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140 * The rescinded channel may be blocked waiting for a response from the host;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);
411 * Upon suspend, an in-use hv_sock channel is marked as "rescinded" and
412 * the relid is invalidated; after hibernation, when the user-space app
413 * destroys the channel, the relid is INVALID_RELID, and in this case
414 * it's unnecessary and unsafe to release the old relid, since the same
415 * relid can refer to a completely different channel now.
417 if (channel->offermsg.child_relid != INVALID_RELID)
418 vmbus_release_relid(channel->offermsg.child_relid);
420 free_channel(channel);
423 void vmbus_free_channels(void)
425 struct vmbus_channel *channel, *tmp;
427 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
429 /* hv_process_channel_removal() needs this */
430 channel->rescind = true;
432 vmbus_device_unregister(channel->device_obj);
436 /* Note: the function can run concurrently for primary/sub channels. */
437 static void vmbus_add_channel_work(struct work_struct *work)
439 struct vmbus_channel *newchannel =
440 container_of(work, struct vmbus_channel, add_channel_work);
441 struct vmbus_channel *primary_channel = newchannel->primary_channel;
446 dev_type = hv_get_dev_type(newchannel);
448 init_vp_index(newchannel, dev_type);
450 if (newchannel->target_cpu != get_cpu()) {
452 smp_call_function_single(newchannel->target_cpu,
456 percpu_channel_enq(newchannel);
461 * This state is used to indicate a successful open
462 * so that when we do close the channel normally, we
463 * can cleanup properly.
465 newchannel->state = CHANNEL_OPEN_STATE;
467 if (primary_channel != NULL) {
468 /* newchannel is a sub-channel. */
469 struct hv_device *dev = primary_channel->device_obj;
471 if (vmbus_add_channel_kobj(dev, newchannel))
474 if (primary_channel->sc_creation_callback != NULL)
475 primary_channel->sc_creation_callback(newchannel);
477 newchannel->probe_done = true;
482 * Start the process of binding the primary channel to the driver
484 newchannel->device_obj = vmbus_device_create(
485 &newchannel->offermsg.offer.if_type,
486 &newchannel->offermsg.offer.if_instance,
488 if (!newchannel->device_obj)
491 newchannel->device_obj->device_id = dev_type;
493 * Add the new device to the bus. This will kick off device-driver
494 * binding which eventually invokes the device driver's AddDevice()
497 ret = vmbus_device_register(newchannel->device_obj);
500 pr_err("unable to add child device object (relid %d)\n",
501 newchannel->offermsg.child_relid);
502 kfree(newchannel->device_obj);
506 newchannel->probe_done = true;
510 mutex_lock(&vmbus_connection.channel_mutex);
513 * We need to set the flag, otherwise
514 * vmbus_onoffer_rescind() can be blocked.
516 newchannel->probe_done = true;
518 if (primary_channel == NULL) {
519 list_del(&newchannel->listentry);
521 spin_lock_irqsave(&primary_channel->lock, flags);
522 list_del(&newchannel->sc_list);
523 spin_unlock_irqrestore(&primary_channel->lock, flags);
526 mutex_unlock(&vmbus_connection.channel_mutex);
528 if (newchannel->target_cpu != get_cpu()) {
530 smp_call_function_single(newchannel->target_cpu,
534 percpu_channel_deq(newchannel);
538 vmbus_release_relid(newchannel->offermsg.child_relid);
540 free_channel(newchannel);
544 * vmbus_process_offer - Process the offer by creating a channel/device
545 * associated with this offer
547 static void vmbus_process_offer(struct vmbus_channel *newchannel)
549 struct vmbus_channel *channel;
550 struct workqueue_struct *wq;
554 mutex_lock(&vmbus_connection.channel_mutex);
556 /* Remember the channels that should be cleaned up upon suspend. */
557 if (is_hvsock_channel(newchannel) || is_sub_channel(newchannel))
558 atomic_inc(&vmbus_connection.nr_chan_close_on_suspend);
561 * Now that we have acquired the channel_mutex,
562 * we can release the potentially racing rescind thread.
564 atomic_dec(&vmbus_connection.offer_in_progress);
566 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
567 if (guid_equal(&channel->offermsg.offer.if_type,
568 &newchannel->offermsg.offer.if_type) &&
569 guid_equal(&channel->offermsg.offer.if_instance,
570 &newchannel->offermsg.offer.if_instance)) {
577 list_add_tail(&newchannel->listentry,
578 &vmbus_connection.chn_list);
581 * Check to see if this is a valid sub-channel.
583 if (newchannel->offermsg.offer.sub_channel_index == 0) {
584 mutex_unlock(&vmbus_connection.channel_mutex);
586 * Don't call free_channel(), because newchannel->kobj
587 * is not initialized yet.
594 * Process the sub-channel.
596 newchannel->primary_channel = channel;
597 spin_lock_irqsave(&channel->lock, flags);
598 list_add_tail(&newchannel->sc_list, &channel->sc_list);
599 spin_unlock_irqrestore(&channel->lock, flags);
602 mutex_unlock(&vmbus_connection.channel_mutex);
605 * vmbus_process_offer() mustn't call channel->sc_creation_callback()
606 * directly for sub-channels, because sc_creation_callback() ->
607 * vmbus_open() may never get the host's response to the
608 * OPEN_CHANNEL message (the host may rescind a channel at any time,
609 * e.g. in the case of hot removing a NIC), and vmbus_onoffer_rescind()
610 * may not wake up the vmbus_open() as it's blocked due to a non-zero
611 * vmbus_connection.offer_in_progress, and finally we have a deadlock.
613 * The above is also true for primary channels, if the related device
614 * drivers use sync probing mode by default.
616 * And, usually the handling of primary channels and sub-channels can
617 * depend on each other, so we should offload them to different
618 * workqueues to avoid possible deadlock, e.g. in sync-probing mode,
619 * NIC1's netvsc_subchan_work() can race with NIC2's netvsc_probe() ->
620 * rtnl_lock(), and causes deadlock: the former gets the rtnl_lock
621 * and waits for all the sub-channels to appear, but the latter
622 * can't get the rtnl_lock and this blocks the handling of
625 INIT_WORK(&newchannel->add_channel_work, vmbus_add_channel_work);
626 wq = fnew ? vmbus_connection.handle_primary_chan_wq :
627 vmbus_connection.handle_sub_chan_wq;
628 queue_work(wq, &newchannel->add_channel_work);
632 * We use this state to statically distribute the channel interrupt load.
634 static int next_numa_node_id;
636 * init_vp_index() accesses global variables like next_numa_node_id, and
637 * it can run concurrently for primary channels and sub-channels: see
638 * vmbus_process_offer(), so we need the lock to protect the global
641 static DEFINE_SPINLOCK(bind_channel_to_cpu_lock);
644 * Starting with Win8, we can statically distribute the incoming
645 * channel interrupt load by binding a channel to VCPU.
646 * We distribute the interrupt loads to one or more NUMA nodes based on
647 * the channel's affinity_policy.
649 * For pre-win8 hosts or non-performance critical channels we assign the
650 * first CPU in the first NUMA node.
652 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
655 bool perf_chn = vmbus_devs[dev_type].perf_device;
656 struct vmbus_channel *primary = channel->primary_channel;
658 cpumask_var_t available_mask;
659 struct cpumask *alloced_mask;
661 if ((vmbus_proto_version == VERSION_WS2008) ||
662 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn) ||
663 !alloc_cpumask_var(&available_mask, GFP_KERNEL)) {
665 * Prior to win8, all channel interrupts are
666 * delivered on cpu 0.
667 * Also if the channel is not a performance critical
668 * channel, bind it to cpu 0.
669 * In case alloc_cpumask_var() fails, bind it to cpu 0.
671 channel->numa_node = 0;
672 channel->target_cpu = 0;
673 channel->target_vp = hv_cpu_number_to_vp_number(0);
677 spin_lock(&bind_channel_to_cpu_lock);
680 * Based on the channel affinity policy, we will assign the NUMA
684 if ((channel->affinity_policy == HV_BALANCED) || (!primary)) {
686 next_node = next_numa_node_id++;
687 if (next_node == nr_node_ids) {
688 next_node = next_numa_node_id = 0;
691 if (cpumask_empty(cpumask_of_node(next_node)))
695 channel->numa_node = next_node;
698 alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
700 if (cpumask_weight(alloced_mask) ==
701 cpumask_weight(cpumask_of_node(primary->numa_node))) {
703 * We have cycled through all the CPUs in the node;
704 * reset the alloced map.
706 cpumask_clear(alloced_mask);
709 cpumask_xor(available_mask, alloced_mask,
710 cpumask_of_node(primary->numa_node));
714 if (primary->affinity_policy == HV_LOCALIZED) {
716 * Normally Hyper-V host doesn't create more subchannels
717 * than there are VCPUs on the node but it is possible when not
718 * all present VCPUs on the node are initialized by guest.
719 * Clear the alloced_cpus_in_node to start over.
721 if (cpumask_equal(&primary->alloced_cpus_in_node,
722 cpumask_of_node(primary->numa_node)))
723 cpumask_clear(&primary->alloced_cpus_in_node);
727 cur_cpu = cpumask_next(cur_cpu, available_mask);
728 if (cur_cpu >= nr_cpu_ids) {
730 cpumask_copy(available_mask,
731 cpumask_of_node(primary->numa_node));
735 if (primary->affinity_policy == HV_LOCALIZED) {
737 * NOTE: in the case of sub-channel, we clear the
738 * sub-channel related bit(s) in
739 * primary->alloced_cpus_in_node in
740 * hv_process_channel_removal(), so when we
741 * reload drivers like hv_netvsc in SMP guest, here
742 * we're able to re-allocate
743 * bit from primary->alloced_cpus_in_node.
745 if (!cpumask_test_cpu(cur_cpu,
746 &primary->alloced_cpus_in_node)) {
747 cpumask_set_cpu(cur_cpu,
748 &primary->alloced_cpus_in_node);
749 cpumask_set_cpu(cur_cpu, alloced_mask);
753 cpumask_set_cpu(cur_cpu, alloced_mask);
758 channel->target_cpu = cur_cpu;
759 channel->target_vp = hv_cpu_number_to_vp_number(cur_cpu);
761 spin_unlock(&bind_channel_to_cpu_lock);
763 free_cpumask_var(available_mask);
766 #define UNLOAD_DELAY_UNIT_MS 10 /* 10 milliseconds */
767 #define UNLOAD_WAIT_MS (100*1000) /* 100 seconds */
768 #define UNLOAD_WAIT_LOOPS (UNLOAD_WAIT_MS/UNLOAD_DELAY_UNIT_MS)
769 #define UNLOAD_MSG_MS (5*1000) /* Every 5 seconds */
770 #define UNLOAD_MSG_LOOPS (UNLOAD_MSG_MS/UNLOAD_DELAY_UNIT_MS)
772 static void vmbus_wait_for_unload(void)
776 struct hv_message *msg;
777 struct vmbus_channel_message_header *hdr;
781 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
782 * used for initial contact or to CPU0 depending on host version. When
783 * we're crashing on a different CPU let's hope that IRQ handler on
784 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
785 * functional and vmbus_unload_response() will complete
786 * vmbus_connection.unload_event. If not, the last thing we can do is
787 * read message pages for all CPUs directly.
789 * Wait up to 100 seconds since an Azure host must writeback any dirty
790 * data in its disk cache before the VMbus UNLOAD request will
791 * complete. This flushing has been empirically observed to take up
792 * to 50 seconds in cases with a lot of dirty data, so allow additional
793 * leeway and for inaccuracies in mdelay(). But eventually time out so
794 * that the panic path can't get hung forever in case the response
795 * message isn't seen.
797 for (i = 1; i <= UNLOAD_WAIT_LOOPS; i++) {
798 if (completion_done(&vmbus_connection.unload_event))
801 for_each_online_cpu(cpu) {
802 struct hv_per_cpu_context *hv_cpu
803 = per_cpu_ptr(hv_context.cpu_context, cpu);
805 page_addr = hv_cpu->synic_message_page;
806 msg = (struct hv_message *)page_addr
807 + VMBUS_MESSAGE_SINT;
809 message_type = READ_ONCE(msg->header.message_type);
810 if (message_type == HVMSG_NONE)
813 hdr = (struct vmbus_channel_message_header *)
816 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
817 complete(&vmbus_connection.unload_event);
819 vmbus_signal_eom(msg, message_type);
823 * Give a notice periodically so someone watching the
824 * serial output won't think it is completely hung.
826 if (!(i % UNLOAD_MSG_LOOPS))
827 pr_notice("Waiting for VMBus UNLOAD to complete\n");
829 mdelay(UNLOAD_DELAY_UNIT_MS);
831 pr_err("Continuing even though VMBus UNLOAD did not complete\n");
835 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
836 * maybe-pending messages on all CPUs to be able to receive new
837 * messages after we reconnect.
839 for_each_online_cpu(cpu) {
840 struct hv_per_cpu_context *hv_cpu
841 = per_cpu_ptr(hv_context.cpu_context, cpu);
843 page_addr = hv_cpu->synic_message_page;
844 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
845 msg->header.message_type = HVMSG_NONE;
850 * vmbus_unload_response - Handler for the unload response.
852 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
855 * This is a global event; just wakeup the waiting thread.
856 * Once we successfully unload, we can cleanup the monitor state.
858 complete(&vmbus_connection.unload_event);
861 void vmbus_initiate_unload(bool crash)
863 struct vmbus_channel_message_header hdr;
865 if (xchg(&vmbus_connection.conn_state, DISCONNECTED) == DISCONNECTED)
868 /* Pre-Win2012R2 hosts don't support reconnect */
869 if (vmbus_proto_version < VERSION_WIN8_1)
872 init_completion(&vmbus_connection.unload_event);
873 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
874 hdr.msgtype = CHANNELMSG_UNLOAD;
875 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
879 * vmbus_initiate_unload() is also called on crash and the crash can be
880 * happening in an interrupt context, where scheduling is impossible.
883 wait_for_completion(&vmbus_connection.unload_event);
885 vmbus_wait_for_unload();
888 static void check_ready_for_resume_event(void)
891 * If all the old primary channels have been fixed up, then it's safe
894 if (atomic_dec_and_test(&vmbus_connection.nr_chan_fixup_on_resume))
895 complete(&vmbus_connection.ready_for_resume_event);
898 static void vmbus_setup_channel_state(struct vmbus_channel *channel,
899 struct vmbus_channel_offer_channel *offer)
902 * Setup state for signalling the host.
904 channel->sig_event = VMBUS_EVENT_CONNECTION_ID;
906 if (vmbus_proto_version != VERSION_WS2008) {
907 channel->is_dedicated_interrupt =
908 (offer->is_dedicated_interrupt != 0);
909 channel->sig_event = offer->connection_id;
912 memcpy(&channel->offermsg, offer,
913 sizeof(struct vmbus_channel_offer_channel));
914 channel->monitor_grp = (u8)offer->monitorid / 32;
915 channel->monitor_bit = (u8)offer->monitorid % 32;
919 * find_primary_channel_by_offer - Get the channel object given the new offer.
920 * This is only used in the resume path of hibernation.
922 static struct vmbus_channel *
923 find_primary_channel_by_offer(const struct vmbus_channel_offer_channel *offer)
925 struct vmbus_channel *channel = NULL, *iter;
926 const guid_t *inst1, *inst2;
928 /* Ignore sub-channel offers. */
929 if (offer->offer.sub_channel_index != 0)
932 mutex_lock(&vmbus_connection.channel_mutex);
934 list_for_each_entry(iter, &vmbus_connection.chn_list, listentry) {
935 inst1 = &iter->offermsg.offer.if_instance;
936 inst2 = &offer->offer.if_instance;
938 if (guid_equal(inst1, inst2)) {
944 mutex_unlock(&vmbus_connection.channel_mutex);
950 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
953 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
955 struct vmbus_channel_offer_channel *offer;
956 struct vmbus_channel *oldchannel, *newchannel;
959 offer = (struct vmbus_channel_offer_channel *)hdr;
961 trace_vmbus_onoffer(offer);
963 oldchannel = find_primary_channel_by_offer(offer);
965 if (oldchannel != NULL) {
966 atomic_dec(&vmbus_connection.offer_in_progress);
969 * We're resuming from hibernation: all the sub-channel and
970 * hv_sock channels we had before the hibernation should have
971 * been cleaned up, and now we must be seeing a re-offered
972 * primary channel that we had before the hibernation.
975 WARN_ON(oldchannel->offermsg.child_relid != INVALID_RELID);
976 /* Fix up the relid. */
977 oldchannel->offermsg.child_relid = offer->child_relid;
979 offer_sz = sizeof(*offer);
980 if (memcmp(offer, &oldchannel->offermsg, offer_sz) == 0) {
981 check_ready_for_resume_event();
986 * This is not an error, since the host can also change the
987 * other field(s) of the offer, e.g. on WS RS5 (Build 17763),
988 * the offer->connection_id of the Mellanox VF vmbus device
989 * can change when the host reoffers the device upon resume.
991 pr_debug("vmbus offer changed: relid=%d\n",
994 print_hex_dump_debug("Old vmbus offer: ", DUMP_PREFIX_OFFSET,
995 16, 4, &oldchannel->offermsg, offer_sz,
997 print_hex_dump_debug("New vmbus offer: ", DUMP_PREFIX_OFFSET,
998 16, 4, offer, offer_sz, false);
1000 /* Fix up the old channel. */
1001 vmbus_setup_channel_state(oldchannel, offer);
1003 check_ready_for_resume_event();
1008 /* Allocate the channel object and save this offer. */
1009 newchannel = alloc_channel();
1011 vmbus_release_relid(offer->child_relid);
1012 atomic_dec(&vmbus_connection.offer_in_progress);
1013 pr_err("Unable to allocate channel object\n");
1017 vmbus_setup_channel_state(newchannel, offer);
1019 vmbus_process_offer(newchannel);
1022 static void check_ready_for_suspend_event(void)
1025 * If all the sub-channels or hv_sock channels have been cleaned up,
1026 * then it's safe to suspend.
1028 if (atomic_dec_and_test(&vmbus_connection.nr_chan_close_on_suspend))
1029 complete(&vmbus_connection.ready_for_suspend_event);
1033 * vmbus_onoffer_rescind - Rescind offer handler.
1035 * We queue a work item to process this offer synchronously
1037 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
1039 struct vmbus_channel_rescind_offer *rescind;
1040 struct vmbus_channel *channel;
1042 bool clean_up_chan_for_suspend;
1044 rescind = (struct vmbus_channel_rescind_offer *)hdr;
1046 trace_vmbus_onoffer_rescind(rescind);
1049 * The offer msg and the corresponding rescind msg
1050 * from the host are guranteed to be ordered -
1051 * offer comes in first and then the rescind.
1052 * Since we process these events in work elements,
1053 * and with preemption, we may end up processing
1054 * the events out of order. Given that we handle these
1055 * work elements on the same CPU, this is possible only
1056 * in the case of preemption. In any case wait here
1057 * until the offer processing has moved beyond the
1058 * point where the channel is discoverable.
1061 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
1063 * We wait here until any channel offer is currently
1069 mutex_lock(&vmbus_connection.channel_mutex);
1070 channel = relid2channel(rescind->child_relid);
1071 mutex_unlock(&vmbus_connection.channel_mutex);
1073 if (channel == NULL) {
1075 * We failed in processing the offer message;
1076 * we would have cleaned up the relid in that
1082 clean_up_chan_for_suspend = is_hvsock_channel(channel) ||
1083 is_sub_channel(channel);
1085 * Before setting channel->rescind in vmbus_rescind_cleanup(), we
1086 * should make sure the channel callback is not running any more.
1088 vmbus_reset_channel_cb(channel);
1091 * Now wait for offer handling to complete.
1093 vmbus_rescind_cleanup(channel);
1094 while (READ_ONCE(channel->probe_done) == false) {
1096 * We wait here until any channel offer is currently
1103 * At this point, the rescind handling can proceed safely.
1106 if (channel->device_obj) {
1107 if (channel->chn_rescind_callback) {
1108 channel->chn_rescind_callback(channel);
1110 if (clean_up_chan_for_suspend)
1111 check_ready_for_suspend_event();
1116 * We will have to unregister this device from the
1119 dev = get_device(&channel->device_obj->device);
1121 vmbus_device_unregister(channel->device_obj);
1124 } else if (channel->primary_channel != NULL) {
1126 * Sub-channel is being rescinded. Following is the channel
1127 * close sequence when initiated from the driveri (refer to
1128 * vmbus_close() for details):
1129 * 1. Close all sub-channels first
1130 * 2. Then close the primary channel.
1132 mutex_lock(&vmbus_connection.channel_mutex);
1133 if (channel->state == CHANNEL_OPEN_STATE) {
1135 * The channel is currently not open;
1136 * it is safe for us to cleanup the channel.
1138 hv_process_channel_removal(channel);
1140 complete(&channel->rescind_event);
1142 mutex_unlock(&vmbus_connection.channel_mutex);
1145 /* The "channel" may have been freed. Do not access it any longer. */
1147 if (clean_up_chan_for_suspend)
1148 check_ready_for_suspend_event();
1151 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
1153 BUG_ON(!is_hvsock_channel(channel));
1155 /* We always get a rescind msg when a connection is closed. */
1156 while (!READ_ONCE(channel->probe_done) || !READ_ONCE(channel->rescind))
1159 vmbus_device_unregister(channel->device_obj);
1161 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
1165 * vmbus_onoffers_delivered -
1166 * This is invoked when all offers have been delivered.
1168 * Nothing to do here.
1170 static void vmbus_onoffers_delivered(
1171 struct vmbus_channel_message_header *hdr)
1176 * vmbus_onopen_result - Open result handler.
1178 * This is invoked when we received a response to our channel open request.
1179 * Find the matching request, copy the response and signal the requesting
1182 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
1184 struct vmbus_channel_open_result *result;
1185 struct vmbus_channel_msginfo *msginfo;
1186 struct vmbus_channel_message_header *requestheader;
1187 struct vmbus_channel_open_channel *openmsg;
1188 unsigned long flags;
1190 result = (struct vmbus_channel_open_result *)hdr;
1192 trace_vmbus_onopen_result(result);
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_OPENCHANNEL) {
1206 (struct vmbus_channel_open_channel *)msginfo->msg;
1207 if (openmsg->child_relid == result->child_relid &&
1208 openmsg->openid == result->openid) {
1209 memcpy(&msginfo->response.open_result,
1212 struct vmbus_channel_open_result));
1213 complete(&msginfo->waitevent);
1218 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1222 * vmbus_ongpadl_created - GPADL created handler.
1224 * This is invoked when we received a response to our gpadl create request.
1225 * Find the matching request, copy the response and signal the requesting
1228 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
1230 struct vmbus_channel_gpadl_created *gpadlcreated;
1231 struct vmbus_channel_msginfo *msginfo;
1232 struct vmbus_channel_message_header *requestheader;
1233 struct vmbus_channel_gpadl_header *gpadlheader;
1234 unsigned long flags;
1236 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
1238 trace_vmbus_ongpadl_created(gpadlcreated);
1241 * Find the establish msg, copy the result and signal/unblock the wait
1244 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1246 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1249 (struct vmbus_channel_message_header *)msginfo->msg;
1251 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
1253 (struct vmbus_channel_gpadl_header *)requestheader;
1255 if ((gpadlcreated->child_relid ==
1256 gpadlheader->child_relid) &&
1257 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1258 memcpy(&msginfo->response.gpadl_created,
1261 struct vmbus_channel_gpadl_created));
1262 complete(&msginfo->waitevent);
1267 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1271 * vmbus_ongpadl_torndown - GPADL torndown handler.
1273 * This is invoked when we received a response to our gpadl teardown request.
1274 * Find the matching request, copy the response and signal the requesting
1277 static void vmbus_ongpadl_torndown(
1278 struct vmbus_channel_message_header *hdr)
1280 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1281 struct vmbus_channel_msginfo *msginfo;
1282 struct vmbus_channel_message_header *requestheader;
1283 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1284 unsigned long flags;
1286 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1288 trace_vmbus_ongpadl_torndown(gpadl_torndown);
1291 * Find the open msg, copy the result and signal/unblock the wait event
1293 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1295 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1298 (struct vmbus_channel_message_header *)msginfo->msg;
1300 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1302 (struct vmbus_channel_gpadl_teardown *)requestheader;
1304 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1305 memcpy(&msginfo->response.gpadl_torndown,
1308 struct vmbus_channel_gpadl_torndown));
1309 complete(&msginfo->waitevent);
1314 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1318 * vmbus_onversion_response - Version response handler
1320 * This is invoked when we received a response to our initiate contact request.
1321 * Find the matching request, copy the response and signal the requesting
1324 static void vmbus_onversion_response(
1325 struct vmbus_channel_message_header *hdr)
1327 struct vmbus_channel_msginfo *msginfo;
1328 struct vmbus_channel_message_header *requestheader;
1329 struct vmbus_channel_version_response *version_response;
1330 unsigned long flags;
1332 version_response = (struct vmbus_channel_version_response *)hdr;
1334 trace_vmbus_onversion_response(version_response);
1336 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1338 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1341 (struct vmbus_channel_message_header *)msginfo->msg;
1343 if (requestheader->msgtype ==
1344 CHANNELMSG_INITIATE_CONTACT) {
1345 memcpy(&msginfo->response.version_response,
1347 sizeof(struct vmbus_channel_version_response));
1348 complete(&msginfo->waitevent);
1351 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1354 /* Channel message dispatch table */
1355 const struct vmbus_channel_message_table_entry
1356 channel_message_table[CHANNELMSG_COUNT] = {
1357 { CHANNELMSG_INVALID, 0, NULL },
1358 { CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer },
1359 { CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind },
1360 { CHANNELMSG_REQUESTOFFERS, 0, NULL },
1361 { CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered },
1362 { CHANNELMSG_OPENCHANNEL, 0, NULL },
1363 { CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result },
1364 { CHANNELMSG_CLOSECHANNEL, 0, NULL },
1365 { CHANNELMSG_GPADL_HEADER, 0, NULL },
1366 { CHANNELMSG_GPADL_BODY, 0, NULL },
1367 { CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created },
1368 { CHANNELMSG_GPADL_TEARDOWN, 0, NULL },
1369 { CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown },
1370 { CHANNELMSG_RELID_RELEASED, 0, NULL },
1371 { CHANNELMSG_INITIATE_CONTACT, 0, NULL },
1372 { CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response },
1373 { CHANNELMSG_UNLOAD, 0, NULL },
1374 { CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response },
1375 { CHANNELMSG_18, 0, NULL },
1376 { CHANNELMSG_19, 0, NULL },
1377 { CHANNELMSG_20, 0, NULL },
1378 { CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL },
1379 { CHANNELMSG_22, 0, NULL },
1380 { CHANNELMSG_TL_CONNECT_RESULT, 0, NULL },
1384 * vmbus_onmessage - Handler for channel protocol messages.
1386 * This is invoked in the vmbus worker thread context.
1388 void vmbus_onmessage(void *context)
1390 struct hv_message *msg = context;
1391 struct vmbus_channel_message_header *hdr;
1393 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1395 trace_vmbus_on_message(hdr);
1398 * vmbus_on_msg_dpc() makes sure the hdr->msgtype here can not go
1399 * out of bound and the message_handler pointer can not be NULL.
1401 channel_message_table[hdr->msgtype].message_handler(hdr);
1405 * vmbus_request_offers - Send a request to get all our pending offers.
1407 int vmbus_request_offers(void)
1409 struct vmbus_channel_message_header *msg;
1410 struct vmbus_channel_msginfo *msginfo;
1413 msginfo = kmalloc(sizeof(*msginfo) +
1414 sizeof(struct vmbus_channel_message_header),
1419 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1421 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1423 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1426 trace_vmbus_request_offers(ret);
1429 pr_err("Unable to request offers - %d\n", ret);
1440 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1442 struct list_head *cur, *tmp;
1443 struct vmbus_channel *cur_channel;
1445 if (primary_channel->sc_creation_callback == NULL)
1448 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1449 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1451 primary_channel->sc_creation_callback(cur_channel);
1455 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1456 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1458 primary_channel->sc_creation_callback = sc_cr_cb;
1460 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1462 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1466 ret = !list_empty(&primary->sc_list);
1470 * Invoke the callback on sub-channel creation.
1471 * This will present a uniform interface to the
1474 invoke_sc_cb(primary);
1479 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1481 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1482 void (*chn_rescind_cb)(struct vmbus_channel *))
1484 channel->chn_rescind_callback = chn_rescind_cb;
1486 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);