescind(rescind);
865 * The offer msg and the corresponding rescind msg
866 * from the host are guranteed to be ordered -
867 * offer comes in first and then the rescind.
868 * Since we process these events in work elements,
869 * and with preemption, we may end up processing
870 * the events out of order. Given that we handle these
871 * work elements on the same CPU, this is possible only
872 * in the case of preemption. In any case wait here
873 * until the offer processing has moved beyond the
874 * point where the channel is discoverable.
877 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
879 * We wait here until any channel offer is currently
885 mutex_lock(&vmbus_connection.channel_mutex);
886 channel = relid2channel(rescind->child_relid);
887 mutex_unlock(&vmbus_connection.channel_mutex);
889 if (channel == NULL) {
891 * We failed in processing the offer message;
892 * we would have cleaned up the relid in that
899 * Now wait for offer handling to complete.
901 while (READ_ONCE(channel->probe_done) == false) {
903 * We wait here until any channel offer is currently
910 * At this point, the rescind handling can proceed safely.
913 if (channel->device_obj) {
914 if (channel->chn_rescind_callback) {
915 channel->chn_rescind_callback(channel);
916 vmbus_rescind_cleanup(channel);
920 * We will have to unregister this device from the
923 dev = get_device(&channel->device_obj->device);
925 vmbus_rescind_cleanup(channel);
926 vmbus_device_unregister(channel->device_obj);
930 if (channel->primary_channel != NULL) {
932 * Sub-channel is being rescinded. Following is the channel
933 * close sequence when initiated from the driveri (refer to
934 * vmbus_close() for details):
935 * 1. Close all sub-channels first
936 * 2. Then close the primary channel.
938 mutex_lock(&vmbus_connection.channel_mutex);
939 vmbus_rescind_cleanup(channel);
940 if (channel->state == CHANNEL_OPEN_STATE) {
942 * The channel is currently not open;
943 * it is safe for us to cleanup the channel.
945 hv_process_channel_removal(rescind->child_relid);
947 mutex_unlock(&vmbus_connection.channel_mutex);
951 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
953 BUG_ON(!is_hvsock_channel(channel));
955 /* We always get a rescind msg when a connection is closed. */
956 while (!READ_ONCE(channel->probe_done) || !READ_ONCE(channel->rescind))
959 vmbus_device_unregister(channel->device_obj);
961 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
965 * vmbus_onoffers_delivered -
966 * This is invoked when all offers have been delivered.
968 * Nothing to do here.
970 static void vmbus_onoffers_delivered(
971 struct vmbus_channel_message_header *hdr)
976 * vmbus_onopen_result - Open result handler.
978 * This is invoked when we received a response to our channel open request.
979 * Find the matching request, copy the response and signal the requesting
982 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
984 struct vmbus_channel_open_result *result;
985 struct vmbus_channel_msginfo *msginfo;
986 struct vmbus_channel_message_header *requestheader;
987 struct vmbus_channel_open_channel *openmsg;
990 result = (struct vmbus_channel_open_result *)hdr;
992 trace_vmbus_onopen_result(result);
995 * Find the open msg, copy the result and signal/unblock the wait event
997 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
999 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1002 (struct vmbus_channel_message_header *)msginfo->msg;
1004 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
1006 (struct vmbus_channel_open_channel *)msginfo->msg;
1007 if (openmsg->child_relid == result->child_relid &&
1008 openmsg->openid == result->openid) {
1009 memcpy(&msginfo->response.open_result,
1012 struct vmbus_channel_open_result));
1013 complete(&msginfo->waitevent);
1018 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1022 * vmbus_ongpadl_created - GPADL created handler.
1024 * This is invoked when we received a response to our gpadl create request.
1025 * Find the matching request, copy the response and signal the requesting
1028 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
1030 struct vmbus_channel_gpadl_created *gpadlcreated;
1031 struct vmbus_channel_msginfo *msginfo;
1032 struct vmbus_channel_message_header *requestheader;
1033 struct vmbus_channel_gpadl_header *gpadlheader;
1034 unsigned long flags;
1036 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
1038 trace_vmbus_ongpadl_created(gpadlcreated);
1041 * Find the establish msg, copy the result and signal/unblock the wait
1044 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1046 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1049 (struct vmbus_channel_message_header *)msginfo->msg;
1051 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
1053 (struct vmbus_channel_gpadl_header *)requestheader;
1055 if ((gpadlcreated->child_relid ==
1056 gpadlheader->child_relid) &&
1057 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1058 memcpy(&msginfo->response.gpadl_created,
1061 struct vmbus_channel_gpadl_created));
1062 complete(&msginfo->waitevent);
1067 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1071 * vmbus_ongpadl_torndown - GPADL torndown handler.
1073 * This is invoked when we received a response to our gpadl teardown request.
1074 * Find the matching request, copy the response and signal the requesting
1077 static void vmbus_ongpadl_torndown(
1078 struct vmbus_channel_message_header *hdr)
1080 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1081 struct vmbus_channel_msginfo *msginfo;
1082 struct vmbus_channel_message_header *requestheader;
1083 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1084 unsigned long flags;
1086 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1088 trace_vmbus_ongpadl_torndown(gpadl_torndown);
1091 * Find the open msg, copy the result and signal/unblock the wait event
1093 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1095 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1098 (struct vmbus_channel_message_header *)msginfo->msg;
1100 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1102 (struct vmbus_channel_gpadl_teardown *)requestheader;
1104 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1105 memcpy(&msginfo->response.gpadl_torndown,
1108 struct vmbus_channel_gpadl_torndown));
1109 complete(&msginfo->waitevent);
1114 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1118 * vmbus_onversion_response - Version response handler
1120 * This is invoked when we received a response to our initiate contact request.
1121 * Find the matching request, copy the response and signal the requesting
1124 static void vmbus_onversion_response(
1125 struct vmbus_channel_message_header *hdr)
1127 struct vmbus_channel_msginfo *msginfo;
1128 struct vmbus_channel_message_header *requestheader;
1129 struct vmbus_channel_version_response *version_response;
1130 unsigned long flags;
1132 version_response = (struct vmbus_channel_version_response *)hdr;
1134 trace_vmbus_onversion_response(version_response);
1136 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1138 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1141 (struct vmbus_channel_message_header *)msginfo->msg;
1143 if (requestheader->msgtype ==
1144 CHANNELMSG_INITIATE_CONTACT) {
1145 memcpy(&msginfo->response.version_response,
1147 sizeof(struct vmbus_channel_version_response));
1148 complete(&msginfo->waitevent);
1151 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1154 /* Channel message dispatch table */
1155 const struct vmbus_channel_message_table_entry
1156 channel_message_table[CHANNELMSG_COUNT] = {
1157 { CHANNELMSG_INVALID, 0, NULL },
1158 { CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer },
1159 { CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind },
1160 { CHANNELMSG_REQUESTOFFERS, 0, NULL },
1161 { CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered },
1162 { CHANNELMSG_OPENCHANNEL, 0, NULL },
1163 { CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result },
1164 { CHANNELMSG_CLOSECHANNEL, 0, NULL },
1165 { CHANNELMSG_GPADL_HEADER, 0, NULL },
1166 { CHANNELMSG_GPADL_BODY, 0, NULL },
1167 { CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created },
1168 { CHANNELMSG_GPADL_TEARDOWN, 0, NULL },
1169 { CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown },
1170 { CHANNELMSG_RELID_RELEASED, 0, NULL },
1171 { CHANNELMSG_INITIATE_CONTACT, 0, NULL },
1172 { CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response },
1173 { CHANNELMSG_UNLOAD, 0, NULL },
1174 { CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response },
1175 { CHANNELMSG_18, 0, NULL },
1176 { CHANNELMSG_19, 0, NULL },
1177 { CHANNELMSG_20, 0, NULL },
1178 { CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL },
1182 * vmbus_onmessage - Handler for channel protocol messages.
1184 * This is invoked in the vmbus worker thread context.
1186 void vmbus_onmessage(void *context)
1188 struct hv_message *msg = context;
1189 struct vmbus_channel_message_header *hdr;
1192 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1193 size = msg->header.payload_size;
1195 trace_vmbus_on_message(hdr);
1197 if (hdr->msgtype >= CHANNELMSG_COUNT) {
1198 pr_err("Received invalid channel message type %d size %d\n",
1199 hdr->msgtype, size);
1200 print_hex_dump_bytes("", DUMP_PREFIX_NONE,
1201 (unsigned char *)msg->u.payload, size);
1205 if (channel_message_table[hdr->msgtype].message_handler)
1206 channel_message_table[hdr->msgtype].message_handler(hdr);
1208 pr_err("Unhandled channel message type %d\n", hdr->msgtype);
1212 * vmbus_request_offers - Send a request to get all our pending offers.
1214 int vmbus_request_offers(void)
1216 struct vmbus_channel_message_header *msg;
1217 struct vmbus_channel_msginfo *msginfo;
1220 msginfo = kmalloc(sizeof(*msginfo) +
1221 sizeof(struct vmbus_channel_message_header),
1226 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1228 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1230 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1233 trace_vmbus_request_offers(ret);
1236 pr_err("Unable to request offers - %d\n", ret);
1248 * Retrieve the (sub) channel on which to send an outgoing request.
1249 * When a primary channel has multiple sub-channels, we try to
1250 * distribute the load equally amongst all available channels.
1252 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary)
1254 struct list_head *cur, *tmp;
1256 struct vmbus_channel *cur_channel;
1257 struct vmbus_channel *outgoing_channel = primary;
1261 if (list_empty(&primary->sc_list))
1262 return outgoing_channel;
1264 next_channel = primary->next_oc++;
1266 if (next_channel > (primary->num_sc)) {
1267 primary->next_oc = 0;
1268 return outgoing_channel;
1271 cur_cpu = hv_cpu_number_to_vp_number(smp_processor_id());
1272 list_for_each_safe(cur, tmp, &primary->sc_list) {
1273 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1274 if (cur_channel->state != CHANNEL_OPENED_STATE)
1277 if (cur_channel->target_vp == cur_cpu)
1280 if (i == next_channel)
1286 return outgoing_channel;
1288 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel);
1290 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1292 struct list_head *cur, *tmp;
1293 struct vmbus_channel *cur_channel;
1295 if (primary_channel->sc_creation_callback == NULL)
1298 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1299 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1301 primary_channel->sc_creation_callback(cur_channel);
1305 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1306 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1308 primary_channel->sc_creation_callback = sc_cr_cb;
1310 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1312 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1316 ret = !list_empty(&primary->sc_list);
1320 * Invoke the callback on sub-channel creation.
1321 * This will present a uniform interface to the
1324 invoke_sc_cb(primary);
1329 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1331 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1332 void (*chn_rescind_cb)(struct vmbus_channel *))
1334 channel->chn_rescind_callback = chn_rescind_cb;
1336 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);