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152 * The rescinded channel may be blocked waiting for a response from the host;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 Negotiate message
202 * @icmsghdrp: Pointer to msg header structure
203 * @buf: Raw buffer channel data
204 * @fw_version: The framework versions we can support.
205 * @fw_vercnt: The size of @fw_version.
206 * @srv_version: The service versions we can support.
207 * @srv_vercnt: The size of @srv_version.
208 * @nego_fw_version: The selected framework version.
209 * @nego_srv_version: The selected service version.
211 * Note: Versions are given in decreasing order.
213 * Set up and fill in default negotiate response message.
214 * Mainly used by Hyper-V drivers.
216 bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp,
217 u8 *buf, const int *fw_version, int fw_vercnt,
218 const int *srv_version, int srv_vercnt,
219 int *nego_fw_version, int *nego_srv_version)
221 int icframe_major, icframe_minor;
222 int icmsg_major, icmsg_minor;
223 int fw_major, fw_minor;
224 int srv_major, srv_minor;
226 bool found_match = false;
227 struct icmsg_negotiate *negop;
229 icmsghdrp->icmsgsize = 0x10;
230 negop = (struct icmsg_negotiate *)&buf[
231 sizeof(struct vmbuspipe_hdr) +
232 sizeof(struct icmsg_hdr)];
234 icframe_major = negop->icframe_vercnt;
237 icmsg_major = negop->icmsg_vercnt;
241 * Select the framework version number we will
245 for (i = 0; i < fw_vercnt; i++) {
246 fw_major = (fw_version[i] >> 16);
247 fw_minor = (fw_version[i] & 0xFFFF);
249 for (j = 0; j < negop->icframe_vercnt; j++) {
250 if ((negop->icversion_data[j].major == fw_major) &&
251 (negop->icversion_data[j].minor == fw_minor)) {
252 icframe_major = negop->icversion_data[j].major;
253 icframe_minor = negop->icversion_data[j].minor;
268 for (i = 0; i < srv_vercnt; i++) {
269 srv_major = (srv_version[i] >> 16);
270 srv_minor = (srv_version[i] & 0xFFFF);
272 for (j = negop->icframe_vercnt;
273 (j < negop->icframe_vercnt + negop->icmsg_vercnt);
276 if ((negop->icversion_data[j].major == srv_major) &&
277 (negop->icversion_data[j].minor == srv_minor)) {
279 icmsg_major = negop->icversion_data[j].major;
280 icmsg_minor = negop->icversion_data[j].minor;
291 * Respond with the framework and service
292 * version numbers we can support.
297 negop->icframe_vercnt = 0;
298 negop->icmsg_vercnt = 0;
300 negop->icframe_vercnt = 1;
301 negop->icmsg_vercnt = 1;
305 *nego_fw_version = (icframe_major << 16) | icframe_minor;
307 if (nego_srv_version)
308 *nego_srv_version = (icmsg_major << 16) | icmsg_minor;
310 negop->icversion_data[0].major = icframe_major;
311 negop->icversion_data[0].minor = icframe_minor;
312 negop->icversion_data[1].major = icmsg_major;
313 negop->icversion_data[1].minor = icmsg_minor;
317 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp);
320 * alloc_channel - Allocate and initialize a vmbus channel object
322 static struct vmbus_channel *alloc_channel(void)
324 struct vmbus_channel *channel;
326 channel = kzalloc(sizeof(*channel), GFP_ATOMIC);
330 spin_lock_init(&channel->lock);
331 init_completion(&channel->rescind_event);
333 INIT_LIST_HEAD(&channel->sc_list);
334 INIT_LIST_HEAD(&channel->percpu_list);
336 tasklet_init(&channel->callback_event,
337 vmbus_on_event, (unsigned long)channel);
343 * free_channel - Release the resources used by the vmbus channel object
345 static void free_channel(struct vmbus_channel *channel)
347 tasklet_kill(&channel->callback_event);
349 kobject_put(&channel->kobj);
352 static void percpu_channel_enq(void *arg)
354 struct vmbus_channel *channel = arg;
355 struct hv_per_cpu_context *hv_cpu
356 = this_cpu_ptr(hv_context.cpu_context);
358 list_add_tail_rcu(&channel->percpu_list, &hv_cpu->chan_list);
361 static void percpu_channel_deq(void *arg)
363 struct vmbus_channel *channel = arg;
365 list_del_rcu(&channel->percpu_list);
369 static void vmbus_release_relid(u32 relid)
371 struct vmbus_channel_relid_released msg;
374 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
375 msg.child_relid = relid;
376 msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
377 ret = vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released),
380 trace_vmbus_release_relid(&msg, ret);
383 void hv_process_channel_removal(struct vmbus_channel *channel)
385 struct vmbus_channel *primary_channel;
388 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
389 BUG_ON(!channel->rescind);
391 if (channel->target_cpu != get_cpu()) {
393 smp_call_function_single(channel->target_cpu,
394 percpu_channel_deq, channel, true);
396 percpu_channel_deq(channel);
400 if (channel->primary_channel == NULL) {
401 list_del(&channel->listentry);
403 primary_channel = channel;
405 primary_channel = channel->primary_channel;
406 spin_lock_irqsave(&primary_channel->lock, flags);
407 list_del(&channel->sc_list);
408 primary_channel->num_sc--;
409 spin_unlock_irqrestore(&primary_channel->lock, flags);
413 * We need to free the bit for init_vp_index() to work in the case
414 * of sub-channel, when we reload drivers like hv_netvsc.
416 if (channel->affinity_policy == HV_LOCALIZED)
417 cpumask_clear_cpu(channel->target_cpu,
418 &primary_channel->alloced_cpus_in_node);
420 vmbus_release_relid(channel->offermsg.child_relid);
422 free_channel(channel);
425 void vmbus_free_channels(void)
427 struct vmbus_channel *channel, *tmp;
429 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
431 /* hv_process_channel_removal() needs this */
432 channel->rescind = true;
434 vmbus_device_unregister(channel->device_obj);
438 /* Note: the function can run concurrently for primary/sub channels. */
439 static void vmbus_add_channel_work(struct work_struct *work)
441 struct vmbus_channel *newchannel =
442 container_of(work, struct vmbus_channel, add_channel_work);
443 struct vmbus_channel *primary_channel = newchannel->primary_channel;
448 dev_type = hv_get_dev_type(newchannel);
450 init_vp_index(newchannel, dev_type);
452 if (newchannel->target_cpu != get_cpu()) {
454 smp_call_function_single(newchannel->target_cpu,
458 percpu_channel_enq(newchannel);
463 * This state is used to indicate a successful open
464 * so that when we do close the channel normally, we
465 * can cleanup properly.
467 newchannel->state = CHANNEL_OPEN_STATE;
469 if (primary_channel != NULL) {
470 /* newchannel is a sub-channel. */
471 struct hv_device *dev = primary_channel->device_obj;
473 if (vmbus_add_channel_kobj(dev, newchannel))
476 if (primary_channel->sc_creation_callback != NULL)
477 primary_channel->sc_creation_callback(newchannel);
479 newchannel->probe_done = true;
484 * Start the process of binding the primary channel to the driver
486 newchannel->device_obj = vmbus_device_create(
487 &newchannel->offermsg.offer.if_type,
488 &newchannel->offermsg.offer.if_instance,
490 if (!newchannel->device_obj)
493 newchannel->device_obj->device_id = dev_type;
495 * Add the new device to the bus. This will kick off device-driver
496 * binding which eventually invokes the device driver's AddDevice()
499 ret = vmbus_device_register(newchannel->device_obj);
502 pr_err("unable to add child device object (relid %d)\n",
503 newchannel->offermsg.child_relid);
504 kfree(newchannel->device_obj);
508 newchannel->probe_done = true;
512 mutex_lock(&vmbus_connection.channel_mutex);
515 * We need to set the flag, otherwise
516 * vmbus_onoffer_rescind() can be blocked.
518 newchannel->probe_done = true;
520 if (primary_channel == NULL) {
521 list_del(&newchannel->listentry);
523 spin_lock_irqsave(&primary_channel->lock, flags);
524 list_del(&newchannel->sc_list);
525 spin_unlock_irqrestore(&primary_channel->lock, flags);
528 mutex_unlock(&vmbus_connection.channel_mutex);
530 if (newchannel->target_cpu != get_cpu()) {
532 smp_call_function_single(newchannel->target_cpu,
536 percpu_channel_deq(newchannel);
540 vmbus_release_relid(newchannel->offermsg.child_relid);
542 free_channel(newchannel);
546 * vmbus_process_offer - Process the offer by creating a channel/device
547 * associated with this offer
549 static void vmbus_process_offer(struct vmbus_channel *newchannel)
551 struct vmbus_channel *channel;
552 struct workqueue_struct *wq;
556 mutex_lock(&vmbus_connection.channel_mutex);
559 * Now that we have acquired the channel_mutex,
560 * we can release the potentially racing rescind thread.
562 atomic_dec(&vmbus_connection.offer_in_progress);
564 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
565 if (!uuid_le_cmp(channel->offermsg.offer.if_type,
566 newchannel->offermsg.offer.if_type) &&
567 !uuid_le_cmp(channel->offermsg.offer.if_instance,
568 newchannel->offermsg.offer.if_instance)) {
575 list_add_tail(&newchannel->listentry,
576 &vmbus_connection.chn_list);
579 * Check to see if this is a valid sub-channel.
581 if (newchannel->offermsg.offer.sub_channel_index == 0) {
582 mutex_unlock(&vmbus_connection.channel_mutex);
584 * Don't call free_channel(), because newchannel->kobj
585 * is not initialized yet.
592 * Process the sub-channel.
594 newchannel->primary_channel = channel;
595 spin_lock_irqsave(&channel->lock, flags);
596 list_add_tail(&newchannel->sc_list, &channel->sc_list);
597 spin_unlock_irqrestore(&channel->lock, flags);
600 mutex_unlock(&vmbus_connection.channel_mutex);
603 * vmbus_process_offer() mustn't call channel->sc_creation_callback()
604 * directly for sub-channels, because sc_creation_callback() ->
605 * vmbus_open() may never get the host's response to the
606 * OPEN_CHANNEL message (the host may rescind a channel at any time,
607 * e.g. in the case of hot removing a NIC), and vmbus_onoffer_rescind()
608 * may not wake up the vmbus_open() as it's blocked due to a non-zero
609 * vmbus_connection.offer_in_progress, and finally we have a deadlock.
611 * The above is also true for primary channels, if the related device
612 * drivers use sync probing mode by default.
614 * And, usually the handling of primary channels and sub-channels can
615 * depend on each other, so we should offload them to different
616 * workqueues to avoid possible deadlock, e.g. in sync-probing mode,
617 * NIC1's netvsc_subchan_work() can race with NIC2's netvsc_probe() ->
618 * rtnl_lock(), and causes deadlock: the former gets the rtnl_lock
619 * and waits for all the sub-channels to appear, but the latter
620 * can't get the rtnl_lock and this blocks the handling of
623 INIT_WORK(&newchannel->add_channel_work, vmbus_add_channel_work);
624 wq = fnew ? vmbus_connection.handle_primary_chan_wq :
625 vmbus_connection.handle_sub_chan_wq;
626 queue_work(wq, &newchannel->add_channel_work);
630 * We use this state to statically distribute the channel interrupt load.
632 static int next_numa_node_id;
634 * init_vp_index() accesses global variables like next_numa_node_id, and
635 * it can run concurrently for primary channels and sub-channels: see
636 * vmbus_process_offer(), so we need the lock to protect the global
639 static DEFINE_SPINLOCK(bind_channel_to_cpu_lock);
642 * Starting with Win8, we can statically distribute the incoming
643 * channel interrupt load by binding a channel to VCPU.
644 * We distribute the interrupt loads to one or more NUMA nodes based on
645 * the channel's affinity_policy.
647 * For pre-win8 hosts or non-performance critical channels we assign the
648 * first CPU in the first NUMA node.
650 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
653 bool perf_chn = vmbus_devs[dev_type].perf_device;
654 struct vmbus_channel *primary = channel->primary_channel;
656 cpumask_var_t available_mask;
657 struct cpumask *alloced_mask;
659 if ((vmbus_proto_version == VERSION_WS2008) ||
660 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn) ||
661 !alloc_cpumask_var(&available_mask, GFP_KERNEL)) {
663 * Prior to win8, all channel interrupts are
664 * delivered on cpu 0.
665 * Also if the channel is not a performance critical
666 * channel, bind it to cpu 0.
667 * In case alloc_cpumask_var() fails, bind it to cpu 0.
669 channel->numa_node = 0;
670 channel->target_cpu = 0;
671 channel->target_vp = hv_cpu_number_to_vp_number(0);
675 spin_lock(&bind_channel_to_cpu_lock);
678 * Based on the channel affinity policy, we will assign the NUMA
682 if ((channel->affinity_policy == HV_BALANCED) || (!primary)) {
684 next_node = next_numa_node_id++;
685 if (next_node == nr_node_ids) {
686 next_node = next_numa_node_id = 0;
689 if (cpumask_empty(cpumask_of_node(next_node)))
693 channel->numa_node = next_node;
696 alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
698 if (cpumask_weight(alloced_mask) ==
699 cpumask_weight(cpumask_of_node(primary->numa_node))) {
701 * We have cycled through all the CPUs in the node;
702 * reset the alloced map.
704 cpumask_clear(alloced_mask);
707 cpumask_xor(available_mask, alloced_mask,
708 cpumask_of_node(primary->numa_node));
712 if (primary->affinity_policy == HV_LOCALIZED) {
714 * Normally Hyper-V host doesn't create more subchannels
715 * than there are VCPUs on the node but it is possible when not
716 * all present VCPUs on the node are initialized by guest.
717 * Clear the alloced_cpus_in_node to start over.
719 if (cpumask_equal(&primary->alloced_cpus_in_node,
720 cpumask_of_node(primary->numa_node)))
721 cpumask_clear(&primary->alloced_cpus_in_node);
725 cur_cpu = cpumask_next(cur_cpu, available_mask);
726 if (cur_cpu >= nr_cpu_ids) {
728 cpumask_copy(available_mask,
729 cpumask_of_node(primary->numa_node));
733 if (primary->affinity_policy == HV_LOCALIZED) {
735 * NOTE: in the case of sub-channel, we clear the
736 * sub-channel related bit(s) in
737 * primary->alloced_cpus_in_node in
738 * hv_process_channel_removal(), so when we
739 * reload drivers like hv_netvsc in SMP guest, here
740 * we're able to re-allocate
741 * bit from primary->alloced_cpus_in_node.
743 if (!cpumask_test_cpu(cur_cpu,
744 &primary->alloced_cpus_in_node)) {
745 cpumask_set_cpu(cur_cpu,
746 &primary->alloced_cpus_in_node);
747 cpumask_set_cpu(cur_cpu, alloced_mask);
751 cpumask_set_cpu(cur_cpu, alloced_mask);
756 channel->target_cpu = cur_cpu;
757 channel->target_vp = hv_cpu_number_to_vp_number(cur_cpu);
759 spin_unlock(&bind_channel_to_cpu_lock);
761 free_cpumask_var(available_mask);
764 static void vmbus_wait_for_unload(void)
768 struct hv_message *msg;
769 struct vmbus_channel_message_header *hdr;
773 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
774 * used for initial contact or to CPU0 depending on host version. When
775 * we're crashing on a different CPU let's hope that IRQ handler on
776 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
777 * functional and vmbus_unload_response() will complete
778 * vmbus_connection.unload_event. If not, the last thing we can do is
779 * read message pages for all CPUs directly.
782 if (completion_done(&vmbus_connection.unload_event))
785 for_each_online_cpu(cpu) {
786 struct hv_per_cpu_context *hv_cpu
787 = per_cpu_ptr(hv_context.cpu_context, cpu);
789 page_addr = hv_cpu->synic_message_page;
790 msg = (struct hv_message *)page_addr
791 + VMBUS_MESSAGE_SINT;
793 message_type = READ_ONCE(msg->header.message_type);
794 if (message_type == HVMSG_NONE)
797 hdr = (struct vmbus_channel_message_header *)
800 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
801 complete(&vmbus_connection.unload_event);
803 vmbus_signal_eom(msg, message_type);
810 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
811 * maybe-pending messages on all CPUs to be able to receive new
812 * messages after we reconnect.
814 for_each_online_cpu(cpu) {
815 struct hv_per_cpu_context *hv_cpu
816 = per_cpu_ptr(hv_context.cpu_context, cpu);
818 page_addr = hv_cpu->synic_message_page;
819 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
820 msg->header.message_type = HVMSG_NONE;
825 * vmbus_unload_response - Handler for the unload response.
827 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
830 * This is a global event; just wakeup the waiting thread.
831 * Once we successfully unload, we can cleanup the monitor state.
833 complete(&vmbus_connection.unload_event);
836 void vmbus_initiate_unload(bool crash)
838 struct vmbus_channel_message_header hdr;
840 /* Pre-Win2012R2 hosts don't support reconnect */
841 if (vmbus_proto_version < VERSION_WIN8_1)
844 init_completion(&vmbus_connection.unload_event);
845 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
846 hdr.msgtype = CHANNELMSG_UNLOAD;
847 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
851 * vmbus_initiate_unload() is also called on crash and the crash can be
852 * happening in an interrupt context, where scheduling is impossible.
855 wait_for_completion(&vmbus_connection.unload_event);
857 vmbus_wait_for_unload();
861 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
864 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
866 struct vmbus_channel_offer_channel *offer;
867 struct vmbus_channel *newchannel;
869 offer = (struct vmbus_channel_offer_channel *)hdr;
871 trace_vmbus_onoffer(offer);
873 /* Allocate the channel object and save this offer. */
874 newchannel = alloc_channel();
876 vmbus_release_relid(offer->child_relid);
877 atomic_dec(&vmbus_connection.offer_in_progress);
878 pr_err("Unable to allocate channel object\n");
883 * Setup state for signalling the host.
885 newchannel->sig_event = VMBUS_EVENT_CONNECTION_ID;
887 if (vmbus_proto_version != VERSION_WS2008) {
888 newchannel->is_dedicated_interrupt =
889 (offer->is_dedicated_interrupt != 0);
890 newchannel->sig_event = offer->connection_id;
893 memcpy(&newchannel->offermsg, offer,
894 sizeof(struct vmbus_channel_offer_channel));
895 newchannel->monitor_grp = (u8)offer->monitorid / 32;
896 newchannel->monitor_bit = (u8)offer->monitorid % 32;
898 vmbus_process_offer(newchannel);
902 * vmbus_onoffer_rescind - Rescind offer handler.
904 * We queue a work item to process this offer synchronously
906 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
908 struct vmbus_channel_rescind_offer *rescind;
909 struct vmbus_channel *channel;
912 rescind = (struct vmbus_channel_rescind_offer *)hdr;
914 trace_vmbus_onoffer_rescind(rescind);
917 * The offer msg and the corresponding rescind msg
918 * from the host are guranteed to be ordered -
919 * offer comes in first and then the rescind.
920 * Since we process these events in work elements,
921 * and with preemption, we may end up processing
922 * the events out of order. Given that we handle these
923 * work elements on the same CPU, this is possible only
924 * in the case of preemption. In any case wait here
925 * until the offer processing has moved beyond the
926 * point where the channel is discoverable.
929 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
931 * We wait here until any channel offer is currently
937 mutex_lock(&vmbus_connection.channel_mutex);
938 channel = relid2channel(rescind->child_relid);
939 mutex_unlock(&vmbus_connection.channel_mutex);
941 if (channel == NULL) {
943 * We failed in processing the offer message;
944 * we would have cleaned up the relid in that
951 * Before setting channel->rescind in vmbus_rescind_cleanup(), we
952 * should make sure the channel callback is not running any more.
954 vmbus_reset_channel_cb(channel);
957 * Now wait for offer handling to complete.
959 vmbus_rescind_cleanup(channel);
960 while (READ_ONCE(channel->probe_done) == false) {
962 * We wait here until any channel offer is currently
969 * At this point, the rescind handling can proceed safely.
972 if (channel->device_obj) {
973 if (channel->chn_rescind_callback) {
974 channel->chn_rescind_callback(channel);
978 * We will have to unregister this device from the
981 dev = get_device(&channel->device_obj->device);
983 vmbus_device_unregister(channel->device_obj);
987 if (channel->primary_channel != NULL) {
989 * Sub-channel is being rescinded. Following is the channel
990 * close sequence when initiated from the driveri (refer to
991 * vmbus_close() for details):
992 * 1. Close all sub-channels first
993 * 2. Then close the primary channel.
995 mutex_lock(&vmbus_connection.channel_mutex);
996 if (channel->state == CHANNEL_OPEN_STATE) {
998 * The channel is currently not open;
999 * it is safe for us to cleanup the channel.
1001 hv_process_channel_removal(channel);
1003 complete(&channel->rescind_event);
1005 mutex_unlock(&vmbus_connection.channel_mutex);
1009 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
1011 BUG_ON(!is_hvsock_channel(channel));
1013 /* We always get a rescind msg when a connection is closed. */
1014 while (!READ_ONCE(channel->probe_done) || !READ_ONCE(channel->rescind))
1017 vmbus_device_unregister(channel->device_obj);
1019 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
1023 * vmbus_onoffers_delivered -
1024 * This is invoked when all offers have been delivered.
1026 * Nothing to do here.
1028 static void vmbus_onoffers_delivered(
1029 struct vmbus_channel_message_header *hdr)
1034 * vmbus_onopen_result - Open result handler.
1036 * This is invoked when we received a response to our channel open request.
1037 * Find the matching request, copy the response and signal the requesting
1040 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
1042 struct vmbus_channel_open_result *result;
1043 struct vmbus_channel_msginfo *msginfo;
1044 struct vmbus_channel_message_header *requestheader;
1045 struct vmbus_channel_open_channel *openmsg;
1046 unsigned long flags;
1048 result = (struct vmbus_channel_open_result *)hdr;
1050 trace_vmbus_onopen_result(result);
1053 * Find the open msg, copy the result and signal/unblock the wait event
1055 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1057 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1060 (struct vmbus_channel_message_header *)msginfo->msg;
1062 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
1064 (struct vmbus_channel_open_channel *)msginfo->msg;
1065 if (openmsg->child_relid == result->child_relid &&
1066 openmsg->openid == result->openid) {
1067 memcpy(&msginfo->response.open_result,
1070 struct vmbus_channel_open_result));
1071 complete(&msginfo->waitevent);
1076 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1080 * vmbus_ongpadl_created - GPADL created handler.
1082 * This is invoked when we received a response to our gpadl create request.
1083 * Find the matching request, copy the response and signal the requesting
1086 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
1088 struct vmbus_channel_gpadl_created *gpadlcreated;
1089 struct vmbus_channel_msginfo *msginfo;
1090 struct vmbus_channel_message_header *requestheader;
1091 struct vmbus_channel_gpadl_header *gpadlheader;
1092 unsigned long flags;
1094 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
1096 trace_vmbus_ongpadl_created(gpadlcreated);
1099 * Find the establish msg, copy the result and signal/unblock the wait
1102 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1104 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1107 (struct vmbus_channel_message_header *)msginfo->msg;
1109 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
1111 (struct vmbus_channel_gpadl_header *)requestheader;
1113 if ((gpadlcreated->child_relid ==
1114 gpadlheader->child_relid) &&
1115 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1116 memcpy(&msginfo->response.gpadl_created,
1119 struct vmbus_channel_gpadl_created));
1120 complete(&msginfo->waitevent);
1125 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1129 * vmbus_ongpadl_torndown - GPADL torndown handler.
1131 * This is invoked when we received a response to our gpadl teardown request.
1132 * Find the matching request, copy the response and signal the requesting
1135 static void vmbus_ongpadl_torndown(
1136 struct vmbus_channel_message_header *hdr)
1138 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1139 struct vmbus_channel_msginfo *msginfo;
1140 struct vmbus_channel_message_header *requestheader;
1141 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1142 unsigned long flags;
1144 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1146 trace_vmbus_ongpadl_torndown(gpadl_torndown);
1149 * Find the open msg, copy the result and signal/unblock the wait event
1151 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1153 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1156 (struct vmbus_channel_message_header *)msginfo->msg;
1158 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1160 (struct vmbus_channel_gpadl_teardown *)requestheader;
1162 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1163 memcpy(&msginfo->response.gpadl_torndown,
1166 struct vmbus_channel_gpadl_torndown));
1167 complete(&msginfo->waitevent);
1172 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1176 * vmbus_onversion_response - Version response handler
1178 * This is invoked when we received a response to our initiate contact request.
1179 * Find the matching request, copy the response and signal the requesting
1182 static void vmbus_onversion_response(
1183 struct vmbus_channel_message_header *hdr)
1185 struct vmbus_channel_msginfo *msginfo;
1186 struct vmbus_channel_message_header *requestheader;
1187 struct vmbus_channel_version_response *version_response;
1188 unsigned long flags;
1190 version_response = (struct vmbus_channel_version_response *)hdr;
1192 trace_vmbus_onversion_response(version_response);
1194 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1196 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1199 (struct vmbus_channel_message_header *)msginfo->msg;
1201 if (requestheader->msgtype ==
1202 CHANNELMSG_INITIATE_CONTACT) {
1203 memcpy(&msginfo->response.version_response,
1205 sizeof(struct vmbus_channel_version_response));
1206 complete(&msginfo->waitevent);
1209 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1212 /* Channel message dispatch table */
1213 const struct vmbus_channel_message_table_entry
1214 channel_message_table[CHANNELMSG_COUNT] = {
1215 { CHANNELMSG_INVALID, 0, NULL },
1216 { CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer },
1217 { CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind },
1218 { CHANNELMSG_REQUESTOFFERS, 0, NULL },
1219 { CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered },
1220 { CHANNELMSG_OPENCHANNEL, 0, NULL },
1221 { CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result },
1222 { CHANNELMSG_CLOSECHANNEL, 0, NULL },
1223 { CHANNELMSG_GPADL_HEADER, 0, NULL },
1224 { CHANNELMSG_GPADL_BODY, 0, NULL },
1225 { CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created },
1226 { CHANNELMSG_GPADL_TEARDOWN, 0, NULL },
1227 { CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown },
1228 { CHANNELMSG_RELID_RELEASED, 0, NULL },
1229 { CHANNELMSG_INITIATE_CONTACT, 0, NULL },
1230 { CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response },
1231 { CHANNELMSG_UNLOAD, 0, NULL },
1232 { CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response },
1233 { CHANNELMSG_18, 0, NULL },
1234 { CHANNELMSG_19, 0, NULL },
1235 { CHANNELMSG_20, 0, NULL },
1236 { CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL },
1240 * vmbus_onmessage - Handler for channel protocol messages.
1242 * This is invoked in the vmbus worker thread context.
1244 void vmbus_onmessage(void *context)
1246 struct hv_message *msg = context;
1247 struct vmbus_channel_message_header *hdr;
1250 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1251 size = msg->header.payload_size;
1253 trace_vmbus_on_message(hdr);
1255 if (hdr->msgtype >= CHANNELMSG_COUNT) {
1256 pr_err("Received invalid channel message type %d size %d\n",
1257 hdr->msgtype, size);
1258 print_hex_dump_bytes("", DUMP_PREFIX_NONE,
1259 (unsigned char *)msg->u.payload, size);
1263 if (channel_message_table[hdr->msgtype].message_handler)
1264 channel_message_table[hdr->msgtype].message_handler(hdr);
1266 pr_err("Unhandled channel message type %d\n", hdr->msgtype);
1270 * vmbus_request_offers - Send a request to get all our pending offers.
1272 int vmbus_request_offers(void)
1274 struct vmbus_channel_message_header *msg;
1275 struct vmbus_channel_msginfo *msginfo;
1278 msginfo = kmalloc(sizeof(*msginfo) +
1279 sizeof(struct vmbus_channel_message_header),
1284 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1286 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1288 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1291 trace_vmbus_request_offers(ret);
1294 pr_err("Unable to request offers - %d\n", ret);
1306 * Retrieve the (sub) channel on which to send an outgoing request.
1307 * When a primary channel has multiple sub-channels, we try to
1308 * distribute the load equally amongst all available channels.
1310 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary)
1312 struct list_head *cur, *tmp;
1314 struct vmbus_channel *cur_channel;
1315 struct vmbus_channel *outgoing_channel = primary;
1319 if (list_empty(&primary->sc_list))
1320 return outgoing_channel;
1322 next_channel = primary->next_oc++;
1324 if (next_channel > (primary->num_sc)) {
1325 primary->next_oc = 0;
1326 return outgoing_channel;
1329 cur_cpu = hv_cpu_number_to_vp_number(smp_processor_id());
1330 list_for_each_safe(cur, tmp, &primary->sc_list) {
1331 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1332 if (cur_channel->state != CHANNEL_OPENED_STATE)
1335 if (cur_channel->target_vp == cur_cpu)
1338 if (i == next_channel)
1344 return outgoing_channel;
1346 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel);
1348 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1350 struct list_head *cur, *tmp;
1351 struct vmbus_channel *cur_channel;
1353 if (primary_channel->sc_creation_callback == NULL)
1356 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1357 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1359 primary_channel->sc_creation_callback(cur_channel);
1363 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1364 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1366 primary_channel->sc_creation_callback = sc_cr_cb;
1368 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1370 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1374 ret = !list_empty(&primary->sc_list);
1378 * Invoke the callback on sub-channel creation.
1379 * This will present a uniform interface to the
1382 invoke_sc_cb(primary);
1387 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1389 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1390 void (*chn_rescind_cb)(struct vmbus_channel *))
1392 channel->chn_rescind_callback = chn_rescind_cb;
1394 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);