2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
18 * Haiyang Zhang <haiyangz@microsoft.com>
19 * Hank Janssen <hjanssen@microsoft.com>
21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
23 #include <linux/kernel.h>
24 #include <linux/interrupt.h>
25 #include <linux/sched.h>
26 #include <linux/wait.h>
28 #include <linux/slab.h>
29 #include <linux/list.h>
30 #include <linux/module.h>
31 #include <linux/completion.h>
32 #include <linux/delay.h>
33 #include <linux/hyperv.h>
34 #include <asm/mshyperv.h>
36 #include "hyperv_vmbus.h"
38 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type);
40 static const struct vmbus_device vmbus_devs[] = {
48 { .dev_type = HV_SCSI,
72 { .dev_type = HV_PCIE,
77 /* Synthetic Frame Buffer */
83 /* Synthetic Keyboard */
90 { .dev_type = HV_MOUSE,
104 .perf_device = false,
110 .perf_device = false,
114 { .dev_type = HV_SHUTDOWN,
116 .perf_device = false,
120 { .dev_type = HV_FCOPY,
122 .perf_device = false,
126 { .dev_type = HV_BACKUP,
128 .perf_device = false,
134 .perf_device = false,
138 { .dev_type = HV_UNKNOWN,
139 .perf_device = false,
143 static const struct {
145 } vmbus_unsupported_devs[] = {
152 * The rescinded channel may be blocked waiting for a response from the host;
155 static void vmbus_rescind_cleanup(struct vmbus_channel *channel)
157 struct vmbus_channel_msginfo *msginfo;
161 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
163 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
166 if (msginfo->waiting_channel == channel) {
167 complete(&msginfo->waitevent);
171 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
174 static bool is_unsupported_vmbus_devs(const uuid_le *guid)
178 for (i = 0; i < ARRAY_SIZE(vmbus_unsupported_devs); i++)
179 if (!uuid_le_cmp(*guid, vmbus_unsupported_devs[i].guid))
184 static u16 hv_get_dev_type(const struct vmbus_channel *channel)
186 const uuid_le *guid = &channel->offermsg.offer.if_type;
189 if (is_hvsock_channel(channel) || is_unsupported_vmbus_devs(guid))
192 for (i = HV_IDE; i < HV_UNKNOWN; i++) {
193 if (!uuid_le_cmp(*guid, vmbus_devs[i].guid))
196 pr_info("Unknown GUID: %pUl\n", guid);
201 * vmbus_prep_negotiate_resp() - Create default response for Hyper-V Negotiate message
202 * @icmsghdrp: Pointer to msg header structure
203 * @icmsg_negotiate: Pointer to negotiate message structure
204 * @buf: Raw buffer channel data
206 * @icmsghdrp is of type &struct icmsg_hdr.
207 * Set up and fill in default negotiate response message.
209 * The fw_version and fw_vercnt specifies the framework version that
212 * The srv_version and srv_vercnt specifies the service
213 * versions we can support.
215 * Versions are given in decreasing order.
217 * nego_fw_version and nego_srv_version store the selected protocol versions.
219 * Mainly used by Hyper-V drivers.
221 bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp,
222 u8 *buf, const int *fw_version, int fw_vercnt,
223 const int *srv_version, int srv_vercnt,
224 int *nego_fw_version, int *nego_srv_version)
226 int icframe_major, icframe_minor;
227 int icmsg_major, icmsg_minor;
228 int fw_major, fw_minor;
229 int srv_major, srv_minor;
231 bool found_match = false;
232 struct icmsg_negotiate *negop;
234 icmsghdrp->icmsgsize = 0x10;
235 negop = (struct icmsg_negotiate *)&buf[
236 sizeof(struct vmbuspipe_hdr) +
237 sizeof(struct icmsg_hdr)];
239 icframe_major = negop->icframe_vercnt;
242 icmsg_major = negop->icmsg_vercnt;
246 * Select the framework version number we will
250 for (i = 0; i < fw_vercnt; i++) {
251 fw_major = (fw_version[i] >> 16);
252 fw_minor = (fw_version[i] & 0xFFFF);
254 for (j = 0; j < negop->icframe_vercnt; j++) {
255 if ((negop->icversion_data[j].major == fw_major) &&
256 (negop->icversion_data[j].minor == fw_minor)) {
257 icframe_major = negop->icversion_data[j].major;
258 icframe_minor = negop->icversion_data[j].minor;
273 for (i = 0; i < srv_vercnt; i++) {
274 srv_major = (srv_version[i] >> 16);
275 srv_minor = (srv_version[i] & 0xFFFF);
277 for (j = negop->icframe_vercnt;
278 (j < negop->icframe_vercnt + negop->icmsg_vercnt);
281 if ((negop->icversion_data[j].major == srv_major) &&
282 (negop->icversion_data[j].minor == srv_minor)) {
284 icmsg_major = negop->icversion_data[j].major;
285 icmsg_minor = negop->icversion_data[j].minor;
296 * Respond with the framework and service
297 * version numbers we can support.
302 negop->icframe_vercnt = 0;
303 negop->icmsg_vercnt = 0;
305 negop->icframe_vercnt = 1;
306 negop->icmsg_vercnt = 1;
310 *nego_fw_version = (icframe_major << 16) | icframe_minor;
312 if (nego_srv_version)
313 *nego_srv_version = (icmsg_major << 16) | icmsg_minor;
315 negop->icversion_data[0].major = icframe_major;
316 negop->icversion_data[0].minor = icframe_minor;
317 negop->icversion_data[1].major = icmsg_major;
318 negop->icversion_data[1].minor = icmsg_minor;
322 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp);
325 * alloc_channel - Allocate and initialize a vmbus channel object
327 static struct vmbus_channel *alloc_channel(void)
329 struct vmbus_channel *channel;
331 channel = kzalloc(sizeof(*channel), GFP_ATOMIC);
335 channel->acquire_ring_lock = true;
336 spin_lock_init(&channel->inbound_lock);
337 spin_lock_init(&channel->lock);
339 INIT_LIST_HEAD(&channel->sc_list);
340 INIT_LIST_HEAD(&channel->percpu_list);
342 tasklet_init(&channel->callback_event,
343 vmbus_on_event, (unsigned long)channel);
349 * free_channel - Release the resources used by the vmbus channel object
351 static void free_channel(struct vmbus_channel *channel)
353 tasklet_kill(&channel->callback_event);
357 static void percpu_channel_enq(void *arg)
359 struct vmbus_channel *channel = arg;
360 struct hv_per_cpu_context *hv_cpu
361 = this_cpu_ptr(hv_context.cpu_context);
363 list_add_tail(&channel->percpu_list, &hv_cpu->chan_list);
366 static void percpu_channel_deq(void *arg)
368 struct vmbus_channel *channel = arg;
370 list_del(&channel->percpu_list);
374 static void vmbus_release_relid(u32 relid)
376 struct vmbus_channel_relid_released msg;
378 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
379 msg.child_relid = relid;
380 msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
381 vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released),
385 void hv_event_tasklet_disable(struct vmbus_channel *channel)
387 tasklet_disable(&channel->callback_event);
390 void hv_event_tasklet_enable(struct vmbus_channel *channel)
392 tasklet_enable(&channel->callback_event);
394 /* In case there is any pending event */
395 tasklet_schedule(&channel->callback_event);
398 void hv_process_channel_removal(struct vmbus_channel *channel, u32 relid)
401 struct vmbus_channel *primary_channel;
403 BUG_ON(!channel->rescind);
404 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
406 hv_event_tasklet_disable(channel);
407 if (channel->target_cpu != get_cpu()) {
409 smp_call_function_single(channel->target_cpu,
410 percpu_channel_deq, channel, true);
412 percpu_channel_deq(channel);
415 hv_event_tasklet_enable(channel);
417 if (channel->primary_channel == NULL) {
418 list_del(&channel->listentry);
420 primary_channel = channel;
422 primary_channel = channel->primary_channel;
423 spin_lock_irqsave(&primary_channel->lock, flags);
424 list_del(&channel->sc_list);
425 primary_channel->num_sc--;
426 spin_unlock_irqrestore(&primary_channel->lock, flags);
430 * We need to free the bit for init_vp_index() to work in the case
431 * of sub-channel, when we reload drivers like hv_netvsc.
433 if (channel->affinity_policy == HV_LOCALIZED)
434 cpumask_clear_cpu(channel->target_cpu,
435 &primary_channel->alloced_cpus_in_node);
437 vmbus_release_relid(relid);
439 free_channel(channel);
442 void vmbus_free_channels(void)
444 struct vmbus_channel *channel, *tmp;
446 mutex_lock(&vmbus_connection.channel_mutex);
447 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
449 /* hv_process_channel_removal() needs this */
450 channel->rescind = true;
452 vmbus_device_unregister(channel->device_obj);
454 mutex_unlock(&vmbus_connection.channel_mutex);
458 * vmbus_process_offer - Process the offer by creating a channel/device
459 * associated with this offer
461 static void vmbus_process_offer(struct vmbus_channel *newchannel)
463 struct vmbus_channel *channel;
469 /* Make sure this is a new offer */
470 mutex_lock(&vmbus_connection.channel_mutex);
472 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
473 if (!uuid_le_cmp(channel->offermsg.offer.if_type,
474 newchannel->offermsg.offer.if_type) &&
475 !uuid_le_cmp(channel->offermsg.offer.if_instance,
476 newchannel->offermsg.offer.if_instance)) {
483 list_add_tail(&newchannel->listentry,
484 &vmbus_connection.chn_list);
486 mutex_unlock(&vmbus_connection.channel_mutex);
490 * Check to see if this is a sub-channel.
492 if (newchannel->offermsg.offer.sub_channel_index != 0) {
494 * Process the sub-channel.
496 newchannel->primary_channel = channel;
497 spin_lock_irqsave(&channel->lock, flags);
498 list_add_tail(&newchannel->sc_list, &channel->sc_list);
500 spin_unlock_irqrestore(&channel->lock, flags);
505 dev_type = hv_get_dev_type(newchannel);
507 init_vp_index(newchannel, dev_type);
509 hv_event_tasklet_disable(newchannel);
510 if (newchannel->target_cpu != get_cpu()) {
512 smp_call_function_single(newchannel->target_cpu,
516 percpu_channel_enq(newchannel);
519 hv_event_tasklet_enable(newchannel);
522 * This state is used to indicate a successful open
523 * so that when we do close the channel normally, we
524 * can cleanup properly
526 newchannel->state = CHANNEL_OPEN_STATE;
529 if (channel->sc_creation_callback != NULL)
530 channel->sc_creation_callback(newchannel);
535 * Start the process of binding this offer to the driver
536 * We need to set the DeviceObject field before calling
537 * vmbus_child_dev_add()
539 newchannel->device_obj = vmbus_device_create(
540 &newchannel->offermsg.offer.if_type,
541 &newchannel->offermsg.offer.if_instance,
543 if (!newchannel->device_obj)
546 newchannel->device_obj->device_id = dev_type;
548 * Add the new device to the bus. This will kick off device-driver
549 * binding which eventually invokes the device driver's AddDevice()
552 mutex_lock(&vmbus_connection.channel_mutex);
553 ret = vmbus_device_register(newchannel->device_obj);
554 mutex_unlock(&vmbus_connection.channel_mutex);
557 pr_err("unable to add child device object (relid %d)\n",
558 newchannel->offermsg.child_relid);
559 kfree(newchannel->device_obj);
565 mutex_lock(&vmbus_connection.channel_mutex);
566 list_del(&newchannel->listentry);
567 mutex_unlock(&vmbus_connection.channel_mutex);
569 hv_event_tasklet_disable(newchannel);
570 if (newchannel->target_cpu != get_cpu()) {
572 smp_call_function_single(newchannel->target_cpu,
573 percpu_channel_deq, newchannel, true);
575 percpu_channel_deq(newchannel);
578 hv_event_tasklet_enable(newchannel);
580 vmbus_release_relid(newchannel->offermsg.child_relid);
583 free_channel(newchannel);
587 * We use this state to statically distribute the channel interrupt load.
589 static int next_numa_node_id;
592 * Starting with Win8, we can statically distribute the incoming
593 * channel interrupt load by binding a channel to VCPU.
594 * We do this in a hierarchical fashion:
595 * First distribute the primary channels across available NUMA nodes
596 * and then distribute the subchannels amongst the CPUs in the NUMA
597 * node assigned to the primary channel.
599 * For pre-win8 hosts or non-performance critical channels we assign the
600 * first CPU in the first NUMA node.
602 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
605 bool perf_chn = vmbus_devs[dev_type].perf_device;
606 struct vmbus_channel *primary = channel->primary_channel;
608 struct cpumask available_mask;
609 struct cpumask *alloced_mask;
611 if ((vmbus_proto_version == VERSION_WS2008) ||
612 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn)) {
614 * Prior to win8, all channel interrupts are
615 * delivered on cpu 0.
616 * Also if the channel is not a performance critical
617 * channel, bind it to cpu 0.
619 channel->numa_node = 0;
620 channel->target_cpu = 0;
621 channel->target_vp = hv_context.vp_index[0];
626 * Based on the channel affinity policy, we will assign the NUMA
630 if ((channel->affinity_policy == HV_BALANCED) || (!primary)) {
632 next_node = next_numa_node_id++;
633 if (next_node == nr_node_ids) {
634 next_node = next_numa_node_id = 0;
637 if (cpumask_empty(cpumask_of_node(next_node)))
641 channel->numa_node = next_node;
644 alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
646 if (cpumask_weight(alloced_mask) ==
647 cpumask_weight(cpumask_of_node(primary->numa_node))) {
649 * We have cycled through all the CPUs in the node;
650 * reset the alloced map.
652 cpumask_clear(alloced_mask);
655 cpumask_xor(&available_mask, alloced_mask,
656 cpumask_of_node(primary->numa_node));
660 if (primary->affinity_policy == HV_LOCALIZED) {
662 * Normally Hyper-V host doesn't create more subchannels
663 * than there are VCPUs on the node but it is possible when not
664 * all present VCPUs on the node are initialized by guest.
665 * Clear the alloced_cpus_in_node to start over.
667 if (cpumask_equal(&primary->alloced_cpus_in_node,
668 cpumask_of_node(primary->numa_node)))
669 cpumask_clear(&primary->alloced_cpus_in_node);
673 cur_cpu = cpumask_next(cur_cpu, &available_mask);
674 if (cur_cpu >= nr_cpu_ids) {
676 cpumask_copy(&available_mask,
677 cpumask_of_node(primary->numa_node));
681 if (primary->affinity_policy == HV_LOCALIZED) {
683 * NOTE: in the case of sub-channel, we clear the
684 * sub-channel related bit(s) in
685 * primary->alloced_cpus_in_node in
686 * hv_process_channel_removal(), so when we
687 * reload drivers like hv_netvsc in SMP guest, here
688 * we're able to re-allocate
689 * bit from primary->alloced_cpus_in_node.
691 if (!cpumask_test_cpu(cur_cpu,
692 &primary->alloced_cpus_in_node)) {
693 cpumask_set_cpu(cur_cpu,
694 &primary->alloced_cpus_in_node);
695 cpumask_set_cpu(cur_cpu, alloced_mask);
699 cpumask_set_cpu(cur_cpu, alloced_mask);
704 channel->target_cpu = cur_cpu;
705 channel->target_vp = hv_context.vp_index[cur_cpu];
708 static void vmbus_wait_for_unload(void)
712 struct hv_message *msg;
713 struct vmbus_channel_message_header *hdr;
717 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
718 * used for initial contact or to CPU0 depending on host version. When
719 * we're crashing on a different CPU let's hope that IRQ handler on
720 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
721 * functional and vmbus_unload_response() will complete
722 * vmbus_connection.unload_event. If not, the last thing we can do is
723 * read message pages for all CPUs directly.
726 if (completion_done(&vmbus_connection.unload_event))
729 for_each_online_cpu(cpu) {
730 struct hv_per_cpu_context *hv_cpu
731 = per_cpu_ptr(hv_context.cpu_context, cpu);
733 page_addr = hv_cpu->synic_message_page;
734 msg = (struct hv_message *)page_addr
735 + VMBUS_MESSAGE_SINT;
737 message_type = READ_ONCE(msg->header.message_type);
738 if (message_type == HVMSG_NONE)
741 hdr = (struct vmbus_channel_message_header *)
744 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
745 complete(&vmbus_connection.unload_event);
747 vmbus_signal_eom(msg, message_type);
754 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
755 * maybe-pending messages on all CPUs to be able to receive new
756 * messages after we reconnect.
758 for_each_online_cpu(cpu) {
759 struct hv_per_cpu_context *hv_cpu
760 = per_cpu_ptr(hv_context.cpu_context, cpu);
762 page_addr = hv_cpu->synic_message_page;
763 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
764 msg->header.message_type = HVMSG_NONE;
769 * vmbus_unload_response - Handler for the unload response.
771 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
774 * This is a global event; just wakeup the waiting thread.
775 * Once we successfully unload, we can cleanup the monitor state.
777 complete(&vmbus_connection.unload_event);
780 void vmbus_initiate_unload(bool crash)
782 struct vmbus_channel_message_header hdr;
784 /* Pre-Win2012R2 hosts don't support reconnect */
785 if (vmbus_proto_version < VERSION_WIN8_1)
788 init_completion(&vmbus_connection.unload_event);
789 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
790 hdr.msgtype = CHANNELMSG_UNLOAD;
791 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
795 * vmbus_initiate_unload() is also called on crash and the crash can be
796 * happening in an interrupt context, where scheduling is impossible.
799 wait_for_completion(&vmbus_connection.unload_event);
801 vmbus_wait_for_unload();
805 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
808 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
810 struct vmbus_channel_offer_channel *offer;
811 struct vmbus_channel *newchannel;
813 offer = (struct vmbus_channel_offer_channel *)hdr;
815 /* Allocate the channel object and save this offer. */
816 newchannel = alloc_channel();
818 pr_err("Unable to allocate channel object\n");
823 * By default we setup state to enable batched
824 * reading. A specific service can choose to
825 * disable this prior to opening the channel.
827 newchannel->batched_reading = true;
830 * Setup state for signalling the host.
832 newchannel->sig_event = (struct hv_input_signal_event *)
833 (ALIGN((unsigned long)
834 &newchannel->sig_buf,
835 HV_HYPERCALL_PARAM_ALIGN));
837 newchannel->sig_event->connectionid.asu32 = 0;
838 newchannel->sig_event->connectionid.u.id = VMBUS_EVENT_CONNECTION_ID;
839 newchannel->sig_event->flag_number = 0;
840 newchannel->sig_event->rsvdz = 0;
842 if (vmbus_proto_version != VERSION_WS2008) {
843 newchannel->is_dedicated_interrupt =
844 (offer->is_dedicated_interrupt != 0);
845 newchannel->sig_event->connectionid.u.id =
846 offer->connection_id;
849 memcpy(&newchannel->offermsg, offer,
850 sizeof(struct vmbus_channel_offer_channel));
851 newchannel->monitor_grp = (u8)offer->monitorid / 32;
852 newchannel->monitor_bit = (u8)offer->monitorid % 32;
854 vmbus_process_offer(newchannel);
858 * vmbus_onoffer_rescind - Rescind offer handler.
860 * We queue a work item to process this offer synchronously
862 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
864 struct vmbus_channel_rescind_offer *rescind;
865 struct vmbus_channel *channel;
869 rescind = (struct vmbus_channel_rescind_offer *)hdr;
871 mutex_lock(&vmbus_connection.channel_mutex);
872 channel = relid2channel(rescind->child_relid);
874 if (channel == NULL) {
876 * This is very impossible, because in
877 * vmbus_process_offer(), we have already invoked
878 * vmbus_release_relid() on error.
883 spin_lock_irqsave(&channel->lock, flags);
884 channel->rescind = true;
885 spin_unlock_irqrestore(&channel->lock, flags);
887 vmbus_rescind_cleanup(channel);
889 if (channel->device_obj) {
890 if (channel->chn_rescind_callback) {
891 channel->chn_rescind_callback(channel);
895 * We will have to unregister this device from the
898 dev = get_device(&channel->device_obj->device);
900 vmbus_device_unregister(channel->device_obj);
904 hv_process_channel_removal(channel,
905 channel->offermsg.child_relid);
909 mutex_unlock(&vmbus_connection.channel_mutex);
912 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
914 mutex_lock(&vmbus_connection.channel_mutex);
916 BUG_ON(!is_hvsock_channel(channel));
918 channel->rescind = true;
919 vmbus_device_unregister(channel->device_obj);
921 mutex_unlock(&vmbus_connection.channel_mutex);
923 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
927 * vmbus_onoffers_delivered -
928 * This is invoked when all offers have been delivered.
930 * Nothing to do here.
932 static void vmbus_onoffers_delivered(
933 struct vmbus_channel_message_header *hdr)
938 * vmbus_onopen_result - Open result handler.
940 * This is invoked when we received a response to our channel open request.
941 * Find the matching request, copy the response and signal the requesting
944 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
946 struct vmbus_channel_open_result *result;
947 struct vmbus_channel_msginfo *msginfo;
948 struct vmbus_channel_message_header *requestheader;
949 struct vmbus_channel_open_channel *openmsg;
952 result = (struct vmbus_channel_open_result *)hdr;
955 * Find the open msg, copy the result and signal/unblock the wait event
957 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
959 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
962 (struct vmbus_channel_message_header *)msginfo->msg;
964 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
966 (struct vmbus_channel_open_channel *)msginfo->msg;
967 if (openmsg->child_relid == result->child_relid &&
968 openmsg->openid == result->openid) {
969 memcpy(&msginfo->response.open_result,
972 struct vmbus_channel_open_result));
973 complete(&msginfo->waitevent);
978 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
982 * vmbus_ongpadl_created - GPADL created handler.
984 * This is invoked when we received a response to our gpadl create request.
985 * Find the matching request, copy the response and signal the requesting
988 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
990 struct vmbus_channel_gpadl_created *gpadlcreated;
991 struct vmbus_channel_msginfo *msginfo;
992 struct vmbus_channel_message_header *requestheader;
993 struct vmbus_channel_gpadl_header *gpadlheader;
996 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
999 * Find the establish msg, copy the result and signal/unblock the wait
1002 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1004 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1007 (struct vmbus_channel_message_header *)msginfo->msg;
1009 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
1011 (struct vmbus_channel_gpadl_header *)requestheader;
1013 if ((gpadlcreated->child_relid ==
1014 gpadlheader->child_relid) &&
1015 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1016 memcpy(&msginfo->response.gpadl_created,
1019 struct vmbus_channel_gpadl_created));
1020 complete(&msginfo->waitevent);
1025 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1029 * vmbus_ongpadl_torndown - GPADL torndown handler.
1031 * This is invoked when we received a response to our gpadl teardown request.
1032 * Find the matching request, copy the response and signal the requesting
1035 static void vmbus_ongpadl_torndown(
1036 struct vmbus_channel_message_header *hdr)
1038 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1039 struct vmbus_channel_msginfo *msginfo;
1040 struct vmbus_channel_message_header *requestheader;
1041 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1042 unsigned long flags;
1044 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1047 * Find the open msg, copy the result and signal/unblock the wait event
1049 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1051 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1054 (struct vmbus_channel_message_header *)msginfo->msg;
1056 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1058 (struct vmbus_channel_gpadl_teardown *)requestheader;
1060 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1061 memcpy(&msginfo->response.gpadl_torndown,
1064 struct vmbus_channel_gpadl_torndown));
1065 complete(&msginfo->waitevent);
1070 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1074 * vmbus_onversion_response - Version response handler
1076 * This is invoked when we received a response to our initiate contact request.
1077 * Find the matching request, copy the response and signal the requesting
1080 static void vmbus_onversion_response(
1081 struct vmbus_channel_message_header *hdr)
1083 struct vmbus_channel_msginfo *msginfo;
1084 struct vmbus_channel_message_header *requestheader;
1085 struct vmbus_channel_version_response *version_response;
1086 unsigned long flags;
1088 version_response = (struct vmbus_channel_version_response *)hdr;
1089 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1091 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1094 (struct vmbus_channel_message_header *)msginfo->msg;
1096 if (requestheader->msgtype ==
1097 CHANNELMSG_INITIATE_CONTACT) {
1098 memcpy(&msginfo->response.version_response,
1100 sizeof(struct vmbus_channel_version_response));
1101 complete(&msginfo->waitevent);
1104 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1107 /* Channel message dispatch table */
1108 struct vmbus_channel_message_table_entry
1109 channel_message_table[CHANNELMSG_COUNT] = {
1110 {CHANNELMSG_INVALID, 0, NULL},
1111 {CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer},
1112 {CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind},
1113 {CHANNELMSG_REQUESTOFFERS, 0, NULL},
1114 {CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered},
1115 {CHANNELMSG_OPENCHANNEL, 0, NULL},
1116 {CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result},
1117 {CHANNELMSG_CLOSECHANNEL, 0, NULL},
1118 {CHANNELMSG_GPADL_HEADER, 0, NULL},
1119 {CHANNELMSG_GPADL_BODY, 0, NULL},
1120 {CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created},
1121 {CHANNELMSG_GPADL_TEARDOWN, 0, NULL},
1122 {CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown},
1123 {CHANNELMSG_RELID_RELEASED, 0, NULL},
1124 {CHANNELMSG_INITIATE_CONTACT, 0, NULL},
1125 {CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response},
1126 {CHANNELMSG_UNLOAD, 0, NULL},
1127 {CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response},
1128 {CHANNELMSG_18, 0, NULL},
1129 {CHANNELMSG_19, 0, NULL},
1130 {CHANNELMSG_20, 0, NULL},
1131 {CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL},
1135 * vmbus_onmessage - Handler for channel protocol messages.
1137 * This is invoked in the vmbus worker thread context.
1139 void vmbus_onmessage(void *context)
1141 struct hv_message *msg = context;
1142 struct vmbus_channel_message_header *hdr;
1145 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1146 size = msg->header.payload_size;
1148 if (hdr->msgtype >= CHANNELMSG_COUNT) {
1149 pr_err("Received invalid channel message type %d size %d\n",
1150 hdr->msgtype, size);
1151 print_hex_dump_bytes("", DUMP_PREFIX_NONE,
1152 (unsigned char *)msg->u.payload, size);
1156 if (channel_message_table[hdr->msgtype].message_handler)
1157 channel_message_table[hdr->msgtype].message_handler(hdr);
1159 pr_err("Unhandled channel message type %d\n", hdr->msgtype);
1163 * vmbus_request_offers - Send a request to get all our pending offers.
1165 int vmbus_request_offers(void)
1167 struct vmbus_channel_message_header *msg;
1168 struct vmbus_channel_msginfo *msginfo;
1171 msginfo = kmalloc(sizeof(*msginfo) +
1172 sizeof(struct vmbus_channel_message_header),
1177 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1179 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1182 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1185 pr_err("Unable to request offers - %d\n", ret);
1197 * Retrieve the (sub) channel on which to send an outgoing request.
1198 * When a primary channel has multiple sub-channels, we try to
1199 * distribute the load equally amongst all available channels.
1201 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary)
1203 struct list_head *cur, *tmp;
1205 struct vmbus_channel *cur_channel;
1206 struct vmbus_channel *outgoing_channel = primary;
1210 if (list_empty(&primary->sc_list))
1211 return outgoing_channel;
1213 next_channel = primary->next_oc++;
1215 if (next_channel > (primary->num_sc)) {
1216 primary->next_oc = 0;
1217 return outgoing_channel;
1220 cur_cpu = hv_context.vp_index[get_cpu()];
1222 list_for_each_safe(cur, tmp, &primary->sc_list) {
1223 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1224 if (cur_channel->state != CHANNEL_OPENED_STATE)
1227 if (cur_channel->target_vp == cur_cpu)
1230 if (i == next_channel)
1236 return outgoing_channel;
1238 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel);
1240 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1242 struct list_head *cur, *tmp;
1243 struct vmbus_channel *cur_channel;
1245 if (primary_channel->sc_creation_callback == NULL)
1248 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1249 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1251 primary_channel->sc_creation_callback(cur_channel);
1255 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1256 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1258 primary_channel->sc_creation_callback = sc_cr_cb;
1260 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1262 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1266 ret = !list_empty(&primary->sc_list);
1270 * Invoke the callback on sub-channel creation.
1271 * This will present a uniform interface to the
1274 invoke_sc_cb(primary);
1279 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1281 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1282 void (*chn_rescind_cb)(struct vmbus_channel *))
1284 channel->chn_rescind_callback = chn_rescind_cb;
1286 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);