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>
35 #include "hyperv_vmbus.h"
37 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type);
39 static const struct vmbus_device vmbus_devs[] = {
47 { .dev_type = HV_SCSI,
71 { .dev_type = HV_PCIE,
76 /* Synthetic Frame Buffer */
82 /* Synthetic Keyboard */
89 { .dev_type = HV_MOUSE,
103 .perf_device = false,
109 .perf_device = false,
113 { .dev_type = HV_SHUTDOWN,
115 .perf_device = false,
119 { .dev_type = HV_FCOPY,
121 .perf_device = false,
125 { .dev_type = HV_BACKUP,
127 .perf_device = false,
133 .perf_device = false,
137 { .dev_type = HV_UNKOWN,
138 .perf_device = false,
142 static const struct {
144 } vmbus_unsupported_devs[] = {
151 * The rescinded channel may be blocked waiting for a response from the host;
154 static void vmbus_rescind_cleanup(struct vmbus_channel *channel)
156 struct vmbus_channel_msginfo *msginfo;
160 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
162 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
165 if (msginfo->waiting_channel == channel) {
166 complete(&msginfo->waitevent);
170 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
173 static bool is_unsupported_vmbus_devs(const uuid_le *guid)
177 for (i = 0; i < ARRAY_SIZE(vmbus_unsupported_devs); i++)
178 if (!uuid_le_cmp(*guid, vmbus_unsupported_devs[i].guid))
183 static u16 hv_get_dev_type(const struct vmbus_channel *channel)
185 const uuid_le *guid = &channel->offermsg.offer.if_type;
188 if (is_hvsock_channel(channel) || is_unsupported_vmbus_devs(guid))
191 for (i = HV_IDE; i < HV_UNKOWN; i++) {
192 if (!uuid_le_cmp(*guid, vmbus_devs[i].guid))
195 pr_info("Unknown GUID: %pUl\n", guid);
200 * vmbus_prep_negotiate_resp() - Create default response for Hyper-V Negotiate message
201 * @icmsghdrp: Pointer to msg header structure
202 * @icmsg_negotiate: Pointer to negotiate message structure
203 * @buf: Raw buffer channel data
205 * @icmsghdrp is of type &struct icmsg_hdr.
206 * @negop is of type &struct icmsg_negotiate.
207 * Set up and fill in default negotiate response message.
209 * The fw_version specifies the framework version that
210 * we can support and srv_version specifies the service
211 * version we can support.
213 * Mainly used by Hyper-V drivers.
215 bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp,
216 struct icmsg_negotiate *negop, u8 *buf,
217 int fw_version, int srv_version)
219 int icframe_major, icframe_minor;
220 int icmsg_major, icmsg_minor;
221 int fw_major, fw_minor;
222 int srv_major, srv_minor;
224 bool found_match = false;
226 icmsghdrp->icmsgsize = 0x10;
227 fw_major = (fw_version >> 16);
228 fw_minor = (fw_version & 0xFFFF);
230 srv_major = (srv_version >> 16);
231 srv_minor = (srv_version & 0xFFFF);
233 negop = (struct icmsg_negotiate *)&buf[
234 sizeof(struct vmbuspipe_hdr) +
235 sizeof(struct icmsg_hdr)];
237 icframe_major = negop->icframe_vercnt;
240 icmsg_major = negop->icmsg_vercnt;
244 * Select the framework version number we will
248 for (i = 0; i < negop->icframe_vercnt; i++) {
249 if ((negop->icversion_data[i].major == fw_major) &&
250 (negop->icversion_data[i].minor == fw_minor)) {
251 icframe_major = negop->icversion_data[i].major;
252 icframe_minor = negop->icversion_data[i].minor;
262 for (i = negop->icframe_vercnt;
263 (i < negop->icframe_vercnt + negop->icmsg_vercnt); i++) {
264 if ((negop->icversion_data[i].major == srv_major) &&
265 (negop->icversion_data[i].minor == srv_minor)) {
266 icmsg_major = negop->icversion_data[i].major;
267 icmsg_minor = negop->icversion_data[i].minor;
273 * Respond with the framework and service
274 * version numbers we can support.
279 negop->icframe_vercnt = 0;
280 negop->icmsg_vercnt = 0;
282 negop->icframe_vercnt = 1;
283 negop->icmsg_vercnt = 1;
286 negop->icversion_data[0].major = icframe_major;
287 negop->icversion_data[0].minor = icframe_minor;
288 negop->icversion_data[1].major = icmsg_major;
289 negop->icversion_data[1].minor = icmsg_minor;
293 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp);
296 * alloc_channel - Allocate and initialize a vmbus channel object
298 static struct vmbus_channel *alloc_channel(void)
300 struct vmbus_channel *channel;
302 channel = kzalloc(sizeof(*channel), GFP_ATOMIC);
306 channel->acquire_ring_lock = true;
307 spin_lock_init(&channel->inbound_lock);
308 spin_lock_init(&channel->lock);
310 INIT_LIST_HEAD(&channel->sc_list);
311 INIT_LIST_HEAD(&channel->percpu_list);
317 * free_channel - Release the resources used by the vmbus channel object
319 static void free_channel(struct vmbus_channel *channel)
324 static void percpu_channel_enq(void *arg)
326 struct vmbus_channel *channel = arg;
327 int cpu = smp_processor_id();
329 list_add_tail(&channel->percpu_list, &hv_context.percpu_list[cpu]);
332 static void percpu_channel_deq(void *arg)
334 struct vmbus_channel *channel = arg;
336 list_del(&channel->percpu_list);
340 static void vmbus_release_relid(u32 relid)
342 struct vmbus_channel_relid_released msg;
344 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
345 msg.child_relid = relid;
346 msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
347 vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released),
351 void hv_event_tasklet_disable(struct vmbus_channel *channel)
353 struct tasklet_struct *tasklet;
354 tasklet = hv_context.event_dpc[channel->target_cpu];
355 tasklet_disable(tasklet);
358 void hv_event_tasklet_enable(struct vmbus_channel *channel)
360 struct tasklet_struct *tasklet;
361 tasklet = hv_context.event_dpc[channel->target_cpu];
362 tasklet_enable(tasklet);
364 /* In case there is any pending event */
365 tasklet_schedule(tasklet);
368 void hv_process_channel_removal(struct vmbus_channel *channel, u32 relid)
371 struct vmbus_channel *primary_channel;
373 BUG_ON(!channel->rescind);
374 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
376 hv_event_tasklet_disable(channel);
377 if (channel->target_cpu != get_cpu()) {
379 smp_call_function_single(channel->target_cpu,
380 percpu_channel_deq, channel, true);
382 percpu_channel_deq(channel);
385 hv_event_tasklet_enable(channel);
387 if (channel->primary_channel == NULL) {
388 list_del(&channel->listentry);
390 primary_channel = channel;
392 primary_channel = channel->primary_channel;
393 spin_lock_irqsave(&primary_channel->lock, flags);
394 list_del(&channel->sc_list);
395 primary_channel->num_sc--;
396 spin_unlock_irqrestore(&primary_channel->lock, flags);
400 * We need to free the bit for init_vp_index() to work in the case
401 * of sub-channel, when we reload drivers like hv_netvsc.
403 if (channel->affinity_policy == HV_LOCALIZED)
404 cpumask_clear_cpu(channel->target_cpu,
405 &primary_channel->alloced_cpus_in_node);
407 vmbus_release_relid(relid);
409 free_channel(channel);
412 void vmbus_free_channels(void)
414 struct vmbus_channel *channel, *tmp;
416 mutex_lock(&vmbus_connection.channel_mutex);
417 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
419 /* hv_process_channel_removal() needs this */
420 channel->rescind = true;
422 vmbus_device_unregister(channel->device_obj);
424 mutex_unlock(&vmbus_connection.channel_mutex);
428 * vmbus_process_offer - Process the offer by creating a channel/device
429 * associated with this offer
431 static void vmbus_process_offer(struct vmbus_channel *newchannel)
433 struct vmbus_channel *channel;
439 /* Make sure this is a new offer */
440 mutex_lock(&vmbus_connection.channel_mutex);
442 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
443 if (!uuid_le_cmp(channel->offermsg.offer.if_type,
444 newchannel->offermsg.offer.if_type) &&
445 !uuid_le_cmp(channel->offermsg.offer.if_instance,
446 newchannel->offermsg.offer.if_instance)) {
453 list_add_tail(&newchannel->listentry,
454 &vmbus_connection.chn_list);
456 mutex_unlock(&vmbus_connection.channel_mutex);
460 * Check to see if this is a sub-channel.
462 if (newchannel->offermsg.offer.sub_channel_index != 0) {
464 * Process the sub-channel.
466 newchannel->primary_channel = channel;
467 spin_lock_irqsave(&channel->lock, flags);
468 list_add_tail(&newchannel->sc_list, &channel->sc_list);
470 spin_unlock_irqrestore(&channel->lock, flags);
475 dev_type = hv_get_dev_type(newchannel);
477 init_vp_index(newchannel, dev_type);
479 hv_event_tasklet_disable(newchannel);
480 if (newchannel->target_cpu != get_cpu()) {
482 smp_call_function_single(newchannel->target_cpu,
486 percpu_channel_enq(newchannel);
489 hv_event_tasklet_enable(newchannel);
492 * This state is used to indicate a successful open
493 * so that when we do close the channel normally, we
494 * can cleanup properly
496 newchannel->state = CHANNEL_OPEN_STATE;
499 if (channel->sc_creation_callback != NULL)
500 channel->sc_creation_callback(newchannel);
505 * Start the process of binding this offer to the driver
506 * We need to set the DeviceObject field before calling
507 * vmbus_child_dev_add()
509 newchannel->device_obj = vmbus_device_create(
510 &newchannel->offermsg.offer.if_type,
511 &newchannel->offermsg.offer.if_instance,
513 if (!newchannel->device_obj)
516 newchannel->device_obj->device_id = dev_type;
518 * Add the new device to the bus. This will kick off device-driver
519 * binding which eventually invokes the device driver's AddDevice()
522 mutex_lock(&vmbus_connection.channel_mutex);
523 ret = vmbus_device_register(newchannel->device_obj);
524 mutex_unlock(&vmbus_connection.channel_mutex);
527 pr_err("unable to add child device object (relid %d)\n",
528 newchannel->offermsg.child_relid);
529 kfree(newchannel->device_obj);
535 mutex_lock(&vmbus_connection.channel_mutex);
536 list_del(&newchannel->listentry);
537 mutex_unlock(&vmbus_connection.channel_mutex);
539 hv_event_tasklet_disable(newchannel);
540 if (newchannel->target_cpu != get_cpu()) {
542 smp_call_function_single(newchannel->target_cpu,
543 percpu_channel_deq, newchannel, true);
545 percpu_channel_deq(newchannel);
548 hv_event_tasklet_enable(newchannel);
550 vmbus_release_relid(newchannel->offermsg.child_relid);
553 free_channel(newchannel);
557 * We use this state to statically distribute the channel interrupt load.
559 static int next_numa_node_id;
562 * Starting with Win8, we can statically distribute the incoming
563 * channel interrupt load by binding a channel to VCPU.
564 * We do this in a hierarchical fashion:
565 * First distribute the primary channels across available NUMA nodes
566 * and then distribute the subchannels amongst the CPUs in the NUMA
567 * node assigned to the primary channel.
569 * For pre-win8 hosts or non-performance critical channels we assign the
570 * first CPU in the first NUMA node.
572 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
575 bool perf_chn = vmbus_devs[dev_type].perf_device;
576 struct vmbus_channel *primary = channel->primary_channel;
578 struct cpumask available_mask;
579 struct cpumask *alloced_mask;
581 if ((vmbus_proto_version == VERSION_WS2008) ||
582 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn)) {
584 * Prior to win8, all channel interrupts are
585 * delivered on cpu 0.
586 * Also if the channel is not a performance critical
587 * channel, bind it to cpu 0.
589 channel->numa_node = 0;
590 channel->target_cpu = 0;
591 channel->target_vp = hv_context.vp_index[0];
596 * Based on the channel affinity policy, we will assign the NUMA
600 if ((channel->affinity_policy == HV_BALANCED) || (!primary)) {
602 next_node = next_numa_node_id++;
603 if (next_node == nr_node_ids) {
604 next_node = next_numa_node_id = 0;
607 if (cpumask_empty(cpumask_of_node(next_node)))
611 channel->numa_node = next_node;
614 alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
616 if (cpumask_weight(alloced_mask) ==
617 cpumask_weight(cpumask_of_node(primary->numa_node))) {
619 * We have cycled through all the CPUs in the node;
620 * reset the alloced map.
622 cpumask_clear(alloced_mask);
625 cpumask_xor(&available_mask, alloced_mask,
626 cpumask_of_node(primary->numa_node));
630 if (primary->affinity_policy == HV_LOCALIZED) {
632 * Normally Hyper-V host doesn't create more subchannels
633 * than there are VCPUs on the node but it is possible when not
634 * all present VCPUs on the node are initialized by guest.
635 * Clear the alloced_cpus_in_node to start over.
637 if (cpumask_equal(&primary->alloced_cpus_in_node,
638 cpumask_of_node(primary->numa_node)))
639 cpumask_clear(&primary->alloced_cpus_in_node);
643 cur_cpu = cpumask_next(cur_cpu, &available_mask);
644 if (cur_cpu >= nr_cpu_ids) {
646 cpumask_copy(&available_mask,
647 cpumask_of_node(primary->numa_node));
651 if (primary->affinity_policy == HV_LOCALIZED) {
653 * NOTE: in the case of sub-channel, we clear the
654 * sub-channel related bit(s) in
655 * primary->alloced_cpus_in_node in
656 * hv_process_channel_removal(), so when we
657 * reload drivers like hv_netvsc in SMP guest, here
658 * we're able to re-allocate
659 * bit from primary->alloced_cpus_in_node.
661 if (!cpumask_test_cpu(cur_cpu,
662 &primary->alloced_cpus_in_node)) {
663 cpumask_set_cpu(cur_cpu,
664 &primary->alloced_cpus_in_node);
665 cpumask_set_cpu(cur_cpu, alloced_mask);
669 cpumask_set_cpu(cur_cpu, alloced_mask);
674 channel->target_cpu = cur_cpu;
675 channel->target_vp = hv_context.vp_index[cur_cpu];
678 #define UNLOAD_DELAY_UNIT_MS 10 /* 10 milliseconds */
679 #define UNLOAD_WAIT_MS (100*1000) /* 100 seconds */
680 #define UNLOAD_WAIT_LOOPS (UNLOAD_WAIT_MS/UNLOAD_DELAY_UNIT_MS)
681 #define UNLOAD_MSG_MS (5*1000) /* Every 5 seconds */
682 #define UNLOAD_MSG_LOOPS (UNLOAD_MSG_MS/UNLOAD_DELAY_UNIT_MS)
684 static void vmbus_wait_for_unload(void)
688 struct hv_message *msg;
689 struct vmbus_channel_message_header *hdr;
693 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
694 * used for initial contact or to CPU0 depending on host version. When
695 * we're crashing on a different CPU let's hope that IRQ handler on
696 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
697 * functional and vmbus_unload_response() will complete
698 * vmbus_connection.unload_event. If not, the last thing we can do is
699 * read message pages for all CPUs directly.
701 * Wait up to 100 seconds since an Azure host must writeback any dirty
702 * data in its disk cache before the VMbus UNLOAD request will
703 * complete. This flushing has been empirically observed to take up
704 * to 50 seconds in cases with a lot of dirty data, so allow additional
705 * leeway and for inaccuracies in mdelay(). But eventually time out so
706 * that the panic path can't get hung forever in case the response
707 * message isn't seen.
709 for (i = 1; i <= UNLOAD_WAIT_LOOPS; i++) {
710 if (completion_done(&vmbus_connection.unload_event))
713 for_each_online_cpu(cpu) {
714 page_addr = hv_context.synic_message_page[cpu];
715 msg = (struct hv_message *)page_addr +
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);
732 * Give a notice periodically so someone watching the
733 * serial output won't think it is completely hung.
735 if (!(i % UNLOAD_MSG_LOOPS))
736 pr_notice("Waiting for VMBus UNLOAD to complete\n");
738 mdelay(UNLOAD_DELAY_UNIT_MS);
740 pr_err("Continuing even though VMBus UNLOAD did not complete\n");
744 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
745 * maybe-pending messages on all CPUs to be able to receive new
746 * messages after we reconnect.
748 for_each_online_cpu(cpu) {
749 page_addr = hv_context.synic_message_page[cpu];
750 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
751 msg->header.message_type = HVMSG_NONE;
756 * vmbus_unload_response - Handler for the unload response.
758 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
761 * This is a global event; just wakeup the waiting thread.
762 * Once we successfully unload, we can cleanup the monitor state.
764 complete(&vmbus_connection.unload_event);
767 void vmbus_initiate_unload(bool crash)
769 struct vmbus_channel_message_header hdr;
771 /* Pre-Win2012R2 hosts don't support reconnect */
772 if (vmbus_proto_version < VERSION_WIN8_1)
775 init_completion(&vmbus_connection.unload_event);
776 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
777 hdr.msgtype = CHANNELMSG_UNLOAD;
778 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
782 * vmbus_initiate_unload() is also called on crash and the crash can be
783 * happening in an interrupt context, where scheduling is impossible.
786 wait_for_completion(&vmbus_connection.unload_event);
788 vmbus_wait_for_unload();
792 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
795 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
797 struct vmbus_channel_offer_channel *offer;
798 struct vmbus_channel *newchannel;
800 offer = (struct vmbus_channel_offer_channel *)hdr;
802 /* Allocate the channel object and save this offer. */
803 newchannel = alloc_channel();
805 vmbus_release_relid(offer->child_relid);
806 pr_err("Unable to allocate channel object\n");
811 * By default we setup state to enable batched
812 * reading. A specific service can choose to
813 * disable this prior to opening the channel.
815 newchannel->batched_reading = true;
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;
857 rescind = (struct vmbus_channel_rescind_offer *)hdr;
859 mutex_lock(&vmbus_connection.channel_mutex);
860 channel = relid2channel(rescind->child_relid);
862 if (channel == NULL) {
864 * This is very impossible, because in
865 * vmbus_process_offer(), we have already invoked
866 * vmbus_release_relid() on error.
871 spin_lock_irqsave(&channel->lock, flags);
872 channel->rescind = true;
873 spin_unlock_irqrestore(&channel->lock, flags);
875 vmbus_rescind_cleanup(channel);
877 if (channel->device_obj) {
878 if (channel->chn_rescind_callback) {
879 channel->chn_rescind_callback(channel);
883 * We will have to unregister this device from the
886 dev = get_device(&channel->device_obj->device);
888 vmbus_device_unregister(channel->device_obj);
892 hv_process_channel_removal(channel,
893 channel->offermsg.child_relid);
897 mutex_unlock(&vmbus_connection.channel_mutex);
900 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
902 mutex_lock(&vmbus_connection.channel_mutex);
904 BUG_ON(!is_hvsock_channel(channel));
906 channel->rescind = true;
907 vmbus_device_unregister(channel->device_obj);
909 mutex_unlock(&vmbus_connection.channel_mutex);
911 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
915 * vmbus_onoffers_delivered -
916 * This is invoked when all offers have been delivered.
918 * Nothing to do here.
920 static void vmbus_onoffers_delivered(
921 struct vmbus_channel_message_header *hdr)
926 * vmbus_onopen_result - Open result handler.
928 * This is invoked when we received a response to our channel open request.
929 * Find the matching request, copy the response and signal the requesting
932 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
934 struct vmbus_channel_open_result *result;
935 struct vmbus_channel_msginfo *msginfo;
936 struct vmbus_channel_message_header *requestheader;
937 struct vmbus_channel_open_channel *openmsg;
940 result = (struct vmbus_channel_open_result *)hdr;
943 * Find the open msg, copy the result and signal/unblock the wait event
945 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
947 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
950 (struct vmbus_channel_message_header *)msginfo->msg;
952 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
954 (struct vmbus_channel_open_channel *)msginfo->msg;
955 if (openmsg->child_relid == result->child_relid &&
956 openmsg->openid == result->openid) {
957 memcpy(&msginfo->response.open_result,
960 struct vmbus_channel_open_result));
961 complete(&msginfo->waitevent);
966 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
970 * vmbus_ongpadl_created - GPADL created handler.
972 * This is invoked when we received a response to our gpadl create request.
973 * Find the matching request, copy the response and signal the requesting
976 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
978 struct vmbus_channel_gpadl_created *gpadlcreated;
979 struct vmbus_channel_msginfo *msginfo;
980 struct vmbus_channel_message_header *requestheader;
981 struct vmbus_channel_gpadl_header *gpadlheader;
984 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
987 * Find the establish msg, copy the result and signal/unblock the wait
990 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
992 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
995 (struct vmbus_channel_message_header *)msginfo->msg;
997 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
999 (struct vmbus_channel_gpadl_header *)requestheader;
1001 if ((gpadlcreated->child_relid ==
1002 gpadlheader->child_relid) &&
1003 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1004 memcpy(&msginfo->response.gpadl_created,
1007 struct vmbus_channel_gpadl_created));
1008 complete(&msginfo->waitevent);
1013 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1017 * vmbus_ongpadl_torndown - GPADL torndown handler.
1019 * This is invoked when we received a response to our gpadl teardown request.
1020 * Find the matching request, copy the response and signal the requesting
1023 static void vmbus_ongpadl_torndown(
1024 struct vmbus_channel_message_header *hdr)
1026 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1027 struct vmbus_channel_msginfo *msginfo;
1028 struct vmbus_channel_message_header *requestheader;
1029 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1030 unsigned long flags;
1032 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1035 * Find the open msg, copy the result and signal/unblock the wait event
1037 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1039 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1042 (struct vmbus_channel_message_header *)msginfo->msg;
1044 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1046 (struct vmbus_channel_gpadl_teardown *)requestheader;
1048 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1049 memcpy(&msginfo->response.gpadl_torndown,
1052 struct vmbus_channel_gpadl_torndown));
1053 complete(&msginfo->waitevent);
1058 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1062 * vmbus_onversion_response - Version response handler
1064 * This is invoked when we received a response to our initiate contact request.
1065 * Find the matching request, copy the response and signal the requesting
1068 static void vmbus_onversion_response(
1069 struct vmbus_channel_message_header *hdr)
1071 struct vmbus_channel_msginfo *msginfo;
1072 struct vmbus_channel_message_header *requestheader;
1073 struct vmbus_channel_version_response *version_response;
1074 unsigned long flags;
1076 version_response = (struct vmbus_channel_version_response *)hdr;
1077 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1079 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1082 (struct vmbus_channel_message_header *)msginfo->msg;
1084 if (requestheader->msgtype ==
1085 CHANNELMSG_INITIATE_CONTACT) {
1086 memcpy(&msginfo->response.version_response,
1088 sizeof(struct vmbus_channel_version_response));
1089 complete(&msginfo->waitevent);
1092 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1095 /* Channel message dispatch table */
1096 struct vmbus_channel_message_table_entry
1097 channel_message_table[CHANNELMSG_COUNT] = {
1098 {CHANNELMSG_INVALID, 0, NULL},
1099 {CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer},
1100 {CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind},
1101 {CHANNELMSG_REQUESTOFFERS, 0, NULL},
1102 {CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered},
1103 {CHANNELMSG_OPENCHANNEL, 0, NULL},
1104 {CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result},
1105 {CHANNELMSG_CLOSECHANNEL, 0, NULL},
1106 {CHANNELMSG_GPADL_HEADER, 0, NULL},
1107 {CHANNELMSG_GPADL_BODY, 0, NULL},
1108 {CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created},
1109 {CHANNELMSG_GPADL_TEARDOWN, 0, NULL},
1110 {CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown},
1111 {CHANNELMSG_RELID_RELEASED, 0, NULL},
1112 {CHANNELMSG_INITIATE_CONTACT, 0, NULL},
1113 {CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response},
1114 {CHANNELMSG_UNLOAD, 0, NULL},
1115 {CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response},
1116 {CHANNELMSG_18, 0, NULL},
1117 {CHANNELMSG_19, 0, NULL},
1118 {CHANNELMSG_20, 0, NULL},
1119 {CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL},
1123 * vmbus_onmessage - Handler for channel protocol messages.
1125 * This is invoked in the vmbus worker thread context.
1127 void vmbus_onmessage(void *context)
1129 struct hv_message *msg = context;
1130 struct vmbus_channel_message_header *hdr;
1133 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1134 size = msg->header.payload_size;
1136 if (hdr->msgtype >= CHANNELMSG_COUNT) {
1137 pr_err("Received invalid channel message type %d size %d\n",
1138 hdr->msgtype, size);
1139 print_hex_dump_bytes("", DUMP_PREFIX_NONE,
1140 (unsigned char *)msg->u.payload, size);
1144 if (channel_message_table[hdr->msgtype].message_handler)
1145 channel_message_table[hdr->msgtype].message_handler(hdr);
1147 pr_err("Unhandled channel message type %d\n", hdr->msgtype);
1151 * vmbus_request_offers - Send a request to get all our pending offers.
1153 int vmbus_request_offers(void)
1155 struct vmbus_channel_message_header *msg;
1156 struct vmbus_channel_msginfo *msginfo;
1159 msginfo = kmalloc(sizeof(*msginfo) +
1160 sizeof(struct vmbus_channel_message_header),
1165 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1167 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1170 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1173 pr_err("Unable to request offers - %d\n", ret);
1185 * Retrieve the (sub) channel on which to send an outgoing request.
1186 * When a primary channel has multiple sub-channels, we try to
1187 * distribute the load equally amongst all available channels.
1189 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary)
1191 struct list_head *cur, *tmp;
1193 struct vmbus_channel *cur_channel;
1194 struct vmbus_channel *outgoing_channel = primary;
1198 if (list_empty(&primary->sc_list))
1199 return outgoing_channel;
1201 next_channel = primary->next_oc++;
1203 if (next_channel > (primary->num_sc)) {
1204 primary->next_oc = 0;
1205 return outgoing_channel;
1208 cur_cpu = hv_context.vp_index[get_cpu()];
1210 list_for_each_safe(cur, tmp, &primary->sc_list) {
1211 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1212 if (cur_channel->state != CHANNEL_OPENED_STATE)
1215 if (cur_channel->target_vp == cur_cpu)
1218 if (i == next_channel)
1224 return outgoing_channel;
1226 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel);
1228 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1230 struct list_head *cur, *tmp;
1231 struct vmbus_channel *cur_channel;
1233 if (primary_channel->sc_creation_callback == NULL)
1236 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1237 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1239 primary_channel->sc_creation_callback(cur_channel);
1243 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1244 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1246 primary_channel->sc_creation_callback = sc_cr_cb;
1248 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1250 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1254 ret = !list_empty(&primary->sc_list);
1258 * Invoke the callback on sub-channel creation.
1259 * This will present a uniform interface to the
1262 invoke_sc_cb(primary);
1267 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1269 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1270 void (*chn_rescind_cb)(struct vmbus_channel *))
1272 channel->chn_rescind_callback = chn_rescind_cb;
1274 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);