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141 * The rescinded channel may be blocked waiting for a response from the host;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 smp_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 smp_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);
610 * Don't call free_channel(), because newchannel->kobj
611 * is not initialized yet.
618 * Process the sub-channel.
620 newchannel->primary_channel = channel;
621 list_add_tail(&newchannel->sc_list, &channel->sc_list);
624 vmbus_channel_map_relid(newchannel);
626 mutex_unlock(&vmbus_connection.channel_mutex);
630 * vmbus_process_offer() mustn't call channel->sc_creation_callback()
631 * directly for sub-channels, because sc_creation_callback() ->
632 * vmbus_open() may never get the host's response to the
633 * OPEN_CHANNEL message (the host may rescind a channel at any time,
634 * e.g. in the case of hot removing a NIC), and vmbus_onoffer_rescind()
635 * may not wake up the vmbus_open() as it's blocked due to a non-zero
636 * vmbus_connection.offer_in_progress, and finally we have a deadlock.
638 * The above is also true for primary channels, if the related device
639 * drivers use sync probing mode by default.
641 * And, usually the handling of primary channels and sub-channels can
642 * depend on each other, so we should offload them to different
643 * workqueues to avoid possible deadlock, e.g. in sync-probing mode,
644 * NIC1's netvsc_subchan_work() can race with NIC2's netvsc_probe() ->
645 * rtnl_lock(), and causes deadlock: the former gets the rtnl_lock
646 * and waits for all the sub-channels to appear, but the latter
647 * can't get the rtnl_lock and this blocks the handling of
650 INIT_WORK(&newchannel->add_channel_work, vmbus_add_channel_work);
651 wq = fnew ? vmbus_connection.handle_primary_chan_wq :
652 vmbus_connection.handle_sub_chan_wq;
653 queue_work(wq, &newchannel->add_channel_work);
657 * We use this state to statically distribute the channel interrupt load.
659 static int next_numa_node_id;
662 * Starting with Win8, we can statically distribute the incoming
663 * channel interrupt load by binding a channel to VCPU.
665 * For pre-win8 hosts or non-performance critical channels we assign the
668 * Starting with win8, performance critical channels will be distributed
669 * evenly among all the available NUMA nodes. Once the node is assigned,
670 * we will assign the CPU based on a simple round robin scheme.
672 static void init_vp_index(struct vmbus_channel *channel)
674 bool perf_chn = hv_is_perf_channel(channel);
675 cpumask_var_t available_mask;
676 struct cpumask *alloced_mask;
680 if ((vmbus_proto_version == VERSION_WS2008) ||
681 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn) ||
682 !alloc_cpumask_var(&available_mask, GFP_KERNEL)) {
684 * Prior to win8, all channel interrupts are
685 * delivered on VMBUS_CONNECT_CPU.
686 * Also if the channel is not a performance critical
687 * channel, bind it to VMBUS_CONNECT_CPU.
688 * In case alloc_cpumask_var() fails, bind it to
691 channel->target_cpu = VMBUS_CONNECT_CPU;
693 hv_set_alloced_cpu(VMBUS_CONNECT_CPU);
698 numa_node = next_numa_node_id++;
699 if (numa_node == nr_node_ids) {
700 next_numa_node_id = 0;
703 if (cpumask_empty(cpumask_of_node(numa_node)))
707 alloced_mask = &hv_context.hv_numa_map[numa_node];
709 if (cpumask_weight(alloced_mask) ==
710 cpumask_weight(cpumask_of_node(numa_node))) {
712 * We have cycled through all the CPUs in the node;
713 * reset the alloced map.
715 cpumask_clear(alloced_mask);
718 cpumask_xor(available_mask, alloced_mask, cpumask_of_node(numa_node));
720 target_cpu = cpumask_first(available_mask);
721 cpumask_set_cpu(target_cpu, alloced_mask);
723 channel->target_cpu = target_cpu;
725 free_cpumask_var(available_mask);
728 static void vmbus_wait_for_unload(void)
732 struct hv_message *msg;
733 struct vmbus_channel_message_header *hdr;
737 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
738 * used for initial contact or to CPU0 depending on host version. When
739 * we're crashing on a different CPU let's hope that IRQ handler on
740 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
741 * functional and vmbus_unload_response() will complete
742 * vmbus_connection.unload_event. If not, the last thing we can do is
743 * read message pages for all CPUs directly.
746 if (completion_done(&vmbus_connection.unload_event))
749 for_each_online_cpu(cpu) {
750 struct hv_per_cpu_context *hv_cpu
751 = per_cpu_ptr(hv_context.cpu_context, cpu);
753 page_addr = hv_cpu->synic_message_page;
754 msg = (struct hv_message *)page_addr
755 + VMBUS_MESSAGE_SINT;
757 message_type = READ_ONCE(msg->header.message_type);
758 if (message_type == HVMSG_NONE)
761 hdr = (struct vmbus_channel_message_header *)
764 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
765 complete(&vmbus_connection.unload_event);
767 vmbus_signal_eom(msg, message_type);
774 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
775 * maybe-pending messages on all CPUs to be able to receive new
776 * messages after we reconnect.
778 for_each_online_cpu(cpu) {
779 struct hv_per_cpu_context *hv_cpu
780 = per_cpu_ptr(hv_context.cpu_context, cpu);
782 page_addr = hv_cpu->synic_message_page;
783 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
784 msg->header.message_type = HVMSG_NONE;
789 * vmbus_unload_response - Handler for the unload response.
791 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
794 * This is a global event; just wakeup the waiting thread.
795 * Once we successfully unload, we can cleanup the monitor state.
797 complete(&vmbus_connection.unload_event);
800 void vmbus_initiate_unload(bool crash)
802 struct vmbus_channel_message_header hdr;
804 if (xchg(&vmbus_connection.conn_state, DISCONNECTED) == DISCONNECTED)
807 /* Pre-Win2012R2 hosts don't support reconnect */
808 if (vmbus_proto_version < VERSION_WIN8_1)
811 init_completion(&vmbus_connection.unload_event);
812 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
813 hdr.msgtype = CHANNELMSG_UNLOAD;
814 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
818 * vmbus_initiate_unload() is also called on crash and the crash can be
819 * happening in an interrupt context, where scheduling is impossible.
822 wait_for_completion(&vmbus_connection.unload_event);
824 vmbus_wait_for_unload();
827 static void check_ready_for_resume_event(void)
830 * If all the old primary channels have been fixed up, then it's safe
833 if (atomic_dec_and_test(&vmbus_connection.nr_chan_fixup_on_resume))
834 complete(&vmbus_connection.ready_for_resume_event);
837 static void vmbus_setup_channel_state(struct vmbus_channel *channel,
838 struct vmbus_channel_offer_channel *offer)
841 * Setup state for signalling the host.
843 channel->sig_event = VMBUS_EVENT_CONNECTION_ID;
845 if (vmbus_proto_version != VERSION_WS2008) {
846 channel->is_dedicated_interrupt =
847 (offer->is_dedicated_interrupt != 0);
848 channel->sig_event = offer->connection_id;
851 memcpy(&channel->offermsg, offer,
852 sizeof(struct vmbus_channel_offer_channel));
853 channel->monitor_grp = (u8)offer->monitorid / 32;
854 channel->monitor_bit = (u8)offer->monitorid % 32;
855 channel->device_id = hv_get_dev_type(channel);
859 * find_primary_channel_by_offer - Get the channel object given the new offer.
860 * This is only used in the resume path of hibernation.
862 static struct vmbus_channel *
863 find_primary_channel_by_offer(const struct vmbus_channel_offer_channel *offer)
865 struct vmbus_channel *channel = NULL, *iter;
866 const guid_t *inst1, *inst2;
868 /* Ignore sub-channel offers. */
869 if (offer->offer.sub_channel_index != 0)
872 mutex_lock(&vmbus_connection.channel_mutex);
874 list_for_each_entry(iter, &vmbus_connection.chn_list, listentry) {
875 inst1 = &iter->offermsg.offer.if_instance;
876 inst2 = &offer->offer.if_instance;
878 if (guid_equal(inst1, inst2)) {
884 mutex_unlock(&vmbus_connection.channel_mutex);
890 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
893 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
895 struct vmbus_channel_offer_channel *offer;
896 struct vmbus_channel *oldchannel, *newchannel;
899 offer = (struct vmbus_channel_offer_channel *)hdr;
901 trace_vmbus_onoffer(offer);
903 oldchannel = find_primary_channel_by_offer(offer);
905 if (oldchannel != NULL) {
907 * We're resuming from hibernation: all the sub-channel and
908 * hv_sock channels we had before the hibernation should have
909 * been cleaned up, and now we must be seeing a re-offered
910 * primary channel that we had before the hibernation.
914 * { Initially: channel relid = INVALID_RELID,
915 * channels[valid_relid] = NULL }
919 * [vmbus_onoffer()] [vmbus_device_release()]
921 * LOCK channel_mutex LOCK channel_mutex
922 * STORE channel relid = valid_relid LOAD r1 = channel relid
923 * MAP_RELID channel if (r1 != INVALID_RELID)
924 * UNLOCK channel_mutex UNMAP_RELID channel
925 * UNLOCK channel_mutex
927 * Forbids: r1 == valid_relid &&
928 * channels[valid_relid] == channel
930 * Note. r1 can be INVALID_RELID only for an hv_sock channel.
931 * None of the hv_sock channels which were present before the
932 * suspend are re-offered upon the resume. See the WARN_ON()
933 * in hv_process_channel_removal().
935 mutex_lock(&vmbus_connection.channel_mutex);
937 atomic_dec(&vmbus_connection.offer_in_progress);
939 WARN_ON(oldchannel->offermsg.child_relid != INVALID_RELID);
940 /* Fix up the relid. */
941 oldchannel->offermsg.child_relid = offer->child_relid;
943 offer_sz = sizeof(*offer);
944 if (memcmp(offer, &oldchannel->offermsg, offer_sz) != 0) {
946 * This is not an error, since the host can also change
947 * the other field(s) of the offer, e.g. on WS RS5
948 * (Build 17763), the offer->connection_id of the
949 * Mellanox VF vmbus device can change when the host
950 * reoffers the device upon resume.
952 pr_debug("vmbus offer changed: relid=%d\n",
955 print_hex_dump_debug("Old vmbus offer: ",
956 DUMP_PREFIX_OFFSET, 16, 4,
957 &oldchannel->offermsg, offer_sz,
959 print_hex_dump_debug("New vmbus offer: ",
960 DUMP_PREFIX_OFFSET, 16, 4,
961 offer, offer_sz, false);
963 /* Fix up the old channel. */
964 vmbus_setup_channel_state(oldchannel, offer);
967 /* Add the channel back to the array of channels. */
968 vmbus_channel_map_relid(oldchannel);
969 check_ready_for_resume_event();
971 mutex_unlock(&vmbus_connection.channel_mutex);
975 /* Allocate the channel object and save this offer. */
976 newchannel = alloc_channel();
978 vmbus_release_relid(offer->child_relid);
979 atomic_dec(&vmbus_connection.offer_in_progress);
980 pr_err("Unable to allocate channel object\n");
984 vmbus_setup_channel_state(newchannel, offer);
986 vmbus_process_offer(newchannel);
989 static void check_ready_for_suspend_event(void)
992 * If all the sub-channels or hv_sock channels have been cleaned up,
993 * then it's safe to suspend.
995 if (atomic_dec_and_test(&vmbus_connection.nr_chan_close_on_suspend))
996 complete(&vmbus_connection.ready_for_suspend_event);
1000 * vmbus_onoffer_rescind - Rescind offer handler.
1002 * We queue a work item to process this offer synchronously
1004 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
1006 struct vmbus_channel_rescind_offer *rescind;
1007 struct vmbus_channel *channel;
1009 bool clean_up_chan_for_suspend;
1011 rescind = (struct vmbus_channel_rescind_offer *)hdr;
1013 trace_vmbus_onoffer_rescind(rescind);
1016 * The offer msg and the corresponding rescind msg
1017 * from the host are guranteed to be ordered -
1018 * offer comes in first and then the rescind.
1019 * Since we process these events in work elements,
1020 * and with preemption, we may end up processing
1021 * the events out of order. We rely on the synchronization
1022 * provided by offer_in_progress and by channel_mutex for
1023 * ordering these events:
1025 * { Initially: offer_in_progress = 1 }
1029 * [vmbus_onoffer()] [vmbus_onoffer_rescind()]
1031 * LOCK channel_mutex WAIT_ON offer_in_progress == 0
1032 * DECREMENT offer_in_progress LOCK channel_mutex
1033 * STORE channels[] LOAD channels[]
1034 * UNLOCK channel_mutex UNLOCK channel_mutex
1036 * Forbids: CPU2's LOAD from *not* seeing CPU1's STORE
1039 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
1041 * We wait here until any channel offer is currently
1047 mutex_lock(&vmbus_connection.channel_mutex);
1048 channel = relid2channel(rescind->child_relid);
1049 mutex_unlock(&vmbus_connection.channel_mutex);
1051 if (channel == NULL) {
1053 * We failed in processing the offer message;
1054 * we would have cleaned up the relid in that
1060 clean_up_chan_for_suspend = is_hvsock_channel(channel) ||
1061 is_sub_channel(channel);
1063 * Before setting channel->rescind in vmbus_rescind_cleanup(), we
1064 * should make sure the channel callback is not running any more.
1066 vmbus_reset_channel_cb(channel);
1069 * Now wait for offer handling to complete.
1071 vmbus_rescind_cleanup(channel);
1072 while (READ_ONCE(channel->probe_done) == false) {
1074 * We wait here until any channel offer is currently
1081 * At this point, the rescind handling can proceed safely.
1084 if (channel->device_obj) {
1085 if (channel->chn_rescind_callback) {
1086 channel->chn_rescind_callback(channel);
1088 if (clean_up_chan_for_suspend)
1089 check_ready_for_suspend_event();
1094 * We will have to unregister this device from the
1097 dev = get_device(&channel->device_obj->device);
1099 vmbus_device_unregister(channel->device_obj);
1103 if (channel->primary_channel != NULL) {
1105 * Sub-channel is being rescinded. Following is the channel
1106 * close sequence when initiated from the driveri (refer to
1107 * vmbus_close() for details):
1108 * 1. Close all sub-channels first
1109 * 2. Then close the primary channel.
1111 mutex_lock(&vmbus_connection.channel_mutex);
1112 if (channel->state == CHANNEL_OPEN_STATE) {
1114 * The channel is currently not open;
1115 * it is safe for us to cleanup the channel.
1117 hv_process_channel_removal(channel);
1119 complete(&channel->rescind_event);
1121 mutex_unlock(&vmbus_connection.channel_mutex);
1124 /* The "channel" may have been freed. Do not access it any longer. */
1126 if (clean_up_chan_for_suspend)
1127 check_ready_for_suspend_event();
1130 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
1132 BUG_ON(!is_hvsock_channel(channel));
1134 /* We always get a rescind msg when a connection is closed. */
1135 while (!READ_ONCE(channel->probe_done) || !READ_ONCE(channel->rescind))
1138 vmbus_device_unregister(channel->device_obj);
1140 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
1144 * vmbus_onoffers_delivered -
1145 * This is invoked when all offers have been delivered.
1147 * Nothing to do here.
1149 static void vmbus_onoffers_delivered(
1150 struct vmbus_channel_message_header *hdr)
1155 * vmbus_onopen_result - Open result handler.
1157 * This is invoked when we received a response to our channel open request.
1158 * Find the matching request, copy the response and signal the requesting
1161 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
1163 struct vmbus_channel_open_result *result;
1164 struct vmbus_channel_msginfo *msginfo;
1165 struct vmbus_channel_message_header *requestheader;
1166 struct vmbus_channel_open_channel *openmsg;
1167 unsigned long flags;
1169 result = (struct vmbus_channel_open_result *)hdr;
1171 trace_vmbus_onopen_result(result);
1174 * Find the open msg, copy the result and signal/unblock the wait event
1176 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1178 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1181 (struct vmbus_channel_message_header *)msginfo->msg;
1183 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
1185 (struct vmbus_channel_open_channel *)msginfo->msg;
1186 if (openmsg->child_relid == result->child_relid &&
1187 openmsg->openid == result->openid) {
1188 memcpy(&msginfo->response.open_result,
1191 struct vmbus_channel_open_result));
1192 complete(&msginfo->waitevent);
1197 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1201 * vmbus_ongpadl_created - GPADL created handler.
1203 * This is invoked when we received a response to our gpadl create request.
1204 * Find the matching request, copy the response and signal the requesting
1207 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
1209 struct vmbus_channel_gpadl_created *gpadlcreated;
1210 struct vmbus_channel_msginfo *msginfo;
1211 struct vmbus_channel_message_header *requestheader;
1212 struct vmbus_channel_gpadl_header *gpadlheader;
1213 unsigned long flags;
1215 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
1217 trace_vmbus_ongpadl_created(gpadlcreated);
1220 * Find the establish msg, copy the result and signal/unblock the wait
1223 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1225 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1228 (struct vmbus_channel_message_header *)msginfo->msg;
1230 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
1232 (struct vmbus_channel_gpadl_header *)requestheader;
1234 if ((gpadlcreated->child_relid ==
1235 gpadlheader->child_relid) &&
1236 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1237 memcpy(&msginfo->response.gpadl_created,
1240 struct vmbus_channel_gpadl_created));
1241 complete(&msginfo->waitevent);
1246 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1250 * vmbus_ongpadl_torndown - GPADL torndown handler.
1252 * This is invoked when we received a response to our gpadl teardown request.
1253 * Find the matching request, copy the response and signal the requesting
1256 static void vmbus_ongpadl_torndown(
1257 struct vmbus_channel_message_header *hdr)
1259 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1260 struct vmbus_channel_msginfo *msginfo;
1261 struct vmbus_channel_message_header *requestheader;
1262 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1263 unsigned long flags;
1265 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1267 trace_vmbus_ongpadl_torndown(gpadl_torndown);
1270 * Find the open msg, copy the result and signal/unblock the wait event
1272 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1274 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1277 (struct vmbus_channel_message_header *)msginfo->msg;
1279 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1281 (struct vmbus_channel_gpadl_teardown *)requestheader;
1283 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1284 memcpy(&msginfo->response.gpadl_torndown,
1287 struct vmbus_channel_gpadl_torndown));
1288 complete(&msginfo->waitevent);
1293 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1297 * vmbus_onversion_response - Version response handler
1299 * This is invoked when we received a response to our initiate contact request.
1300 * Find the matching request, copy the response and signal the requesting
1303 static void vmbus_onversion_response(
1304 struct vmbus_channel_message_header *hdr)
1306 struct vmbus_channel_msginfo *msginfo;
1307 struct vmbus_channel_message_header *requestheader;
1308 struct vmbus_channel_version_response *version_response;
1309 unsigned long flags;
1311 version_response = (struct vmbus_channel_version_response *)hdr;
1313 trace_vmbus_onversion_response(version_response);
1315 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1317 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1320 (struct vmbus_channel_message_header *)msginfo->msg;
1322 if (requestheader->msgtype ==
1323 CHANNELMSG_INITIATE_CONTACT) {
1324 memcpy(&msginfo->response.version_response,
1326 sizeof(struct vmbus_channel_version_response));
1327 complete(&msginfo->waitevent);
1330 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1333 /* Channel message dispatch table */
1334 const struct vmbus_channel_message_table_entry
1335 channel_message_table[CHANNELMSG_COUNT] = {
1336 { CHANNELMSG_INVALID, 0, NULL, 0},
1337 { CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer,
1338 sizeof(struct vmbus_channel_offer_channel)},
1339 { CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind,
1340 sizeof(struct vmbus_channel_rescind_offer) },
1341 { CHANNELMSG_REQUESTOFFERS, 0, NULL, 0},
1342 { CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered, 0},
1343 { CHANNELMSG_OPENCHANNEL, 0, NULL, 0},
1344 { CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result,
1345 sizeof(struct vmbus_channel_open_result)},
1346 { CHANNELMSG_CLOSECHANNEL, 0, NULL, 0},
1347 { CHANNELMSG_GPADL_HEADER, 0, NULL, 0},
1348 { CHANNELMSG_GPADL_BODY, 0, NULL, 0},
1349 { CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created,
1350 sizeof(struct vmbus_channel_gpadl_created)},
1351 { CHANNELMSG_GPADL_TEARDOWN, 0, NULL, 0},
1352 { CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown,
1353 sizeof(struct vmbus_channel_gpadl_torndown) },
1354 { CHANNELMSG_RELID_RELEASED, 0, NULL, 0},
1355 { CHANNELMSG_INITIATE_CONTACT, 0, NULL, 0},
1356 { CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response,
1357 sizeof(struct vmbus_channel_version_response)},
1358 { CHANNELMSG_UNLOAD, 0, NULL, 0},
1359 { CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response, 0},
1360 { CHANNELMSG_18, 0, NULL, 0},
1361 { CHANNELMSG_19, 0, NULL, 0},
1362 { CHANNELMSG_20, 0, NULL, 0},
1363 { CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL, 0},
1364 { CHANNELMSG_MODIFYCHANNEL, 0, NULL, 0},
1365 { CHANNELMSG_TL_CONNECT_RESULT, 0, NULL, 0},
1369 * vmbus_onmessage - Handler for channel protocol messages.
1371 * This is invoked in the vmbus worker thread context.
1373 void vmbus_onmessage(struct vmbus_channel_message_header *hdr)
1375 trace_vmbus_on_message(hdr);
1378 * vmbus_on_msg_dpc() makes sure the hdr->msgtype here can not go
1379 * out of bound and the message_handler pointer can not be NULL.
1381 channel_message_table[hdr->msgtype].message_handler(hdr);
1385 * vmbus_request_offers - Send a request to get all our pending offers.
1387 int vmbus_request_offers(void)
1389 struct vmbus_channel_message_header *msg;
1390 struct vmbus_channel_msginfo *msginfo;
1393 msginfo = kmalloc(sizeof(*msginfo) +
1394 sizeof(struct vmbus_channel_message_header),
1399 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1401 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1403 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1406 trace_vmbus_request_offers(ret);
1409 pr_err("Unable to request offers - %d\n", ret);
1420 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1422 struct list_head *cur, *tmp;
1423 struct vmbus_channel *cur_channel;
1425 if (primary_channel->sc_creation_callback == NULL)
1428 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1429 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1431 primary_channel->sc_creation_callback(cur_channel);
1435 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1436 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1438 primary_channel->sc_creation_callback = sc_cr_cb;
1440 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1442 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1446 ret = !list_empty(&primary->sc_list);
1450 * Invoke the callback on sub-channel creation.
1451 * This will present a uniform interface to the
1454 invoke_sc_cb(primary);
1459 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1461 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1462 void (*chn_rescind_cb)(struct vmbus_channel *))
1464 channel->chn_rescind_callback = chn_rescind_cb;
1466 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);