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152 * The rescinded channel may be blocked waiting for a response from the host;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);
353 vmbus_remove_channel_attr_group(channel);
355 kobject_put(&channel->kobj);
358 static void percpu_channel_enq(void *arg)
360 struct vmbus_channel *channel = arg;
361 struct hv_per_cpu_context *hv_cpu
362 = this_cpu_ptr(hv_context.cpu_context);
364 list_add_tail_rcu(&channel->percpu_list, &hv_cpu->chan_list);
367 static void percpu_channel_deq(void *arg)
369 struct vmbus_channel *channel = arg;
371 list_del_rcu(&channel->percpu_list);
375 static void vmbus_release_relid(u32 relid)
377 struct vmbus_channel_relid_released msg;
380 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
381 msg.child_relid = relid;
382 msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
383 ret = vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released),
386 trace_vmbus_release_relid(&msg, ret);
389 void hv_process_channel_removal(u32 relid)
392 struct vmbus_channel *primary_channel, *channel;
394 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
397 * Make sure channel is valid as we may have raced.
399 channel = relid2channel(relid);
403 BUG_ON(!channel->rescind);
404 if (channel->target_cpu != get_cpu()) {
406 smp_call_function_single(channel->target_cpu,
407 percpu_channel_deq, channel, true);
409 percpu_channel_deq(channel);
413 if (channel->primary_channel == NULL) {
414 list_del(&channel->listentry);
416 primary_channel = channel;
418 primary_channel = channel->primary_channel;
419 spin_lock_irqsave(&primary_channel->lock, flags);
420 list_del(&channel->sc_list);
421 primary_channel->num_sc--;
422 spin_unlock_irqrestore(&primary_channel->lock, flags);
426 * We need to free the bit for init_vp_index() to work in the case
427 * of sub-channel, when we reload drivers like hv_netvsc.
429 if (channel->affinity_policy == HV_LOCALIZED)
430 cpumask_clear_cpu(channel->target_cpu,
431 &primary_channel->alloced_cpus_in_node);
433 vmbus_release_relid(relid);
435 free_channel(channel);
438 void vmbus_free_channels(void)
440 struct vmbus_channel *channel, *tmp;
442 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
444 /* hv_process_channel_removal() needs this */
445 channel->rescind = true;
447 vmbus_device_unregister(channel->device_obj);
451 /* Note: the function can run concurrently for primary/sub channels. */
452 static void vmbus_add_channel_work(struct work_struct *work)
454 struct vmbus_channel *newchannel =
455 container_of(work, struct vmbus_channel, add_channel_work);
456 struct vmbus_channel *primary_channel = newchannel->primary_channel;
461 dev_type = hv_get_dev_type(newchannel);
463 init_vp_index(newchannel, dev_type);
465 if (newchannel->target_cpu != get_cpu()) {
467 smp_call_function_single(newchannel->target_cpu,
471 percpu_channel_enq(newchannel);
476 * This state is used to indicate a successful open
477 * so that when we do close the channel normally, we
478 * can cleanup properly.
480 newchannel->state = CHANNEL_OPEN_STATE;
482 if (primary_channel != NULL) {
483 /* newchannel is a sub-channel. */
484 struct hv_device *dev = primary_channel->device_obj;
486 if (vmbus_add_channel_kobj(dev, newchannel))
489 if (primary_channel->sc_creation_callback != NULL)
490 primary_channel->sc_creation_callback(newchannel);
492 newchannel->probe_done = true;
497 * Start the process of binding the primary channel to the driver
499 newchannel->device_obj = vmbus_device_create(
500 &newchannel->offermsg.offer.if_type,
501 &newchannel->offermsg.offer.if_instance,
503 if (!newchannel->device_obj)
506 newchannel->device_obj->device_id = dev_type;
508 * Add the new device to the bus. This will kick off device-driver
509 * binding which eventually invokes the device driver's AddDevice()
512 ret = vmbus_device_register(newchannel->device_obj);
515 pr_err("unable to add child device object (relid %d)\n",
516 newchannel->offermsg.child_relid);
517 kfree(newchannel->device_obj);
521 newchannel->probe_done = true;
525 mutex_lock(&vmbus_connection.channel_mutex);
528 * We need to set the flag, otherwise
529 * vmbus_onoffer_rescind() can be blocked.
531 newchannel->probe_done = true;
533 if (primary_channel == NULL) {
534 list_del(&newchannel->listentry);
536 spin_lock_irqsave(&primary_channel->lock, flags);
537 list_del(&newchannel->sc_list);
538 spin_unlock_irqrestore(&primary_channel->lock, flags);
541 mutex_unlock(&vmbus_connection.channel_mutex);
543 if (newchannel->target_cpu != get_cpu()) {
545 smp_call_function_single(newchannel->target_cpu,
549 percpu_channel_deq(newchannel);
553 vmbus_release_relid(newchannel->offermsg.child_relid);
555 free_channel(newchannel);
559 * vmbus_process_offer - Process the offer by creating a channel/device
560 * associated with this offer
562 static void vmbus_process_offer(struct vmbus_channel *newchannel)
564 struct vmbus_channel *channel;
565 struct workqueue_struct *wq;
569 mutex_lock(&vmbus_connection.channel_mutex);
572 * Now that we have acquired the channel_mutex,
573 * we can release the potentially racing rescind thread.
575 atomic_dec(&vmbus_connection.offer_in_progress);
577 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
578 if (!uuid_le_cmp(channel->offermsg.offer.if_type,
579 newchannel->offermsg.offer.if_type) &&
580 !uuid_le_cmp(channel->offermsg.offer.if_instance,
581 newchannel->offermsg.offer.if_instance)) {
588 list_add_tail(&newchannel->listentry,
589 &vmbus_connection.chn_list);
592 * Check to see if this is a valid sub-channel.
594 if (newchannel->offermsg.offer.sub_channel_index == 0) {
595 mutex_unlock(&vmbus_connection.channel_mutex);
597 * Don't call free_channel(), because newchannel->kobj
598 * is not initialized yet.
605 * Process the sub-channel.
607 newchannel->primary_channel = channel;
608 spin_lock_irqsave(&channel->lock, flags);
609 list_add_tail(&newchannel->sc_list, &channel->sc_list);
610 spin_unlock_irqrestore(&channel->lock, flags);
613 mutex_unlock(&vmbus_connection.channel_mutex);
616 * vmbus_process_offer() mustn't call channel->sc_creation_callback()
617 * directly for sub-channels, because sc_creation_callback() ->
618 * vmbus_open() may never get the host's response to the
619 * OPEN_CHANNEL message (the host may rescind a channel at any time,
620 * e.g. in the case of hot removing a NIC), and vmbus_onoffer_rescind()
621 * may not wake up the vmbus_open() as it's blocked due to a non-zero
622 * vmbus_connection.offer_in_progress, and finally we have a deadlock.
624 * The above is also true for primary channels, if the related device
625 * drivers use sync probing mode by default.
627 * And, usually the handling of primary channels and sub-channels can
628 * depend on each other, so we should offload them to different
629 * workqueues to avoid possible deadlock, e.g. in sync-probing mode,
630 * NIC1's netvsc_subchan_work() can race with NIC2's netvsc_probe() ->
631 * rtnl_lock(), and causes deadlock: the former gets the rtnl_lock
632 * and waits for all the sub-channels to appear, but the latter
633 * can't get the rtnl_lock and this blocks the handling of
636 INIT_WORK(&newchannel->add_channel_work, vmbus_add_channel_work);
637 wq = fnew ? vmbus_connection.handle_primary_chan_wq :
638 vmbus_connection.handle_sub_chan_wq;
639 queue_work(wq, &newchannel->add_channel_work);
643 * We use this state to statically distribute the channel interrupt load.
645 static int next_numa_node_id;
647 * init_vp_index() accesses global variables like next_numa_node_id, and
648 * it can run concurrently for primary channels and sub-channels: see
649 * vmbus_process_offer(), so we need the lock to protect the global
652 static DEFINE_SPINLOCK(bind_channel_to_cpu_lock);
655 * Starting with Win8, we can statically distribute the incoming
656 * channel interrupt load by binding a channel to VCPU.
657 * We distribute the interrupt loads to one or more NUMA nodes based on
658 * the channel's affinity_policy.
660 * For pre-win8 hosts or non-performance critical channels we assign the
661 * first CPU in the first NUMA node.
663 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
666 bool perf_chn = vmbus_devs[dev_type].perf_device;
667 struct vmbus_channel *primary = channel->primary_channel;
669 cpumask_var_t available_mask;
670 struct cpumask *alloced_mask;
672 if ((vmbus_proto_version == VERSION_WS2008) ||
673 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn) ||
674 !alloc_cpumask_var(&available_mask, GFP_KERNEL)) {
676 * Prior to win8, all channel interrupts are
677 * delivered on cpu 0.
678 * Also if the channel is not a performance critical
679 * channel, bind it to cpu 0.
680 * In case alloc_cpumask_var() fails, bind it to cpu 0.
682 channel->numa_node = 0;
683 channel->target_cpu = 0;
684 channel->target_vp = hv_cpu_number_to_vp_number(0);
688 spin_lock(&bind_channel_to_cpu_lock);
691 * Based on the channel affinity policy, we will assign the NUMA
695 if ((channel->affinity_policy == HV_BALANCED) || (!primary)) {
697 next_node = next_numa_node_id++;
698 if (next_node == nr_node_ids) {
699 next_node = next_numa_node_id = 0;
702 if (cpumask_empty(cpumask_of_node(next_node)))
706 channel->numa_node = next_node;
709 alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
711 if (cpumask_weight(alloced_mask) ==
712 cpumask_weight(cpumask_of_node(primary->numa_node))) {
714 * We have cycled through all the CPUs in the node;
715 * reset the alloced map.
717 cpumask_clear(alloced_mask);
720 cpumask_xor(available_mask, alloced_mask,
721 cpumask_of_node(primary->numa_node));
725 if (primary->affinity_policy == HV_LOCALIZED) {
727 * Normally Hyper-V host doesn't create more subchannels
728 * than there are VCPUs on the node but it is possible when not
729 * all present VCPUs on the node are initialized by guest.
730 * Clear the alloced_cpus_in_node to start over.
732 if (cpumask_equal(&primary->alloced_cpus_in_node,
733 cpumask_of_node(primary->numa_node)))
734 cpumask_clear(&primary->alloced_cpus_in_node);
738 cur_cpu = cpumask_next(cur_cpu, available_mask);
739 if (cur_cpu >= nr_cpu_ids) {
741 cpumask_copy(available_mask,
742 cpumask_of_node(primary->numa_node));
746 if (primary->affinity_policy == HV_LOCALIZED) {
748 * NOTE: in the case of sub-channel, we clear the
749 * sub-channel related bit(s) in
750 * primary->alloced_cpus_in_node in
751 * hv_process_channel_removal(), so when we
752 * reload drivers like hv_netvsc in SMP guest, here
753 * we're able to re-allocate
754 * bit from primary->alloced_cpus_in_node.
756 if (!cpumask_test_cpu(cur_cpu,
757 &primary->alloced_cpus_in_node)) {
758 cpumask_set_cpu(cur_cpu,
759 &primary->alloced_cpus_in_node);
760 cpumask_set_cpu(cur_cpu, alloced_mask);
764 cpumask_set_cpu(cur_cpu, alloced_mask);
769 channel->target_cpu = cur_cpu;
770 channel->target_vp = hv_cpu_number_to_vp_number(cur_cpu);
772 spin_unlock(&bind_channel_to_cpu_lock);
774 free_cpumask_var(available_mask);
777 #define UNLOAD_DELAY_UNIT_MS 10 /* 10 milliseconds */
778 #define UNLOAD_WAIT_MS (100*1000) /* 100 seconds */
779 #define UNLOAD_WAIT_LOOPS (UNLOAD_WAIT_MS/UNLOAD_DELAY_UNIT_MS)
780 #define UNLOAD_MSG_MS (5*1000) /* Every 5 seconds */
781 #define UNLOAD_MSG_LOOPS (UNLOAD_MSG_MS/UNLOAD_DELAY_UNIT_MS)
783 static void vmbus_wait_for_unload(void)
787 struct hv_message *msg;
788 struct vmbus_channel_message_header *hdr;
792 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
793 * used for initial contact or to CPU0 depending on host version. When
794 * we're crashing on a different CPU let's hope that IRQ handler on
795 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
796 * functional and vmbus_unload_response() will complete
797 * vmbus_connection.unload_event. If not, the last thing we can do is
798 * read message pages for all CPUs directly.
800 * Wait up to 100 seconds since an Azure host must writeback any dirty
801 * data in its disk cache before the VMbus UNLOAD request will
802 * complete. This flushing has been empirically observed to take up
803 * to 50 seconds in cases with a lot of dirty data, so allow additional
804 * leeway and for inaccuracies in mdelay(). But eventually time out so
805 * that the panic path can't get hung forever in case the response
806 * message isn't seen.
808 for (i = 1; i <= UNLOAD_WAIT_LOOPS; i++) {
809 if (completion_done(&vmbus_connection.unload_event))
812 for_each_online_cpu(cpu) {
813 struct hv_per_cpu_context *hv_cpu
814 = per_cpu_ptr(hv_context.cpu_context, cpu);
816 page_addr = hv_cpu->synic_message_page;
817 msg = (struct hv_message *)page_addr
818 + VMBUS_MESSAGE_SINT;
820 message_type = READ_ONCE(msg->header.message_type);
821 if (message_type == HVMSG_NONE)
824 hdr = (struct vmbus_channel_message_header *)
827 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
828 complete(&vmbus_connection.unload_event);
830 vmbus_signal_eom(msg, message_type);
834 * Give a notice periodically so someone watching the
835 * serial output won't think it is completely hung.
837 if (!(i % UNLOAD_MSG_LOOPS))
838 pr_notice("Waiting for VMBus UNLOAD to complete\n");
840 mdelay(UNLOAD_DELAY_UNIT_MS);
842 pr_err("Continuing even though VMBus UNLOAD did not complete\n");
846 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
847 * maybe-pending messages on all CPUs to be able to receive new
848 * messages after we reconnect.
850 for_each_online_cpu(cpu) {
851 struct hv_per_cpu_context *hv_cpu
852 = per_cpu_ptr(hv_context.cpu_context, cpu);
854 page_addr = hv_cpu->synic_message_page;
855 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
856 msg->header.message_type = HVMSG_NONE;
861 * vmbus_unload_response - Handler for the unload response.
863 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
866 * This is a global event; just wakeup the waiting thread.
867 * Once we successfully unload, we can cleanup the monitor state.
869 complete(&vmbus_connection.unload_event);
872 void vmbus_initiate_unload(bool crash)
874 struct vmbus_channel_message_header hdr;
876 if (xchg(&vmbus_connection.conn_state, DISCONNECTED) == DISCONNECTED)
879 /* Pre-Win2012R2 hosts don't support reconnect */
880 if (vmbus_proto_version < VERSION_WIN8_1)
883 init_completion(&vmbus_connection.unload_event);
884 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
885 hdr.msgtype = CHANNELMSG_UNLOAD;
886 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
890 * vmbus_initiate_unload() is also called on crash and the crash can be
891 * happening in an interrupt context, where scheduling is impossible.
894 wait_for_completion(&vmbus_connection.unload_event);
896 vmbus_wait_for_unload();
900 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
903 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
905 struct vmbus_channel_offer_channel *offer;
906 struct vmbus_channel *newchannel;
908 offer = (struct vmbus_channel_offer_channel *)hdr;
910 trace_vmbus_onoffer(offer);
912 /* Allocate the channel object and save this offer. */
913 newchannel = alloc_channel();
915 vmbus_release_relid(offer->child_relid);
916 atomic_dec(&vmbus_connection.offer_in_progress);
917 pr_err("Unable to allocate channel object\n");
922 * Setup state for signalling the host.
924 newchannel->sig_event = VMBUS_EVENT_CONNECTION_ID;
926 if (vmbus_proto_version != VERSION_WS2008) {
927 newchannel->is_dedicated_interrupt =
928 (offer->is_dedicated_interrupt != 0);
929 newchannel->sig_event = offer->connection_id;
932 memcpy(&newchannel->offermsg, offer,
933 sizeof(struct vmbus_channel_offer_channel));
934 newchannel->monitor_grp = (u8)offer->monitorid / 32;
935 newchannel->monitor_bit = (u8)offer->monitorid % 32;
937 vmbus_process_offer(newchannel);
941 * vmbus_onoffer_rescind - Rescind offer handler.
943 * We queue a work item to process this offer synchronously
945 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
947 struct vmbus_channel_rescind_offer *rescind;
948 struct vmbus_channel *channel;
951 rescind = (struct vmbus_channel_rescind_offer *)hdr;
953 trace_vmbus_onoffer_rescind(rescind);
956 * The offer msg and the corresponding rescind msg
957 * from the host are guranteed to be ordered -
958 * offer comes in first and then the rescind.
959 * Since we process these events in work elements,
960 * and with preemption, we may end up processing
961 * the events out of order. Given that we handle these
962 * work elements on the same CPU, this is possible only
963 * in the case of preemption. In any case wait here
964 * until the offer processing has moved beyond the
965 * point where the channel is discoverable.
968 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
970 * We wait here until any channel offer is currently
976 mutex_lock(&vmbus_connection.channel_mutex);
977 channel = relid2channel(rescind->child_relid);
978 mutex_unlock(&vmbus_connection.channel_mutex);
980 if (channel == NULL) {
982 * We failed in processing the offer message;
983 * we would have cleaned up the relid in that
990 * Before setting channel->rescind in vmbus_rescind_cleanup(), we
991 * should make sure the channel callback is not running any more.
993 vmbus_reset_channel_cb(channel);
996 * Now wait for offer handling to complete.
998 vmbus_rescind_cleanup(channel);
999 while (READ_ONCE(channel->probe_done) == false) {
1001 * We wait here until any channel offer is currently
1008 * At this point, the rescind handling can proceed safely.
1011 if (channel->device_obj) {
1012 if (channel->chn_rescind_callback) {
1013 channel->chn_rescind_callback(channel);
1017 * We will have to unregister this device from the
1020 dev = get_device(&channel->device_obj->device);
1022 vmbus_device_unregister(channel->device_obj);
1025 } else if (channel->primary_channel != NULL) {
1027 * Sub-channel is being rescinded. Following is the channel
1028 * close sequence when initiated from the driveri (refer to
1029 * vmbus_close() for details):
1030 * 1. Close all sub-channels first
1031 * 2. Then close the primary channel.
1033 mutex_lock(&vmbus_connection.channel_mutex);
1034 if (channel->state == CHANNEL_OPEN_STATE) {
1036 * The channel is currently not open;
1037 * it is safe for us to cleanup the channel.
1039 hv_process_channel_removal(rescind->child_relid);
1041 complete(&channel->rescind_event);
1043 mutex_unlock(&vmbus_connection.channel_mutex);
1047 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
1049 BUG_ON(!is_hvsock_channel(channel));
1051 /* We always get a rescind msg when a connection is closed. */
1052 while (!READ_ONCE(channel->probe_done) || !READ_ONCE(channel->rescind))
1055 vmbus_device_unregister(channel->device_obj);
1057 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
1061 * vmbus_onoffers_delivered -
1062 * This is invoked when all offers have been delivered.
1064 * Nothing to do here.
1066 static void vmbus_onoffers_delivered(
1067 struct vmbus_channel_message_header *hdr)
1072 * vmbus_onopen_result - Open result handler.
1074 * This is invoked when we received a response to our channel open request.
1075 * Find the matching request, copy the response and signal the requesting
1078 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
1080 struct vmbus_channel_open_result *result;
1081 struct vmbus_channel_msginfo *msginfo;
1082 struct vmbus_channel_message_header *requestheader;
1083 struct vmbus_channel_open_channel *openmsg;
1084 unsigned long flags;
1086 result = (struct vmbus_channel_open_result *)hdr;
1088 trace_vmbus_onopen_result(result);
1091 * Find the open msg, copy the result and signal/unblock the wait event
1093 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1095 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1098 (struct vmbus_channel_message_header *)msginfo->msg;
1100 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
1102 (struct vmbus_channel_open_channel *)msginfo->msg;
1103 if (openmsg->child_relid == result->child_relid &&
1104 openmsg->openid == result->openid) {
1105 memcpy(&msginfo->response.open_result,
1108 struct vmbus_channel_open_result));
1109 complete(&msginfo->waitevent);
1114 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1118 * vmbus_ongpadl_created - GPADL created handler.
1120 * This is invoked when we received a response to our gpadl create request.
1121 * Find the matching request, copy the response and signal the requesting
1124 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
1126 struct vmbus_channel_gpadl_created *gpadlcreated;
1127 struct vmbus_channel_msginfo *msginfo;
1128 struct vmbus_channel_message_header *requestheader;
1129 struct vmbus_channel_gpadl_header *gpadlheader;
1130 unsigned long flags;
1132 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
1134 trace_vmbus_ongpadl_created(gpadlcreated);
1137 * Find the establish msg, copy the result and signal/unblock the wait
1140 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1142 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1145 (struct vmbus_channel_message_header *)msginfo->msg;
1147 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
1149 (struct vmbus_channel_gpadl_header *)requestheader;
1151 if ((gpadlcreated->child_relid ==
1152 gpadlheader->child_relid) &&
1153 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1154 memcpy(&msginfo->response.gpadl_created,
1157 struct vmbus_channel_gpadl_created));
1158 complete(&msginfo->waitevent);
1163 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1167 * vmbus_ongpadl_torndown - GPADL torndown handler.
1169 * This is invoked when we received a response to our gpadl teardown request.
1170 * Find the matching request, copy the response and signal the requesting
1173 static void vmbus_ongpadl_torndown(
1174 struct vmbus_channel_message_header *hdr)
1176 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1177 struct vmbus_channel_msginfo *msginfo;
1178 struct vmbus_channel_message_header *requestheader;
1179 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1180 unsigned long flags;
1182 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1184 trace_vmbus_ongpadl_torndown(gpadl_torndown);
1187 * Find the open msg, copy the result and signal/unblock the wait event
1189 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1191 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1194 (struct vmbus_channel_message_header *)msginfo->msg;
1196 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1198 (struct vmbus_channel_gpadl_teardown *)requestheader;
1200 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1201 memcpy(&msginfo->response.gpadl_torndown,
1204 struct vmbus_channel_gpadl_torndown));
1205 complete(&msginfo->waitevent);
1210 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1214 * vmbus_onversion_response - Version response handler
1216 * This is invoked when we received a response to our initiate contact request.
1217 * Find the matching request, copy the response and signal the requesting
1220 static void vmbus_onversion_response(
1221 struct vmbus_channel_message_header *hdr)
1223 struct vmbus_channel_msginfo *msginfo;
1224 struct vmbus_channel_message_header *requestheader;
1225 struct vmbus_channel_version_response *version_response;
1226 unsigned long flags;
1228 version_response = (struct vmbus_channel_version_response *)hdr;
1230 trace_vmbus_onversion_response(version_response);
1232 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1234 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1237 (struct vmbus_channel_message_header *)msginfo->msg;
1239 if (requestheader->msgtype ==
1240 CHANNELMSG_INITIATE_CONTACT) {
1241 memcpy(&msginfo->response.version_response,
1243 sizeof(struct vmbus_channel_version_response));
1244 complete(&msginfo->waitevent);
1247 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1250 /* Channel message dispatch table */
1251 const struct vmbus_channel_message_table_entry
1252 channel_message_table[CHANNELMSG_COUNT] = {
1253 { CHANNELMSG_INVALID, 0, NULL },
1254 { CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer },
1255 { CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind },
1256 { CHANNELMSG_REQUESTOFFERS, 0, NULL },
1257 { CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered },
1258 { CHANNELMSG_OPENCHANNEL, 0, NULL },
1259 { CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result },
1260 { CHANNELMSG_CLOSECHANNEL, 0, NULL },
1261 { CHANNELMSG_GPADL_HEADER, 0, NULL },
1262 { CHANNELMSG_GPADL_BODY, 0, NULL },
1263 { CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created },
1264 { CHANNELMSG_GPADL_TEARDOWN, 0, NULL },
1265 { CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown },
1266 { CHANNELMSG_RELID_RELEASED, 0, NULL },
1267 { CHANNELMSG_INITIATE_CONTACT, 0, NULL },
1268 { CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response },
1269 { CHANNELMSG_UNLOAD, 0, NULL },
1270 { CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response },
1271 { CHANNELMSG_18, 0, NULL },
1272 { CHANNELMSG_19, 0, NULL },
1273 { CHANNELMSG_20, 0, NULL },
1274 { CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL },
1275 { CHANNELMSG_22, 0, NULL },
1276 { CHANNELMSG_TL_CONNECT_RESULT, 0, NULL },
1280 * vmbus_onmessage - Handler for channel protocol messages.
1282 * This is invoked in the vmbus worker thread context.
1284 void vmbus_onmessage(void *context)
1286 struct hv_message *msg = context;
1287 struct vmbus_channel_message_header *hdr;
1289 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1291 trace_vmbus_on_message(hdr);
1294 * vmbus_on_msg_dpc() makes sure the hdr->msgtype here can not go
1295 * out of bound and the message_handler pointer can not be NULL.
1297 channel_message_table[hdr->msgtype].message_handler(hdr);
1301 * vmbus_request_offers - Send a request to get all our pending offers.
1303 int vmbus_request_offers(void)
1305 struct vmbus_channel_message_header *msg;
1306 struct vmbus_channel_msginfo *msginfo;
1309 msginfo = kmalloc(sizeof(*msginfo) +
1310 sizeof(struct vmbus_channel_message_header),
1315 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1317 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1319 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1322 trace_vmbus_request_offers(ret);
1325 pr_err("Unable to request offers - %d\n", ret);
1337 * Retrieve the (sub) channel on which to send an outgoing request.
1338 * When a primary channel has multiple sub-channels, we try to
1339 * distribute the load equally amongst all available channels.
1341 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary)
1343 struct list_head *cur, *tmp;
1345 struct vmbus_channel *cur_channel;
1346 struct vmbus_channel *outgoing_channel = primary;
1350 if (list_empty(&primary->sc_list))
1351 return outgoing_channel;
1353 next_channel = primary->next_oc++;
1355 if (next_channel > (primary->num_sc)) {
1356 primary->next_oc = 0;
1357 return outgoing_channel;
1360 cur_cpu = hv_cpu_number_to_vp_number(smp_processor_id());
1361 list_for_each_safe(cur, tmp, &primary->sc_list) {
1362 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1363 if (cur_channel->state != CHANNEL_OPENED_STATE)
1366 if (cur_channel->target_vp == cur_cpu)
1369 if (i == next_channel)
1375 return outgoing_channel;
1377 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel);
1379 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1381 struct list_head *cur, *tmp;
1382 struct vmbus_channel *cur_channel;
1384 if (primary_channel->sc_creation_callback == NULL)
1387 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1388 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1390 primary_channel->sc_creation_callback(cur_channel);
1394 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1395 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1397 primary_channel->sc_creation_callback = sc_cr_cb;
1399 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1401 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1405 ret = !list_empty(&primary->sc_list);
1409 * Invoke the callback on sub-channel creation.
1410 * This will present a uniform interface to the
1413 invoke_sc_cb(primary);
1418 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1420 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1421 void (*chn_rescind_cb)(struct vmbus_channel *))
1423 channel->chn_rescind_callback = chn_rescind_cb;
1425 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);