1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2009, Microsoft Corporation.
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 #include <linux/kernel.h>
12 #include <linux/interrupt.h>
13 #include <linux/sched.h>
14 #include <linux/wait.h>
16 #include <linux/slab.h>
17 #include <linux/list.h>
18 #include <linux/module.h>
19 #include <linux/completion.h>
20 #include <linux/delay.h>
21 #include <linux/hyperv.h>
22 #include <asm/mshyperv.h>
24 #include "hyperv_vmbus.h"
26 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type);
28 static const struct vmbus_device vmbus_devs[] = {
36 { .dev_type = HV_SCSI,
60 { .dev_type = HV_PCIE,
65 /* Synthetic Frame Buffer */
71 /* Synthetic Keyboard */
78 { .dev_type = HV_MOUSE,
102 { .dev_type = HV_SHUTDOWN,
104 .perf_device = false,
108 { .dev_type = HV_FCOPY,
110 .perf_device = false,
114 { .dev_type = HV_BACKUP,
116 .perf_device = false,
122 .perf_device = false,
126 { .dev_type = HV_UNKNOWN,
127 .perf_device = false,
131 static const struct {
133 } vmbus_unsupported_devs[] = {
140 * The rescinded channel may be blocked waiting for a response from the host;
143 static void vmbus_rescind_cleanup(struct vmbus_channel *channel)
145 struct vmbus_channel_msginfo *msginfo;
149 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
150 channel->rescind = true;
151 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
154 if (msginfo->waiting_channel == channel) {
155 complete(&msginfo->waitevent);
159 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
162 static bool is_unsupported_vmbus_devs(const guid_t *guid)
166 for (i = 0; i < ARRAY_SIZE(vmbus_unsupported_devs); i++)
167 if (guid_equal(guid, &vmbus_unsupported_devs[i].guid))
172 static u16 hv_get_dev_type(const struct vmbus_channel *channel)
174 const guid_t *guid = &channel->offermsg.offer.if_type;
177 if (is_hvsock_channel(channel) || is_unsupported_vmbus_devs(guid))
180 for (i = HV_IDE; i < HV_UNKNOWN; i++) {
181 if (guid_equal(guid, &vmbus_devs[i].guid))
184 pr_info("Unknown GUID: %pUl\n", guid);
189 * vmbus_prep_negotiate_resp() - Create default response for Negotiate message
190 * @icmsghdrp: Pointer to msg header structure
191 * @buf: Raw buffer channel data
192 * @fw_version: The framework versions we can support.
193 * @fw_vercnt: The size of @fw_version.
194 * @srv_version: The service versions we can support.
195 * @srv_vercnt: The size of @srv_version.
196 * @nego_fw_version: The selected framework version.
197 * @nego_srv_version: The selected service version.
199 * Note: Versions are given in decreasing order.
201 * Set up and fill in default negotiate response message.
202 * Mainly used by Hyper-V drivers.
204 bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp,
205 u8 *buf, const int *fw_version, int fw_vercnt,
206 const int *srv_version, int srv_vercnt,
207 int *nego_fw_version, int *nego_srv_version)
209 int icframe_major, icframe_minor;
210 int icmsg_major, icmsg_minor;
211 int fw_major, fw_minor;
212 int srv_major, srv_minor;
214 bool found_match = false;
215 struct icmsg_negotiate *negop;
217 icmsghdrp->icmsgsize = 0x10;
218 negop = (struct icmsg_negotiate *)&buf[
219 sizeof(struct vmbuspipe_hdr) +
220 sizeof(struct icmsg_hdr)];
222 icframe_major = negop->icframe_vercnt;
225 icmsg_major = negop->icmsg_vercnt;
229 * Select the framework version number we will
233 for (i = 0; i < fw_vercnt; i++) {
234 fw_major = (fw_version[i] >> 16);
235 fw_minor = (fw_version[i] & 0xFFFF);
237 for (j = 0; j < negop->icframe_vercnt; j++) {
238 if ((negop->icversion_data[j].major == fw_major) &&
239 (negop->icversion_data[j].minor == fw_minor)) {
240 icframe_major = negop->icversion_data[j].major;
241 icframe_minor = negop->icversion_data[j].minor;
256 for (i = 0; i < srv_vercnt; i++) {
257 srv_major = (srv_version[i] >> 16);
258 srv_minor = (srv_version[i] & 0xFFFF);
260 for (j = negop->icframe_vercnt;
261 (j < negop->icframe_vercnt + negop->icmsg_vercnt);
264 if ((negop->icversion_data[j].major == srv_major) &&
265 (negop->icversion_data[j].minor == srv_minor)) {
267 icmsg_major = negop->icversion_data[j].major;
268 icmsg_minor = negop->icversion_data[j].minor;
279 * Respond with the framework and service
280 * version numbers we can support.
285 negop->icframe_vercnt = 0;
286 negop->icmsg_vercnt = 0;
288 negop->icframe_vercnt = 1;
289 negop->icmsg_vercnt = 1;
293 *nego_fw_version = (icframe_major << 16) | icframe_minor;
295 if (nego_srv_version)
296 *nego_srv_version = (icmsg_major << 16) | icmsg_minor;
298 negop->icversion_data[0].major = icframe_major;
299 negop->icversion_data[0].minor = icframe_minor;
300 negop->icversion_data[1].major = icmsg_major;
301 negop->icversion_data[1].minor = icmsg_minor;
305 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp);
308 * alloc_channel - Allocate and initialize a vmbus channel object
310 static struct vmbus_channel *alloc_channel(void)
312 struct vmbus_channel *channel;
314 channel = kzalloc(sizeof(*channel), GFP_ATOMIC);
318 spin_lock_init(&channel->lock);
319 init_completion(&channel->rescind_event);
321 INIT_LIST_HEAD(&channel->sc_list);
322 INIT_LIST_HEAD(&channel->percpu_list);
324 tasklet_init(&channel->callback_event,
325 vmbus_on_event, (unsigned long)channel);
327 hv_ringbuffer_pre_init(channel);
333 * free_channel - Release the resources used by the vmbus channel object
335 static void free_channel(struct vmbus_channel *channel)
337 tasklet_kill(&channel->callback_event);
338 vmbus_remove_channel_attr_group(channel);
340 kobject_put(&channel->kobj);
343 static void percpu_channel_enq(void *arg)
345 struct vmbus_channel *channel = arg;
346 struct hv_per_cpu_context *hv_cpu
347 = this_cpu_ptr(hv_context.cpu_context);
349 list_add_tail_rcu(&channel->percpu_list, &hv_cpu->chan_list);
352 static void percpu_channel_deq(void *arg)
354 struct vmbus_channel *channel = arg;
356 list_del_rcu(&channel->percpu_list);
360 static void vmbus_release_relid(u32 relid)
362 struct vmbus_channel_relid_released msg;
365 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
366 msg.child_relid = relid;
367 msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
368 ret = vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released),
371 trace_vmbus_release_relid(&msg, ret);
374 void hv_process_channel_removal(struct vmbus_channel *channel)
376 struct vmbus_channel *primary_channel;
379 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
380 BUG_ON(!channel->rescind);
382 if (channel->target_cpu != get_cpu()) {
384 smp_call_function_single(channel->target_cpu,
385 percpu_channel_deq, channel, true);
387 percpu_channel_deq(channel);
391 if (channel->primary_channel == NULL) {
392 list_del(&channel->listentry);
394 primary_channel = channel;
396 primary_channel = channel->primary_channel;
397 spin_lock_irqsave(&primary_channel->lock, flags);
398 list_del(&channel->sc_list);
399 spin_unlock_irqrestore(&primary_channel->lock, flags);
403 * We need to free the bit for init_vp_index() to work in the case
404 * of sub-channel, when we reload drivers like hv_netvsc.
406 if (channel->affinity_policy == HV_LOCALIZED)
407 cpumask_clear_cpu(channel->target_cpu,
408 &primary_channel->alloced_cpus_in_node);
410 vmbus_release_relid(channel->offermsg.child_relid);
412 free_channel(channel);
415 void vmbus_free_channels(void)
417 struct vmbus_channel *channel, *tmp;
419 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
421 /* hv_process_channel_removal() needs this */
422 channel->rescind = true;
424 vmbus_device_unregister(channel->device_obj);
428 /* Note: the function can run concurrently for primary/sub channels. */
429 static void vmbus_add_channel_work(struct work_struct *work)
431 struct vmbus_channel *newchannel =
432 container_of(work, struct vmbus_channel, add_channel_work);
433 struct vmbus_channel *primary_channel = newchannel->primary_channel;
438 dev_type = hv_get_dev_type(newchannel);
440 init_vp_index(newchannel, dev_type);
442 if (newchannel->target_cpu != get_cpu()) {
444 smp_call_function_single(newchannel->target_cpu,
448 percpu_channel_enq(newchannel);
453 * This state is used to indicate a successful open
454 * so that when we do close the channel normally, we
455 * can cleanup properly.
457 newchannel->state = CHANNEL_OPEN_STATE;
459 if (primary_channel != NULL) {
460 /* newchannel is a sub-channel. */
461 struct hv_device *dev = primary_channel->device_obj;
463 if (vmbus_add_channel_kobj(dev, newchannel))
466 if (primary_channel->sc_creation_callback != NULL)
467 primary_channel->sc_creation_callback(newchannel);
469 newchannel->probe_done = true;
474 * Start the process of binding the primary channel to the driver
476 newchannel->device_obj = vmbus_device_create(
477 &newchannel->offermsg.offer.if_type,
478 &newchannel->offermsg.offer.if_instance,
480 if (!newchannel->device_obj)
483 newchannel->device_obj->device_id = dev_type;
485 * Add the new device to the bus. This will kick off device-driver
486 * binding which eventually invokes the device driver's AddDevice()
489 ret = vmbus_device_register(newchannel->device_obj);
492 pr_err("unable to add child device object (relid %d)\n",
493 newchannel->offermsg.child_relid);
494 kfree(newchannel->device_obj);
498 newchannel->probe_done = true;
502 mutex_lock(&vmbus_connection.channel_mutex);
505 * We need to set the flag, otherwise
506 * vmbus_onoffer_rescind() can be blocked.
508 newchannel->probe_done = true;
510 if (primary_channel == NULL) {
511 list_del(&newchannel->listentry);
513 spin_lock_irqsave(&primary_channel->lock, flags);
514 list_del(&newchannel->sc_list);
515 spin_unlock_irqrestore(&primary_channel->lock, flags);
518 mutex_unlock(&vmbus_connection.channel_mutex);
520 if (newchannel->target_cpu != get_cpu()) {
522 smp_call_function_single(newchannel->target_cpu,
526 percpu_channel_deq(newchannel);
530 vmbus_release_relid(newchannel->offermsg.child_relid);
532 free_channel(newchannel);
536 * vmbus_process_offer - Process the offer by creating a channel/device
537 * associated with this offer
539 static void vmbus_process_offer(struct vmbus_channel *newchannel)
541 struct vmbus_channel *channel;
542 struct workqueue_struct *wq;
546 mutex_lock(&vmbus_connection.channel_mutex);
549 * Now that we have acquired the channel_mutex,
550 * we can release the potentially racing rescind thread.
552 atomic_dec(&vmbus_connection.offer_in_progress);
554 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
555 if (guid_equal(&channel->offermsg.offer.if_type,
556 &newchannel->offermsg.offer.if_type) &&
557 guid_equal(&channel->offermsg.offer.if_instance,
558 &newchannel->offermsg.offer.if_instance)) {
565 list_add_tail(&newchannel->listentry,
566 &vmbus_connection.chn_list);
569 * Check to see if this is a valid sub-channel.
571 if (newchannel->offermsg.offer.sub_channel_index == 0) {
572 mutex_unlock(&vmbus_connection.channel_mutex);
574 * Don't call free_channel(), because newchannel->kobj
575 * is not initialized yet.
582 * Process the sub-channel.
584 newchannel->primary_channel = channel;
585 spin_lock_irqsave(&channel->lock, flags);
586 list_add_tail(&newchannel->sc_list, &channel->sc_list);
587 spin_unlock_irqrestore(&channel->lock, flags);
590 mutex_unlock(&vmbus_connection.channel_mutex);
593 * vmbus_process_offer() mustn't call channel->sc_creation_callback()
594 * directly for sub-channels, because sc_creation_callback() ->
595 * vmbus_open() may never get the host's response to the
596 * OPEN_CHANNEL message (the host may rescind a channel at any time,
597 * e.g. in the case of hot removing a NIC), and vmbus_onoffer_rescind()
598 * may not wake up the vmbus_open() as it's blocked due to a non-zero
599 * vmbus_connection.offer_in_progress, and finally we have a deadlock.
601 * The above is also true for primary channels, if the related device
602 * drivers use sync probing mode by default.
604 * And, usually the handling of primary channels and sub-channels can
605 * depend on each other, so we should offload them to different
606 * workqueues to avoid possible deadlock, e.g. in sync-probing mode,
607 * NIC1's netvsc_subchan_work() can race with NIC2's netvsc_probe() ->
608 * rtnl_lock(), and causes deadlock: the former gets the rtnl_lock
609 * and waits for all the sub-channels to appear, but the latter
610 * can't get the rtnl_lock and this blocks the handling of
613 INIT_WORK(&newchannel->add_channel_work, vmbus_add_channel_work);
614 wq = fnew ? vmbus_connection.handle_primary_chan_wq :
615 vmbus_connection.handle_sub_chan_wq;
616 queue_work(wq, &newchannel->add_channel_work);
620 * We use this state to statically distribute the channel interrupt load.
622 static int next_numa_node_id;
624 * init_vp_index() accesses global variables like next_numa_node_id, and
625 * it can run concurrently for primary channels and sub-channels: see
626 * vmbus_process_offer(), so we need the lock to protect the global
629 static DEFINE_SPINLOCK(bind_channel_to_cpu_lock);
632 * Starting with Win8, we can statically distribute the incoming
633 * channel interrupt load by binding a channel to VCPU.
634 * We distribute the interrupt loads to one or more NUMA nodes based on
635 * the channel's affinity_policy.
637 * For pre-win8 hosts or non-performance critical channels we assign the
638 * first CPU in the first NUMA node.
640 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
643 bool perf_chn = vmbus_devs[dev_type].perf_device;
644 struct vmbus_channel *primary = channel->primary_channel;
646 cpumask_var_t available_mask;
647 struct cpumask *alloced_mask;
649 if ((vmbus_proto_version == VERSION_WS2008) ||
650 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn) ||
651 !alloc_cpumask_var(&available_mask, GFP_KERNEL)) {
653 * Prior to win8, all channel interrupts are
654 * delivered on cpu 0.
655 * Also if the channel is not a performance critical
656 * channel, bind it to cpu 0.
657 * In case alloc_cpumask_var() fails, bind it to cpu 0.
659 channel->numa_node = 0;
660 channel->target_cpu = 0;
661 channel->target_vp = hv_cpu_number_to_vp_number(0);
665 spin_lock(&bind_channel_to_cpu_lock);
668 * Based on the channel affinity policy, we will assign the NUMA
672 if ((channel->affinity_policy == HV_BALANCED) || (!primary)) {
674 next_node = next_numa_node_id++;
675 if (next_node == nr_node_ids) {
676 next_node = next_numa_node_id = 0;
679 if (cpumask_empty(cpumask_of_node(next_node)))
683 channel->numa_node = next_node;
686 alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
688 if (cpumask_weight(alloced_mask) ==
689 cpumask_weight(cpumask_of_node(primary->numa_node))) {
691 * We have cycled through all the CPUs in the node;
692 * reset the alloced map.
694 cpumask_clear(alloced_mask);
697 cpumask_xor(available_mask, alloced_mask,
698 cpumask_of_node(primary->numa_node));
702 if (primary->affinity_policy == HV_LOCALIZED) {
704 * Normally Hyper-V host doesn't create more subchannels
705 * than there are VCPUs on the node but it is possible when not
706 * all present VCPUs on the node are initialized by guest.
707 * Clear the alloced_cpus_in_node to start over.
709 if (cpumask_equal(&primary->alloced_cpus_in_node,
710 cpumask_of_node(primary->numa_node)))
711 cpumask_clear(&primary->alloced_cpus_in_node);
715 cur_cpu = cpumask_next(cur_cpu, available_mask);
716 if (cur_cpu >= nr_cpu_ids) {
718 cpumask_copy(available_mask,
719 cpumask_of_node(primary->numa_node));
723 if (primary->affinity_policy == HV_LOCALIZED) {
725 * NOTE: in the case of sub-channel, we clear the
726 * sub-channel related bit(s) in
727 * primary->alloced_cpus_in_node in
728 * hv_process_channel_removal(), so when we
729 * reload drivers like hv_netvsc in SMP guest, here
730 * we're able to re-allocate
731 * bit from primary->alloced_cpus_in_node.
733 if (!cpumask_test_cpu(cur_cpu,
734 &primary->alloced_cpus_in_node)) {
735 cpumask_set_cpu(cur_cpu,
736 &primary->alloced_cpus_in_node);
737 cpumask_set_cpu(cur_cpu, alloced_mask);
741 cpumask_set_cpu(cur_cpu, alloced_mask);
746 channel->target_cpu = cur_cpu;
747 channel->target_vp = hv_cpu_number_to_vp_number(cur_cpu);
749 spin_unlock(&bind_channel_to_cpu_lock);
751 free_cpumask_var(available_mask);
754 static void vmbus_wait_for_unload(void)
758 struct hv_message *msg;
759 struct vmbus_channel_message_header *hdr;
763 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
764 * used for initial contact or to CPU0 depending on host version. When
765 * we're crashing on a different CPU let's hope that IRQ handler on
766 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
767 * functional and vmbus_unload_response() will complete
768 * vmbus_connection.unload_event. If not, the last thing we can do is
769 * read message pages for all CPUs directly.
772 if (completion_done(&vmbus_connection.unload_event))
775 for_each_online_cpu(cpu) {
776 struct hv_per_cpu_context *hv_cpu
777 = per_cpu_ptr(hv_context.cpu_context, cpu);
779 page_addr = hv_cpu->synic_message_page;
780 msg = (struct hv_message *)page_addr
781 + VMBUS_MESSAGE_SINT;
783 message_type = READ_ONCE(msg->header.message_type);
784 if (message_type == HVMSG_NONE)
787 hdr = (struct vmbus_channel_message_header *)
790 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
791 complete(&vmbus_connection.unload_event);
793 vmbus_signal_eom(msg, message_type);
800 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
801 * maybe-pending messages on all CPUs to be able to receive new
802 * messages after we reconnect.
804 for_each_online_cpu(cpu) {
805 struct hv_per_cpu_context *hv_cpu
806 = per_cpu_ptr(hv_context.cpu_context, cpu);
808 page_addr = hv_cpu->synic_message_page;
809 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
810 msg->header.message_type = HVMSG_NONE;
815 * vmbus_unload_response - Handler for the unload response.
817 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
820 * This is a global event; just wakeup the waiting thread.
821 * Once we successfully unload, we can cleanup the monitor state.
823 complete(&vmbus_connection.unload_event);
826 void vmbus_initiate_unload(bool crash)
828 struct vmbus_channel_message_header hdr;
830 /* Pre-Win2012R2 hosts don't support reconnect */
831 if (vmbus_proto_version < VERSION_WIN8_1)
834 init_completion(&vmbus_connection.unload_event);
835 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
836 hdr.msgtype = CHANNELMSG_UNLOAD;
837 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
841 * vmbus_initiate_unload() is also called on crash and the crash can be
842 * happening in an interrupt context, where scheduling is impossible.
845 wait_for_completion(&vmbus_connection.unload_event);
847 vmbus_wait_for_unload();
851 * find_primary_channel_by_offer - Get the channel object given the new offer.
852 * This is only used in the resume path of hibernation.
854 static struct vmbus_channel *
855 find_primary_channel_by_offer(const struct vmbus_channel_offer_channel *offer)
857 struct vmbus_channel *channel = NULL, *iter;
858 const guid_t *inst1, *inst2;
860 /* Ignore sub-channel offers. */
861 if (offer->offer.sub_channel_index != 0)
864 mutex_lock(&vmbus_connection.channel_mutex);
866 list_for_each_entry(iter, &vmbus_connection.chn_list, listentry) {
867 inst1 = &iter->offermsg.offer.if_instance;
868 inst2 = &offer->offer.if_instance;
870 if (guid_equal(inst1, inst2)) {
876 mutex_unlock(&vmbus_connection.channel_mutex);
882 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
885 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
887 struct vmbus_channel_offer_channel *offer;
888 struct vmbus_channel *oldchannel, *newchannel;
891 offer = (struct vmbus_channel_offer_channel *)hdr;
893 trace_vmbus_onoffer(offer);
895 oldchannel = find_primary_channel_by_offer(offer);
897 if (oldchannel != NULL) {
898 atomic_dec(&vmbus_connection.offer_in_progress);
901 * We're resuming from hibernation: we expect the host to send
902 * exactly the same offers that we had before the hibernation.
904 offer_sz = sizeof(*offer);
905 if (memcmp(offer, &oldchannel->offermsg, offer_sz) == 0)
908 pr_debug("Mismatched offer from the host (relid=%d)\n",
911 print_hex_dump_debug("Old vmbus offer: ", DUMP_PREFIX_OFFSET,
912 16, 4, &oldchannel->offermsg, offer_sz,
914 print_hex_dump_debug("New vmbus offer: ", DUMP_PREFIX_OFFSET,
915 16, 4, offer, offer_sz, false);
919 /* Allocate the channel object and save this offer. */
920 newchannel = alloc_channel();
922 vmbus_release_relid(offer->child_relid);
923 atomic_dec(&vmbus_connection.offer_in_progress);
924 pr_err("Unable to allocate channel object\n");
929 * Setup state for signalling the host.
931 newchannel->sig_event = VMBUS_EVENT_CONNECTION_ID;
933 if (vmbus_proto_version != VERSION_WS2008) {
934 newchannel->is_dedicated_interrupt =
935 (offer->is_dedicated_interrupt != 0);
936 newchannel->sig_event = offer->connection_id;
939 memcpy(&newchannel->offermsg, offer,
940 sizeof(struct vmbus_channel_offer_channel));
941 newchannel->monitor_grp = (u8)offer->monitorid / 32;
942 newchannel->monitor_bit = (u8)offer->monitorid % 32;
944 vmbus_process_offer(newchannel);
948 * vmbus_onoffer_rescind - Rescind offer handler.
950 * We queue a work item to process this offer synchronously
952 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
954 struct vmbus_channel_rescind_offer *rescind;
955 struct vmbus_channel *channel;
958 rescind = (struct vmbus_channel_rescind_offer *)hdr;
960 trace_vmbus_onoffer_rescind(rescind);
963 * The offer msg and the corresponding rescind msg
964 * from the host are guranteed to be ordered -
965 * offer comes in first and then the rescind.
966 * Since we process these events in work elements,
967 * and with preemption, we may end up processing
968 * the events out of order. Given that we handle these
969 * work elements on the same CPU, this is possible only
970 * in the case of preemption. In any case wait here
971 * until the offer processing has moved beyond the
972 * point where the channel is discoverable.
975 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
977 * We wait here until any channel offer is currently
983 mutex_lock(&vmbus_connection.channel_mutex);
984 channel = relid2channel(rescind->child_relid);
985 mutex_unlock(&vmbus_connection.channel_mutex);
987 if (channel == NULL) {
989 * We failed in processing the offer message;
990 * we would have cleaned up the relid in that
997 * Before setting channel->rescind in vmbus_rescind_cleanup(), we
998 * should make sure the channel callback is not running any more.
1000 vmbus_reset_channel_cb(channel);
1003 * Now wait for offer handling to complete.
1005 vmbus_rescind_cleanup(channel);
1006 while (READ_ONCE(channel->probe_done) == false) {
1008 * We wait here until any channel offer is currently
1015 * At this point, the rescind handling can proceed safely.
1018 if (channel->device_obj) {
1019 if (channel->chn_rescind_callback) {
1020 channel->chn_rescind_callback(channel);
1024 * We will have to unregister this device from the
1027 dev = get_device(&channel->device_obj->device);
1029 vmbus_device_unregister(channel->device_obj);
1033 if (channel->primary_channel != NULL) {
1035 * Sub-channel is being rescinded. Following is the channel
1036 * close sequence when initiated from the driveri (refer to
1037 * vmbus_close() for details):
1038 * 1. Close all sub-channels first
1039 * 2. Then close the primary channel.
1041 mutex_lock(&vmbus_connection.channel_mutex);
1042 if (channel->state == CHANNEL_OPEN_STATE) {
1044 * The channel is currently not open;
1045 * it is safe for us to cleanup the channel.
1047 hv_process_channel_removal(channel);
1049 complete(&channel->rescind_event);
1051 mutex_unlock(&vmbus_connection.channel_mutex);
1055 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
1057 BUG_ON(!is_hvsock_channel(channel));
1059 /* We always get a rescind msg when a connection is closed. */
1060 while (!READ_ONCE(channel->probe_done) || !READ_ONCE(channel->rescind))
1063 vmbus_device_unregister(channel->device_obj);
1065 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
1069 * vmbus_onoffers_delivered -
1070 * This is invoked when all offers have been delivered.
1072 * Nothing to do here.
1074 static void vmbus_onoffers_delivered(
1075 struct vmbus_channel_message_header *hdr)
1080 * vmbus_onopen_result - Open result handler.
1082 * This is invoked when we received a response to our channel open request.
1083 * Find the matching request, copy the response and signal the requesting
1086 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
1088 struct vmbus_channel_open_result *result;
1089 struct vmbus_channel_msginfo *msginfo;
1090 struct vmbus_channel_message_header *requestheader;
1091 struct vmbus_channel_open_channel *openmsg;
1092 unsigned long flags;
1094 result = (struct vmbus_channel_open_result *)hdr;
1096 trace_vmbus_onopen_result(result);
1099 * Find the open msg, copy the result and signal/unblock the wait event
1101 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1103 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1106 (struct vmbus_channel_message_header *)msginfo->msg;
1108 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
1110 (struct vmbus_channel_open_channel *)msginfo->msg;
1111 if (openmsg->child_relid == result->child_relid &&
1112 openmsg->openid == result->openid) {
1113 memcpy(&msginfo->response.open_result,
1116 struct vmbus_channel_open_result));
1117 complete(&msginfo->waitevent);
1122 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1126 * vmbus_ongpadl_created - GPADL created handler.
1128 * This is invoked when we received a response to our gpadl create request.
1129 * Find the matching request, copy the response and signal the requesting
1132 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
1134 struct vmbus_channel_gpadl_created *gpadlcreated;
1135 struct vmbus_channel_msginfo *msginfo;
1136 struct vmbus_channel_message_header *requestheader;
1137 struct vmbus_channel_gpadl_header *gpadlheader;
1138 unsigned long flags;
1140 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
1142 trace_vmbus_ongpadl_created(gpadlcreated);
1145 * Find the establish msg, copy the result and signal/unblock the wait
1148 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1150 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1153 (struct vmbus_channel_message_header *)msginfo->msg;
1155 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
1157 (struct vmbus_channel_gpadl_header *)requestheader;
1159 if ((gpadlcreated->child_relid ==
1160 gpadlheader->child_relid) &&
1161 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1162 memcpy(&msginfo->response.gpadl_created,
1165 struct vmbus_channel_gpadl_created));
1166 complete(&msginfo->waitevent);
1171 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1175 * vmbus_ongpadl_torndown - GPADL torndown handler.
1177 * This is invoked when we received a response to our gpadl teardown request.
1178 * Find the matching request, copy the response and signal the requesting
1181 static void vmbus_ongpadl_torndown(
1182 struct vmbus_channel_message_header *hdr)
1184 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1185 struct vmbus_channel_msginfo *msginfo;
1186 struct vmbus_channel_message_header *requestheader;
1187 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1188 unsigned long flags;
1190 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1192 trace_vmbus_ongpadl_torndown(gpadl_torndown);
1195 * Find the open msg, copy the result and signal/unblock the wait event
1197 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1199 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1202 (struct vmbus_channel_message_header *)msginfo->msg;
1204 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1206 (struct vmbus_channel_gpadl_teardown *)requestheader;
1208 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1209 memcpy(&msginfo->response.gpadl_torndown,
1212 struct vmbus_channel_gpadl_torndown));
1213 complete(&msginfo->waitevent);
1218 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1222 * vmbus_onversion_response - Version response handler
1224 * This is invoked when we received a response to our initiate contact request.
1225 * Find the matching request, copy the response and signal the requesting
1228 static void vmbus_onversion_response(
1229 struct vmbus_channel_message_header *hdr)
1231 struct vmbus_channel_msginfo *msginfo;
1232 struct vmbus_channel_message_header *requestheader;
1233 struct vmbus_channel_version_response *version_response;
1234 unsigned long flags;
1236 version_response = (struct vmbus_channel_version_response *)hdr;
1238 trace_vmbus_onversion_response(version_response);
1240 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1242 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1245 (struct vmbus_channel_message_header *)msginfo->msg;
1247 if (requestheader->msgtype ==
1248 CHANNELMSG_INITIATE_CONTACT) {
1249 memcpy(&msginfo->response.version_response,
1251 sizeof(struct vmbus_channel_version_response));
1252 complete(&msginfo->waitevent);
1255 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1258 /* Channel message dispatch table */
1259 const struct vmbus_channel_message_table_entry
1260 channel_message_table[CHANNELMSG_COUNT] = {
1261 { CHANNELMSG_INVALID, 0, NULL },
1262 { CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer },
1263 { CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind },
1264 { CHANNELMSG_REQUESTOFFERS, 0, NULL },
1265 { CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered },
1266 { CHANNELMSG_OPENCHANNEL, 0, NULL },
1267 { CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result },
1268 { CHANNELMSG_CLOSECHANNEL, 0, NULL },
1269 { CHANNELMSG_GPADL_HEADER, 0, NULL },
1270 { CHANNELMSG_GPADL_BODY, 0, NULL },
1271 { CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created },
1272 { CHANNELMSG_GPADL_TEARDOWN, 0, NULL },
1273 { CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown },
1274 { CHANNELMSG_RELID_RELEASED, 0, NULL },
1275 { CHANNELMSG_INITIATE_CONTACT, 0, NULL },
1276 { CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response },
1277 { CHANNELMSG_UNLOAD, 0, NULL },
1278 { CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response },
1279 { CHANNELMSG_18, 0, NULL },
1280 { CHANNELMSG_19, 0, NULL },
1281 { CHANNELMSG_20, 0, NULL },
1282 { CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL },
1286 * vmbus_onmessage - Handler for channel protocol messages.
1288 * This is invoked in the vmbus worker thread context.
1290 void vmbus_onmessage(void *context)
1292 struct hv_message *msg = context;
1293 struct vmbus_channel_message_header *hdr;
1296 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1297 size = msg->header.payload_size;
1299 trace_vmbus_on_message(hdr);
1301 if (hdr->msgtype >= CHANNELMSG_COUNT) {
1302 pr_err("Received invalid channel message type %d size %d\n",
1303 hdr->msgtype, size);
1304 print_hex_dump_bytes("", DUMP_PREFIX_NONE,
1305 (unsigned char *)msg->u.payload, size);
1309 if (channel_message_table[hdr->msgtype].message_handler)
1310 channel_message_table[hdr->msgtype].message_handler(hdr);
1312 pr_err("Unhandled channel message type %d\n", hdr->msgtype);
1316 * vmbus_request_offers - Send a request to get all our pending offers.
1318 int vmbus_request_offers(void)
1320 struct vmbus_channel_message_header *msg;
1321 struct vmbus_channel_msginfo *msginfo;
1324 msginfo = kmalloc(sizeof(*msginfo) +
1325 sizeof(struct vmbus_channel_message_header),
1330 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1332 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1334 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1337 trace_vmbus_request_offers(ret);
1340 pr_err("Unable to request offers - %d\n", ret);
1351 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1353 struct list_head *cur, *tmp;
1354 struct vmbus_channel *cur_channel;
1356 if (primary_channel->sc_creation_callback == NULL)
1359 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1360 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1362 primary_channel->sc_creation_callback(cur_channel);
1366 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1367 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1369 primary_channel->sc_creation_callback = sc_cr_cb;
1371 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1373 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1377 ret = !list_empty(&primary->sc_list);
1381 * Invoke the callback on sub-channel creation.
1382 * This will present a uniform interface to the
1385 invoke_sc_cb(primary);
1390 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1392 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1393 void (*chn_rescind_cb)(struct vmbus_channel *))
1395 channel->chn_rescind_callback = chn_rescind_cb;
1397 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);