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141 * The rescinded channel may be blocked waiting for a response from the host;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 init_completion(&channel->rescind_event);
322 INIT_LIST_HEAD(&channel->sc_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 void vmbus_channel_map_relid(struct vmbus_channel *channel)
345 if (WARN_ON(channel->offermsg.child_relid >= MAX_CHANNEL_RELIDS))
348 * The mapping of the channel's relid is visible from the CPUs that
349 * execute vmbus_chan_sched() by the time that vmbus_chan_sched() will
352 * (a) In the "normal (i.e., not resuming from hibernation)" path,
353 * the full barrier in virt_store_mb() guarantees that the store
354 * is propagated to all CPUs before the add_channel_work work
355 * is queued. In turn, add_channel_work is queued before the
356 * channel's ring buffer is allocated/initialized and the
357 * OPENCHANNEL message for the channel is sent in vmbus_open().
358 * Hyper-V won't start sending the interrupts for the channel
359 * before the OPENCHANNEL message is acked. The memory barrier
360 * in vmbus_chan_sched() -> sync_test_and_clear_bit() ensures
361 * that vmbus_chan_sched() must find the channel's relid in
362 * recv_int_page before retrieving the channel pointer from the
365 * (b) In the "resuming from hibernation" path, the virt_store_mb()
366 * guarantees that the store is propagated to all CPUs before
367 * the VMBus connection is marked as ready for the resume event
368 * (cf. check_ready_for_resume_event()). The interrupt handler
369 * of the VMBus driver and vmbus_chan_sched() can not run before
370 * vmbus_bus_resume() has completed execution (cf. resume_noirq).
373 vmbus_connection.channels[channel->offermsg.child_relid],
377 void vmbus_channel_unmap_relid(struct vmbus_channel *channel)
379 if (WARN_ON(channel->offermsg.child_relid >= MAX_CHANNEL_RELIDS))
382 vmbus_connection.channels[channel->offermsg.child_relid],
386 static void vmbus_release_relid(u32 relid)
388 struct vmbus_channel_relid_released msg;
391 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
392 msg.child_relid = relid;
393 msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
394 ret = vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released),
397 trace_vmbus_release_relid(&msg, ret);
400 void hv_process_channel_removal(struct vmbus_channel *channel)
402 lockdep_assert_held(&vmbus_connection.channel_mutex);
403 BUG_ON(!channel->rescind);
406 * hv_process_channel_removal() could find INVALID_RELID only for
407 * hv_sock channels. See the inline comments in vmbus_onoffer().
409 WARN_ON(channel->offermsg.child_relid == INVALID_RELID &&
410 !is_hvsock_channel(channel));
413 * Upon suspend, an in-use hv_sock channel is removed from the array of
414 * channels and the relid is invalidated. After hibernation, when the
415 * user-space appplication destroys the channel, it's unnecessary and
416 * unsafe to remove the channel from the array of channels. See also
417 * the inline comments before the call of vmbus_release_relid() below.
419 if (channel->offermsg.child_relid != INVALID_RELID)
420 vmbus_channel_unmap_relid(channel);
422 if (channel->primary_channel == NULL)
423 list_del(&channel->listentry);
425 list_del(&channel->sc_list);
428 * If this is a "perf" channel, updates the hv_numa_map[] masks so that
429 * init_vp_index() can (re-)use the CPU.
431 if (hv_is_perf_channel(channel))
432 hv_clear_alloced_cpu(channel->target_cpu);
435 * Upon suspend, an in-use hv_sock channel is marked as "rescinded" and
436 * the relid is invalidated; after hibernation, when the user-space app
437 * destroys the channel, the relid is INVALID_RELID, and in this case
438 * it's unnecessary and unsafe to release the old relid, since the same
439 * relid can refer to a completely different channel now.
441 if (channel->offermsg.child_relid != INVALID_RELID)
442 vmbus_release_relid(channel->offermsg.child_relid);
444 free_channel(channel);
447 void vmbus_free_channels(void)
449 struct vmbus_channel *channel, *tmp;
451 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
453 /* hv_process_channel_removal() needs this */
454 channel->rescind = true;
456 vmbus_device_unregister(channel->device_obj);
460 /* Note: the function can run concurrently for primary/sub channels. */
461 static void vmbus_add_channel_work(struct work_struct *work)
463 struct vmbus_channel *newchannel =
464 container_of(work, struct vmbus_channel, add_channel_work);
465 struct vmbus_channel *primary_channel = newchannel->primary_channel;
469 * This state is used to indicate a successful open
470 * so that when we do close the channel normally, we
471 * can cleanup properly.
473 newchannel->state = CHANNEL_OPEN_STATE;
475 if (primary_channel != NULL) {
476 /* newchannel is a sub-channel. */
477 struct hv_device *dev = primary_channel->device_obj;
479 if (vmbus_add_channel_kobj(dev, newchannel))
482 if (primary_channel->sc_creation_callback != NULL)
483 primary_channel->sc_creation_callback(newchannel);
485 newchannel->probe_done = true;
490 * Start the process of binding the primary channel to the driver
492 newchannel->device_obj = vmbus_device_create(
493 &newchannel->offermsg.offer.if_type,
494 &newchannel->offermsg.offer.if_instance,
496 if (!newchannel->device_obj)
499 newchannel->device_obj->device_id = newchannel->device_id;
501 * Add the new device to the bus. This will kick off device-driver
502 * binding which eventually invokes the device driver's AddDevice()
505 ret = vmbus_device_register(newchannel->device_obj);
508 pr_err("unable to add child device object (relid %d)\n",
509 newchannel->offermsg.child_relid);
510 kfree(newchannel->device_obj);
514 newchannel->probe_done = true;
518 mutex_lock(&vmbus_connection.channel_mutex);
521 * We need to set the flag, otherwise
522 * vmbus_onoffer_rescind() can be blocked.
524 newchannel->probe_done = true;
526 if (primary_channel == NULL)
527 list_del(&newchannel->listentry);
529 list_del(&newchannel->sc_list);
531 /* vmbus_process_offer() has mapped the channel. */
532 vmbus_channel_unmap_relid(newchannel);
534 mutex_unlock(&vmbus_connection.channel_mutex);
536 vmbus_release_relid(newchannel->offermsg.child_relid);
538 free_channel(newchannel);
542 * vmbus_process_offer - Process the offer by creating a channel/device
543 * associated with this offer
545 static void vmbus_process_offer(struct vmbus_channel *newchannel)
547 struct vmbus_channel *channel;
548 struct workqueue_struct *wq;
552 * Synchronize vmbus_process_offer() and CPU hotplugging:
556 * [vmbus_process_offer()] [Hot removal of the CPU]
558 * CPU_READ_LOCK CPUS_WRITE_LOCK
559 * LOAD cpu_online_mask SEARCH chn_list
560 * STORE target_cpu LOAD target_cpu
561 * INSERT chn_list STORE cpu_online_mask
562 * CPUS_READ_UNLOCK CPUS_WRITE_UNLOCK
564 * Forbids: CPU1's LOAD from *not* seing CPU2's STORE &&
565 * CPU2's SEARCH from *not* seeing CPU1's INSERT
567 * Forbids: CPU2's SEARCH from seeing CPU1's INSERT &&
568 * CPU2's LOAD from *not* seing CPU1's STORE
573 * Serializes the modifications of the chn_list list as well as
574 * the accesses to next_numa_node_id in init_vp_index().
576 mutex_lock(&vmbus_connection.channel_mutex);
578 init_vp_index(newchannel);
580 /* Remember the channels that should be cleaned up upon suspend. */
581 if (is_hvsock_channel(newchannel) || is_sub_channel(newchannel))
582 atomic_inc(&vmbus_connection.nr_chan_close_on_suspend);
585 * Now that we have acquired the channel_mutex,
586 * we can release the potentially racing rescind thread.
588 atomic_dec(&vmbus_connection.offer_in_progress);
590 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
591 if (guid_equal(&channel->offermsg.offer.if_type,
592 &newchannel->offermsg.offer.if_type) &&
593 guid_equal(&channel->offermsg.offer.if_instance,
594 &newchannel->offermsg.offer.if_instance)) {
601 list_add_tail(&newchannel->listentry,
602 &vmbus_connection.chn_list);
605 * Check to see if this is a valid sub-channel.
607 if (newchannel->offermsg.offer.sub_channel_index == 0) {
608 mutex_unlock(&vmbus_connection.channel_mutex);
611 * Don't call free_channel(), because newchannel->kobj
612 * is not initialized yet.
619 * Process the sub-channel.
621 newchannel->primary_channel = channel;
622 list_add_tail(&newchannel->sc_list, &channel->sc_list);
625 vmbus_channel_map_relid(newchannel);
627 mutex_unlock(&vmbus_connection.channel_mutex);
631 * vmbus_process_offer() mustn't call channel->sc_creation_callback()
632 * directly for sub-channels, because sc_creation_callback() ->
633 * vmbus_open() may never get the host's response to the
634 * OPEN_CHANNEL message (the host may rescind a channel at any time,
635 * e.g. in the case of hot removing a NIC), and vmbus_onoffer_rescind()
636 * may not wake up the vmbus_open() as it's blocked due to a non-zero
637 * vmbus_connection.offer_in_progress, and finally we have a deadlock.
639 * The above is also true for primary channels, if the related device
640 * drivers use sync probing mode by default.
642 * And, usually the handling of primary channels and sub-channels can
643 * depend on each other, so we should offload them to different
644 * workqueues to avoid possible deadlock, e.g. in sync-probing mode,
645 * NIC1's netvsc_subchan_work() can race with NIC2's netvsc_probe() ->
646 * rtnl_lock(), and causes deadlock: the former gets the rtnl_lock
647 * and waits for all the sub-channels to appear, but the latter
648 * can't get the rtnl_lock and this blocks the handling of
651 INIT_WORK(&newchannel->add_channel_work, vmbus_add_channel_work);
652 wq = fnew ? vmbus_connection.handle_primary_chan_wq :
653 vmbus_connection.handle_sub_chan_wq;
654 queue_work(wq, &newchannel->add_channel_work);
658 * We use this state to statically distribute the channel interrupt load.
660 static int next_numa_node_id;
663 * Starting with Win8, we can statically distribute the incoming
664 * channel interrupt load by binding a channel to VCPU.
666 * For pre-win8 hosts or non-performance critical channels we assign the
669 * Starting with win8, performance critical channels will be distributed
670 * evenly among all the available NUMA nodes. Once the node is assigned,
671 * we will assign the CPU based on a simple round robin scheme.
673 static void init_vp_index(struct vmbus_channel *channel)
675 bool perf_chn = hv_is_perf_channel(channel);
676 cpumask_var_t available_mask;
677 struct cpumask *alloced_mask;
681 if ((vmbus_proto_version == VERSION_WS2008) ||
682 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn) ||
683 !alloc_cpumask_var(&available_mask, GFP_KERNEL)) {
685 * Prior to win8, all channel interrupts are
686 * delivered on VMBUS_CONNECT_CPU.
687 * Also if the channel is not a performance critical
688 * channel, bind it to VMBUS_CONNECT_CPU.
689 * In case alloc_cpumask_var() fails, bind it to
692 channel->target_cpu = VMBUS_CONNECT_CPU;
694 hv_set_alloced_cpu(VMBUS_CONNECT_CPU);
699 numa_node = next_numa_node_id++;
700 if (numa_node == nr_node_ids) {
701 next_numa_node_id = 0;
704 if (cpumask_empty(cpumask_of_node(numa_node)))
708 alloced_mask = &hv_context.hv_numa_map[numa_node];
710 if (cpumask_weight(alloced_mask) ==
711 cpumask_weight(cpumask_of_node(numa_node))) {
713 * We have cycled through all the CPUs in the node;
714 * reset the alloced map.
716 cpumask_clear(alloced_mask);
719 cpumask_xor(available_mask, alloced_mask, cpumask_of_node(numa_node));
721 target_cpu = cpumask_first(available_mask);
722 cpumask_set_cpu(target_cpu, alloced_mask);
724 channel->target_cpu = target_cpu;
726 free_cpumask_var(available_mask);
729 #define UNLOAD_DELAY_UNIT_MS 10 /* 10 milliseconds */
730 #define UNLOAD_WAIT_MS (100*1000) /* 100 seconds */
731 #define UNLOAD_WAIT_LOOPS (UNLOAD_WAIT_MS/UNLOAD_DELAY_UNIT_MS)
732 #define UNLOAD_MSG_MS (5*1000) /* Every 5 seconds */
733 #define UNLOAD_MSG_LOOPS (UNLOAD_MSG_MS/UNLOAD_DELAY_UNIT_MS)
735 static void vmbus_wait_for_unload(void)
739 struct hv_message *msg;
740 struct vmbus_channel_message_header *hdr;
744 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
745 * used for initial contact or to CPU0 depending on host version. When
746 * we're crashing on a different CPU let's hope that IRQ handler on
747 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
748 * functional and vmbus_unload_response() will complete
749 * vmbus_connection.unload_event. If not, the last thing we can do is
750 * read message pages for all CPUs directly.
752 * Wait up to 100 seconds since an Azure host must writeback any dirty
753 * data in its disk cache before the VMbus UNLOAD request will
754 * complete. This flushing has been empirically observed to take up
755 * to 50 seconds in cases with a lot of dirty data, so allow additional
756 * leeway and for inaccuracies in mdelay(). But eventually time out so
757 * that the panic path can't get hung forever in case the response
758 * message isn't seen.
760 for (i = 1; i <= UNLOAD_WAIT_LOOPS; i++) {
761 if (completion_done(&vmbus_connection.unload_event))
764 for_each_online_cpu(cpu) {
765 struct hv_per_cpu_context *hv_cpu
766 = per_cpu_ptr(hv_context.cpu_context, cpu);
768 page_addr = hv_cpu->synic_message_page;
769 msg = (struct hv_message *)page_addr
770 + VMBUS_MESSAGE_SINT;
772 message_type = READ_ONCE(msg->header.message_type);
773 if (message_type == HVMSG_NONE)
776 hdr = (struct vmbus_channel_message_header *)
779 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
780 complete(&vmbus_connection.unload_event);
782 vmbus_signal_eom(msg, message_type);
786 * Give a notice periodically so someone watching the
787 * serial output won't think it is completely hung.
789 if (!(i % UNLOAD_MSG_LOOPS))
790 pr_notice("Waiting for VMBus UNLOAD to complete\n");
792 mdelay(UNLOAD_DELAY_UNIT_MS);
794 pr_err("Continuing even though VMBus UNLOAD did not complete\n");
798 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
799 * maybe-pending messages on all CPUs to be able to receive new
800 * messages after we reconnect.
802 for_each_online_cpu(cpu) {
803 struct hv_per_cpu_context *hv_cpu
804 = per_cpu_ptr(hv_context.cpu_context, cpu);
806 page_addr = hv_cpu->synic_message_page;
807 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
808 msg->header.message_type = HVMSG_NONE;
813 * vmbus_unload_response - Handler for the unload response.
815 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
818 * This is a global event; just wakeup the waiting thread.
819 * Once we successfully unload, we can cleanup the monitor state.
821 complete(&vmbus_connection.unload_event);
824 void vmbus_initiate_unload(bool crash)
826 struct vmbus_channel_message_header hdr;
828 if (xchg(&vmbus_connection.conn_state, DISCONNECTED) == DISCONNECTED)
831 /* Pre-Win2012R2 hosts don't support reconnect */
832 if (vmbus_proto_version < VERSION_WIN8_1)
835 init_completion(&vmbus_connection.unload_event);
836 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
837 hdr.msgtype = CHANNELMSG_UNLOAD;
838 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
842 * vmbus_initiate_unload() is also called on crash and the crash can be
843 * happening in an interrupt context, where scheduling is impossible.
846 wait_for_completion(&vmbus_connection.unload_event);
848 vmbus_wait_for_unload();
851 static void check_ready_for_resume_event(void)
854 * If all the old primary channels have been fixed up, then it's safe
857 if (atomic_dec_and_test(&vmbus_connection.nr_chan_fixup_on_resume))
858 complete(&vmbus_connection.ready_for_resume_event);
861 static void vmbus_setup_channel_state(struct vmbus_channel *channel,
862 struct vmbus_channel_offer_channel *offer)
865 * Setup state for signalling the host.
867 channel->sig_event = VMBUS_EVENT_CONNECTION_ID;
869 if (vmbus_proto_version != VERSION_WS2008) {
870 channel->is_dedicated_interrupt =
871 (offer->is_dedicated_interrupt != 0);
872 channel->sig_event = offer->connection_id;
875 memcpy(&channel->offermsg, offer,
876 sizeof(struct vmbus_channel_offer_channel));
877 channel->monitor_grp = (u8)offer->monitorid / 32;
878 channel->monitor_bit = (u8)offer->monitorid % 32;
879 channel->device_id = hv_get_dev_type(channel);
883 * find_primary_channel_by_offer - Get the channel object given the new offer.
884 * This is only used in the resume path of hibernation.
886 static struct vmbus_channel *
887 find_primary_channel_by_offer(const struct vmbus_channel_offer_channel *offer)
889 struct vmbus_channel *channel = NULL, *iter;
890 const guid_t *inst1, *inst2;
892 /* Ignore sub-channel offers. */
893 if (offer->offer.sub_channel_index != 0)
896 mutex_lock(&vmbus_connection.channel_mutex);
898 list_for_each_entry(iter, &vmbus_connection.chn_list, listentry) {
899 inst1 = &iter->offermsg.offer.if_instance;
900 inst2 = &offer->offer.if_instance;
902 if (guid_equal(inst1, inst2)) {
908 mutex_unlock(&vmbus_connection.channel_mutex);
914 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
917 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
919 struct vmbus_channel_offer_channel *offer;
920 struct vmbus_channel *oldchannel, *newchannel;
923 offer = (struct vmbus_channel_offer_channel *)hdr;
925 trace_vmbus_onoffer(offer);
927 oldchannel = find_primary_channel_by_offer(offer);
929 if (oldchannel != NULL) {
931 * We're resuming from hibernation: all the sub-channel and
932 * hv_sock channels we had before the hibernation should have
933 * been cleaned up, and now we must be seeing a re-offered
934 * primary channel that we had before the hibernation.
938 * { Initially: channel relid = INVALID_RELID,
939 * channels[valid_relid] = NULL }
943 * [vmbus_onoffer()] [vmbus_device_release()]
945 * LOCK channel_mutex LOCK channel_mutex
946 * STORE channel relid = valid_relid LOAD r1 = channel relid
947 * MAP_RELID channel if (r1 != INVALID_RELID)
948 * UNLOCK channel_mutex UNMAP_RELID channel
949 * UNLOCK channel_mutex
951 * Forbids: r1 == valid_relid &&
952 * channels[valid_relid] == channel
954 * Note. r1 can be INVALID_RELID only for an hv_sock channel.
955 * None of the hv_sock channels which were present before the
956 * suspend are re-offered upon the resume. See the WARN_ON()
957 * in hv_process_channel_removal().
959 mutex_lock(&vmbus_connection.channel_mutex);
961 atomic_dec(&vmbus_connection.offer_in_progress);
963 WARN_ON(oldchannel->offermsg.child_relid != INVALID_RELID);
964 /* Fix up the relid. */
965 oldchannel->offermsg.child_relid = offer->child_relid;
967 offer_sz = sizeof(*offer);
968 if (memcmp(offer, &oldchannel->offermsg, offer_sz) != 0) {
970 * This is not an error, since the host can also change
971 * the other field(s) of the offer, e.g. on WS RS5
972 * (Build 17763), the offer->connection_id of the
973 * Mellanox VF vmbus device can change when the host
974 * reoffers the device upon resume.
976 pr_debug("vmbus offer changed: relid=%d\n",
979 print_hex_dump_debug("Old vmbus offer: ",
980 DUMP_PREFIX_OFFSET, 16, 4,
981 &oldchannel->offermsg, offer_sz,
983 print_hex_dump_debug("New vmbus offer: ",
984 DUMP_PREFIX_OFFSET, 16, 4,
985 offer, offer_sz, false);
987 /* Fix up the old channel. */
988 vmbus_setup_channel_state(oldchannel, offer);
991 /* Add the channel back to the array of channels. */
992 vmbus_channel_map_relid(oldchannel);
993 check_ready_for_resume_event();
995 mutex_unlock(&vmbus_connection.channel_mutex);
999 /* Allocate the channel object and save this offer. */
1000 newchannel = alloc_channel();
1002 vmbus_release_relid(offer->child_relid);
1003 atomic_dec(&vmbus_connection.offer_in_progress);
1004 pr_err("Unable to allocate channel object\n");
1008 vmbus_setup_channel_state(newchannel, offer);
1010 vmbus_process_offer(newchannel);
1013 static void check_ready_for_suspend_event(void)
1016 * If all the sub-channels or hv_sock channels have been cleaned up,
1017 * then it's safe to suspend.
1019 if (atomic_dec_and_test(&vmbus_connection.nr_chan_close_on_suspend))
1020 complete(&vmbus_connection.ready_for_suspend_event);
1024 * vmbus_onoffer_rescind - Rescind offer handler.
1026 * We queue a work item to process this offer synchronously
1028 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
1030 struct vmbus_channel_rescind_offer *rescind;
1031 struct vmbus_channel *channel;
1033 bool clean_up_chan_for_suspend;
1035 rescind = (struct vmbus_channel_rescind_offer *)hdr;
1037 trace_vmbus_onoffer_rescind(rescind);
1040 * The offer msg and the corresponding rescind msg
1041 * from the host are guranteed to be ordered -
1042 * offer comes in first and then the rescind.
1043 * Since we process these events in work elements,
1044 * and with preemption, we may end up processing
1045 * the events out of order. We rely on the synchronization
1046 * provided by offer_in_progress and by channel_mutex for
1047 * ordering these events:
1049 * { Initially: offer_in_progress = 1 }
1053 * [vmbus_onoffer()] [vmbus_onoffer_rescind()]
1055 * LOCK channel_mutex WAIT_ON offer_in_progress == 0
1056 * DECREMENT offer_in_progress LOCK channel_mutex
1057 * STORE channels[] LOAD channels[]
1058 * UNLOCK channel_mutex UNLOCK channel_mutex
1060 * Forbids: CPU2's LOAD from *not* seeing CPU1's STORE
1063 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
1065 * We wait here until any channel offer is currently
1071 mutex_lock(&vmbus_connection.channel_mutex);
1072 channel = relid2channel(rescind->child_relid);
1073 mutex_unlock(&vmbus_connection.channel_mutex);
1075 if (channel == NULL) {
1077 * We failed in processing the offer message;
1078 * we would have cleaned up the relid in that
1084 clean_up_chan_for_suspend = is_hvsock_channel(channel) ||
1085 is_sub_channel(channel);
1087 * Before setting channel->rescind in vmbus_rescind_cleanup(), we
1088 * should make sure the channel callback is not running any more.
1090 vmbus_reset_channel_cb(channel);
1093 * Now wait for offer handling to complete.
1095 vmbus_rescind_cleanup(channel);
1096 while (READ_ONCE(channel->probe_done) == false) {
1098 * We wait here until any channel offer is currently
1105 * At this point, the rescind handling can proceed safely.
1108 if (channel->device_obj) {
1109 if (channel->chn_rescind_callback) {
1110 channel->chn_rescind_callback(channel);
1112 if (clean_up_chan_for_suspend)
1113 check_ready_for_suspend_event();
1118 * We will have to unregister this device from the
1121 dev = get_device(&channel->device_obj->device);
1123 vmbus_device_unregister(channel->device_obj);
1126 } else if (channel->primary_channel != NULL) {
1128 * Sub-channel is being rescinded. Following is the channel
1129 * close sequence when initiated from the driveri (refer to
1130 * vmbus_close() for details):
1131 * 1. Close all sub-channels first
1132 * 2. Then close the primary channel.
1134 mutex_lock(&vmbus_connection.channel_mutex);
1135 if (channel->state == CHANNEL_OPEN_STATE) {
1137 * The channel is currently not open;
1138 * it is safe for us to cleanup the channel.
1140 hv_process_channel_removal(channel);
1142 complete(&channel->rescind_event);
1144 mutex_unlock(&vmbus_connection.channel_mutex);
1147 /* The "channel" may have been freed. Do not access it any longer. */
1149 if (clean_up_chan_for_suspend)
1150 check_ready_for_suspend_event();
1153 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
1155 BUG_ON(!is_hvsock_channel(channel));
1157 /* We always get a rescind msg when a connection is closed. */
1158 while (!READ_ONCE(channel->probe_done) || !READ_ONCE(channel->rescind))
1161 vmbus_device_unregister(channel->device_obj);
1163 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
1167 * vmbus_onoffers_delivered -
1168 * This is invoked when all offers have been delivered.
1170 * Nothing to do here.
1172 static void vmbus_onoffers_delivered(
1173 struct vmbus_channel_message_header *hdr)
1178 * vmbus_onopen_result - Open result handler.
1180 * This is invoked when we received a response to our channel open request.
1181 * Find the matching request, copy the response and signal the requesting
1184 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
1186 struct vmbus_channel_open_result *result;
1187 struct vmbus_channel_msginfo *msginfo;
1188 struct vmbus_channel_message_header *requestheader;
1189 struct vmbus_channel_open_channel *openmsg;
1190 unsigned long flags;
1192 result = (struct vmbus_channel_open_result *)hdr;
1194 trace_vmbus_onopen_result(result);
1197 * Find the open msg, copy the result and signal/unblock the wait event
1199 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1201 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1204 (struct vmbus_channel_message_header *)msginfo->msg;
1206 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
1208 (struct vmbus_channel_open_channel *)msginfo->msg;
1209 if (openmsg->child_relid == result->child_relid &&
1210 openmsg->openid == result->openid) {
1211 memcpy(&msginfo->response.open_result,
1214 struct vmbus_channel_open_result));
1215 complete(&msginfo->waitevent);
1220 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1224 * vmbus_ongpadl_created - GPADL created handler.
1226 * This is invoked when we received a response to our gpadl create request.
1227 * Find the matching request, copy the response and signal the requesting
1230 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
1232 struct vmbus_channel_gpadl_created *gpadlcreated;
1233 struct vmbus_channel_msginfo *msginfo;
1234 struct vmbus_channel_message_header *requestheader;
1235 struct vmbus_channel_gpadl_header *gpadlheader;
1236 unsigned long flags;
1238 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
1240 trace_vmbus_ongpadl_created(gpadlcreated);
1243 * Find the establish msg, copy the result and signal/unblock the wait
1246 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1248 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1251 (struct vmbus_channel_message_header *)msginfo->msg;
1253 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
1255 (struct vmbus_channel_gpadl_header *)requestheader;
1257 if ((gpadlcreated->child_relid ==
1258 gpadlheader->child_relid) &&
1259 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1260 memcpy(&msginfo->response.gpadl_created,
1263 struct vmbus_channel_gpadl_created));
1264 complete(&msginfo->waitevent);
1269 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1273 * vmbus_ongpadl_torndown - GPADL torndown handler.
1275 * This is invoked when we received a response to our gpadl teardown request.
1276 * Find the matching request, copy the response and signal the requesting
1279 static void vmbus_ongpadl_torndown(
1280 struct vmbus_channel_message_header *hdr)
1282 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1283 struct vmbus_channel_msginfo *msginfo;
1284 struct vmbus_channel_message_header *requestheader;
1285 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1286 unsigned long flags;
1288 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1290 trace_vmbus_ongpadl_torndown(gpadl_torndown);
1293 * Find the open msg, copy the result and signal/unblock the wait event
1295 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1297 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1300 (struct vmbus_channel_message_header *)msginfo->msg;
1302 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1304 (struct vmbus_channel_gpadl_teardown *)requestheader;
1306 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1307 memcpy(&msginfo->response.gpadl_torndown,
1310 struct vmbus_channel_gpadl_torndown));
1311 complete(&msginfo->waitevent);
1316 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1320 * vmbus_onversion_response - Version response handler
1322 * This is invoked when we received a response to our initiate contact request.
1323 * Find the matching request, copy the response and signal the requesting
1326 static void vmbus_onversion_response(
1327 struct vmbus_channel_message_header *hdr)
1329 struct vmbus_channel_msginfo *msginfo;
1330 struct vmbus_channel_message_header *requestheader;
1331 struct vmbus_channel_version_response *version_response;
1332 unsigned long flags;
1334 version_response = (struct vmbus_channel_version_response *)hdr;
1336 trace_vmbus_onversion_response(version_response);
1338 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1340 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1343 (struct vmbus_channel_message_header *)msginfo->msg;
1345 if (requestheader->msgtype ==
1346 CHANNELMSG_INITIATE_CONTACT) {
1347 memcpy(&msginfo->response.version_response,
1349 sizeof(struct vmbus_channel_version_response));
1350 complete(&msginfo->waitevent);
1353 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1356 /* Channel message dispatch table */
1357 const struct vmbus_channel_message_table_entry
1358 channel_message_table[CHANNELMSG_COUNT] = {
1359 { CHANNELMSG_INVALID, 0, NULL, 0},
1360 { CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer,
1361 sizeof(struct vmbus_channel_offer_channel)},
1362 { CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind,
1363 sizeof(struct vmbus_channel_rescind_offer) },
1364 { CHANNELMSG_REQUESTOFFERS, 0, NULL, 0},
1365 { CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered, 0},
1366 { CHANNELMSG_OPENCHANNEL, 0, NULL, 0},
1367 { CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result,
1368 sizeof(struct vmbus_channel_open_result)},
1369 { CHANNELMSG_CLOSECHANNEL, 0, NULL, 0},
1370 { CHANNELMSG_GPADL_HEADER, 0, NULL, 0},
1371 { CHANNELMSG_GPADL_BODY, 0, NULL, 0},
1372 { CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created,
1373 sizeof(struct vmbus_channel_gpadl_created)},
1374 { CHANNELMSG_GPADL_TEARDOWN, 0, NULL, 0},
1375 { CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown,
1376 sizeof(struct vmbus_channel_gpadl_torndown) },
1377 { CHANNELMSG_RELID_RELEASED, 0, NULL, 0},
1378 { CHANNELMSG_INITIATE_CONTACT, 0, NULL, 0},
1379 { CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response,
1380 sizeof(struct vmbus_channel_version_response)},
1381 { CHANNELMSG_UNLOAD, 0, NULL, 0},
1382 { CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response, 0},
1383 { CHANNELMSG_18, 0, NULL, 0},
1384 { CHANNELMSG_19, 0, NULL, 0},
1385 { CHANNELMSG_20, 0, NULL, 0},
1386 { CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL, 0},
1387 { CHANNELMSG_MODIFYCHANNEL, 0, NULL, 0},
1388 { CHANNELMSG_TL_CONNECT_RESULT, 0, NULL, 0},
1392 * vmbus_onmessage - Handler for channel protocol messages.
1394 * This is invoked in the vmbus worker thread context.
1396 void vmbus_onmessage(struct vmbus_channel_message_header *hdr)
1398 trace_vmbus_on_message(hdr);
1401 * vmbus_on_msg_dpc() makes sure the hdr->msgtype here can not go
1402 * out of bound and the message_handler pointer can not be NULL.
1404 channel_message_table[hdr->msgtype].message_handler(hdr);
1408 * vmbus_request_offers - Send a request to get all our pending offers.
1410 int vmbus_request_offers(void)
1412 struct vmbus_channel_message_header *msg;
1413 struct vmbus_channel_msginfo *msginfo;
1416 msginfo = kmalloc(sizeof(*msginfo) +
1417 sizeof(struct vmbus_channel_message_header),
1422 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1424 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1426 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1429 trace_vmbus_request_offers(ret);
1432 pr_err("Unable to request offers - %d\n", ret);
1443 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1445 struct list_head *cur, *tmp;
1446 struct vmbus_channel *cur_channel;
1448 if (primary_channel->sc_creation_callback == NULL)
1451 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1452 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1454 primary_channel->sc_creation_callback(cur_channel);
1458 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1459 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1461 primary_channel->sc_creation_callback = sc_cr_cb;
1463 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1465 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1469 ret = !list_empty(&primary->sc_list);
1473 * Invoke the callback on sub-channel creation.
1474 * This will present a uniform interface to the
1477 invoke_sc_cb(primary);
1482 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1484 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1485 void (*chn_rescind_cb)(struct vmbus_channel *))
1487 channel->chn_rescind_callback = chn_rescind_cb;
1489 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);