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/cpu.h>
22 #include <linux/hyperv.h>
23 #include <asm/mshyperv.h>
25 #include "hyperv_vmbus.h"
27 static void init_vp_index(struct vmbus_channel *channel);
29 const struct vmbus_device vmbus_devs[] = {
37 { .dev_type = HV_SCSI,
61 { .dev_type = HV_PCIE,
66 /* Synthetic Frame Buffer */
72 /* Synthetic Keyboard */
79 { .dev_type = HV_MOUSE,
103 { .dev_type = HV_SHUTDOWN,
105 .perf_device = false,
109 { .dev_type = HV_FCOPY,
111 .perf_device = false,
115 { .dev_type = HV_BACKUP,
117 .perf_device = false,
123 .perf_device = false,
127 { .dev_type = HV_UNKNOWN,
128 .perf_device = false,
132 static const struct {
134 } vmbus_unsupported_devs[] = {
141 * The rescinded channel may be blocked waiting for a response from the host;
144 static void vmbus_rescind_cleanup(struct vmbus_channel *channel)
146 struct vmbus_channel_msginfo *msginfo;
150 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
151 channel->rescind = true;
152 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
155 if (msginfo->waiting_channel == channel) {
156 complete(&msginfo->waitevent);
160 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
163 static bool is_unsupported_vmbus_devs(const guid_t *guid)
167 for (i = 0; i < ARRAY_SIZE(vmbus_unsupported_devs); i++)
168 if (guid_equal(guid, &vmbus_unsupported_devs[i].guid))
173 static u16 hv_get_dev_type(const struct vmbus_channel *channel)
175 const guid_t *guid = &channel->offermsg.offer.if_type;
178 if (is_hvsock_channel(channel) || is_unsupported_vmbus_devs(guid))
181 for (i = HV_IDE; i < HV_UNKNOWN; i++) {
182 if (guid_equal(guid, &vmbus_devs[i].guid))
185 pr_info("Unknown GUID: %pUl\n", guid);
190 * vmbus_prep_negotiate_resp() - Create default response for Negotiate message
191 * @icmsghdrp: Pointer to msg header structure
192 * @buf: Raw buffer channel data
193 * @fw_version: The framework versions we can support.
194 * @fw_vercnt: The size of @fw_version.
195 * @srv_version: The service versions we can support.
196 * @srv_vercnt: The size of @srv_version.
197 * @nego_fw_version: The selected framework version.
198 * @nego_srv_version: The selected service version.
200 * Note: Versions are given in decreasing order.
202 * Set up and fill in default negotiate response message.
203 * Mainly used by Hyper-V drivers.
205 bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp,
206 u8 *buf, const int *fw_version, int fw_vercnt,
207 const int *srv_version, int srv_vercnt,
208 int *nego_fw_version, int *nego_srv_version)
210 int icframe_major, icframe_minor;
211 int icmsg_major, icmsg_minor;
212 int fw_major, fw_minor;
213 int srv_major, srv_minor;
215 bool found_match = false;
216 struct icmsg_negotiate *negop;
218 icmsghdrp->icmsgsize = 0x10;
219 negop = (struct icmsg_negotiate *)&buf[
220 sizeof(struct vmbuspipe_hdr) +
221 sizeof(struct icmsg_hdr)];
223 icframe_major = negop->icframe_vercnt;
226 icmsg_major = negop->icmsg_vercnt;
230 * Select the framework version number we will
234 for (i = 0; i < fw_vercnt; i++) {
235 fw_major = (fw_version[i] >> 16);
236 fw_minor = (fw_version[i] & 0xFFFF);
238 for (j = 0; j < negop->icframe_vercnt; j++) {
239 if ((negop->icversion_data[j].major == fw_major) &&
240 (negop->icversion_data[j].minor == fw_minor)) {
241 icframe_major = negop->icversion_data[j].major;
242 icframe_minor = negop->icversion_data[j].minor;
257 for (i = 0; i < srv_vercnt; i++) {
258 srv_major = (srv_version[i] >> 16);
259 srv_minor = (srv_version[i] & 0xFFFF);
261 for (j = negop->icframe_vercnt;
262 (j < negop->icframe_vercnt + negop->icmsg_vercnt);
265 if ((negop->icversion_data[j].major == srv_major) &&
266 (negop->icversion_data[j].minor == srv_minor)) {
268 icmsg_major = negop->icversion_data[j].major;
269 icmsg_minor = negop->icversion_data[j].minor;
280 * Respond with the framework and service
281 * version numbers we can support.
286 negop->icframe_vercnt = 0;
287 negop->icmsg_vercnt = 0;
289 negop->icframe_vercnt = 1;
290 negop->icmsg_vercnt = 1;
294 *nego_fw_version = (icframe_major << 16) | icframe_minor;
296 if (nego_srv_version)
297 *nego_srv_version = (icmsg_major << 16) | icmsg_minor;
299 negop->icversion_data[0].major = icframe_major;
300 negop->icversion_data[0].minor = icframe_minor;
301 negop->icversion_data[1].major = icmsg_major;
302 negop->icversion_data[1].minor = icmsg_minor;
306 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp);
309 * alloc_channel - Allocate and initialize a vmbus channel object
311 static struct vmbus_channel *alloc_channel(void)
313 struct vmbus_channel *channel;
315 channel = kzalloc(sizeof(*channel), GFP_ATOMIC);
319 spin_lock_init(&channel->sched_lock);
320 spin_lock_init(&channel->lock);
321 init_completion(&channel->rescind_event);
323 INIT_LIST_HEAD(&channel->sc_list);
325 tasklet_init(&channel->callback_event,
326 vmbus_on_event, (unsigned long)channel);
328 hv_ringbuffer_pre_init(channel);
334 * free_channel - Release the resources used by the vmbus channel object
336 static void free_channel(struct vmbus_channel *channel)
338 tasklet_kill(&channel->callback_event);
339 vmbus_remove_channel_attr_group(channel);
341 kobject_put(&channel->kobj);
344 void vmbus_channel_map_relid(struct vmbus_channel *channel)
346 if (WARN_ON(channel->offermsg.child_relid >= MAX_CHANNEL_RELIDS))
349 * The mapping of the channel's relid is visible from the CPUs that
350 * execute vmbus_chan_sched() by the time that vmbus_chan_sched() will
353 * (a) In the "normal (i.e., not resuming from hibernation)" path,
354 * the full barrier in smp_store_mb() guarantees that the store
355 * is propagated to all CPUs before the add_channel_work work
356 * is queued. In turn, add_channel_work is queued before the
357 * channel's ring buffer is allocated/initialized and the
358 * OPENCHANNEL message for the channel is sent in vmbus_open().
359 * Hyper-V won't start sending the interrupts for the channel
360 * before the OPENCHANNEL message is acked. The memory barrier
361 * in vmbus_chan_sched() -> sync_test_and_clear_bit() ensures
362 * that vmbus_chan_sched() must find the channel's relid in
363 * recv_int_page before retrieving the channel pointer from the
366 * (b) In the "resuming from hibernation" path, the smp_store_mb()
367 * guarantees that the store is propagated to all CPUs before
368 * the VMBus connection is marked as ready for the resume event
369 * (cf. check_ready_for_resume_event()). The interrupt handler
370 * of the VMBus driver and vmbus_chan_sched() can not run before
371 * vmbus_bus_resume() has completed execution (cf. resume_noirq).
374 vmbus_connection.channels[channel->offermsg.child_relid],
378 void vmbus_channel_unmap_relid(struct vmbus_channel *channel)
380 if (WARN_ON(channel->offermsg.child_relid >= MAX_CHANNEL_RELIDS))
383 vmbus_connection.channels[channel->offermsg.child_relid],
387 static void vmbus_release_relid(u32 relid)
389 struct vmbus_channel_relid_released msg;
392 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
393 msg.child_relid = relid;
394 msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
395 ret = vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released),
398 trace_vmbus_release_relid(&msg, ret);
401 void hv_process_channel_removal(struct vmbus_channel *channel)
405 lockdep_assert_held(&vmbus_connection.channel_mutex);
406 BUG_ON(!channel->rescind);
409 * hv_process_channel_removal() could find INVALID_RELID only for
410 * hv_sock channels. See the inline comments in vmbus_onoffer().
412 WARN_ON(channel->offermsg.child_relid == INVALID_RELID &&
413 !is_hvsock_channel(channel));
416 * Upon suspend, an in-use hv_sock channel is removed from the array of
417 * channels and the relid is invalidated. After hibernation, when the
418 * user-space appplication destroys the channel, it's unnecessary and
419 * unsafe to remove the channel from the array of channels. See also
420 * the inline comments before the call of vmbus_release_relid() below.
422 if (channel->offermsg.child_relid != INVALID_RELID)
423 vmbus_channel_unmap_relid(channel);
425 if (channel->primary_channel == NULL) {
426 list_del(&channel->listentry);
428 struct vmbus_channel *primary_channel = channel->primary_channel;
429 spin_lock_irqsave(&primary_channel->lock, flags);
430 list_del(&channel->sc_list);
431 spin_unlock_irqrestore(&primary_channel->lock, flags);
435 * If this is a "perf" channel, updates the hv_numa_map[] masks so that
436 * init_vp_index() can (re-)use the CPU.
438 if (hv_is_perf_channel(channel))
439 hv_clear_alloced_cpu(channel->target_cpu);
442 * Upon suspend, an in-use hv_sock channel is marked as "rescinded" and
443 * the relid is invalidated; after hibernation, when the user-space app
444 * destroys the channel, the relid is INVALID_RELID, and in this case
445 * it's unnecessary and unsafe to release the old relid, since the same
446 * relid can refer to a completely different channel now.
448 if (channel->offermsg.child_relid != INVALID_RELID)
449 vmbus_release_relid(channel->offermsg.child_relid);
451 free_channel(channel);
454 void vmbus_free_channels(void)
456 struct vmbus_channel *channel, *tmp;
458 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
460 /* hv_process_channel_removal() needs this */
461 channel->rescind = true;
463 vmbus_device_unregister(channel->device_obj);
467 /* Note: the function can run concurrently for primary/sub channels. */
468 static void vmbus_add_channel_work(struct work_struct *work)
470 struct vmbus_channel *newchannel =
471 container_of(work, struct vmbus_channel, add_channel_work);
472 struct vmbus_channel *primary_channel = newchannel->primary_channel;
477 * This state is used to indicate a successful open
478 * so that when we do close the channel normally, we
479 * can cleanup properly.
481 newchannel->state = CHANNEL_OPEN_STATE;
483 if (primary_channel != NULL) {
484 /* newchannel is a sub-channel. */
485 struct hv_device *dev = primary_channel->device_obj;
487 if (vmbus_add_channel_kobj(dev, newchannel))
490 if (primary_channel->sc_creation_callback != NULL)
491 primary_channel->sc_creation_callback(newchannel);
493 newchannel->probe_done = true;
498 * Start the process of binding the primary channel to the driver
500 newchannel->device_obj = vmbus_device_create(
501 &newchannel->offermsg.offer.if_type,
502 &newchannel->offermsg.offer.if_instance,
504 if (!newchannel->device_obj)
507 newchannel->device_obj->device_id = newchannel->device_id;
509 * Add the new device to the bus. This will kick off device-driver
510 * binding which eventually invokes the device driver's AddDevice()
513 ret = vmbus_device_register(newchannel->device_obj);
516 pr_err("unable to add child device object (relid %d)\n",
517 newchannel->offermsg.child_relid);
518 kfree(newchannel->device_obj);
522 newchannel->probe_done = true;
526 mutex_lock(&vmbus_connection.channel_mutex);
529 * We need to set the flag, otherwise
530 * vmbus_onoffer_rescind() can be blocked.
532 newchannel->probe_done = true;
534 if (primary_channel == NULL) {
535 list_del(&newchannel->listentry);
537 spin_lock_irqsave(&primary_channel->lock, flags);
538 list_del(&newchannel->sc_list);
539 spin_unlock_irqrestore(&primary_channel->lock, flags);
542 /* vmbus_process_offer() has mapped the channel. */
543 vmbus_channel_unmap_relid(newchannel);
545 mutex_unlock(&vmbus_connection.channel_mutex);
547 vmbus_release_relid(newchannel->offermsg.child_relid);
549 free_channel(newchannel);
553 * vmbus_process_offer - Process the offer by creating a channel/device
554 * associated with this offer
556 static void vmbus_process_offer(struct vmbus_channel *newchannel)
558 struct vmbus_channel *channel;
559 struct workqueue_struct *wq;
564 * Synchronize vmbus_process_offer() and CPU hotplugging:
568 * [vmbus_process_offer()] [Hot removal of the CPU]
570 * CPU_READ_LOCK CPUS_WRITE_LOCK
571 * LOAD cpu_online_mask SEARCH chn_list
572 * STORE target_cpu LOAD target_cpu
573 * INSERT chn_list STORE cpu_online_mask
574 * CPUS_READ_UNLOCK CPUS_WRITE_UNLOCK
576 * Forbids: CPU1's LOAD from *not* seing CPU2's STORE &&
577 * CPU2's SEARCH from *not* seeing CPU1's INSERT
579 * Forbids: CPU2's SEARCH from seeing CPU1's INSERT &&
580 * CPU2's LOAD from *not* seing CPU1's STORE
585 * Serializes the modifications of the chn_list list as well as
586 * the accesses to next_numa_node_id in init_vp_index().
588 mutex_lock(&vmbus_connection.channel_mutex);
590 init_vp_index(newchannel);
592 /* Remember the channels that should be cleaned up upon suspend. */
593 if (is_hvsock_channel(newchannel) || is_sub_channel(newchannel))
594 atomic_inc(&vmbus_connection.nr_chan_close_on_suspend);
597 * Now that we have acquired the channel_mutex,
598 * we can release the potentially racing rescind thread.
600 atomic_dec(&vmbus_connection.offer_in_progress);
602 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
603 if (guid_equal(&channel->offermsg.offer.if_type,
604 &newchannel->offermsg.offer.if_type) &&
605 guid_equal(&channel->offermsg.offer.if_instance,
606 &newchannel->offermsg.offer.if_instance)) {
613 list_add_tail(&newchannel->listentry,
614 &vmbus_connection.chn_list);
617 * Check to see if this is a valid sub-channel.
619 if (newchannel->offermsg.offer.sub_channel_index == 0) {
620 mutex_unlock(&vmbus_connection.channel_mutex);
622 * Don't call free_channel(), because newchannel->kobj
623 * is not initialized yet.
630 * Process the sub-channel.
632 newchannel->primary_channel = channel;
633 spin_lock_irqsave(&channel->lock, flags);
634 list_add_tail(&newchannel->sc_list, &channel->sc_list);
635 spin_unlock_irqrestore(&channel->lock, flags);
638 vmbus_channel_map_relid(newchannel);
640 mutex_unlock(&vmbus_connection.channel_mutex);
644 * vmbus_process_offer() mustn't call channel->sc_creation_callback()
645 * directly for sub-channels, because sc_creation_callback() ->
646 * vmbus_open() may never get the host's response to the
647 * OPEN_CHANNEL message (the host may rescind a channel at any time,
648 * e.g. in the case of hot removing a NIC), and vmbus_onoffer_rescind()
649 * may not wake up the vmbus_open() as it's blocked due to a non-zero
650 * vmbus_connection.offer_in_progress, and finally we have a deadlock.
652 * The above is also true for primary channels, if the related device
653 * drivers use sync probing mode by default.
655 * And, usually the handling of primary channels and sub-channels can
656 * depend on each other, so we should offload them to different
657 * workqueues to avoid possible deadlock, e.g. in sync-probing mode,
658 * NIC1's netvsc_subchan_work() can race with NIC2's netvsc_probe() ->
659 * rtnl_lock(), and causes deadlock: the former gets the rtnl_lock
660 * and waits for all the sub-channels to appear, but the latter
661 * can't get the rtnl_lock and this blocks the handling of
664 INIT_WORK(&newchannel->add_channel_work, vmbus_add_channel_work);
665 wq = fnew ? vmbus_connection.handle_primary_chan_wq :
666 vmbus_connection.handle_sub_chan_wq;
667 queue_work(wq, &newchannel->add_channel_work);
671 * We use this state to statically distribute the channel interrupt load.
673 static int next_numa_node_id;
676 * Starting with Win8, we can statically distribute the incoming
677 * channel interrupt load by binding a channel to VCPU.
679 * For pre-win8 hosts or non-performance critical channels we assign the
682 * Starting with win8, performance critical channels will be distributed
683 * evenly among all the available NUMA nodes. Once the node is assigned,
684 * we will assign the CPU based on a simple round robin scheme.
686 static void init_vp_index(struct vmbus_channel *channel)
688 bool perf_chn = hv_is_perf_channel(channel);
689 cpumask_var_t available_mask;
690 struct cpumask *alloced_mask;
694 if ((vmbus_proto_version == VERSION_WS2008) ||
695 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn) ||
696 !alloc_cpumask_var(&available_mask, GFP_KERNEL)) {
698 * Prior to win8, all channel interrupts are
699 * delivered on VMBUS_CONNECT_CPU.
700 * Also if the channel is not a performance critical
701 * channel, bind it to VMBUS_CONNECT_CPU.
702 * In case alloc_cpumask_var() fails, bind it to
705 channel->numa_node = cpu_to_node(VMBUS_CONNECT_CPU);
706 channel->target_cpu = VMBUS_CONNECT_CPU;
708 hv_cpu_number_to_vp_number(VMBUS_CONNECT_CPU);
710 hv_set_alloced_cpu(VMBUS_CONNECT_CPU);
715 numa_node = next_numa_node_id++;
716 if (numa_node == nr_node_ids) {
717 next_numa_node_id = 0;
720 if (cpumask_empty(cpumask_of_node(numa_node)))
724 channel->numa_node = numa_node;
725 alloced_mask = &hv_context.hv_numa_map[numa_node];
727 if (cpumask_weight(alloced_mask) ==
728 cpumask_weight(cpumask_of_node(numa_node))) {
730 * We have cycled through all the CPUs in the node;
731 * reset the alloced map.
733 cpumask_clear(alloced_mask);
736 cpumask_xor(available_mask, alloced_mask, cpumask_of_node(numa_node));
738 target_cpu = cpumask_first(available_mask);
739 cpumask_set_cpu(target_cpu, alloced_mask);
741 channel->target_cpu = target_cpu;
742 channel->target_vp = hv_cpu_number_to_vp_number(target_cpu);
744 free_cpumask_var(available_mask);
747 static void vmbus_wait_for_unload(void)
751 struct hv_message *msg;
752 struct vmbus_channel_message_header *hdr;
756 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
757 * used for initial contact or to CPU0 depending on host version. When
758 * we're crashing on a different CPU let's hope that IRQ handler on
759 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
760 * functional and vmbus_unload_response() will complete
761 * vmbus_connection.unload_event. If not, the last thing we can do is
762 * read message pages for all CPUs directly.
765 if (completion_done(&vmbus_connection.unload_event))
768 for_each_online_cpu(cpu) {
769 struct hv_per_cpu_context *hv_cpu
770 = per_cpu_ptr(hv_context.cpu_context, cpu);
772 page_addr = hv_cpu->synic_message_page;
773 msg = (struct hv_message *)page_addr
774 + VMBUS_MESSAGE_SINT;
776 message_type = READ_ONCE(msg->header.message_type);
777 if (message_type == HVMSG_NONE)
780 hdr = (struct vmbus_channel_message_header *)
783 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
784 complete(&vmbus_connection.unload_event);
786 vmbus_signal_eom(msg, message_type);
793 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
794 * maybe-pending messages on all CPUs to be able to receive new
795 * messages after we reconnect.
797 for_each_online_cpu(cpu) {
798 struct hv_per_cpu_context *hv_cpu
799 = per_cpu_ptr(hv_context.cpu_context, cpu);
801 page_addr = hv_cpu->synic_message_page;
802 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
803 msg->header.message_type = HVMSG_NONE;
808 * vmbus_unload_response - Handler for the unload response.
810 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
813 * This is a global event; just wakeup the waiting thread.
814 * Once we successfully unload, we can cleanup the monitor state.
816 complete(&vmbus_connection.unload_event);
819 void vmbus_initiate_unload(bool crash)
821 struct vmbus_channel_message_header hdr;
823 if (xchg(&vmbus_connection.conn_state, DISCONNECTED) == DISCONNECTED)
826 /* Pre-Win2012R2 hosts don't support reconnect */
827 if (vmbus_proto_version < VERSION_WIN8_1)
830 init_completion(&vmbus_connection.unload_event);
831 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
832 hdr.msgtype = CHANNELMSG_UNLOAD;
833 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
837 * vmbus_initiate_unload() is also called on crash and the crash can be
838 * happening in an interrupt context, where scheduling is impossible.
841 wait_for_completion(&vmbus_connection.unload_event);
843 vmbus_wait_for_unload();
846 static void check_ready_for_resume_event(void)
849 * If all the old primary channels have been fixed up, then it's safe
852 if (atomic_dec_and_test(&vmbus_connection.nr_chan_fixup_on_resume))
853 complete(&vmbus_connection.ready_for_resume_event);
856 static void vmbus_setup_channel_state(struct vmbus_channel *channel,
857 struct vmbus_channel_offer_channel *offer)
860 * Setup state for signalling the host.
862 channel->sig_event = VMBUS_EVENT_CONNECTION_ID;
864 if (vmbus_proto_version != VERSION_WS2008) {
865 channel->is_dedicated_interrupt =
866 (offer->is_dedicated_interrupt != 0);
867 channel->sig_event = offer->connection_id;
870 memcpy(&channel->offermsg, offer,
871 sizeof(struct vmbus_channel_offer_channel));
872 channel->monitor_grp = (u8)offer->monitorid / 32;
873 channel->monitor_bit = (u8)offer->monitorid % 32;
874 channel->device_id = hv_get_dev_type(channel);
878 * find_primary_channel_by_offer - Get the channel object given the new offer.
879 * This is only used in the resume path of hibernation.
881 static struct vmbus_channel *
882 find_primary_channel_by_offer(const struct vmbus_channel_offer_channel *offer)
884 struct vmbus_channel *channel = NULL, *iter;
885 const guid_t *inst1, *inst2;
887 /* Ignore sub-channel offers. */
888 if (offer->offer.sub_channel_index != 0)
891 mutex_lock(&vmbus_connection.channel_mutex);
893 list_for_each_entry(iter, &vmbus_connection.chn_list, listentry) {
894 inst1 = &iter->offermsg.offer.if_instance;
895 inst2 = &offer->offer.if_instance;
897 if (guid_equal(inst1, inst2)) {
903 mutex_unlock(&vmbus_connection.channel_mutex);
909 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
912 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
914 struct vmbus_channel_offer_channel *offer;
915 struct vmbus_channel *oldchannel, *newchannel;
918 offer = (struct vmbus_channel_offer_channel *)hdr;
920 trace_vmbus_onoffer(offer);
922 oldchannel = find_primary_channel_by_offer(offer);
924 if (oldchannel != NULL) {
926 * We're resuming from hibernation: all the sub-channel and
927 * hv_sock channels we had before the hibernation should have
928 * been cleaned up, and now we must be seeing a re-offered
929 * primary channel that we had before the hibernation.
933 * { Initially: channel relid = INVALID_RELID,
934 * channels[valid_relid] = NULL }
938 * [vmbus_onoffer()] [vmbus_device_release()]
940 * LOCK channel_mutex LOCK channel_mutex
941 * STORE channel relid = valid_relid LOAD r1 = channel relid
942 * MAP_RELID channel if (r1 != INVALID_RELID)
943 * UNLOCK channel_mutex UNMAP_RELID channel
944 * UNLOCK channel_mutex
946 * Forbids: r1 == valid_relid &&
947 * channels[valid_relid] == channel
949 * Note. r1 can be INVALID_RELID only for an hv_sock channel.
950 * None of the hv_sock channels which were present before the
951 * suspend are re-offered upon the resume. See the WARN_ON()
952 * in hv_process_channel_removal().
954 mutex_lock(&vmbus_connection.channel_mutex);
956 atomic_dec(&vmbus_connection.offer_in_progress);
958 WARN_ON(oldchannel->offermsg.child_relid != INVALID_RELID);
959 /* Fix up the relid. */
960 oldchannel->offermsg.child_relid = offer->child_relid;
962 offer_sz = sizeof(*offer);
963 if (memcmp(offer, &oldchannel->offermsg, offer_sz) != 0) {
965 * This is not an error, since the host can also change
966 * the other field(s) of the offer, e.g. on WS RS5
967 * (Build 17763), the offer->connection_id of the
968 * Mellanox VF vmbus device can change when the host
969 * reoffers the device upon resume.
971 pr_debug("vmbus offer changed: relid=%d\n",
974 print_hex_dump_debug("Old vmbus offer: ",
975 DUMP_PREFIX_OFFSET, 16, 4,
976 &oldchannel->offermsg, offer_sz,
978 print_hex_dump_debug("New vmbus offer: ",
979 DUMP_PREFIX_OFFSET, 16, 4,
980 offer, offer_sz, false);
982 /* Fix up the old channel. */
983 vmbus_setup_channel_state(oldchannel, offer);
986 /* Add the channel back to the array of channels. */
987 vmbus_channel_map_relid(oldchannel);
988 check_ready_for_resume_event();
990 mutex_unlock(&vmbus_connection.channel_mutex);
994 /* Allocate the channel object and save this offer. */
995 newchannel = alloc_channel();
997 vmbus_release_relid(offer->child_relid);
998 atomic_dec(&vmbus_connection.offer_in_progress);
999 pr_err("Unable to allocate channel object\n");
1003 vmbus_setup_channel_state(newchannel, offer);
1005 vmbus_process_offer(newchannel);
1008 static void check_ready_for_suspend_event(void)
1011 * If all the sub-channels or hv_sock channels have been cleaned up,
1012 * then it's safe to suspend.
1014 if (atomic_dec_and_test(&vmbus_connection.nr_chan_close_on_suspend))
1015 complete(&vmbus_connection.ready_for_suspend_event);
1019 * vmbus_onoffer_rescind - Rescind offer handler.
1021 * We queue a work item to process this offer synchronously
1023 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
1025 struct vmbus_channel_rescind_offer *rescind;
1026 struct vmbus_channel *channel;
1028 bool clean_up_chan_for_suspend;
1030 rescind = (struct vmbus_channel_rescind_offer *)hdr;
1032 trace_vmbus_onoffer_rescind(rescind);
1035 * The offer msg and the corresponding rescind msg
1036 * from the host are guranteed to be ordered -
1037 * offer comes in first and then the rescind.
1038 * Since we process these events in work elements,
1039 * and with preemption, we may end up processing
1040 * the events out of order. We rely on the synchronization
1041 * provided by offer_in_progress and by channel_mutex for
1042 * ordering these events:
1044 * { Initially: offer_in_progress = 1 }
1048 * [vmbus_onoffer()] [vmbus_onoffer_rescind()]
1050 * LOCK channel_mutex WAIT_ON offer_in_progress == 0
1051 * DECREMENT offer_in_progress LOCK channel_mutex
1052 * STORE channels[] LOAD channels[]
1053 * UNLOCK channel_mutex UNLOCK channel_mutex
1055 * Forbids: CPU2's LOAD from *not* seeing CPU1's STORE
1058 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
1060 * We wait here until any channel offer is currently
1066 mutex_lock(&vmbus_connection.channel_mutex);
1067 channel = relid2channel(rescind->child_relid);
1068 mutex_unlock(&vmbus_connection.channel_mutex);
1070 if (channel == NULL) {
1072 * We failed in processing the offer message;
1073 * we would have cleaned up the relid in that
1079 clean_up_chan_for_suspend = is_hvsock_channel(channel) ||
1080 is_sub_channel(channel);
1082 * Before setting channel->rescind in vmbus_rescind_cleanup(), we
1083 * should make sure the channel callback is not running any more.
1085 vmbus_reset_channel_cb(channel);
1088 * Now wait for offer handling to complete.
1090 vmbus_rescind_cleanup(channel);
1091 while (READ_ONCE(channel->probe_done) == false) {
1093 * We wait here until any channel offer is currently
1100 * At this point, the rescind handling can proceed safely.
1103 if (channel->device_obj) {
1104 if (channel->chn_rescind_callback) {
1105 channel->chn_rescind_callback(channel);
1107 if (clean_up_chan_for_suspend)
1108 check_ready_for_suspend_event();
1113 * We will have to unregister this device from the
1116 dev = get_device(&channel->device_obj->device);
1118 vmbus_device_unregister(channel->device_obj);
1122 if (channel->primary_channel != NULL) {
1124 * Sub-channel is being rescinded. Following is the channel
1125 * close sequence when initiated from the driveri (refer to
1126 * vmbus_close() for details):
1127 * 1. Close all sub-channels first
1128 * 2. Then close the primary channel.
1130 mutex_lock(&vmbus_connection.channel_mutex);
1131 if (channel->state == CHANNEL_OPEN_STATE) {
1133 * The channel is currently not open;
1134 * it is safe for us to cleanup the channel.
1136 hv_process_channel_removal(channel);
1138 complete(&channel->rescind_event);
1140 mutex_unlock(&vmbus_connection.channel_mutex);
1143 /* The "channel" may have been freed. Do not access it any longer. */
1145 if (clean_up_chan_for_suspend)
1146 check_ready_for_suspend_event();
1149 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
1151 BUG_ON(!is_hvsock_channel(channel));
1153 /* We always get a rescind msg when a connection is closed. */
1154 while (!READ_ONCE(channel->probe_done) || !READ_ONCE(channel->rescind))
1157 vmbus_device_unregister(channel->device_obj);
1159 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
1163 * vmbus_onoffers_delivered -
1164 * This is invoked when all offers have been delivered.
1166 * Nothing to do here.
1168 static void vmbus_onoffers_delivered(
1169 struct vmbus_channel_message_header *hdr)
1174 * vmbus_onopen_result - Open result handler.
1176 * This is invoked when we received a response to our channel open request.
1177 * Find the matching request, copy the response and signal the requesting
1180 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
1182 struct vmbus_channel_open_result *result;
1183 struct vmbus_channel_msginfo *msginfo;
1184 struct vmbus_channel_message_header *requestheader;
1185 struct vmbus_channel_open_channel *openmsg;
1186 unsigned long flags;
1188 result = (struct vmbus_channel_open_result *)hdr;
1190 trace_vmbus_onopen_result(result);
1193 * Find the open msg, copy the result and signal/unblock the wait event
1195 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1197 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1200 (struct vmbus_channel_message_header *)msginfo->msg;
1202 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
1204 (struct vmbus_channel_open_channel *)msginfo->msg;
1205 if (openmsg->child_relid == result->child_relid &&
1206 openmsg->openid == result->openid) {
1207 memcpy(&msginfo->response.open_result,
1210 struct vmbus_channel_open_result));
1211 complete(&msginfo->waitevent);
1216 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1220 * vmbus_ongpadl_created - GPADL created handler.
1222 * This is invoked when we received a response to our gpadl create request.
1223 * Find the matching request, copy the response and signal the requesting
1226 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
1228 struct vmbus_channel_gpadl_created *gpadlcreated;
1229 struct vmbus_channel_msginfo *msginfo;
1230 struct vmbus_channel_message_header *requestheader;
1231 struct vmbus_channel_gpadl_header *gpadlheader;
1232 unsigned long flags;
1234 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
1236 trace_vmbus_ongpadl_created(gpadlcreated);
1239 * Find the establish msg, copy the result and signal/unblock the wait
1242 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1244 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1247 (struct vmbus_channel_message_header *)msginfo->msg;
1249 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
1251 (struct vmbus_channel_gpadl_header *)requestheader;
1253 if ((gpadlcreated->child_relid ==
1254 gpadlheader->child_relid) &&
1255 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1256 memcpy(&msginfo->response.gpadl_created,
1259 struct vmbus_channel_gpadl_created));
1260 complete(&msginfo->waitevent);
1265 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1269 * vmbus_ongpadl_torndown - GPADL torndown handler.
1271 * This is invoked when we received a response to our gpadl teardown request.
1272 * Find the matching request, copy the response and signal the requesting
1275 static void vmbus_ongpadl_torndown(
1276 struct vmbus_channel_message_header *hdr)
1278 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1279 struct vmbus_channel_msginfo *msginfo;
1280 struct vmbus_channel_message_header *requestheader;
1281 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1282 unsigned long flags;
1284 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1286 trace_vmbus_ongpadl_torndown(gpadl_torndown);
1289 * Find the open msg, copy the result and signal/unblock the wait event
1291 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1293 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1296 (struct vmbus_channel_message_header *)msginfo->msg;
1298 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1300 (struct vmbus_channel_gpadl_teardown *)requestheader;
1302 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1303 memcpy(&msginfo->response.gpadl_torndown,
1306 struct vmbus_channel_gpadl_torndown));
1307 complete(&msginfo->waitevent);
1312 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1316 * vmbus_onversion_response - Version response handler
1318 * This is invoked when we received a response to our initiate contact request.
1319 * Find the matching request, copy the response and signal the requesting
1322 static void vmbus_onversion_response(
1323 struct vmbus_channel_message_header *hdr)
1325 struct vmbus_channel_msginfo *msginfo;
1326 struct vmbus_channel_message_header *requestheader;
1327 struct vmbus_channel_version_response *version_response;
1328 unsigned long flags;
1330 version_response = (struct vmbus_channel_version_response *)hdr;
1332 trace_vmbus_onversion_response(version_response);
1334 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1336 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1339 (struct vmbus_channel_message_header *)msginfo->msg;
1341 if (requestheader->msgtype ==
1342 CHANNELMSG_INITIATE_CONTACT) {
1343 memcpy(&msginfo->response.version_response,
1345 sizeof(struct vmbus_channel_version_response));
1346 complete(&msginfo->waitevent);
1349 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1352 /* Channel message dispatch table */
1353 const struct vmbus_channel_message_table_entry
1354 channel_message_table[CHANNELMSG_COUNT] = {
1355 { CHANNELMSG_INVALID, 0, NULL, 0},
1356 { CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer,
1357 sizeof(struct vmbus_channel_offer_channel)},
1358 { CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind,
1359 sizeof(struct vmbus_channel_rescind_offer) },
1360 { CHANNELMSG_REQUESTOFFERS, 0, NULL, 0},
1361 { CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered, 0},
1362 { CHANNELMSG_OPENCHANNEL, 0, NULL, 0},
1363 { CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result,
1364 sizeof(struct vmbus_channel_open_result)},
1365 { CHANNELMSG_CLOSECHANNEL, 0, NULL, 0},
1366 { CHANNELMSG_GPADL_HEADER, 0, NULL, 0},
1367 { CHANNELMSG_GPADL_BODY, 0, NULL, 0},
1368 { CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created,
1369 sizeof(struct vmbus_channel_gpadl_created)},
1370 { CHANNELMSG_GPADL_TEARDOWN, 0, NULL, 0},
1371 { CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown,
1372 sizeof(struct vmbus_channel_gpadl_torndown) },
1373 { CHANNELMSG_RELID_RELEASED, 0, NULL, 0},
1374 { CHANNELMSG_INITIATE_CONTACT, 0, NULL, 0},
1375 { CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response,
1376 sizeof(struct vmbus_channel_version_response)},
1377 { CHANNELMSG_UNLOAD, 0, NULL, 0},
1378 { CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response, 0},
1379 { CHANNELMSG_18, 0, NULL, 0},
1380 { CHANNELMSG_19, 0, NULL, 0},
1381 { CHANNELMSG_20, 0, NULL, 0},
1382 { CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL, 0},
1383 { CHANNELMSG_MODIFYCHANNEL, 0, NULL, 0},
1384 { CHANNELMSG_TL_CONNECT_RESULT, 0, NULL, 0},
1388 * vmbus_onmessage - Handler for channel protocol messages.
1390 * This is invoked in the vmbus worker thread context.
1392 void vmbus_onmessage(struct vmbus_channel_message_header *hdr)
1394 trace_vmbus_on_message(hdr);
1397 * vmbus_on_msg_dpc() makes sure the hdr->msgtype here can not go
1398 * out of bound and the message_handler pointer can not be NULL.
1400 channel_message_table[hdr->msgtype].message_handler(hdr);
1404 * vmbus_request_offers - Send a request to get all our pending offers.
1406 int vmbus_request_offers(void)
1408 struct vmbus_channel_message_header *msg;
1409 struct vmbus_channel_msginfo *msginfo;
1412 msginfo = kmalloc(sizeof(*msginfo) +
1413 sizeof(struct vmbus_channel_message_header),
1418 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1420 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1422 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1425 trace_vmbus_request_offers(ret);
1428 pr_err("Unable to request offers - %d\n", ret);
1439 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1441 struct list_head *cur, *tmp;
1442 struct vmbus_channel *cur_channel;
1444 if (primary_channel->sc_creation_callback == NULL)
1447 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1448 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1450 primary_channel->sc_creation_callback(cur_channel);
1454 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1455 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1457 primary_channel->sc_creation_callback = sc_cr_cb;
1459 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1461 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1465 ret = !list_empty(&primary->sc_list);
1469 * Invoke the callback on sub-channel creation.
1470 * This will present a uniform interface to the
1473 invoke_sc_cb(primary);
1478 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1480 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1481 void (*chn_rescind_cb)(struct vmbus_channel *))
1483 channel->chn_rescind_callback = chn_rescind_cb;
1485 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);