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[] = {
34 .allowed_in_isolated = false,
38 { .dev_type = HV_SCSI,
41 .allowed_in_isolated = true,
48 .allowed_in_isolated = false,
55 .allowed_in_isolated = true,
62 .allowed_in_isolated = false,
66 { .dev_type = HV_PCIE,
69 .allowed_in_isolated = false,
72 /* Synthetic Frame Buffer */
76 .allowed_in_isolated = false,
79 /* Synthetic Keyboard */
83 .allowed_in_isolated = false,
87 { .dev_type = HV_MOUSE,
90 .allowed_in_isolated = false,
97 .allowed_in_isolated = false,
103 .perf_device = false,
104 .allowed_in_isolated = true,
110 .perf_device = false,
111 .allowed_in_isolated = true,
115 { .dev_type = HV_SHUTDOWN,
117 .perf_device = false,
118 .allowed_in_isolated = true,
122 { .dev_type = HV_FCOPY,
124 .perf_device = false,
125 .allowed_in_isolated = false,
129 { .dev_type = HV_BACKUP,
131 .perf_device = false,
132 .allowed_in_isolated = false,
138 .perf_device = false,
139 .allowed_in_isolated = false,
143 { .dev_type = HV_UNKNOWN,
144 .perf_device = false,
145 .allowed_in_isolated = false,
149 static const struct {
151 } vmbus_unsupported_devs[] = {
158 * The rescinded channel may be blocked waiting for a response from the host;
161 static void vmbus_rescind_cleanup(struct vmbus_channel *channel)
163 struct vmbus_channel_msginfo *msginfo;
167 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
168 channel->rescind = true;
169 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
172 if (msginfo->waiting_channel == channel) {
173 complete(&msginfo->waitevent);
177 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
180 static bool is_unsupported_vmbus_devs(const guid_t *guid)
184 for (i = 0; i < ARRAY_SIZE(vmbus_unsupported_devs); i++)
185 if (guid_equal(guid, &vmbus_unsupported_devs[i].guid))
190 static u16 hv_get_dev_type(const struct vmbus_channel *channel)
192 const guid_t *guid = &channel->offermsg.offer.if_type;
195 if (is_hvsock_channel(channel) || is_unsupported_vmbus_devs(guid))
198 for (i = HV_IDE; i < HV_UNKNOWN; i++) {
199 if (guid_equal(guid, &vmbus_devs[i].guid))
202 pr_info("Unknown GUID: %pUl\n", guid);
207 * vmbus_prep_negotiate_resp() - Create default response for Negotiate message
208 * @icmsghdrp: Pointer to msg header structure
209 * @buf: Raw buffer channel data
210 * @buflen: Length of the raw buffer channel data.
211 * @fw_version: The framework versions we can support.
212 * @fw_vercnt: The size of @fw_version.
213 * @srv_version: The service versions we can support.
214 * @srv_vercnt: The size of @srv_version.
215 * @nego_fw_version: The selected framework version.
216 * @nego_srv_version: The selected service version.
218 * Note: Versions are given in decreasing order.
220 * Set up and fill in default negotiate response message.
221 * Mainly used by Hyper-V drivers.
223 bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, u8 *buf,
224 u32 buflen, const int *fw_version, int fw_vercnt,
225 const int *srv_version, int srv_vercnt,
226 int *nego_fw_version, int *nego_srv_version)
228 int icframe_major, icframe_minor;
229 int icmsg_major, icmsg_minor;
230 int fw_major, fw_minor;
231 int srv_major, srv_minor;
233 bool found_match = false;
234 struct icmsg_negotiate *negop;
236 /* Check that there's enough space for icframe_vercnt, icmsg_vercnt */
237 if (buflen < ICMSG_HDR + offsetof(struct icmsg_negotiate, reserved)) {
238 pr_err_ratelimited("Invalid icmsg negotiate\n");
242 icmsghdrp->icmsgsize = 0x10;
243 negop = (struct icmsg_negotiate *)&buf[ICMSG_HDR];
245 icframe_major = negop->icframe_vercnt;
248 icmsg_major = negop->icmsg_vercnt;
251 /* Validate negop packet */
252 if (icframe_major > IC_VERSION_NEGOTIATION_MAX_VER_COUNT ||
253 icmsg_major > IC_VERSION_NEGOTIATION_MAX_VER_COUNT ||
254 ICMSG_NEGOTIATE_PKT_SIZE(icframe_major, icmsg_major) > buflen) {
255 pr_err_ratelimited("Invalid icmsg negotiate - icframe_major: %u, icmsg_major: %u\n",
256 icframe_major, icmsg_major);
261 * Select the framework version number we will
265 for (i = 0; i < fw_vercnt; i++) {
266 fw_major = (fw_version[i] >> 16);
267 fw_minor = (fw_version[i] & 0xFFFF);
269 for (j = 0; j < negop->icframe_vercnt; j++) {
270 if ((negop->icversion_data[j].major == fw_major) &&
271 (negop->icversion_data[j].minor == fw_minor)) {
272 icframe_major = negop->icversion_data[j].major;
273 icframe_minor = negop->icversion_data[j].minor;
288 for (i = 0; i < srv_vercnt; i++) {
289 srv_major = (srv_version[i] >> 16);
290 srv_minor = (srv_version[i] & 0xFFFF);
292 for (j = negop->icframe_vercnt;
293 (j < negop->icframe_vercnt + negop->icmsg_vercnt);
296 if ((negop->icversion_data[j].major == srv_major) &&
297 (negop->icversion_data[j].minor == srv_minor)) {
299 icmsg_major = negop->icversion_data[j].major;
300 icmsg_minor = negop->icversion_data[j].minor;
311 * Respond with the framework and service
312 * version numbers we can support.
317 negop->icframe_vercnt = 0;
318 negop->icmsg_vercnt = 0;
320 negop->icframe_vercnt = 1;
321 negop->icmsg_vercnt = 1;
325 *nego_fw_version = (icframe_major << 16) | icframe_minor;
327 if (nego_srv_version)
328 *nego_srv_version = (icmsg_major << 16) | icmsg_minor;
330 negop->icversion_data[0].major = icframe_major;
331 negop->icversion_data[0].minor = icframe_minor;
332 negop->icversion_data[1].major = icmsg_major;
333 negop->icversion_data[1].minor = icmsg_minor;
336 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp);
339 * alloc_channel - Allocate and initialize a vmbus channel object
341 static struct vmbus_channel *alloc_channel(void)
343 struct vmbus_channel *channel;
345 channel = kzalloc(sizeof(*channel), GFP_ATOMIC);
349 spin_lock_init(&channel->sched_lock);
350 init_completion(&channel->rescind_event);
352 INIT_LIST_HEAD(&channel->sc_list);
354 tasklet_init(&channel->callback_event,
355 vmbus_on_event, (unsigned long)channel);
357 hv_ringbuffer_pre_init(channel);
363 * free_channel - Release the resources used by the vmbus channel object
365 static void free_channel(struct vmbus_channel *channel)
367 tasklet_kill(&channel->callback_event);
368 vmbus_remove_channel_attr_group(channel);
370 kobject_put(&channel->kobj);
373 void vmbus_channel_map_relid(struct vmbus_channel *channel)
375 if (WARN_ON(channel->offermsg.child_relid >= MAX_CHANNEL_RELIDS))
378 * The mapping of the channel's relid is visible from the CPUs that
379 * execute vmbus_chan_sched() by the time that vmbus_chan_sched() will
382 * (a) In the "normal (i.e., not resuming from hibernation)" path,
383 * the full barrier in smp_store_mb() guarantees that the store
384 * is propagated to all CPUs before the add_channel_work work
385 * is queued. In turn, add_channel_work is queued before the
386 * channel's ring buffer is allocated/initialized and the
387 * OPENCHANNEL message for the channel is sent in vmbus_open().
388 * Hyper-V won't start sending the interrupts for the channel
389 * before the OPENCHANNEL message is acked. The memory barrier
390 * in vmbus_chan_sched() -> sync_test_and_clear_bit() ensures
391 * that vmbus_chan_sched() must find the channel's relid in
392 * recv_int_page before retrieving the channel pointer from the
395 * (b) In the "resuming from hibernation" path, the smp_store_mb()
396 * guarantees that the store is propagated to all CPUs before
397 * the VMBus connection is marked as ready for the resume event
398 * (cf. check_ready_for_resume_event()). The interrupt handler
399 * of the VMBus driver and vmbus_chan_sched() can not run before
400 * vmbus_bus_resume() has completed execution (cf. resume_noirq).
403 vmbus_connection.channels[channel->offermsg.child_relid],
407 void vmbus_channel_unmap_relid(struct vmbus_channel *channel)
409 if (WARN_ON(channel->offermsg.child_relid >= MAX_CHANNEL_RELIDS))
412 vmbus_connection.channels[channel->offermsg.child_relid],
416 static void vmbus_release_relid(u32 relid)
418 struct vmbus_channel_relid_released msg;
421 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
422 msg.child_relid = relid;
423 msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
424 ret = vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released),
427 trace_vmbus_release_relid(&msg, ret);
430 void hv_process_channel_removal(struct vmbus_channel *channel)
432 lockdep_assert_held(&vmbus_connection.channel_mutex);
433 BUG_ON(!channel->rescind);
436 * hv_process_channel_removal() could find INVALID_RELID only for
437 * hv_sock channels. See the inline comments in vmbus_onoffer().
439 WARN_ON(channel->offermsg.child_relid == INVALID_RELID &&
440 !is_hvsock_channel(channel));
443 * Upon suspend, an in-use hv_sock channel is removed from the array of
444 * channels and the relid is invalidated. After hibernation, when the
445 * user-space appplication destroys the channel, it's unnecessary and
446 * unsafe to remove the channel from the array of channels. See also
447 * the inline comments before the call of vmbus_release_relid() below.
449 if (channel->offermsg.child_relid != INVALID_RELID)
450 vmbus_channel_unmap_relid(channel);
452 if (channel->primary_channel == NULL)
453 list_del(&channel->listentry);
455 list_del(&channel->sc_list);
458 * If this is a "perf" channel, updates the hv_numa_map[] masks so that
459 * init_vp_index() can (re-)use the CPU.
461 if (hv_is_perf_channel(channel))
462 hv_clear_alloced_cpu(channel->target_cpu);
465 * Upon suspend, an in-use hv_sock channel is marked as "rescinded" and
466 * the relid is invalidated; after hibernation, when the user-space app
467 * destroys the channel, the relid is INVALID_RELID, and in this case
468 * it's unnecessary and unsafe to release the old relid, since the same
469 * relid can refer to a completely different channel now.
471 if (channel->offermsg.child_relid != INVALID_RELID)
472 vmbus_release_relid(channel->offermsg.child_relid);
474 free_channel(channel);
477 void vmbus_free_channels(void)
479 struct vmbus_channel *channel, *tmp;
481 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
483 /* hv_process_channel_removal() needs this */
484 channel->rescind = true;
486 vmbus_device_unregister(channel->device_obj);
490 /* Note: the function can run concurrently for primary/sub channels. */
491 static void vmbus_add_channel_work(struct work_struct *work)
493 struct vmbus_channel *newchannel =
494 container_of(work, struct vmbus_channel, add_channel_work);
495 struct vmbus_channel *primary_channel = newchannel->primary_channel;
499 * This state is used to indicate a successful open
500 * so that when we do close the channel normally, we
501 * can cleanup properly.
503 newchannel->state = CHANNEL_OPEN_STATE;
505 if (primary_channel != NULL) {
506 /* newchannel is a sub-channel. */
507 struct hv_device *dev = primary_channel->device_obj;
509 if (vmbus_add_channel_kobj(dev, newchannel))
512 if (primary_channel->sc_creation_callback != NULL)
513 primary_channel->sc_creation_callback(newchannel);
515 newchannel->probe_done = true;
520 * Start the process of binding the primary channel to the driver
522 newchannel->device_obj = vmbus_device_create(
523 &newchannel->offermsg.offer.if_type,
524 &newchannel->offermsg.offer.if_instance,
526 if (!newchannel->device_obj)
529 newchannel->device_obj->device_id = newchannel->device_id;
531 * Add the new device to the bus. This will kick off device-driver
532 * binding which eventually invokes the device driver's AddDevice()
535 ret = vmbus_device_register(newchannel->device_obj);
538 pr_err("unable to add child device object (relid %d)\n",
539 newchannel->offermsg.child_relid);
540 kfree(newchannel->device_obj);
544 newchannel->probe_done = true;
548 mutex_lock(&vmbus_connection.channel_mutex);
551 * We need to set the flag, otherwise
552 * vmbus_onoffer_rescind() can be blocked.
554 newchannel->probe_done = true;
556 if (primary_channel == NULL)
557 list_del(&newchannel->listentry);
559 list_del(&newchannel->sc_list);
561 /* vmbus_process_offer() has mapped the channel. */
562 vmbus_channel_unmap_relid(newchannel);
564 mutex_unlock(&vmbus_connection.channel_mutex);
566 vmbus_release_relid(newchannel->offermsg.child_relid);
568 free_channel(newchannel);
572 * vmbus_process_offer - Process the offer by creating a channel/device
573 * associated with this offer
575 static void vmbus_process_offer(struct vmbus_channel *newchannel)
577 struct vmbus_channel *channel;
578 struct workqueue_struct *wq;
582 * Synchronize vmbus_process_offer() and CPU hotplugging:
586 * [vmbus_process_offer()] [Hot removal of the CPU]
588 * CPU_READ_LOCK CPUS_WRITE_LOCK
589 * LOAD cpu_online_mask SEARCH chn_list
590 * STORE target_cpu LOAD target_cpu
591 * INSERT chn_list STORE cpu_online_mask
592 * CPUS_READ_UNLOCK CPUS_WRITE_UNLOCK
594 * Forbids: CPU1's LOAD from *not* seing CPU2's STORE &&
595 * CPU2's SEARCH from *not* seeing CPU1's INSERT
597 * Forbids: CPU2's SEARCH from seeing CPU1's INSERT &&
598 * CPU2's LOAD from *not* seing CPU1's STORE
603 * Serializes the modifications of the chn_list list as well as
604 * the accesses to next_numa_node_id in init_vp_index().
606 mutex_lock(&vmbus_connection.channel_mutex);
608 init_vp_index(newchannel);
610 /* Remember the channels that should be cleaned up upon suspend. */
611 if (is_hvsock_channel(newchannel) || is_sub_channel(newchannel))
612 atomic_inc(&vmbus_connection.nr_chan_close_on_suspend);
615 * Now that we have acquired the channel_mutex,
616 * we can release the potentially racing rescind thread.
618 atomic_dec(&vmbus_connection.offer_in_progress);
620 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
621 if (guid_equal(&channel->offermsg.offer.if_type,
622 &newchannel->offermsg.offer.if_type) &&
623 guid_equal(&channel->offermsg.offer.if_instance,
624 &newchannel->offermsg.offer.if_instance)) {
631 list_add_tail(&newchannel->listentry,
632 &vmbus_connection.chn_list);
635 * Check to see if this is a valid sub-channel.
637 if (newchannel->offermsg.offer.sub_channel_index == 0) {
638 mutex_unlock(&vmbus_connection.channel_mutex);
640 * Don't call free_channel(), because newchannel->kobj
641 * is not initialized yet.
648 * Process the sub-channel.
650 newchannel->primary_channel = channel;
651 list_add_tail(&newchannel->sc_list, &channel->sc_list);
654 vmbus_channel_map_relid(newchannel);
656 mutex_unlock(&vmbus_connection.channel_mutex);
660 * vmbus_process_offer() mustn't call channel->sc_creation_callback()
661 * directly for sub-channels, because sc_creation_callback() ->
662 * vmbus_open() may never get the host's response to the
663 * OPEN_CHANNEL message (the host may rescind a channel at any time,
664 * e.g. in the case of hot removing a NIC), and vmbus_onoffer_rescind()
665 * may not wake up the vmbus_open() as it's blocked due to a non-zero
666 * vmbus_connection.offer_in_progress, and finally we have a deadlock.
668 * The above is also true for primary channels, if the related device
669 * drivers use sync probing mode by default.
671 * And, usually the handling of primary channels and sub-channels can
672 * depend on each other, so we should offload them to different
673 * workqueues to avoid possible deadlock, e.g. in sync-probing mode,
674 * NIC1's netvsc_subchan_work() can race with NIC2's netvsc_probe() ->
675 * rtnl_lock(), and causes deadlock: the former gets the rtnl_lock
676 * and waits for all the sub-channels to appear, but the latter
677 * can't get the rtnl_lock and this blocks the handling of
680 INIT_WORK(&newchannel->add_channel_work, vmbus_add_channel_work);
681 wq = fnew ? vmbus_connection.handle_primary_chan_wq :
682 vmbus_connection.handle_sub_chan_wq;
683 queue_work(wq, &newchannel->add_channel_work);
687 * We use this state to statically distribute the channel interrupt load.
689 static int next_numa_node_id;
692 * Starting with Win8, we can statically distribute the incoming
693 * channel interrupt load by binding a channel to VCPU.
695 * For pre-win8 hosts or non-performance critical channels we assign the
698 * Starting with win8, performance critical channels will be distributed
699 * evenly among all the available NUMA nodes. Once the node is assigned,
700 * we will assign the CPU based on a simple round robin scheme.
702 static void init_vp_index(struct vmbus_channel *channel)
704 bool perf_chn = hv_is_perf_channel(channel);
705 cpumask_var_t available_mask;
706 struct cpumask *alloced_mask;
710 if ((vmbus_proto_version == VERSION_WS2008) ||
711 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn) ||
712 !alloc_cpumask_var(&available_mask, GFP_KERNEL)) {
714 * Prior to win8, all channel interrupts are
715 * delivered on VMBUS_CONNECT_CPU.
716 * Also if the channel is not a performance critical
717 * channel, bind it to VMBUS_CONNECT_CPU.
718 * In case alloc_cpumask_var() fails, bind it to
721 channel->target_cpu = VMBUS_CONNECT_CPU;
723 hv_set_alloced_cpu(VMBUS_CONNECT_CPU);
728 numa_node = next_numa_node_id++;
729 if (numa_node == nr_node_ids) {
730 next_numa_node_id = 0;
733 if (cpumask_empty(cpumask_of_node(numa_node)))
737 alloced_mask = &hv_context.hv_numa_map[numa_node];
739 if (cpumask_weight(alloced_mask) ==
740 cpumask_weight(cpumask_of_node(numa_node))) {
742 * We have cycled through all the CPUs in the node;
743 * reset the alloced map.
745 cpumask_clear(alloced_mask);
748 cpumask_xor(available_mask, alloced_mask, cpumask_of_node(numa_node));
750 target_cpu = cpumask_first(available_mask);
751 cpumask_set_cpu(target_cpu, alloced_mask);
753 channel->target_cpu = target_cpu;
755 free_cpumask_var(available_mask);
758 static void vmbus_wait_for_unload(void)
762 struct hv_message *msg;
763 struct vmbus_channel_message_header *hdr;
767 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
768 * used for initial contact or to CPU0 depending on host version. When
769 * we're crashing on a different CPU let's hope that IRQ handler on
770 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
771 * functional and vmbus_unload_response() will complete
772 * vmbus_connection.unload_event. If not, the last thing we can do is
773 * read message pages for all CPUs directly.
775 * Wait no more than 10 seconds so that the panic path can't get
776 * hung forever in case the response message isn't seen.
778 for (i = 0; i < 1000; i++) {
779 if (completion_done(&vmbus_connection.unload_event))
782 for_each_online_cpu(cpu) {
783 struct hv_per_cpu_context *hv_cpu
784 = per_cpu_ptr(hv_context.cpu_context, cpu);
786 page_addr = hv_cpu->synic_message_page;
787 msg = (struct hv_message *)page_addr
788 + VMBUS_MESSAGE_SINT;
790 message_type = READ_ONCE(msg->header.message_type);
791 if (message_type == HVMSG_NONE)
794 hdr = (struct vmbus_channel_message_header *)
797 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
798 complete(&vmbus_connection.unload_event);
800 vmbus_signal_eom(msg, message_type);
807 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
808 * maybe-pending messages on all CPUs to be able to receive new
809 * messages after we reconnect.
811 for_each_online_cpu(cpu) {
812 struct hv_per_cpu_context *hv_cpu
813 = per_cpu_ptr(hv_context.cpu_context, cpu);
815 page_addr = hv_cpu->synic_message_page;
816 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
817 msg->header.message_type = HVMSG_NONE;
822 * vmbus_unload_response - Handler for the unload response.
824 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
827 * This is a global event; just wakeup the waiting thread.
828 * Once we successfully unload, we can cleanup the monitor state.
830 complete(&vmbus_connection.unload_event);
833 void vmbus_initiate_unload(bool crash)
835 struct vmbus_channel_message_header hdr;
837 if (xchg(&vmbus_connection.conn_state, DISCONNECTED) == DISCONNECTED)
840 /* Pre-Win2012R2 hosts don't support reconnect */
841 if (vmbus_proto_version < VERSION_WIN8_1)
844 init_completion(&vmbus_connection.unload_event);
845 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
846 hdr.msgtype = CHANNELMSG_UNLOAD;
847 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
851 * vmbus_initiate_unload() is also called on crash and the crash can be
852 * happening in an interrupt context, where scheduling is impossible.
855 wait_for_completion(&vmbus_connection.unload_event);
857 vmbus_wait_for_unload();
860 static void check_ready_for_resume_event(void)
863 * If all the old primary channels have been fixed up, then it's safe
866 if (atomic_dec_and_test(&vmbus_connection.nr_chan_fixup_on_resume))
867 complete(&vmbus_connection.ready_for_resume_event);
870 static void vmbus_setup_channel_state(struct vmbus_channel *channel,
871 struct vmbus_channel_offer_channel *offer)
874 * Setup state for signalling the host.
876 channel->sig_event = VMBUS_EVENT_CONNECTION_ID;
878 if (vmbus_proto_version != VERSION_WS2008) {
879 channel->is_dedicated_interrupt =
880 (offer->is_dedicated_interrupt != 0);
881 channel->sig_event = offer->connection_id;
884 memcpy(&channel->offermsg, offer,
885 sizeof(struct vmbus_channel_offer_channel));
886 channel->monitor_grp = (u8)offer->monitorid / 32;
887 channel->monitor_bit = (u8)offer->monitorid % 32;
888 channel->device_id = hv_get_dev_type(channel);
892 * find_primary_channel_by_offer - Get the channel object given the new offer.
893 * This is only used in the resume path of hibernation.
895 static struct vmbus_channel *
896 find_primary_channel_by_offer(const struct vmbus_channel_offer_channel *offer)
898 struct vmbus_channel *channel = NULL, *iter;
899 const guid_t *inst1, *inst2;
901 /* Ignore sub-channel offers. */
902 if (offer->offer.sub_channel_index != 0)
905 mutex_lock(&vmbus_connection.channel_mutex);
907 list_for_each_entry(iter, &vmbus_connection.chn_list, listentry) {
908 inst1 = &iter->offermsg.offer.if_instance;
909 inst2 = &offer->offer.if_instance;
911 if (guid_equal(inst1, inst2)) {
917 mutex_unlock(&vmbus_connection.channel_mutex);
922 static bool vmbus_is_valid_device(const guid_t *guid)
926 if (!hv_is_isolation_supported())
929 for (i = 0; i < ARRAY_SIZE(vmbus_devs); i++) {
930 if (guid_equal(guid, &vmbus_devs[i].guid))
931 return vmbus_devs[i].allowed_in_isolated;
937 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
940 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
942 struct vmbus_channel_offer_channel *offer;
943 struct vmbus_channel *oldchannel, *newchannel;
946 offer = (struct vmbus_channel_offer_channel *)hdr;
948 trace_vmbus_onoffer(offer);
950 if (!vmbus_is_valid_device(&offer->offer.if_type)) {
951 pr_err_ratelimited("Invalid offer %d from the host supporting isolation\n",
953 atomic_dec(&vmbus_connection.offer_in_progress);
957 oldchannel = find_primary_channel_by_offer(offer);
959 if (oldchannel != NULL) {
961 * We're resuming from hibernation: all the sub-channel and
962 * hv_sock channels we had before the hibernation should have
963 * been cleaned up, and now we must be seeing a re-offered
964 * primary channel that we had before the hibernation.
968 * { Initially: channel relid = INVALID_RELID,
969 * channels[valid_relid] = NULL }
973 * [vmbus_onoffer()] [vmbus_device_release()]
975 * LOCK channel_mutex LOCK channel_mutex
976 * STORE channel relid = valid_relid LOAD r1 = channel relid
977 * MAP_RELID channel if (r1 != INVALID_RELID)
978 * UNLOCK channel_mutex UNMAP_RELID channel
979 * UNLOCK channel_mutex
981 * Forbids: r1 == valid_relid &&
982 * channels[valid_relid] == channel
984 * Note. r1 can be INVALID_RELID only for an hv_sock channel.
985 * None of the hv_sock channels which were present before the
986 * suspend are re-offered upon the resume. See the WARN_ON()
987 * in hv_process_channel_removal().
989 mutex_lock(&vmbus_connection.channel_mutex);
991 atomic_dec(&vmbus_connection.offer_in_progress);
993 WARN_ON(oldchannel->offermsg.child_relid != INVALID_RELID);
994 /* Fix up the relid. */
995 oldchannel->offermsg.child_relid = offer->child_relid;
997 offer_sz = sizeof(*offer);
998 if (memcmp(offer, &oldchannel->offermsg, offer_sz) != 0) {
1000 * This is not an error, since the host can also change
1001 * the other field(s) of the offer, e.g. on WS RS5
1002 * (Build 17763), the offer->connection_id of the
1003 * Mellanox VF vmbus device can change when the host
1004 * reoffers the device upon resume.
1006 pr_debug("vmbus offer changed: relid=%d\n",
1007 offer->child_relid);
1009 print_hex_dump_debug("Old vmbus offer: ",
1010 DUMP_PREFIX_OFFSET, 16, 4,
1011 &oldchannel->offermsg, offer_sz,
1013 print_hex_dump_debug("New vmbus offer: ",
1014 DUMP_PREFIX_OFFSET, 16, 4,
1015 offer, offer_sz, false);
1017 /* Fix up the old channel. */
1018 vmbus_setup_channel_state(oldchannel, offer);
1021 /* Add the channel back to the array of channels. */
1022 vmbus_channel_map_relid(oldchannel);
1023 check_ready_for_resume_event();
1025 mutex_unlock(&vmbus_connection.channel_mutex);
1029 /* Allocate the channel object and save this offer. */
1030 newchannel = alloc_channel();
1032 vmbus_release_relid(offer->child_relid);
1033 atomic_dec(&vmbus_connection.offer_in_progress);
1034 pr_err("Unable to allocate channel object\n");
1038 vmbus_setup_channel_state(newchannel, offer);
1040 vmbus_process_offer(newchannel);
1043 static void check_ready_for_suspend_event(void)
1046 * If all the sub-channels or hv_sock channels have been cleaned up,
1047 * then it's safe to suspend.
1049 if (atomic_dec_and_test(&vmbus_connection.nr_chan_close_on_suspend))
1050 complete(&vmbus_connection.ready_for_suspend_event);
1054 * vmbus_onoffer_rescind - Rescind offer handler.
1056 * We queue a work item to process this offer synchronously
1058 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
1060 struct vmbus_channel_rescind_offer *rescind;
1061 struct vmbus_channel *channel;
1063 bool clean_up_chan_for_suspend;
1065 rescind = (struct vmbus_channel_rescind_offer *)hdr;
1067 trace_vmbus_onoffer_rescind(rescind);
1070 * The offer msg and the corresponding rescind msg
1071 * from the host are guranteed to be ordered -
1072 * offer comes in first and then the rescind.
1073 * Since we process these events in work elements,
1074 * and with preemption, we may end up processing
1075 * the events out of order. We rely on the synchronization
1076 * provided by offer_in_progress and by channel_mutex for
1077 * ordering these events:
1079 * { Initially: offer_in_progress = 1 }
1083 * [vmbus_onoffer()] [vmbus_onoffer_rescind()]
1085 * LOCK channel_mutex WAIT_ON offer_in_progress == 0
1086 * DECREMENT offer_in_progress LOCK channel_mutex
1087 * STORE channels[] LOAD channels[]
1088 * UNLOCK channel_mutex UNLOCK channel_mutex
1090 * Forbids: CPU2's LOAD from *not* seeing CPU1's STORE
1093 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
1095 * We wait here until any channel offer is currently
1101 mutex_lock(&vmbus_connection.channel_mutex);
1102 channel = relid2channel(rescind->child_relid);
1103 if (channel != NULL) {
1105 * Guarantee that no other instance of vmbus_onoffer_rescind()
1106 * has got a reference to the channel object. Synchronize on
1107 * &vmbus_connection.channel_mutex.
1109 if (channel->rescind_ref) {
1110 mutex_unlock(&vmbus_connection.channel_mutex);
1113 channel->rescind_ref = true;
1115 mutex_unlock(&vmbus_connection.channel_mutex);
1117 if (channel == NULL) {
1119 * We failed in processing the offer message;
1120 * we would have cleaned up the relid in that
1126 clean_up_chan_for_suspend = is_hvsock_channel(channel) ||
1127 is_sub_channel(channel);
1129 * Before setting channel->rescind in vmbus_rescind_cleanup(), we
1130 * should make sure the channel callback is not running any more.
1132 vmbus_reset_channel_cb(channel);
1135 * Now wait for offer handling to complete.
1137 vmbus_rescind_cleanup(channel);
1138 while (READ_ONCE(channel->probe_done) == false) {
1140 * We wait here until any channel offer is currently
1147 * At this point, the rescind handling can proceed safely.
1150 if (channel->device_obj) {
1151 if (channel->chn_rescind_callback) {
1152 channel->chn_rescind_callback(channel);
1154 if (clean_up_chan_for_suspend)
1155 check_ready_for_suspend_event();
1160 * We will have to unregister this device from the
1163 dev = get_device(&channel->device_obj->device);
1165 vmbus_device_unregister(channel->device_obj);
1168 } else if (channel->primary_channel != NULL) {
1170 * Sub-channel is being rescinded. Following is the channel
1171 * close sequence when initiated from the driveri (refer to
1172 * vmbus_close() for details):
1173 * 1. Close all sub-channels first
1174 * 2. Then close the primary channel.
1176 mutex_lock(&vmbus_connection.channel_mutex);
1177 if (channel->state == CHANNEL_OPEN_STATE) {
1179 * The channel is currently not open;
1180 * it is safe for us to cleanup the channel.
1182 hv_process_channel_removal(channel);
1184 complete(&channel->rescind_event);
1186 mutex_unlock(&vmbus_connection.channel_mutex);
1189 /* The "channel" may have been freed. Do not access it any longer. */
1191 if (clean_up_chan_for_suspend)
1192 check_ready_for_suspend_event();
1195 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
1197 BUG_ON(!is_hvsock_channel(channel));
1199 /* We always get a rescind msg when a connection is closed. */
1200 while (!READ_ONCE(channel->probe_done) || !READ_ONCE(channel->rescind))
1203 vmbus_device_unregister(channel->device_obj);
1205 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
1209 * vmbus_onoffers_delivered -
1210 * This is invoked when all offers have been delivered.
1212 * Nothing to do here.
1214 static void vmbus_onoffers_delivered(
1215 struct vmbus_channel_message_header *hdr)
1220 * vmbus_onopen_result - Open result handler.
1222 * This is invoked when we received a response to our channel open request.
1223 * Find the matching request, copy the response and signal the requesting
1226 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
1228 struct vmbus_channel_open_result *result;
1229 struct vmbus_channel_msginfo *msginfo;
1230 struct vmbus_channel_message_header *requestheader;
1231 struct vmbus_channel_open_channel *openmsg;
1232 unsigned long flags;
1234 result = (struct vmbus_channel_open_result *)hdr;
1236 trace_vmbus_onopen_result(result);
1239 * Find the open msg, copy the result and signal/unblock the wait event
1241 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1243 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1246 (struct vmbus_channel_message_header *)msginfo->msg;
1248 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
1250 (struct vmbus_channel_open_channel *)msginfo->msg;
1251 if (openmsg->child_relid == result->child_relid &&
1252 openmsg->openid == result->openid) {
1253 memcpy(&msginfo->response.open_result,
1256 struct vmbus_channel_open_result));
1257 complete(&msginfo->waitevent);
1262 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1266 * vmbus_ongpadl_created - GPADL created handler.
1268 * This is invoked when we received a response to our gpadl create request.
1269 * Find the matching request, copy the response and signal the requesting
1272 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
1274 struct vmbus_channel_gpadl_created *gpadlcreated;
1275 struct vmbus_channel_msginfo *msginfo;
1276 struct vmbus_channel_message_header *requestheader;
1277 struct vmbus_channel_gpadl_header *gpadlheader;
1278 unsigned long flags;
1280 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
1282 trace_vmbus_ongpadl_created(gpadlcreated);
1285 * Find the establish msg, copy the result and signal/unblock the wait
1288 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1290 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1293 (struct vmbus_channel_message_header *)msginfo->msg;
1295 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
1297 (struct vmbus_channel_gpadl_header *)requestheader;
1299 if ((gpadlcreated->child_relid ==
1300 gpadlheader->child_relid) &&
1301 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1302 memcpy(&msginfo->response.gpadl_created,
1305 struct vmbus_channel_gpadl_created));
1306 complete(&msginfo->waitevent);
1311 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1315 * vmbus_onmodifychannel_response - Modify Channel response handler.
1317 * This is invoked when we received a response to our channel modify request.
1318 * Find the matching request, copy the response and signal the requesting thread.
1320 static void vmbus_onmodifychannel_response(struct vmbus_channel_message_header *hdr)
1322 struct vmbus_channel_modifychannel_response *response;
1323 struct vmbus_channel_msginfo *msginfo;
1324 unsigned long flags;
1326 response = (struct vmbus_channel_modifychannel_response *)hdr;
1328 trace_vmbus_onmodifychannel_response(response);
1331 * Find the modify msg, copy the response and signal/unblock the wait event.
1333 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1335 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list, msglistentry) {
1336 struct vmbus_channel_message_header *responseheader =
1337 (struct vmbus_channel_message_header *)msginfo->msg;
1339 if (responseheader->msgtype == CHANNELMSG_MODIFYCHANNEL) {
1340 struct vmbus_channel_modifychannel *modifymsg;
1342 modifymsg = (struct vmbus_channel_modifychannel *)msginfo->msg;
1343 if (modifymsg->child_relid == response->child_relid) {
1344 memcpy(&msginfo->response.modify_response, response,
1346 complete(&msginfo->waitevent);
1351 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1355 * vmbus_ongpadl_torndown - GPADL torndown handler.
1357 * This is invoked when we received a response to our gpadl teardown request.
1358 * Find the matching request, copy the response and signal the requesting
1361 static void vmbus_ongpadl_torndown(
1362 struct vmbus_channel_message_header *hdr)
1364 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1365 struct vmbus_channel_msginfo *msginfo;
1366 struct vmbus_channel_message_header *requestheader;
1367 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1368 unsigned long flags;
1370 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1372 trace_vmbus_ongpadl_torndown(gpadl_torndown);
1375 * Find the open msg, copy the result and signal/unblock the wait event
1377 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1379 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1382 (struct vmbus_channel_message_header *)msginfo->msg;
1384 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1386 (struct vmbus_channel_gpadl_teardown *)requestheader;
1388 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1389 memcpy(&msginfo->response.gpadl_torndown,
1392 struct vmbus_channel_gpadl_torndown));
1393 complete(&msginfo->waitevent);
1398 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1402 * vmbus_onversion_response - Version response handler
1404 * This is invoked when we received a response to our initiate contact request.
1405 * Find the matching request, copy the response and signal the requesting
1408 static void vmbus_onversion_response(
1409 struct vmbus_channel_message_header *hdr)
1411 struct vmbus_channel_msginfo *msginfo;
1412 struct vmbus_channel_message_header *requestheader;
1413 struct vmbus_channel_version_response *version_response;
1414 unsigned long flags;
1416 version_response = (struct vmbus_channel_version_response *)hdr;
1418 trace_vmbus_onversion_response(version_response);
1420 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1422 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1425 (struct vmbus_channel_message_header *)msginfo->msg;
1427 if (requestheader->msgtype ==
1428 CHANNELMSG_INITIATE_CONTACT) {
1429 memcpy(&msginfo->response.version_response,
1431 sizeof(struct vmbus_channel_version_response));
1432 complete(&msginfo->waitevent);
1435 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1438 /* Channel message dispatch table */
1439 const struct vmbus_channel_message_table_entry
1440 channel_message_table[CHANNELMSG_COUNT] = {
1441 { CHANNELMSG_INVALID, 0, NULL, 0},
1442 { CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer,
1443 sizeof(struct vmbus_channel_offer_channel)},
1444 { CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind,
1445 sizeof(struct vmbus_channel_rescind_offer) },
1446 { CHANNELMSG_REQUESTOFFERS, 0, NULL, 0},
1447 { CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered, 0},
1448 { CHANNELMSG_OPENCHANNEL, 0, NULL, 0},
1449 { CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result,
1450 sizeof(struct vmbus_channel_open_result)},
1451 { CHANNELMSG_CLOSECHANNEL, 0, NULL, 0},
1452 { CHANNELMSG_GPADL_HEADER, 0, NULL, 0},
1453 { CHANNELMSG_GPADL_BODY, 0, NULL, 0},
1454 { CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created,
1455 sizeof(struct vmbus_channel_gpadl_created)},
1456 { CHANNELMSG_GPADL_TEARDOWN, 0, NULL, 0},
1457 { CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown,
1458 sizeof(struct vmbus_channel_gpadl_torndown) },
1459 { CHANNELMSG_RELID_RELEASED, 0, NULL, 0},
1460 { CHANNELMSG_INITIATE_CONTACT, 0, NULL, 0},
1461 { CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response,
1462 sizeof(struct vmbus_channel_version_response)},
1463 { CHANNELMSG_UNLOAD, 0, NULL, 0},
1464 { CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response, 0},
1465 { CHANNELMSG_18, 0, NULL, 0},
1466 { CHANNELMSG_19, 0, NULL, 0},
1467 { CHANNELMSG_20, 0, NULL, 0},
1468 { CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL, 0},
1469 { CHANNELMSG_MODIFYCHANNEL, 0, NULL, 0},
1470 { CHANNELMSG_TL_CONNECT_RESULT, 0, NULL, 0},
1471 { CHANNELMSG_MODIFYCHANNEL_RESPONSE, 1, vmbus_onmodifychannel_response,
1472 sizeof(struct vmbus_channel_modifychannel_response)},
1476 * vmbus_onmessage - Handler for channel protocol messages.
1478 * This is invoked in the vmbus worker thread context.
1480 void vmbus_onmessage(struct vmbus_channel_message_header *hdr)
1482 trace_vmbus_on_message(hdr);
1485 * vmbus_on_msg_dpc() makes sure the hdr->msgtype here can not go
1486 * out of bound and the message_handler pointer can not be NULL.
1488 channel_message_table[hdr->msgtype].message_handler(hdr);
1492 * vmbus_request_offers - Send a request to get all our pending offers.
1494 int vmbus_request_offers(void)
1496 struct vmbus_channel_message_header *msg;
1497 struct vmbus_channel_msginfo *msginfo;
1500 msginfo = kmalloc(sizeof(*msginfo) +
1501 sizeof(struct vmbus_channel_message_header),
1506 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1508 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1510 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1513 trace_vmbus_request_offers(ret);
1516 pr_err("Unable to request offers - %d\n", ret);
1527 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1529 struct list_head *cur, *tmp;
1530 struct vmbus_channel *cur_channel;
1532 if (primary_channel->sc_creation_callback == NULL)
1535 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1536 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1538 primary_channel->sc_creation_callback(cur_channel);
1542 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1543 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1545 primary_channel->sc_creation_callback = sc_cr_cb;
1547 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1549 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1553 ret = !list_empty(&primary->sc_list);
1557 * Invoke the callback on sub-channel creation.
1558 * This will present a uniform interface to the
1561 invoke_sc_cb(primary);
1566 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1568 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1569 void (*chn_rescind_cb)(struct vmbus_channel *))
1571 channel->chn_rescind_callback = chn_rescind_cb;
1573 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);