2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
18 * Haiyang Zhang <haiyangz@microsoft.com>
19 * Hank Janssen <hjanssen@microsoft.com>
21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
23 #include <linux/kernel.h>
24 #include <linux/interrupt.h>
25 #include <linux/sched.h>
26 #include <linux/wait.h>
28 #include <linux/slab.h>
29 #include <linux/list.h>
30 #include <linux/module.h>
31 #include <linux/completion.h>
32 #include <linux/delay.h>
33 #include <linux/hyperv.h>
34 #include <asm/mshyperv.h>
36 #include "hyperv_vmbus.h"
38 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type);
40 static const struct vmbus_device vmbus_devs[] = {
48 { .dev_type = HV_SCSI,
72 { .dev_type = HV_PCIE,
77 /* Synthetic Frame Buffer */
83 /* Synthetic Keyboard */
90 { .dev_type = HV_MOUSE,
104 .perf_device = false,
110 .perf_device = false,
114 { .dev_type = HV_SHUTDOWN,
116 .perf_device = false,
120 { .dev_type = HV_FCOPY,
122 .perf_device = false,
126 { .dev_type = HV_BACKUP,
128 .perf_device = false,
134 .perf_device = false,
138 { .dev_type = HV_UNKNOWN,
139 .perf_device = false,
143 static const struct {
145 } vmbus_unsupported_devs[] = {
152 * The rescinded channel may be blocked waiting for a response from the host;
155 static void vmbus_rescind_cleanup(struct vmbus_channel *channel)
157 struct vmbus_channel_msginfo *msginfo;
161 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
162 channel->rescind = true;
163 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
166 if (msginfo->waiting_channel == channel) {
167 complete(&msginfo->waitevent);
171 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
174 static bool is_unsupported_vmbus_devs(const uuid_le *guid)
178 for (i = 0; i < ARRAY_SIZE(vmbus_unsupported_devs); i++)
179 if (!uuid_le_cmp(*guid, vmbus_unsupported_devs[i].guid))
184 static u16 hv_get_dev_type(const struct vmbus_channel *channel)
186 const uuid_le *guid = &channel->offermsg.offer.if_type;
189 if (is_hvsock_channel(channel) || is_unsupported_vmbus_devs(guid))
192 for (i = HV_IDE; i < HV_UNKNOWN; i++) {
193 if (!uuid_le_cmp(*guid, vmbus_devs[i].guid))
196 pr_info("Unknown GUID: %pUl\n", guid);
201 * vmbus_prep_negotiate_resp() - Create default response for Hyper-V Negotiate message
202 * @icmsghdrp: Pointer to msg header structure
203 * @icmsg_negotiate: Pointer to negotiate message structure
204 * @buf: Raw buffer channel data
206 * @icmsghdrp is of type &struct icmsg_hdr.
207 * Set up and fill in default negotiate response message.
209 * The fw_version and fw_vercnt specifies the framework version that
212 * The srv_version and srv_vercnt specifies the service
213 * versions we can support.
215 * Versions are given in decreasing order.
217 * nego_fw_version and nego_srv_version store the selected protocol versions.
219 * Mainly used by Hyper-V drivers.
221 bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp,
222 u8 *buf, const int *fw_version, int fw_vercnt,
223 const int *srv_version, int srv_vercnt,
224 int *nego_fw_version, int *nego_srv_version)
226 int icframe_major, icframe_minor;
227 int icmsg_major, icmsg_minor;
228 int fw_major, fw_minor;
229 int srv_major, srv_minor;
231 bool found_match = false;
232 struct icmsg_negotiate *negop;
234 icmsghdrp->icmsgsize = 0x10;
235 negop = (struct icmsg_negotiate *)&buf[
236 sizeof(struct vmbuspipe_hdr) +
237 sizeof(struct icmsg_hdr)];
239 icframe_major = negop->icframe_vercnt;
242 icmsg_major = negop->icmsg_vercnt;
246 * Select the framework version number we will
250 for (i = 0; i < fw_vercnt; i++) {
251 fw_major = (fw_version[i] >> 16);
252 fw_minor = (fw_version[i] & 0xFFFF);
254 for (j = 0; j < negop->icframe_vercnt; j++) {
255 if ((negop->icversion_data[j].major == fw_major) &&
256 (negop->icversion_data[j].minor == fw_minor)) {
257 icframe_major = negop->icversion_data[j].major;
258 icframe_minor = negop->icversion_data[j].minor;
273 for (i = 0; i < srv_vercnt; i++) {
274 srv_major = (srv_version[i] >> 16);
275 srv_minor = (srv_version[i] & 0xFFFF);
277 for (j = negop->icframe_vercnt;
278 (j < negop->icframe_vercnt + negop->icmsg_vercnt);
281 if ((negop->icversion_data[j].major == srv_major) &&
282 (negop->icversion_data[j].minor == srv_minor)) {
284 icmsg_major = negop->icversion_data[j].major;
285 icmsg_minor = negop->icversion_data[j].minor;
296 * Respond with the framework and service
297 * version numbers we can support.
302 negop->icframe_vercnt = 0;
303 negop->icmsg_vercnt = 0;
305 negop->icframe_vercnt = 1;
306 negop->icmsg_vercnt = 1;
310 *nego_fw_version = (icframe_major << 16) | icframe_minor;
312 if (nego_srv_version)
313 *nego_srv_version = (icmsg_major << 16) | icmsg_minor;
315 negop->icversion_data[0].major = icframe_major;
316 negop->icversion_data[0].minor = icframe_minor;
317 negop->icversion_data[1].major = icmsg_major;
318 negop->icversion_data[1].minor = icmsg_minor;
322 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp);
325 * alloc_channel - Allocate and initialize a vmbus channel object
327 static struct vmbus_channel *alloc_channel(void)
329 struct vmbus_channel *channel;
331 channel = kzalloc(sizeof(*channel), GFP_ATOMIC);
335 spin_lock_init(&channel->lock);
336 init_completion(&channel->rescind_event);
338 INIT_LIST_HEAD(&channel->sc_list);
339 INIT_LIST_HEAD(&channel->percpu_list);
341 tasklet_init(&channel->callback_event,
342 vmbus_on_event, (unsigned long)channel);
348 * free_channel - Release the resources used by the vmbus channel object
350 static void free_channel(struct vmbus_channel *channel)
352 tasklet_kill(&channel->callback_event);
354 kfree_rcu(channel, rcu);
357 static void percpu_channel_enq(void *arg)
359 struct vmbus_channel *channel = arg;
360 struct hv_per_cpu_context *hv_cpu
361 = this_cpu_ptr(hv_context.cpu_context);
363 list_add_tail_rcu(&channel->percpu_list, &hv_cpu->chan_list);
366 static void percpu_channel_deq(void *arg)
368 struct vmbus_channel *channel = arg;
370 list_del_rcu(&channel->percpu_list);
374 static void vmbus_release_relid(u32 relid)
376 struct vmbus_channel_relid_released msg;
378 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
379 msg.child_relid = relid;
380 msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
381 vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released),
385 void hv_process_channel_removal(u32 relid)
388 struct vmbus_channel *primary_channel, *channel;
390 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
393 * Make sure channel is valid as we may have raced.
395 channel = relid2channel(relid);
399 BUG_ON(!channel->rescind);
400 if (channel->target_cpu != get_cpu()) {
402 smp_call_function_single(channel->target_cpu,
403 percpu_channel_deq, channel, true);
405 percpu_channel_deq(channel);
409 if (channel->primary_channel == NULL) {
410 list_del(&channel->listentry);
412 primary_channel = channel;
414 primary_channel = channel->primary_channel;
415 spin_lock_irqsave(&primary_channel->lock, flags);
416 list_del(&channel->sc_list);
417 primary_channel->num_sc--;
418 spin_unlock_irqrestore(&primary_channel->lock, flags);
422 * We need to free the bit for init_vp_index() to work in the case
423 * of sub-channel, when we reload drivers like hv_netvsc.
425 if (channel->affinity_policy == HV_LOCALIZED)
426 cpumask_clear_cpu(channel->target_cpu,
427 &primary_channel->alloced_cpus_in_node);
429 vmbus_release_relid(relid);
431 free_channel(channel);
434 void vmbus_free_channels(void)
436 struct vmbus_channel *channel, *tmp;
438 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
440 /* hv_process_channel_removal() needs this */
441 channel->rescind = true;
443 vmbus_device_unregister(channel->device_obj);
447 /* Note: the function can run concurrently for primary/sub channels. */
448 static void vmbus_add_channel_work(struct work_struct *work)
450 struct vmbus_channel *newchannel =
451 container_of(work, struct vmbus_channel, add_channel_work);
452 struct vmbus_channel *primary_channel = newchannel->primary_channel;
457 dev_type = hv_get_dev_type(newchannel);
459 init_vp_index(newchannel, dev_type);
461 if (newchannel->target_cpu != get_cpu()) {
463 smp_call_function_single(newchannel->target_cpu,
467 percpu_channel_enq(newchannel);
472 * This state is used to indicate a successful open
473 * so that when we do close the channel normally, we
474 * can cleanup properly.
476 newchannel->state = CHANNEL_OPEN_STATE;
478 if (primary_channel != NULL) {
479 /* newchannel is a sub-channel. */
481 if (primary_channel->sc_creation_callback != NULL)
482 primary_channel->sc_creation_callback(newchannel);
484 newchannel->probe_done = true;
489 * Start the process of binding the primary channel to the driver
491 newchannel->device_obj = vmbus_device_create(
492 &newchannel->offermsg.offer.if_type,
493 &newchannel->offermsg.offer.if_instance,
495 if (!newchannel->device_obj)
498 newchannel->device_obj->device_id = dev_type;
500 * Add the new device to the bus. This will kick off device-driver
501 * binding which eventually invokes the device driver's AddDevice()
504 ret = vmbus_device_register(newchannel->device_obj);
507 pr_err("unable to add child device object (relid %d)\n",
508 newchannel->offermsg.child_relid);
509 kfree(newchannel->device_obj);
513 newchannel->probe_done = true;
517 mutex_lock(&vmbus_connection.channel_mutex);
520 * We need to set the flag, otherwise
521 * vmbus_onoffer_rescind() can be blocked.
523 newchannel->probe_done = true;
525 if (primary_channel == NULL) {
526 list_del(&newchannel->listentry);
528 spin_lock_irqsave(&primary_channel->lock, flags);
529 list_del(&newchannel->sc_list);
530 spin_unlock_irqrestore(&primary_channel->lock, flags);
533 mutex_unlock(&vmbus_connection.channel_mutex);
535 if (newchannel->target_cpu != get_cpu()) {
537 smp_call_function_single(newchannel->target_cpu,
541 percpu_channel_deq(newchannel);
545 vmbus_release_relid(newchannel->offermsg.child_relid);
547 free_channel(newchannel);
551 * vmbus_process_offer - Process the offer by creating a channel/device
552 * associated with this offer
554 static void vmbus_process_offer(struct vmbus_channel *newchannel)
556 struct vmbus_channel *channel;
557 struct workqueue_struct *wq;
561 mutex_lock(&vmbus_connection.channel_mutex);
564 * Now that we have acquired the channel_mutex,
565 * we can release the potentially racing rescind thread.
567 atomic_dec(&vmbus_connection.offer_in_progress);
569 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
570 if (!uuid_le_cmp(channel->offermsg.offer.if_type,
571 newchannel->offermsg.offer.if_type) &&
572 !uuid_le_cmp(channel->offermsg.offer.if_instance,
573 newchannel->offermsg.offer.if_instance)) {
580 list_add_tail(&newchannel->listentry,
581 &vmbus_connection.chn_list);
584 * Check to see if this is a valid sub-channel.
586 if (newchannel->offermsg.offer.sub_channel_index == 0) {
587 mutex_unlock(&vmbus_connection.channel_mutex);
589 * Don't call free_channel(), because newchannel->kobj
590 * is not initialized yet.
597 * Process the sub-channel.
599 newchannel->primary_channel = channel;
600 spin_lock_irqsave(&channel->lock, flags);
601 list_add_tail(&newchannel->sc_list, &channel->sc_list);
602 spin_unlock_irqrestore(&channel->lock, flags);
605 mutex_unlock(&vmbus_connection.channel_mutex);
608 * vmbus_process_offer() mustn't call channel->sc_creation_callback()
609 * directly for sub-channels, because sc_creation_callback() ->
610 * vmbus_open() may never get the host's response to the
611 * OPEN_CHANNEL message (the host may rescind a channel at any time,
612 * e.g. in the case of hot removing a NIC), and vmbus_onoffer_rescind()
613 * may not wake up the vmbus_open() as it's blocked due to a non-zero
614 * vmbus_connection.offer_in_progress, and finally we have a deadlock.
616 * The above is also true for primary channels, if the related device
617 * drivers use sync probing mode by default.
619 * And, usually the handling of primary channels and sub-channels can
620 * depend on each other, so we should offload them to different
621 * workqueues to avoid possible deadlock, e.g. in sync-probing mode,
622 * NIC1's netvsc_subchan_work() can race with NIC2's netvsc_probe() ->
623 * rtnl_lock(), and causes deadlock: the former gets the rtnl_lock
624 * and waits for all the sub-channels to appear, but the latter
625 * can't get the rtnl_lock and this blocks the handling of
628 INIT_WORK(&newchannel->add_channel_work, vmbus_add_channel_work);
629 wq = fnew ? vmbus_connection.handle_primary_chan_wq :
630 vmbus_connection.handle_sub_chan_wq;
631 queue_work(wq, &newchannel->add_channel_work);
635 * We use this state to statically distribute the channel interrupt load.
637 static int next_numa_node_id;
639 * init_vp_index() accesses global variables like next_numa_node_id, and
640 * it can run concurrently for primary channels and sub-channels: see
641 * vmbus_process_offer(), so we need the lock to protect the global
644 static DEFINE_SPINLOCK(bind_channel_to_cpu_lock);
647 * Starting with Win8, we can statically distribute the incoming
648 * channel interrupt load by binding a channel to VCPU.
649 * We do this in a hierarchical fashion:
650 * First distribute the primary channels across available NUMA nodes
651 * and then distribute the subchannels amongst the CPUs in the NUMA
652 * node assigned to the primary channel.
654 * For pre-win8 hosts or non-performance critical channels we assign the
655 * first CPU in the first NUMA node.
657 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
660 bool perf_chn = vmbus_devs[dev_type].perf_device;
661 struct vmbus_channel *primary = channel->primary_channel;
663 cpumask_var_t available_mask;
664 struct cpumask *alloced_mask;
666 if ((vmbus_proto_version == VERSION_WS2008) ||
667 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn) ||
668 !alloc_cpumask_var(&available_mask, GFP_KERNEL)) {
670 * Prior to win8, all channel interrupts are
671 * delivered on cpu 0.
672 * Also if the channel is not a performance critical
673 * channel, bind it to cpu 0.
674 * In case alloc_cpumask_var() fails, bind it to cpu 0.
676 channel->numa_node = 0;
677 channel->target_cpu = 0;
678 channel->target_vp = hv_cpu_number_to_vp_number(0);
682 spin_lock(&bind_channel_to_cpu_lock);
685 * Based on the channel affinity policy, we will assign the NUMA
689 if ((channel->affinity_policy == HV_BALANCED) || (!primary)) {
691 next_node = next_numa_node_id++;
692 if (next_node == nr_node_ids) {
693 next_node = next_numa_node_id = 0;
696 if (cpumask_empty(cpumask_of_node(next_node)))
700 channel->numa_node = next_node;
703 alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
705 if (cpumask_weight(alloced_mask) ==
706 cpumask_weight(cpumask_of_node(primary->numa_node))) {
708 * We have cycled through all the CPUs in the node;
709 * reset the alloced map.
711 cpumask_clear(alloced_mask);
714 cpumask_xor(available_mask, alloced_mask,
715 cpumask_of_node(primary->numa_node));
719 if (primary->affinity_policy == HV_LOCALIZED) {
721 * Normally Hyper-V host doesn't create more subchannels
722 * than there are VCPUs on the node but it is possible when not
723 * all present VCPUs on the node are initialized by guest.
724 * Clear the alloced_cpus_in_node to start over.
726 if (cpumask_equal(&primary->alloced_cpus_in_node,
727 cpumask_of_node(primary->numa_node)))
728 cpumask_clear(&primary->alloced_cpus_in_node);
732 cur_cpu = cpumask_next(cur_cpu, available_mask);
733 if (cur_cpu >= nr_cpu_ids) {
735 cpumask_copy(available_mask,
736 cpumask_of_node(primary->numa_node));
740 if (primary->affinity_policy == HV_LOCALIZED) {
742 * NOTE: in the case of sub-channel, we clear the
743 * sub-channel related bit(s) in
744 * primary->alloced_cpus_in_node in
745 * hv_process_channel_removal(), so when we
746 * reload drivers like hv_netvsc in SMP guest, here
747 * we're able to re-allocate
748 * bit from primary->alloced_cpus_in_node.
750 if (!cpumask_test_cpu(cur_cpu,
751 &primary->alloced_cpus_in_node)) {
752 cpumask_set_cpu(cur_cpu,
753 &primary->alloced_cpus_in_node);
754 cpumask_set_cpu(cur_cpu, alloced_mask);
758 cpumask_set_cpu(cur_cpu, alloced_mask);
763 channel->target_cpu = cur_cpu;
764 channel->target_vp = hv_cpu_number_to_vp_number(cur_cpu);
766 spin_unlock(&bind_channel_to_cpu_lock);
768 free_cpumask_var(available_mask);
771 #define UNLOAD_DELAY_UNIT_MS 10 /* 10 milliseconds */
772 #define UNLOAD_WAIT_MS (100*1000) /* 100 seconds */
773 #define UNLOAD_WAIT_LOOPS (UNLOAD_WAIT_MS/UNLOAD_DELAY_UNIT_MS)
774 #define UNLOAD_MSG_MS (5*1000) /* Every 5 seconds */
775 #define UNLOAD_MSG_LOOPS (UNLOAD_MSG_MS/UNLOAD_DELAY_UNIT_MS)
777 static void vmbus_wait_for_unload(void)
781 struct hv_message *msg;
782 struct vmbus_channel_message_header *hdr;
786 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
787 * used for initial contact or to CPU0 depending on host version. When
788 * we're crashing on a different CPU let's hope that IRQ handler on
789 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
790 * functional and vmbus_unload_response() will complete
791 * vmbus_connection.unload_event. If not, the last thing we can do is
792 * read message pages for all CPUs directly.
794 * Wait up to 100 seconds since an Azure host must writeback any dirty
795 * data in its disk cache before the VMbus UNLOAD request will
796 * complete. This flushing has been empirically observed to take up
797 * to 50 seconds in cases with a lot of dirty data, so allow additional
798 * leeway and for inaccuracies in mdelay(). But eventually time out so
799 * that the panic path can't get hung forever in case the response
800 * message isn't seen.
802 for (i = 1; i <= UNLOAD_WAIT_LOOPS; i++) {
803 if (completion_done(&vmbus_connection.unload_event))
806 for_each_present_cpu(cpu) {
807 struct hv_per_cpu_context *hv_cpu
808 = per_cpu_ptr(hv_context.cpu_context, cpu);
811 * In a CoCo VM the synic_message_page is not allocated
812 * in hv_synic_alloc(). Instead it is set/cleared in
813 * hv_synic_enable_regs() and hv_synic_disable_regs()
814 * such that it is set only when the CPU is online. If
815 * not all present CPUs are online, the message page
816 * might be NULL, so skip such CPUs.
818 page_addr = hv_cpu->synic_message_page;
822 msg = (struct hv_message *)page_addr
823 + VMBUS_MESSAGE_SINT;
825 message_type = READ_ONCE(msg->header.message_type);
826 if (message_type == HVMSG_NONE)
829 hdr = (struct vmbus_channel_message_header *)
832 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
833 complete(&vmbus_connection.unload_event);
835 vmbus_signal_eom(msg, message_type);
839 * Give a notice periodically so someone watching the
840 * serial output won't think it is completely hung.
842 if (!(i % UNLOAD_MSG_LOOPS))
843 pr_notice("Waiting for VMBus UNLOAD to complete\n");
845 mdelay(UNLOAD_DELAY_UNIT_MS);
847 pr_err("Continuing even though VMBus UNLOAD did not complete\n");
851 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
852 * maybe-pending messages on all CPUs to be able to receive new
853 * messages after we reconnect.
855 for_each_present_cpu(cpu) {
856 struct hv_per_cpu_context *hv_cpu
857 = per_cpu_ptr(hv_context.cpu_context, cpu);
859 page_addr = hv_cpu->synic_message_page;
863 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
864 msg->header.message_type = HVMSG_NONE;
869 * vmbus_unload_response - Handler for the unload response.
871 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
874 * This is a global event; just wakeup the waiting thread.
875 * Once we successfully unload, we can cleanup the monitor state.
877 complete(&vmbus_connection.unload_event);
880 void vmbus_initiate_unload(bool crash)
882 struct vmbus_channel_message_header hdr;
884 /* Pre-Win2012R2 hosts don't support reconnect */
885 if (vmbus_proto_version < VERSION_WIN8_1)
888 init_completion(&vmbus_connection.unload_event);
889 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
890 hdr.msgtype = CHANNELMSG_UNLOAD;
891 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
895 * vmbus_initiate_unload() is also called on crash and the crash can be
896 * happening in an interrupt context, where scheduling is impossible.
899 wait_for_completion(&vmbus_connection.unload_event);
901 vmbus_wait_for_unload();
905 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
908 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
910 struct vmbus_channel_offer_channel *offer;
911 struct vmbus_channel *newchannel;
913 offer = (struct vmbus_channel_offer_channel *)hdr;
915 /* Allocate the channel object and save this offer. */
916 newchannel = alloc_channel();
918 vmbus_release_relid(offer->child_relid);
919 atomic_dec(&vmbus_connection.offer_in_progress);
920 pr_err("Unable to allocate channel object\n");
925 * Setup state for signalling the host.
927 newchannel->sig_event = VMBUS_EVENT_CONNECTION_ID;
929 if (vmbus_proto_version != VERSION_WS2008) {
930 newchannel->is_dedicated_interrupt =
931 (offer->is_dedicated_interrupt != 0);
932 newchannel->sig_event = offer->connection_id;
935 memcpy(&newchannel->offermsg, offer,
936 sizeof(struct vmbus_channel_offer_channel));
937 newchannel->monitor_grp = (u8)offer->monitorid / 32;
938 newchannel->monitor_bit = (u8)offer->monitorid % 32;
940 vmbus_process_offer(newchannel);
944 * vmbus_onoffer_rescind - Rescind offer handler.
946 * We queue a work item to process this offer synchronously
948 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
950 struct vmbus_channel_rescind_offer *rescind;
951 struct vmbus_channel *channel;
954 rescind = (struct vmbus_channel_rescind_offer *)hdr;
957 * The offer msg and the corresponding rescind msg
958 * from the host are guranteed to be ordered -
959 * offer comes in first and then the rescind.
960 * Since we process these events in work elements,
961 * and with preemption, we may end up processing
962 * the events out of order. Given that we handle these
963 * work elements on the same CPU, this is possible only
964 * in the case of preemption. In any case wait here
965 * until the offer processing has moved beyond the
966 * point where the channel is discoverable.
969 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
971 * We wait here until any channel offer is currently
977 mutex_lock(&vmbus_connection.channel_mutex);
978 channel = relid2channel(rescind->child_relid);
979 mutex_unlock(&vmbus_connection.channel_mutex);
981 if (channel == NULL) {
983 * We failed in processing the offer message;
984 * we would have cleaned up the relid in that
991 * Before setting channel->rescind in vmbus_rescind_cleanup(), we
992 * should make sure the channel callback is not running any more.
994 vmbus_reset_channel_cb(channel);
997 * Now wait for offer handling to complete.
999 vmbus_rescind_cleanup(channel);
1000 while (READ_ONCE(channel->probe_done) == false) {
1002 * We wait here until any channel offer is currently
1009 * At this point, the rescind handling can proceed safely.
1012 if (channel->device_obj) {
1013 if (channel->chn_rescind_callback) {
1014 channel->chn_rescind_callback(channel);
1018 * We will have to unregister this device from the
1021 dev = get_device(&channel->device_obj->device);
1023 vmbus_device_unregister(channel->device_obj);
1026 } else if (channel->primary_channel != NULL) {
1028 * Sub-channel is being rescinded. Following is the channel
1029 * close sequence when initiated from the driveri (refer to
1030 * vmbus_close() for details):
1031 * 1. Close all sub-channels first
1032 * 2. Then close the primary channel.
1034 mutex_lock(&vmbus_connection.channel_mutex);
1035 if (channel->state == CHANNEL_OPEN_STATE) {
1037 * The channel is currently not open;
1038 * it is safe for us to cleanup the channel.
1040 hv_process_channel_removal(rescind->child_relid);
1042 complete(&channel->rescind_event);
1044 mutex_unlock(&vmbus_connection.channel_mutex);
1048 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
1050 BUG_ON(!is_hvsock_channel(channel));
1052 /* We always get a rescind msg when a connection is closed. */
1053 while (!READ_ONCE(channel->probe_done) || !READ_ONCE(channel->rescind))
1056 vmbus_device_unregister(channel->device_obj);
1058 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
1062 * vmbus_onoffers_delivered -
1063 * This is invoked when all offers have been delivered.
1065 * Nothing to do here.
1067 static void vmbus_onoffers_delivered(
1068 struct vmbus_channel_message_header *hdr)
1073 * vmbus_onopen_result - Open result handler.
1075 * This is invoked when we received a response to our channel open request.
1076 * Find the matching request, copy the response and signal the requesting
1079 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
1081 struct vmbus_channel_open_result *result;
1082 struct vmbus_channel_msginfo *msginfo;
1083 struct vmbus_channel_message_header *requestheader;
1084 struct vmbus_channel_open_channel *openmsg;
1085 unsigned long flags;
1087 result = (struct vmbus_channel_open_result *)hdr;
1090 * Find the open msg, copy the result and signal/unblock the wait event
1092 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1094 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1097 (struct vmbus_channel_message_header *)msginfo->msg;
1099 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
1101 (struct vmbus_channel_open_channel *)msginfo->msg;
1102 if (openmsg->child_relid == result->child_relid &&
1103 openmsg->openid == result->openid) {
1104 memcpy(&msginfo->response.open_result,
1107 struct vmbus_channel_open_result));
1108 complete(&msginfo->waitevent);
1113 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1117 * vmbus_ongpadl_created - GPADL created handler.
1119 * This is invoked when we received a response to our gpadl create request.
1120 * Find the matching request, copy the response and signal the requesting
1123 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
1125 struct vmbus_channel_gpadl_created *gpadlcreated;
1126 struct vmbus_channel_msginfo *msginfo;
1127 struct vmbus_channel_message_header *requestheader;
1128 struct vmbus_channel_gpadl_header *gpadlheader;
1129 unsigned long flags;
1131 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
1134 * Find the establish msg, copy the result and signal/unblock the wait
1137 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1139 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1142 (struct vmbus_channel_message_header *)msginfo->msg;
1144 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
1146 (struct vmbus_channel_gpadl_header *)requestheader;
1148 if ((gpadlcreated->child_relid ==
1149 gpadlheader->child_relid) &&
1150 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1151 memcpy(&msginfo->response.gpadl_created,
1154 struct vmbus_channel_gpadl_created));
1155 complete(&msginfo->waitevent);
1160 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1164 * vmbus_ongpadl_torndown - GPADL torndown handler.
1166 * This is invoked when we received a response to our gpadl teardown request.
1167 * Find the matching request, copy the response and signal the requesting
1170 static void vmbus_ongpadl_torndown(
1171 struct vmbus_channel_message_header *hdr)
1173 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1174 struct vmbus_channel_msginfo *msginfo;
1175 struct vmbus_channel_message_header *requestheader;
1176 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1177 unsigned long flags;
1179 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1182 * Find the open msg, copy the result and signal/unblock the wait event
1184 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1186 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1189 (struct vmbus_channel_message_header *)msginfo->msg;
1191 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1193 (struct vmbus_channel_gpadl_teardown *)requestheader;
1195 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1196 memcpy(&msginfo->response.gpadl_torndown,
1199 struct vmbus_channel_gpadl_torndown));
1200 complete(&msginfo->waitevent);
1205 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1209 * vmbus_onversion_response - Version response handler
1211 * This is invoked when we received a response to our initiate contact request.
1212 * Find the matching request, copy the response and signal the requesting
1215 static void vmbus_onversion_response(
1216 struct vmbus_channel_message_header *hdr)
1218 struct vmbus_channel_msginfo *msginfo;
1219 struct vmbus_channel_message_header *requestheader;
1220 struct vmbus_channel_version_response *version_response;
1221 unsigned long flags;
1223 version_response = (struct vmbus_channel_version_response *)hdr;
1224 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1226 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1229 (struct vmbus_channel_message_header *)msginfo->msg;
1231 if (requestheader->msgtype ==
1232 CHANNELMSG_INITIATE_CONTACT) {
1233 memcpy(&msginfo->response.version_response,
1235 sizeof(struct vmbus_channel_version_response));
1236 complete(&msginfo->waitevent);
1239 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1242 /* Channel message dispatch table */
1243 const struct vmbus_channel_message_table_entry
1244 channel_message_table[CHANNELMSG_COUNT] = {
1245 { CHANNELMSG_INVALID, 0, NULL },
1246 { CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer },
1247 { CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind },
1248 { CHANNELMSG_REQUESTOFFERS, 0, NULL },
1249 { CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered },
1250 { CHANNELMSG_OPENCHANNEL, 0, NULL },
1251 { CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result },
1252 { CHANNELMSG_CLOSECHANNEL, 0, NULL },
1253 { CHANNELMSG_GPADL_HEADER, 0, NULL },
1254 { CHANNELMSG_GPADL_BODY, 0, NULL },
1255 { CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created },
1256 { CHANNELMSG_GPADL_TEARDOWN, 0, NULL },
1257 { CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown },
1258 { CHANNELMSG_RELID_RELEASED, 0, NULL },
1259 { CHANNELMSG_INITIATE_CONTACT, 0, NULL },
1260 { CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response },
1261 { CHANNELMSG_UNLOAD, 0, NULL },
1262 { CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response },
1263 { CHANNELMSG_18, 0, NULL },
1264 { CHANNELMSG_19, 0, NULL },
1265 { CHANNELMSG_20, 0, NULL },
1266 { CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL },
1267 { CHANNELMSG_22, 0, NULL },
1268 { CHANNELMSG_TL_CONNECT_RESULT, 0, NULL },
1272 * vmbus_onmessage - Handler for channel protocol messages.
1274 * This is invoked in the vmbus worker thread context.
1276 void vmbus_onmessage(void *context)
1278 struct hv_message *msg = context;
1279 struct vmbus_channel_message_header *hdr;
1281 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1284 * vmbus_on_msg_dpc() makes sure the hdr->msgtype here can not go
1285 * out of bound and the message_handler pointer can not be NULL.
1287 channel_message_table[hdr->msgtype].message_handler(hdr);
1291 * vmbus_request_offers - Send a request to get all our pending offers.
1293 int vmbus_request_offers(void)
1295 struct vmbus_channel_message_header *msg;
1296 struct vmbus_channel_msginfo *msginfo;
1299 msginfo = kmalloc(sizeof(*msginfo) +
1300 sizeof(struct vmbus_channel_message_header),
1305 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1307 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1310 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1313 pr_err("Unable to request offers - %d\n", ret);
1325 * Retrieve the (sub) channel on which to send an outgoing request.
1326 * When a primary channel has multiple sub-channels, we try to
1327 * distribute the load equally amongst all available channels.
1329 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary)
1331 struct list_head *cur, *tmp;
1333 struct vmbus_channel *cur_channel;
1334 struct vmbus_channel *outgoing_channel = primary;
1338 if (list_empty(&primary->sc_list))
1339 return outgoing_channel;
1341 next_channel = primary->next_oc++;
1343 if (next_channel > (primary->num_sc)) {
1344 primary->next_oc = 0;
1345 return outgoing_channel;
1348 cur_cpu = hv_cpu_number_to_vp_number(smp_processor_id());
1349 list_for_each_safe(cur, tmp, &primary->sc_list) {
1350 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1351 if (cur_channel->state != CHANNEL_OPENED_STATE)
1354 if (cur_channel->target_vp == cur_cpu)
1357 if (i == next_channel)
1363 return outgoing_channel;
1365 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel);
1367 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1369 struct list_head *cur, *tmp;
1370 struct vmbus_channel *cur_channel;
1372 if (primary_channel->sc_creation_callback == NULL)
1375 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1376 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1378 primary_channel->sc_creation_callback(cur_channel);
1382 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1383 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1385 primary_channel->sc_creation_callback = sc_cr_cb;
1387 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1389 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1393 ret = !list_empty(&primary->sc_list);
1397 * Invoke the callback on sub-channel creation.
1398 * This will present a uniform interface to the
1401 invoke_sc_cb(primary);
1406 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1408 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1409 void (*chn_rescind_cb)(struct vmbus_channel *))
1411 channel->chn_rescind_callback = chn_rescind_cb;
1413 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);