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>
8 * K. Y. Srinivasan <kys@microsoft.com>
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/device.h>
15 #include <linux/interrupt.h>
16 #include <linux/sysctl.h>
17 #include <linux/slab.h>
18 #include <linux/acpi.h>
19 #include <linux/completion.h>
20 #include <linux/hyperv.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/clockchips.h>
23 #include <linux/cpu.h>
24 #include <linux/sched/isolation.h>
25 #include <linux/sched/task_stack.h>
27 #include <linux/delay.h>
28 #include <linux/notifier.h>
29 #include <linux/panic_notifier.h>
30 #include <linux/ptrace.h>
31 #include <linux/screen_info.h>
32 #include <linux/kdebug.h>
33 #include <linux/efi.h>
34 #include <linux/random.h>
35 #include <linux/kernel.h>
36 #include <linux/syscore_ops.h>
37 #include <linux/dma-map-ops.h>
38 #include <clocksource/hyperv_timer.h>
39 #include "hyperv_vmbus.h"
42 struct list_head node;
43 struct hv_vmbus_device_id id;
46 static struct acpi_device *hv_acpi_dev;
48 static struct completion probe_event;
50 static int hyperv_cpuhp_online;
52 static void *hv_panic_page;
54 static long __percpu *vmbus_evt;
56 /* Values parsed from ACPI DSDT */
61 * Boolean to control whether to report panic messages over Hyper-V.
63 * It can be set via /proc/sys/kernel/hyperv_record_panic_msg
65 static int sysctl_record_panic_msg = 1;
67 static int hyperv_report_reg(void)
69 return !sysctl_record_panic_msg || !hv_panic_page;
72 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
77 vmbus_initiate_unload(true);
80 * Hyper-V should be notified only once about a panic. If we will be
81 * doing hv_kmsg_dump() with kmsg data later, don't do the notification
84 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE
85 && hyperv_report_reg()) {
86 regs = current_pt_regs();
87 hyperv_report_panic(regs, val, false);
92 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
95 struct die_args *die = args;
96 struct pt_regs *regs = die->regs;
98 /* Don't notify Hyper-V if the die event is other than oops */
103 * Hyper-V should be notified only once about a panic. If we will be
104 * doing hv_kmsg_dump() with kmsg data later, don't do the notification
107 if (hyperv_report_reg())
108 hyperv_report_panic(regs, val, true);
112 static struct notifier_block hyperv_die_block = {
113 .notifier_call = hyperv_die_event,
115 static struct notifier_block hyperv_panic_block = {
116 .notifier_call = hyperv_panic_event,
119 static const char *fb_mmio_name = "fb_range";
120 static struct resource *fb_mmio;
121 static struct resource *hyperv_mmio;
122 static DEFINE_MUTEX(hyperv_mmio_lock);
124 static int vmbus_exists(void)
126 if (hv_acpi_dev == NULL)
132 static u8 channel_monitor_group(const struct vmbus_channel *channel)
134 return (u8)channel->offermsg.monitorid / 32;
137 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
139 return (u8)channel->offermsg.monitorid % 32;
142 static u32 channel_pending(const struct vmbus_channel *channel,
143 const struct hv_monitor_page *monitor_page)
145 u8 monitor_group = channel_monitor_group(channel);
147 return monitor_page->trigger_group[monitor_group].pending;
150 static u32 channel_latency(const struct vmbus_channel *channel,
151 const struct hv_monitor_page *monitor_page)
153 u8 monitor_group = channel_monitor_group(channel);
154 u8 monitor_offset = channel_monitor_offset(channel);
156 return monitor_page->latency[monitor_group][monitor_offset];
159 static u32 channel_conn_id(struct vmbus_channel *channel,
160 struct hv_monitor_page *monitor_page)
162 u8 monitor_group = channel_monitor_group(channel);
163 u8 monitor_offset = channel_monitor_offset(channel);
165 return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
168 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
171 struct hv_device *hv_dev = device_to_hv_device(dev);
173 if (!hv_dev->channel)
175 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
177 static DEVICE_ATTR_RO(id);
179 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
182 struct hv_device *hv_dev = device_to_hv_device(dev);
184 if (!hv_dev->channel)
186 return sprintf(buf, "%d\n", hv_dev->channel->state);
188 static DEVICE_ATTR_RO(state);
190 static ssize_t monitor_id_show(struct device *dev,
191 struct device_attribute *dev_attr, char *buf)
193 struct hv_device *hv_dev = device_to_hv_device(dev);
195 if (!hv_dev->channel)
197 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
199 static DEVICE_ATTR_RO(monitor_id);
201 static ssize_t class_id_show(struct device *dev,
202 struct device_attribute *dev_attr, char *buf)
204 struct hv_device *hv_dev = device_to_hv_device(dev);
206 if (!hv_dev->channel)
208 return sprintf(buf, "{%pUl}\n",
209 &hv_dev->channel->offermsg.offer.if_type);
211 static DEVICE_ATTR_RO(class_id);
213 static ssize_t device_id_show(struct device *dev,
214 struct device_attribute *dev_attr, char *buf)
216 struct hv_device *hv_dev = device_to_hv_device(dev);
218 if (!hv_dev->channel)
220 return sprintf(buf, "{%pUl}\n",
221 &hv_dev->channel->offermsg.offer.if_instance);
223 static DEVICE_ATTR_RO(device_id);
225 static ssize_t modalias_show(struct device *dev,
226 struct device_attribute *dev_attr, char *buf)
228 struct hv_device *hv_dev = device_to_hv_device(dev);
230 return sprintf(buf, "vmbus:%*phN\n", UUID_SIZE, &hv_dev->dev_type);
232 static DEVICE_ATTR_RO(modalias);
235 static ssize_t numa_node_show(struct device *dev,
236 struct device_attribute *attr, char *buf)
238 struct hv_device *hv_dev = device_to_hv_device(dev);
240 if (!hv_dev->channel)
243 return sprintf(buf, "%d\n", cpu_to_node(hv_dev->channel->target_cpu));
245 static DEVICE_ATTR_RO(numa_node);
248 static ssize_t server_monitor_pending_show(struct device *dev,
249 struct device_attribute *dev_attr,
252 struct hv_device *hv_dev = device_to_hv_device(dev);
254 if (!hv_dev->channel)
256 return sprintf(buf, "%d\n",
257 channel_pending(hv_dev->channel,
258 vmbus_connection.monitor_pages[0]));
260 static DEVICE_ATTR_RO(server_monitor_pending);
262 static ssize_t client_monitor_pending_show(struct device *dev,
263 struct device_attribute *dev_attr,
266 struct hv_device *hv_dev = device_to_hv_device(dev);
268 if (!hv_dev->channel)
270 return sprintf(buf, "%d\n",
271 channel_pending(hv_dev->channel,
272 vmbus_connection.monitor_pages[1]));
274 static DEVICE_ATTR_RO(client_monitor_pending);
276 static ssize_t server_monitor_latency_show(struct device *dev,
277 struct device_attribute *dev_attr,
280 struct hv_device *hv_dev = device_to_hv_device(dev);
282 if (!hv_dev->channel)
284 return sprintf(buf, "%d\n",
285 channel_latency(hv_dev->channel,
286 vmbus_connection.monitor_pages[0]));
288 static DEVICE_ATTR_RO(server_monitor_latency);
290 static ssize_t client_monitor_latency_show(struct device *dev,
291 struct device_attribute *dev_attr,
294 struct hv_device *hv_dev = device_to_hv_device(dev);
296 if (!hv_dev->channel)
298 return sprintf(buf, "%d\n",
299 channel_latency(hv_dev->channel,
300 vmbus_connection.monitor_pages[1]));
302 static DEVICE_ATTR_RO(client_monitor_latency);
304 static ssize_t server_monitor_conn_id_show(struct device *dev,
305 struct device_attribute *dev_attr,
308 struct hv_device *hv_dev = device_to_hv_device(dev);
310 if (!hv_dev->channel)
312 return sprintf(buf, "%d\n",
313 channel_conn_id(hv_dev->channel,
314 vmbus_connection.monitor_pages[0]));
316 static DEVICE_ATTR_RO(server_monitor_conn_id);
318 static ssize_t client_monitor_conn_id_show(struct device *dev,
319 struct device_attribute *dev_attr,
322 struct hv_device *hv_dev = device_to_hv_device(dev);
324 if (!hv_dev->channel)
326 return sprintf(buf, "%d\n",
327 channel_conn_id(hv_dev->channel,
328 vmbus_connection.monitor_pages[1]));
330 static DEVICE_ATTR_RO(client_monitor_conn_id);
332 static ssize_t out_intr_mask_show(struct device *dev,
333 struct device_attribute *dev_attr, char *buf)
335 struct hv_device *hv_dev = device_to_hv_device(dev);
336 struct hv_ring_buffer_debug_info outbound;
339 if (!hv_dev->channel)
342 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
347 return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
349 static DEVICE_ATTR_RO(out_intr_mask);
351 static ssize_t out_read_index_show(struct device *dev,
352 struct device_attribute *dev_attr, char *buf)
354 struct hv_device *hv_dev = device_to_hv_device(dev);
355 struct hv_ring_buffer_debug_info outbound;
358 if (!hv_dev->channel)
361 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
365 return sprintf(buf, "%d\n", outbound.current_read_index);
367 static DEVICE_ATTR_RO(out_read_index);
369 static ssize_t out_write_index_show(struct device *dev,
370 struct device_attribute *dev_attr,
373 struct hv_device *hv_dev = device_to_hv_device(dev);
374 struct hv_ring_buffer_debug_info outbound;
377 if (!hv_dev->channel)
380 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
384 return sprintf(buf, "%d\n", outbound.current_write_index);
386 static DEVICE_ATTR_RO(out_write_index);
388 static ssize_t out_read_bytes_avail_show(struct device *dev,
389 struct device_attribute *dev_attr,
392 struct hv_device *hv_dev = device_to_hv_device(dev);
393 struct hv_ring_buffer_debug_info outbound;
396 if (!hv_dev->channel)
399 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
403 return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
405 static DEVICE_ATTR_RO(out_read_bytes_avail);
407 static ssize_t out_write_bytes_avail_show(struct device *dev,
408 struct device_attribute *dev_attr,
411 struct hv_device *hv_dev = device_to_hv_device(dev);
412 struct hv_ring_buffer_debug_info outbound;
415 if (!hv_dev->channel)
418 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
422 return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
424 static DEVICE_ATTR_RO(out_write_bytes_avail);
426 static ssize_t in_intr_mask_show(struct device *dev,
427 struct device_attribute *dev_attr, char *buf)
429 struct hv_device *hv_dev = device_to_hv_device(dev);
430 struct hv_ring_buffer_debug_info inbound;
433 if (!hv_dev->channel)
436 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
440 return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
442 static DEVICE_ATTR_RO(in_intr_mask);
444 static ssize_t in_read_index_show(struct device *dev,
445 struct device_attribute *dev_attr, char *buf)
447 struct hv_device *hv_dev = device_to_hv_device(dev);
448 struct hv_ring_buffer_debug_info inbound;
451 if (!hv_dev->channel)
454 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
458 return sprintf(buf, "%d\n", inbound.current_read_index);
460 static DEVICE_ATTR_RO(in_read_index);
462 static ssize_t in_write_index_show(struct device *dev,
463 struct device_attribute *dev_attr, char *buf)
465 struct hv_device *hv_dev = device_to_hv_device(dev);
466 struct hv_ring_buffer_debug_info inbound;
469 if (!hv_dev->channel)
472 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
476 return sprintf(buf, "%d\n", inbound.current_write_index);
478 static DEVICE_ATTR_RO(in_write_index);
480 static ssize_t in_read_bytes_avail_show(struct device *dev,
481 struct device_attribute *dev_attr,
484 struct hv_device *hv_dev = device_to_hv_device(dev);
485 struct hv_ring_buffer_debug_info inbound;
488 if (!hv_dev->channel)
491 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
495 return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
497 static DEVICE_ATTR_RO(in_read_bytes_avail);
499 static ssize_t in_write_bytes_avail_show(struct device *dev,
500 struct device_attribute *dev_attr,
503 struct hv_device *hv_dev = device_to_hv_device(dev);
504 struct hv_ring_buffer_debug_info inbound;
507 if (!hv_dev->channel)
510 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
514 return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
516 static DEVICE_ATTR_RO(in_write_bytes_avail);
518 static ssize_t channel_vp_mapping_show(struct device *dev,
519 struct device_attribute *dev_attr,
522 struct hv_device *hv_dev = device_to_hv_device(dev);
523 struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
524 int buf_size = PAGE_SIZE, n_written, tot_written;
525 struct list_head *cur;
530 mutex_lock(&vmbus_connection.channel_mutex);
532 tot_written = snprintf(buf, buf_size, "%u:%u\n",
533 channel->offermsg.child_relid, channel->target_cpu);
535 list_for_each(cur, &channel->sc_list) {
536 if (tot_written >= buf_size - 1)
539 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
540 n_written = scnprintf(buf + tot_written,
541 buf_size - tot_written,
543 cur_sc->offermsg.child_relid,
545 tot_written += n_written;
548 mutex_unlock(&vmbus_connection.channel_mutex);
552 static DEVICE_ATTR_RO(channel_vp_mapping);
554 static ssize_t vendor_show(struct device *dev,
555 struct device_attribute *dev_attr,
558 struct hv_device *hv_dev = device_to_hv_device(dev);
560 return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
562 static DEVICE_ATTR_RO(vendor);
564 static ssize_t device_show(struct device *dev,
565 struct device_attribute *dev_attr,
568 struct hv_device *hv_dev = device_to_hv_device(dev);
570 return sprintf(buf, "0x%x\n", hv_dev->device_id);
572 static DEVICE_ATTR_RO(device);
574 static ssize_t driver_override_store(struct device *dev,
575 struct device_attribute *attr,
576 const char *buf, size_t count)
578 struct hv_device *hv_dev = device_to_hv_device(dev);
581 ret = driver_set_override(dev, &hv_dev->driver_override, buf, count);
588 static ssize_t driver_override_show(struct device *dev,
589 struct device_attribute *attr, char *buf)
591 struct hv_device *hv_dev = device_to_hv_device(dev);
595 len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
600 static DEVICE_ATTR_RW(driver_override);
602 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
603 static struct attribute *vmbus_dev_attrs[] = {
605 &dev_attr_state.attr,
606 &dev_attr_monitor_id.attr,
607 &dev_attr_class_id.attr,
608 &dev_attr_device_id.attr,
609 &dev_attr_modalias.attr,
611 &dev_attr_numa_node.attr,
613 &dev_attr_server_monitor_pending.attr,
614 &dev_attr_client_monitor_pending.attr,
615 &dev_attr_server_monitor_latency.attr,
616 &dev_attr_client_monitor_latency.attr,
617 &dev_attr_server_monitor_conn_id.attr,
618 &dev_attr_client_monitor_conn_id.attr,
619 &dev_attr_out_intr_mask.attr,
620 &dev_attr_out_read_index.attr,
621 &dev_attr_out_write_index.attr,
622 &dev_attr_out_read_bytes_avail.attr,
623 &dev_attr_out_write_bytes_avail.attr,
624 &dev_attr_in_intr_mask.attr,
625 &dev_attr_in_read_index.attr,
626 &dev_attr_in_write_index.attr,
627 &dev_attr_in_read_bytes_avail.attr,
628 &dev_attr_in_write_bytes_avail.attr,
629 &dev_attr_channel_vp_mapping.attr,
630 &dev_attr_vendor.attr,
631 &dev_attr_device.attr,
632 &dev_attr_driver_override.attr,
637 * Device-level attribute_group callback function. Returns the permission for
638 * each attribute, and returns 0 if an attribute is not visible.
640 static umode_t vmbus_dev_attr_is_visible(struct kobject *kobj,
641 struct attribute *attr, int idx)
643 struct device *dev = kobj_to_dev(kobj);
644 const struct hv_device *hv_dev = device_to_hv_device(dev);
646 /* Hide the monitor attributes if the monitor mechanism is not used. */
647 if (!hv_dev->channel->offermsg.monitor_allocated &&
648 (attr == &dev_attr_monitor_id.attr ||
649 attr == &dev_attr_server_monitor_pending.attr ||
650 attr == &dev_attr_client_monitor_pending.attr ||
651 attr == &dev_attr_server_monitor_latency.attr ||
652 attr == &dev_attr_client_monitor_latency.attr ||
653 attr == &dev_attr_server_monitor_conn_id.attr ||
654 attr == &dev_attr_client_monitor_conn_id.attr))
660 static const struct attribute_group vmbus_dev_group = {
661 .attrs = vmbus_dev_attrs,
662 .is_visible = vmbus_dev_attr_is_visible
664 __ATTRIBUTE_GROUPS(vmbus_dev);
666 /* Set up the attribute for /sys/bus/vmbus/hibernation */
667 static ssize_t hibernation_show(struct bus_type *bus, char *buf)
669 return sprintf(buf, "%d\n", !!hv_is_hibernation_supported());
672 static BUS_ATTR_RO(hibernation);
674 static struct attribute *vmbus_bus_attrs[] = {
675 &bus_attr_hibernation.attr,
678 static const struct attribute_group vmbus_bus_group = {
679 .attrs = vmbus_bus_attrs,
681 __ATTRIBUTE_GROUPS(vmbus_bus);
684 * vmbus_uevent - add uevent for our device
686 * This routine is invoked when a device is added or removed on the vmbus to
687 * generate a uevent to udev in the userspace. The udev will then look at its
688 * rule and the uevent generated here to load the appropriate driver
690 * The alias string will be of the form vmbus:guid where guid is the string
691 * representation of the device guid (each byte of the guid will be
692 * represented with two hex characters.
694 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
696 struct hv_device *dev = device_to_hv_device(device);
697 const char *format = "MODALIAS=vmbus:%*phN";
699 return add_uevent_var(env, format, UUID_SIZE, &dev->dev_type);
702 static const struct hv_vmbus_device_id *
703 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const guid_t *guid)
706 return NULL; /* empty device table */
708 for (; !guid_is_null(&id->guid); id++)
709 if (guid_equal(&id->guid, guid))
715 static const struct hv_vmbus_device_id *
716 hv_vmbus_dynid_match(struct hv_driver *drv, const guid_t *guid)
718 const struct hv_vmbus_device_id *id = NULL;
719 struct vmbus_dynid *dynid;
721 spin_lock(&drv->dynids.lock);
722 list_for_each_entry(dynid, &drv->dynids.list, node) {
723 if (guid_equal(&dynid->id.guid, guid)) {
728 spin_unlock(&drv->dynids.lock);
733 static const struct hv_vmbus_device_id vmbus_device_null;
736 * Return a matching hv_vmbus_device_id pointer.
737 * If there is no match, return NULL.
739 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
740 struct hv_device *dev)
742 const guid_t *guid = &dev->dev_type;
743 const struct hv_vmbus_device_id *id;
745 /* When driver_override is set, only bind to the matching driver */
746 if (dev->driver_override && strcmp(dev->driver_override, drv->name))
749 /* Look at the dynamic ids first, before the static ones */
750 id = hv_vmbus_dynid_match(drv, guid);
752 id = hv_vmbus_dev_match(drv->id_table, guid);
754 /* driver_override will always match, send a dummy id */
755 if (!id && dev->driver_override)
756 id = &vmbus_device_null;
761 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
762 static int vmbus_add_dynid(struct hv_driver *drv, guid_t *guid)
764 struct vmbus_dynid *dynid;
766 dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
770 dynid->id.guid = *guid;
772 spin_lock(&drv->dynids.lock);
773 list_add_tail(&dynid->node, &drv->dynids.list);
774 spin_unlock(&drv->dynids.lock);
776 return driver_attach(&drv->driver);
779 static void vmbus_free_dynids(struct hv_driver *drv)
781 struct vmbus_dynid *dynid, *n;
783 spin_lock(&drv->dynids.lock);
784 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
785 list_del(&dynid->node);
788 spin_unlock(&drv->dynids.lock);
792 * store_new_id - sysfs frontend to vmbus_add_dynid()
794 * Allow GUIDs to be added to an existing driver via sysfs.
796 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
799 struct hv_driver *drv = drv_to_hv_drv(driver);
803 retval = guid_parse(buf, &guid);
807 if (hv_vmbus_dynid_match(drv, &guid))
810 retval = vmbus_add_dynid(drv, &guid);
815 static DRIVER_ATTR_WO(new_id);
818 * store_remove_id - remove a PCI device ID from this driver
820 * Removes a dynamic pci device ID to this driver.
822 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
825 struct hv_driver *drv = drv_to_hv_drv(driver);
826 struct vmbus_dynid *dynid, *n;
830 retval = guid_parse(buf, &guid);
835 spin_lock(&drv->dynids.lock);
836 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
837 struct hv_vmbus_device_id *id = &dynid->id;
839 if (guid_equal(&id->guid, &guid)) {
840 list_del(&dynid->node);
846 spin_unlock(&drv->dynids.lock);
850 static DRIVER_ATTR_WO(remove_id);
852 static struct attribute *vmbus_drv_attrs[] = {
853 &driver_attr_new_id.attr,
854 &driver_attr_remove_id.attr,
857 ATTRIBUTE_GROUPS(vmbus_drv);
861 * vmbus_match - Attempt to match the specified device to the specified driver
863 static int vmbus_match(struct device *device, struct device_driver *driver)
865 struct hv_driver *drv = drv_to_hv_drv(driver);
866 struct hv_device *hv_dev = device_to_hv_device(device);
868 /* The hv_sock driver handles all hv_sock offers. */
869 if (is_hvsock_channel(hv_dev->channel))
872 if (hv_vmbus_get_id(drv, hv_dev))
879 * vmbus_probe - Add the new vmbus's child device
881 static int vmbus_probe(struct device *child_device)
884 struct hv_driver *drv =
885 drv_to_hv_drv(child_device->driver);
886 struct hv_device *dev = device_to_hv_device(child_device);
887 const struct hv_vmbus_device_id *dev_id;
889 dev_id = hv_vmbus_get_id(drv, dev);
891 ret = drv->probe(dev, dev_id);
893 pr_err("probe failed for device %s (%d)\n",
894 dev_name(child_device), ret);
897 pr_err("probe not set for driver %s\n",
898 dev_name(child_device));
905 * vmbus_dma_configure -- Configure DMA coherence for VMbus device
907 static int vmbus_dma_configure(struct device *child_device)
910 * On ARM64, propagate the DMA coherence setting from the top level
911 * VMbus ACPI device to the child VMbus device being added here.
912 * On x86/x64 coherence is assumed and these calls have no effect.
914 hv_setup_dma_ops(child_device,
915 device_get_dma_attr(&hv_acpi_dev->dev) == DEV_DMA_COHERENT);
920 * vmbus_remove - Remove a vmbus device
922 static void vmbus_remove(struct device *child_device)
924 struct hv_driver *drv;
925 struct hv_device *dev = device_to_hv_device(child_device);
927 if (child_device->driver) {
928 drv = drv_to_hv_drv(child_device->driver);
935 * vmbus_shutdown - Shutdown a vmbus device
937 static void vmbus_shutdown(struct device *child_device)
939 struct hv_driver *drv;
940 struct hv_device *dev = device_to_hv_device(child_device);
943 /* The device may not be attached yet */
944 if (!child_device->driver)
947 drv = drv_to_hv_drv(child_device->driver);
953 #ifdef CONFIG_PM_SLEEP
955 * vmbus_suspend - Suspend a vmbus device
957 static int vmbus_suspend(struct device *child_device)
959 struct hv_driver *drv;
960 struct hv_device *dev = device_to_hv_device(child_device);
962 /* The device may not be attached yet */
963 if (!child_device->driver)
966 drv = drv_to_hv_drv(child_device->driver);
970 return drv->suspend(dev);
974 * vmbus_resume - Resume a vmbus device
976 static int vmbus_resume(struct device *child_device)
978 struct hv_driver *drv;
979 struct hv_device *dev = device_to_hv_device(child_device);
981 /* The device may not be attached yet */
982 if (!child_device->driver)
985 drv = drv_to_hv_drv(child_device->driver);
989 return drv->resume(dev);
992 #define vmbus_suspend NULL
993 #define vmbus_resume NULL
994 #endif /* CONFIG_PM_SLEEP */
997 * vmbus_device_release - Final callback release of the vmbus child device
999 static void vmbus_device_release(struct device *device)
1001 struct hv_device *hv_dev = device_to_hv_device(device);
1002 struct vmbus_channel *channel = hv_dev->channel;
1004 hv_debug_rm_dev_dir(hv_dev);
1006 mutex_lock(&vmbus_connection.channel_mutex);
1007 hv_process_channel_removal(channel);
1008 mutex_unlock(&vmbus_connection.channel_mutex);
1013 * Note: we must use the "noirq" ops: see the comment before vmbus_bus_pm.
1015 * suspend_noirq/resume_noirq are set to NULL to support Suspend-to-Idle: we
1016 * shouldn't suspend the vmbus devices upon Suspend-to-Idle, otherwise there
1017 * is no way to wake up a Generation-2 VM.
1019 * The other 4 ops are for hibernation.
1022 static const struct dev_pm_ops vmbus_pm = {
1023 .suspend_noirq = NULL,
1024 .resume_noirq = NULL,
1025 .freeze_noirq = vmbus_suspend,
1026 .thaw_noirq = vmbus_resume,
1027 .poweroff_noirq = vmbus_suspend,
1028 .restore_noirq = vmbus_resume,
1031 /* The one and only one */
1032 static struct bus_type hv_bus = {
1034 .match = vmbus_match,
1035 .shutdown = vmbus_shutdown,
1036 .remove = vmbus_remove,
1037 .probe = vmbus_probe,
1038 .uevent = vmbus_uevent,
1039 .dma_configure = vmbus_dma_configure,
1040 .dev_groups = vmbus_dev_groups,
1041 .drv_groups = vmbus_drv_groups,
1042 .bus_groups = vmbus_bus_groups,
1046 struct onmessage_work_context {
1047 struct work_struct work;
1049 struct hv_message_header header;
1054 static void vmbus_onmessage_work(struct work_struct *work)
1056 struct onmessage_work_context *ctx;
1058 /* Do not process messages if we're in DISCONNECTED state */
1059 if (vmbus_connection.conn_state == DISCONNECTED)
1062 ctx = container_of(work, struct onmessage_work_context,
1064 vmbus_onmessage((struct vmbus_channel_message_header *)
1069 void vmbus_on_msg_dpc(unsigned long data)
1071 struct hv_per_cpu_context *hv_cpu = (void *)data;
1072 void *page_addr = hv_cpu->synic_message_page;
1073 struct hv_message msg_copy, *msg = (struct hv_message *)page_addr +
1075 struct vmbus_channel_message_header *hdr;
1076 enum vmbus_channel_message_type msgtype;
1077 const struct vmbus_channel_message_table_entry *entry;
1078 struct onmessage_work_context *ctx;
1083 * 'enum vmbus_channel_message_type' is supposed to always be 'u32' as
1084 * it is being used in 'struct vmbus_channel_message_header' definition
1085 * which is supposed to match hypervisor ABI.
1087 BUILD_BUG_ON(sizeof(enum vmbus_channel_message_type) != sizeof(u32));
1090 * Since the message is in memory shared with the host, an erroneous or
1091 * malicious Hyper-V could modify the message while vmbus_on_msg_dpc()
1092 * or individual message handlers are executing; to prevent this, copy
1093 * the message into private memory.
1095 memcpy(&msg_copy, msg, sizeof(struct hv_message));
1097 message_type = msg_copy.header.message_type;
1098 if (message_type == HVMSG_NONE)
1102 hdr = (struct vmbus_channel_message_header *)msg_copy.u.payload;
1103 msgtype = hdr->msgtype;
1105 trace_vmbus_on_msg_dpc(hdr);
1107 if (msgtype >= CHANNELMSG_COUNT) {
1108 WARN_ONCE(1, "unknown msgtype=%d\n", msgtype);
1112 payload_size = msg_copy.header.payload_size;
1113 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) {
1114 WARN_ONCE(1, "payload size is too large (%d)\n", payload_size);
1118 entry = &channel_message_table[msgtype];
1120 if (!entry->message_handler)
1123 if (payload_size < entry->min_payload_len) {
1124 WARN_ONCE(1, "message too short: msgtype=%d len=%d\n", msgtype, payload_size);
1128 if (entry->handler_type == VMHT_BLOCKING) {
1129 ctx = kmalloc(struct_size(ctx, msg.payload, payload_size), GFP_ATOMIC);
1133 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1134 memcpy(&ctx->msg, &msg_copy, sizeof(msg->header) + payload_size);
1137 * The host can generate a rescind message while we
1138 * may still be handling the original offer. We deal with
1139 * this condition by relying on the synchronization provided
1140 * by offer_in_progress and by channel_mutex. See also the
1141 * inline comments in vmbus_onoffer_rescind().
1144 case CHANNELMSG_RESCIND_CHANNELOFFER:
1146 * If we are handling the rescind message;
1147 * schedule the work on the global work queue.
1149 * The OFFER message and the RESCIND message should
1150 * not be handled by the same serialized work queue,
1151 * because the OFFER handler may call vmbus_open(),
1152 * which tries to open the channel by sending an
1153 * OPEN_CHANNEL message to the host and waits for
1154 * the host's response; however, if the host has
1155 * rescinded the channel before it receives the
1156 * OPEN_CHANNEL message, the host just silently
1157 * ignores the OPEN_CHANNEL message; as a result,
1158 * the guest's OFFER handler hangs for ever, if we
1159 * handle the RESCIND message in the same serialized
1160 * work queue: the RESCIND handler can not start to
1161 * run before the OFFER handler finishes.
1163 if (vmbus_connection.ignore_any_offer_msg)
1165 queue_work(vmbus_connection.rescind_work_queue, &ctx->work);
1168 case CHANNELMSG_OFFERCHANNEL:
1170 * The host sends the offer message of a given channel
1171 * before sending the rescind message of the same
1172 * channel. These messages are sent to the guest's
1173 * connect CPU; the guest then starts processing them
1174 * in the tasklet handler on this CPU:
1178 * [vmbus_on_msg_dpc()]
1179 * atomic_inc() // CHANNELMSG_OFFERCHANNEL
1182 * [vmbus_on_msg_dpc()]
1183 * schedule_work() // CHANNELMSG_RESCIND_CHANNELOFFER
1185 * We rely on the memory-ordering properties of the
1186 * queue_work() and schedule_work() primitives, which
1187 * guarantee that the atomic increment will be visible
1188 * to the CPUs which will execute the offer & rescind
1189 * works by the time these works will start execution.
1191 if (vmbus_connection.ignore_any_offer_msg)
1193 atomic_inc(&vmbus_connection.offer_in_progress);
1197 queue_work(vmbus_connection.work_queue, &ctx->work);
1200 entry->message_handler(hdr);
1203 vmbus_signal_eom(msg, message_type);
1206 #ifdef CONFIG_PM_SLEEP
1208 * Fake RESCIND_CHANNEL messages to clean up hv_sock channels by force for
1209 * hibernation, because hv_sock connections can not persist across hibernation.
1211 static void vmbus_force_channel_rescinded(struct vmbus_channel *channel)
1213 struct onmessage_work_context *ctx;
1214 struct vmbus_channel_rescind_offer *rescind;
1216 WARN_ON(!is_hvsock_channel(channel));
1219 * Allocation size is small and the allocation should really not fail,
1220 * otherwise the state of the hv_sock connections ends up in limbo.
1222 ctx = kzalloc(sizeof(*ctx) + sizeof(*rescind),
1223 GFP_KERNEL | __GFP_NOFAIL);
1226 * So far, these are not really used by Linux. Just set them to the
1227 * reasonable values conforming to the definitions of the fields.
1229 ctx->msg.header.message_type = 1;
1230 ctx->msg.header.payload_size = sizeof(*rescind);
1232 /* These values are actually used by Linux. */
1233 rescind = (struct vmbus_channel_rescind_offer *)ctx->msg.payload;
1234 rescind->header.msgtype = CHANNELMSG_RESCIND_CHANNELOFFER;
1235 rescind->child_relid = channel->offermsg.child_relid;
1237 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1239 queue_work(vmbus_connection.work_queue, &ctx->work);
1241 #endif /* CONFIG_PM_SLEEP */
1244 * Schedule all channels with events pending
1246 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1248 unsigned long *recv_int_page;
1252 * The event page can be directly checked to get the id of
1253 * the channel that has the interrupt pending.
1255 void *page_addr = hv_cpu->synic_event_page;
1256 union hv_synic_event_flags *event
1257 = (union hv_synic_event_flags *)page_addr +
1260 maxbits = HV_EVENT_FLAGS_COUNT;
1261 recv_int_page = event->flags;
1263 if (unlikely(!recv_int_page))
1266 for_each_set_bit(relid, recv_int_page, maxbits) {
1267 void (*callback_fn)(void *context);
1268 struct vmbus_channel *channel;
1270 if (!sync_test_and_clear_bit(relid, recv_int_page))
1273 /* Special case - vmbus channel protocol msg */
1278 * Pairs with the kfree_rcu() in vmbus_chan_release().
1279 * Guarantees that the channel data structure doesn't
1280 * get freed while the channel pointer below is being
1285 /* Find channel based on relid */
1286 channel = relid2channel(relid);
1287 if (channel == NULL)
1288 goto sched_unlock_rcu;
1290 if (channel->rescind)
1291 goto sched_unlock_rcu;
1294 * Make sure that the ring buffer data structure doesn't get
1295 * freed while we dereference the ring buffer pointer. Test
1296 * for the channel's onchannel_callback being NULL within a
1297 * sched_lock critical section. See also the inline comments
1298 * in vmbus_reset_channel_cb().
1300 spin_lock(&channel->sched_lock);
1302 callback_fn = channel->onchannel_callback;
1303 if (unlikely(callback_fn == NULL))
1306 trace_vmbus_chan_sched(channel);
1308 ++channel->interrupts;
1310 switch (channel->callback_mode) {
1312 (*callback_fn)(channel->channel_callback_context);
1315 case HV_CALL_BATCHED:
1316 hv_begin_read(&channel->inbound);
1318 case HV_CALL_DIRECT:
1319 tasklet_schedule(&channel->callback_event);
1323 spin_unlock(&channel->sched_lock);
1329 static void vmbus_isr(void)
1331 struct hv_per_cpu_context *hv_cpu
1332 = this_cpu_ptr(hv_context.cpu_context);
1334 struct hv_message *msg;
1336 vmbus_chan_sched(hv_cpu);
1338 page_addr = hv_cpu->synic_message_page;
1339 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1341 /* Check if there are actual msgs to be processed */
1342 if (msg->header.message_type != HVMSG_NONE) {
1343 if (msg->header.message_type == HVMSG_TIMER_EXPIRED) {
1345 vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
1347 tasklet_schedule(&hv_cpu->msg_dpc);
1350 add_interrupt_randomness(vmbus_interrupt);
1353 static irqreturn_t vmbus_percpu_isr(int irq, void *dev_id)
1360 * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1361 * buffer and call into Hyper-V to transfer the data.
1363 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1364 enum kmsg_dump_reason reason)
1366 struct kmsg_dump_iter iter;
1367 size_t bytes_written;
1369 /* We are only interested in panics. */
1370 if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1374 * Write dump contents to the page. No need to synchronize; panic should
1375 * be single-threaded.
1377 kmsg_dump_rewind(&iter);
1378 kmsg_dump_get_buffer(&iter, false, hv_panic_page, HV_HYP_PAGE_SIZE,
1383 * P3 to contain the physical address of the panic page & P4 to
1384 * contain the size of the panic data in that page. Rest of the
1385 * registers are no-op when the NOTIFY_MSG flag is set.
1387 hv_set_register(HV_REGISTER_CRASH_P0, 0);
1388 hv_set_register(HV_REGISTER_CRASH_P1, 0);
1389 hv_set_register(HV_REGISTER_CRASH_P2, 0);
1390 hv_set_register(HV_REGISTER_CRASH_P3, virt_to_phys(hv_panic_page));
1391 hv_set_register(HV_REGISTER_CRASH_P4, bytes_written);
1394 * Let Hyper-V know there is crash data available along with
1395 * the panic message.
1397 hv_set_register(HV_REGISTER_CRASH_CTL,
1398 (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG));
1401 static struct kmsg_dumper hv_kmsg_dumper = {
1402 .dump = hv_kmsg_dump,
1405 static void hv_kmsg_dump_register(void)
1409 hv_panic_page = hv_alloc_hyperv_zeroed_page();
1410 if (!hv_panic_page) {
1411 pr_err("Hyper-V: panic message page memory allocation failed\n");
1415 ret = kmsg_dump_register(&hv_kmsg_dumper);
1417 pr_err("Hyper-V: kmsg dump register error 0x%x\n", ret);
1418 hv_free_hyperv_page((unsigned long)hv_panic_page);
1419 hv_panic_page = NULL;
1423 static struct ctl_table_header *hv_ctl_table_hdr;
1426 * sysctl option to allow the user to control whether kmsg data should be
1427 * reported to Hyper-V on panic.
1429 static struct ctl_table hv_ctl_table[] = {
1431 .procname = "hyperv_record_panic_msg",
1432 .data = &sysctl_record_panic_msg,
1433 .maxlen = sizeof(int),
1435 .proc_handler = proc_dointvec_minmax,
1436 .extra1 = SYSCTL_ZERO,
1437 .extra2 = SYSCTL_ONE
1442 static struct ctl_table hv_root_table[] = {
1444 .procname = "kernel",
1446 .child = hv_ctl_table
1452 * vmbus_bus_init -Main vmbus driver initialization routine.
1455 * - initialize the vmbus driver context
1456 * - invoke the vmbus hv main init routine
1457 * - retrieve the channel offers
1459 static int vmbus_bus_init(void)
1465 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1469 ret = bus_register(&hv_bus);
1474 * VMbus interrupts are best modeled as per-cpu interrupts. If
1475 * on an architecture with support for per-cpu IRQs (e.g. ARM64),
1476 * allocate a per-cpu IRQ using standard Linux kernel functionality.
1477 * If not on such an architecture (e.g., x86/x64), then rely on
1478 * code in the arch-specific portion of the code tree to connect
1479 * the VMbus interrupt handler.
1482 if (vmbus_irq == -1) {
1483 hv_setup_vmbus_handler(vmbus_isr);
1485 vmbus_evt = alloc_percpu(long);
1486 ret = request_percpu_irq(vmbus_irq, vmbus_percpu_isr,
1487 "Hyper-V VMbus", vmbus_evt);
1489 pr_err("Can't request Hyper-V VMbus IRQ %d, Err %d",
1491 free_percpu(vmbus_evt);
1496 ret = hv_synic_alloc();
1501 * Initialize the per-cpu interrupt state and stimer state.
1502 * Then connect to the host.
1504 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1505 hv_synic_init, hv_synic_cleanup);
1508 hyperv_cpuhp_online = ret;
1510 ret = vmbus_connect();
1514 if (hv_is_isolation_supported())
1515 sysctl_record_panic_msg = 0;
1518 * Only register if the crash MSRs are available
1520 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1521 u64 hyperv_crash_ctl;
1523 * Panic message recording (sysctl_record_panic_msg)
1524 * is enabled by default in non-isolated guests and
1525 * disabled by default in isolated guests; the panic
1526 * message recording won't be available in isolated
1527 * guests should the following registration fail.
1529 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1530 if (!hv_ctl_table_hdr)
1531 pr_err("Hyper-V: sysctl table register error");
1534 * Register for panic kmsg callback only if the right
1535 * capability is supported by the hypervisor.
1537 hyperv_crash_ctl = hv_get_register(HV_REGISTER_CRASH_CTL);
1538 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG)
1539 hv_kmsg_dump_register();
1541 register_die_notifier(&hyperv_die_block);
1545 * Always register the panic notifier because we need to unload
1546 * the VMbus channel connection to prevent any VMbus
1547 * activity after the VM panics.
1549 atomic_notifier_chain_register(&panic_notifier_list,
1550 &hyperv_panic_block);
1552 vmbus_request_offers();
1557 cpuhp_remove_state(hyperv_cpuhp_online);
1561 if (vmbus_irq == -1) {
1562 hv_remove_vmbus_handler();
1564 free_percpu_irq(vmbus_irq, vmbus_evt);
1565 free_percpu(vmbus_evt);
1568 bus_unregister(&hv_bus);
1569 unregister_sysctl_table(hv_ctl_table_hdr);
1570 hv_ctl_table_hdr = NULL;
1575 * __vmbus_child_driver_register() - Register a vmbus's driver
1576 * @hv_driver: Pointer to driver structure you want to register
1577 * @owner: owner module of the drv
1578 * @mod_name: module name string
1580 * Registers the given driver with Linux through the 'driver_register()' call
1581 * and sets up the hyper-v vmbus handling for this driver.
1582 * It will return the state of the 'driver_register()' call.
1585 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1589 pr_info("registering driver %s\n", hv_driver->name);
1591 ret = vmbus_exists();
1595 hv_driver->driver.name = hv_driver->name;
1596 hv_driver->driver.owner = owner;
1597 hv_driver->driver.mod_name = mod_name;
1598 hv_driver->driver.bus = &hv_bus;
1600 spin_lock_init(&hv_driver->dynids.lock);
1601 INIT_LIST_HEAD(&hv_driver->dynids.list);
1603 ret = driver_register(&hv_driver->driver);
1607 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1610 * vmbus_driver_unregister() - Unregister a vmbus's driver
1611 * @hv_driver: Pointer to driver structure you want to
1614 * Un-register the given driver that was previous registered with a call to
1615 * vmbus_driver_register()
1617 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1619 pr_info("unregistering driver %s\n", hv_driver->name);
1621 if (!vmbus_exists()) {
1622 driver_unregister(&hv_driver->driver);
1623 vmbus_free_dynids(hv_driver);
1626 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1630 * Called when last reference to channel is gone.
1632 static void vmbus_chan_release(struct kobject *kobj)
1634 struct vmbus_channel *channel
1635 = container_of(kobj, struct vmbus_channel, kobj);
1637 kfree_rcu(channel, rcu);
1640 struct vmbus_chan_attribute {
1641 struct attribute attr;
1642 ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1643 ssize_t (*store)(struct vmbus_channel *chan,
1644 const char *buf, size_t count);
1646 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1647 struct vmbus_chan_attribute chan_attr_##_name \
1648 = __ATTR(_name, _mode, _show, _store)
1649 #define VMBUS_CHAN_ATTR_RW(_name) \
1650 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1651 #define VMBUS_CHAN_ATTR_RO(_name) \
1652 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1653 #define VMBUS_CHAN_ATTR_WO(_name) \
1654 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1656 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1657 struct attribute *attr, char *buf)
1659 const struct vmbus_chan_attribute *attribute
1660 = container_of(attr, struct vmbus_chan_attribute, attr);
1661 struct vmbus_channel *chan
1662 = container_of(kobj, struct vmbus_channel, kobj);
1664 if (!attribute->show)
1667 return attribute->show(chan, buf);
1670 static ssize_t vmbus_chan_attr_store(struct kobject *kobj,
1671 struct attribute *attr, const char *buf,
1674 const struct vmbus_chan_attribute *attribute
1675 = container_of(attr, struct vmbus_chan_attribute, attr);
1676 struct vmbus_channel *chan
1677 = container_of(kobj, struct vmbus_channel, kobj);
1679 if (!attribute->store)
1682 return attribute->store(chan, buf, count);
1685 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1686 .show = vmbus_chan_attr_show,
1687 .store = vmbus_chan_attr_store,
1690 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1692 struct hv_ring_buffer_info *rbi = &channel->outbound;
1695 mutex_lock(&rbi->ring_buffer_mutex);
1696 if (!rbi->ring_buffer) {
1697 mutex_unlock(&rbi->ring_buffer_mutex);
1701 ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1702 mutex_unlock(&rbi->ring_buffer_mutex);
1705 static VMBUS_CHAN_ATTR_RO(out_mask);
1707 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1709 struct hv_ring_buffer_info *rbi = &channel->inbound;
1712 mutex_lock(&rbi->ring_buffer_mutex);
1713 if (!rbi->ring_buffer) {
1714 mutex_unlock(&rbi->ring_buffer_mutex);
1718 ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1719 mutex_unlock(&rbi->ring_buffer_mutex);
1722 static VMBUS_CHAN_ATTR_RO(in_mask);
1724 static ssize_t read_avail_show(struct vmbus_channel *channel, char *buf)
1726 struct hv_ring_buffer_info *rbi = &channel->inbound;
1729 mutex_lock(&rbi->ring_buffer_mutex);
1730 if (!rbi->ring_buffer) {
1731 mutex_unlock(&rbi->ring_buffer_mutex);
1735 ret = sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1736 mutex_unlock(&rbi->ring_buffer_mutex);
1739 static VMBUS_CHAN_ATTR_RO(read_avail);
1741 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1743 struct hv_ring_buffer_info *rbi = &channel->outbound;
1746 mutex_lock(&rbi->ring_buffer_mutex);
1747 if (!rbi->ring_buffer) {
1748 mutex_unlock(&rbi->ring_buffer_mutex);
1752 ret = sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1753 mutex_unlock(&rbi->ring_buffer_mutex);
1756 static VMBUS_CHAN_ATTR_RO(write_avail);
1758 static ssize_t target_cpu_show(struct vmbus_channel *channel, char *buf)
1760 return sprintf(buf, "%u\n", channel->target_cpu);
1762 static ssize_t target_cpu_store(struct vmbus_channel *channel,
1763 const char *buf, size_t count)
1765 u32 target_cpu, origin_cpu;
1766 ssize_t ret = count;
1768 if (vmbus_proto_version < VERSION_WIN10_V4_1)
1771 if (sscanf(buf, "%uu", &target_cpu) != 1)
1774 /* Validate target_cpu for the cpumask_test_cpu() operation below. */
1775 if (target_cpu >= nr_cpumask_bits)
1778 if (!cpumask_test_cpu(target_cpu, housekeeping_cpumask(HK_TYPE_MANAGED_IRQ)))
1781 /* No CPUs should come up or down during this. */
1784 if (!cpu_online(target_cpu)) {
1790 * Synchronizes target_cpu_store() and channel closure:
1792 * { Initially: state = CHANNEL_OPENED }
1796 * [target_cpu_store()] [vmbus_disconnect_ring()]
1798 * LOCK channel_mutex LOCK channel_mutex
1799 * LOAD r1 = state LOAD r2 = state
1800 * IF (r1 == CHANNEL_OPENED) IF (r2 == CHANNEL_OPENED)
1801 * SEND MODIFYCHANNEL STORE state = CHANNEL_OPEN
1802 * [...] SEND CLOSECHANNEL
1803 * UNLOCK channel_mutex UNLOCK channel_mutex
1805 * Forbids: r1 == r2 == CHANNEL_OPENED (i.e., CPU1's LOCK precedes
1806 * CPU2's LOCK) && CPU2's SEND precedes CPU1's SEND
1808 * Note. The host processes the channel messages "sequentially", in
1809 * the order in which they are received on a per-partition basis.
1811 mutex_lock(&vmbus_connection.channel_mutex);
1814 * Hyper-V will ignore MODIFYCHANNEL messages for "non-open" channels;
1815 * avoid sending the message and fail here for such channels.
1817 if (channel->state != CHANNEL_OPENED_STATE) {
1819 goto cpu_store_unlock;
1822 origin_cpu = channel->target_cpu;
1823 if (target_cpu == origin_cpu)
1824 goto cpu_store_unlock;
1826 if (vmbus_send_modifychannel(channel,
1827 hv_cpu_number_to_vp_number(target_cpu))) {
1829 goto cpu_store_unlock;
1833 * For version before VERSION_WIN10_V5_3, the following warning holds:
1835 * Warning. At this point, there is *no* guarantee that the host will
1836 * have successfully processed the vmbus_send_modifychannel() request.
1837 * See the header comment of vmbus_send_modifychannel() for more info.
1839 * Lags in the processing of the above vmbus_send_modifychannel() can
1840 * result in missed interrupts if the "old" target CPU is taken offline
1841 * before Hyper-V starts sending interrupts to the "new" target CPU.
1842 * But apart from this offlining scenario, the code tolerates such
1843 * lags. It will function correctly even if a channel interrupt comes
1844 * in on a CPU that is different from the channel target_cpu value.
1847 channel->target_cpu = target_cpu;
1849 /* See init_vp_index(). */
1850 if (hv_is_perf_channel(channel))
1851 hv_update_allocated_cpus(origin_cpu, target_cpu);
1853 /* Currently set only for storvsc channels. */
1854 if (channel->change_target_cpu_callback) {
1855 (*channel->change_target_cpu_callback)(channel,
1856 origin_cpu, target_cpu);
1860 mutex_unlock(&vmbus_connection.channel_mutex);
1864 static VMBUS_CHAN_ATTR(cpu, 0644, target_cpu_show, target_cpu_store);
1866 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1869 return sprintf(buf, "%d\n",
1870 channel_pending(channel,
1871 vmbus_connection.monitor_pages[1]));
1873 static VMBUS_CHAN_ATTR(pending, 0444, channel_pending_show, NULL);
1875 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1878 return sprintf(buf, "%d\n",
1879 channel_latency(channel,
1880 vmbus_connection.monitor_pages[1]));
1882 static VMBUS_CHAN_ATTR(latency, 0444, channel_latency_show, NULL);
1884 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1886 return sprintf(buf, "%llu\n", channel->interrupts);
1888 static VMBUS_CHAN_ATTR(interrupts, 0444, channel_interrupts_show, NULL);
1890 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1892 return sprintf(buf, "%llu\n", channel->sig_events);
1894 static VMBUS_CHAN_ATTR(events, 0444, channel_events_show, NULL);
1896 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1899 return sprintf(buf, "%llu\n",
1900 (unsigned long long)channel->intr_in_full);
1902 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1904 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1907 return sprintf(buf, "%llu\n",
1908 (unsigned long long)channel->intr_out_empty);
1910 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1912 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1915 return sprintf(buf, "%llu\n",
1916 (unsigned long long)channel->out_full_first);
1918 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1920 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1923 return sprintf(buf, "%llu\n",
1924 (unsigned long long)channel->out_full_total);
1926 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1928 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1931 return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1933 static VMBUS_CHAN_ATTR(monitor_id, 0444, subchannel_monitor_id_show, NULL);
1935 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1938 return sprintf(buf, "%u\n",
1939 channel->offermsg.offer.sub_channel_index);
1941 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1943 static struct attribute *vmbus_chan_attrs[] = {
1944 &chan_attr_out_mask.attr,
1945 &chan_attr_in_mask.attr,
1946 &chan_attr_read_avail.attr,
1947 &chan_attr_write_avail.attr,
1948 &chan_attr_cpu.attr,
1949 &chan_attr_pending.attr,
1950 &chan_attr_latency.attr,
1951 &chan_attr_interrupts.attr,
1952 &chan_attr_events.attr,
1953 &chan_attr_intr_in_full.attr,
1954 &chan_attr_intr_out_empty.attr,
1955 &chan_attr_out_full_first.attr,
1956 &chan_attr_out_full_total.attr,
1957 &chan_attr_monitor_id.attr,
1958 &chan_attr_subchannel_id.attr,
1963 * Channel-level attribute_group callback function. Returns the permission for
1964 * each attribute, and returns 0 if an attribute is not visible.
1966 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1967 struct attribute *attr, int idx)
1969 const struct vmbus_channel *channel =
1970 container_of(kobj, struct vmbus_channel, kobj);
1972 /* Hide the monitor attributes if the monitor mechanism is not used. */
1973 if (!channel->offermsg.monitor_allocated &&
1974 (attr == &chan_attr_pending.attr ||
1975 attr == &chan_attr_latency.attr ||
1976 attr == &chan_attr_monitor_id.attr))
1982 static struct attribute_group vmbus_chan_group = {
1983 .attrs = vmbus_chan_attrs,
1984 .is_visible = vmbus_chan_attr_is_visible
1987 static struct kobj_type vmbus_chan_ktype = {
1988 .sysfs_ops = &vmbus_chan_sysfs_ops,
1989 .release = vmbus_chan_release,
1993 * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1995 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1997 const struct device *device = &dev->device;
1998 struct kobject *kobj = &channel->kobj;
1999 u32 relid = channel->offermsg.child_relid;
2002 kobj->kset = dev->channels_kset;
2003 ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
2010 ret = sysfs_create_group(kobj, &vmbus_chan_group);
2014 * The calling functions' error handling paths will cleanup the
2015 * empty channel directory.
2018 dev_err(device, "Unable to set up channel sysfs files\n");
2022 kobject_uevent(kobj, KOBJ_ADD);
2028 * vmbus_remove_channel_attr_group - remove the channel's attribute group
2030 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
2032 sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
2036 * vmbus_device_create - Creates and registers a new child device
2039 struct hv_device *vmbus_device_create(const guid_t *type,
2040 const guid_t *instance,
2041 struct vmbus_channel *channel)
2043 struct hv_device *child_device_obj;
2045 child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
2046 if (!child_device_obj) {
2047 pr_err("Unable to allocate device object for child device\n");
2051 child_device_obj->channel = channel;
2052 guid_copy(&child_device_obj->dev_type, type);
2053 guid_copy(&child_device_obj->dev_instance, instance);
2054 child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
2056 return child_device_obj;
2060 * vmbus_device_register - Register the child device
2062 int vmbus_device_register(struct hv_device *child_device_obj)
2064 struct kobject *kobj = &child_device_obj->device.kobj;
2067 dev_set_name(&child_device_obj->device, "%pUl",
2068 &child_device_obj->channel->offermsg.offer.if_instance);
2070 child_device_obj->device.bus = &hv_bus;
2071 child_device_obj->device.parent = &hv_acpi_dev->dev;
2072 child_device_obj->device.release = vmbus_device_release;
2074 child_device_obj->device.dma_parms = &child_device_obj->dma_parms;
2075 child_device_obj->device.dma_mask = &child_device_obj->dma_mask;
2076 dma_set_mask(&child_device_obj->device, DMA_BIT_MASK(64));
2079 * Register with the LDM. This will kick off the driver/device
2080 * binding...which will eventually call vmbus_match() and vmbus_probe()
2082 ret = device_register(&child_device_obj->device);
2084 pr_err("Unable to register child device\n");
2088 child_device_obj->channels_kset = kset_create_and_add("channels",
2090 if (!child_device_obj->channels_kset) {
2092 goto err_dev_unregister;
2095 ret = vmbus_add_channel_kobj(child_device_obj,
2096 child_device_obj->channel);
2098 pr_err("Unable to register primary channeln");
2099 goto err_kset_unregister;
2101 hv_debug_add_dev_dir(child_device_obj);
2105 err_kset_unregister:
2106 kset_unregister(child_device_obj->channels_kset);
2109 device_unregister(&child_device_obj->device);
2114 * vmbus_device_unregister - Remove the specified child device
2117 void vmbus_device_unregister(struct hv_device *device_obj)
2119 pr_debug("child device %s unregistered\n",
2120 dev_name(&device_obj->device));
2122 kset_unregister(device_obj->channels_kset);
2125 * Kick off the process of unregistering the device.
2126 * This will call vmbus_remove() and eventually vmbus_device_release()
2128 device_unregister(&device_obj->device);
2133 * VMBUS is an acpi enumerated device. Get the information we
2136 #define VTPM_BASE_ADDRESS 0xfed40000
2137 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
2139 resource_size_t start = 0;
2140 resource_size_t end = 0;
2141 struct resource *new_res;
2142 struct resource **old_res = &hyperv_mmio;
2143 struct resource **prev_res = NULL;
2146 switch (res->type) {
2149 * "Address" descriptors are for bus windows. Ignore
2150 * "memory" descriptors, which are for registers on
2153 case ACPI_RESOURCE_TYPE_ADDRESS32:
2154 start = res->data.address32.address.minimum;
2155 end = res->data.address32.address.maximum;
2158 case ACPI_RESOURCE_TYPE_ADDRESS64:
2159 start = res->data.address64.address.minimum;
2160 end = res->data.address64.address.maximum;
2164 * The IRQ information is needed only on ARM64, which Hyper-V
2165 * sets up in the extended format. IRQ information is present
2166 * on x86/x64 in the non-extended format but it is not used by
2167 * Linux. So don't bother checking for the non-extended format.
2169 case ACPI_RESOURCE_TYPE_EXTENDED_IRQ:
2170 if (!acpi_dev_resource_interrupt(res, 0, &r)) {
2171 pr_err("Unable to parse Hyper-V ACPI interrupt\n");
2174 /* ARM64 INTID for VMbus */
2175 vmbus_interrupt = res->data.extended_irq.interrupts[0];
2176 /* Linux IRQ number */
2177 vmbus_irq = r.start;
2181 /* Unused resource type */
2186 * Ignore ranges that are below 1MB, as they're not
2187 * necessary or useful here.
2192 new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
2194 return AE_NO_MEMORY;
2196 /* If this range overlaps the virtual TPM, truncate it. */
2197 if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
2198 end = VTPM_BASE_ADDRESS;
2200 new_res->name = "hyperv mmio";
2201 new_res->flags = IORESOURCE_MEM;
2202 new_res->start = start;
2206 * If two ranges are adjacent, merge them.
2214 if (((*old_res)->end + 1) == new_res->start) {
2215 (*old_res)->end = new_res->end;
2220 if ((*old_res)->start == new_res->end + 1) {
2221 (*old_res)->start = new_res->start;
2226 if ((*old_res)->start > new_res->end) {
2227 new_res->sibling = *old_res;
2229 (*prev_res)->sibling = new_res;
2235 old_res = &(*old_res)->sibling;
2242 static int vmbus_acpi_remove(struct acpi_device *device)
2244 struct resource *cur_res;
2245 struct resource *next_res;
2249 __release_region(hyperv_mmio, fb_mmio->start,
2250 resource_size(fb_mmio));
2254 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
2255 next_res = cur_res->sibling;
2263 static void vmbus_reserve_fb(void)
2267 * Make a claim for the frame buffer in the resource tree under the
2268 * first node, which will be the one below 4GB. The length seems to
2269 * be underreported, particularly in a Generation 1 VM. So start out
2270 * reserving a larger area and make it smaller until it succeeds.
2273 if (screen_info.lfb_base) {
2274 if (efi_enabled(EFI_BOOT))
2275 size = max_t(__u32, screen_info.lfb_size, 0x800000);
2277 size = max_t(__u32, screen_info.lfb_size, 0x4000000);
2279 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
2280 fb_mmio = __request_region(hyperv_mmio,
2281 screen_info.lfb_base, size,
2288 * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
2289 * @new: If successful, supplied a pointer to the
2290 * allocated MMIO space.
2291 * @device_obj: Identifies the caller
2292 * @min: Minimum guest physical address of the
2294 * @max: Maximum guest physical address
2295 * @size: Size of the range to be allocated
2296 * @align: Alignment of the range to be allocated
2297 * @fb_overlap_ok: Whether this allocation can be allowed
2298 * to overlap the video frame buffer.
2300 * This function walks the resources granted to VMBus by the
2301 * _CRS object in the ACPI namespace underneath the parent
2302 * "bridge" whether that's a root PCI bus in the Generation 1
2303 * case or a Module Device in the Generation 2 case. It then
2304 * attempts to allocate from the global MMIO pool in a way that
2305 * matches the constraints supplied in these parameters and by
2308 * Return: 0 on success, -errno on failure
2310 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
2311 resource_size_t min, resource_size_t max,
2312 resource_size_t size, resource_size_t align,
2315 struct resource *iter, *shadow;
2316 resource_size_t range_min, range_max, start;
2317 const char *dev_n = dev_name(&device_obj->device);
2321 mutex_lock(&hyperv_mmio_lock);
2324 * If overlaps with frame buffers are allowed, then first attempt to
2325 * make the allocation from within the reserved region. Because it
2326 * is already reserved, no shadow allocation is necessary.
2328 if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
2329 !(max < fb_mmio->start)) {
2331 range_min = fb_mmio->start;
2332 range_max = fb_mmio->end;
2333 start = (range_min + align - 1) & ~(align - 1);
2334 for (; start + size - 1 <= range_max; start += align) {
2335 *new = request_mem_region_exclusive(start, size, dev_n);
2343 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2344 if ((iter->start >= max) || (iter->end <= min))
2347 range_min = iter->start;
2348 range_max = iter->end;
2349 start = (range_min + align - 1) & ~(align - 1);
2350 for (; start + size - 1 <= range_max; start += align) {
2351 shadow = __request_region(iter, start, size, NULL,
2356 *new = request_mem_region_exclusive(start, size, dev_n);
2358 shadow->name = (char *)*new;
2363 __release_region(iter, start, size);
2368 mutex_unlock(&hyperv_mmio_lock);
2371 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
2374 * vmbus_free_mmio() - Free a memory-mapped I/O range.
2375 * @start: Base address of region to release.
2376 * @size: Size of the range to be allocated
2378 * This function releases anything requested by
2379 * vmbus_mmio_allocate().
2381 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
2383 struct resource *iter;
2385 mutex_lock(&hyperv_mmio_lock);
2386 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2387 if ((iter->start >= start + size) || (iter->end <= start))
2390 __release_region(iter, start, size);
2392 release_mem_region(start, size);
2393 mutex_unlock(&hyperv_mmio_lock);
2396 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2398 static int vmbus_acpi_add(struct acpi_device *device)
2401 int ret_val = -ENODEV;
2402 struct acpi_device *ancestor;
2404 hv_acpi_dev = device;
2407 * Older versions of Hyper-V for ARM64 fail to include the _CCA
2408 * method on the top level VMbus device in the DSDT. But devices
2409 * are hardware coherent in all current Hyper-V use cases, so fix
2410 * up the ACPI device to behave as if _CCA is present and indicates
2411 * hardware coherence.
2413 ACPI_COMPANION_SET(&device->dev, device);
2414 if (IS_ENABLED(CONFIG_ACPI_CCA_REQUIRED) &&
2415 device_get_dma_attr(&device->dev) == DEV_DMA_NOT_SUPPORTED) {
2416 pr_info("No ACPI _CCA found; assuming coherent device I/O\n");
2417 device->flags.cca_seen = true;
2418 device->flags.coherent_dma = true;
2421 result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2422 vmbus_walk_resources, NULL);
2424 if (ACPI_FAILURE(result))
2427 * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2428 * firmware) is the VMOD that has the mmio ranges. Get that.
2430 for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
2431 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2432 vmbus_walk_resources, NULL);
2434 if (ACPI_FAILURE(result))
2444 complete(&probe_event);
2446 vmbus_acpi_remove(device);
2450 #ifdef CONFIG_PM_SLEEP
2451 static int vmbus_bus_suspend(struct device *dev)
2453 struct hv_per_cpu_context *hv_cpu = per_cpu_ptr(
2454 hv_context.cpu_context, VMBUS_CONNECT_CPU);
2455 struct vmbus_channel *channel, *sc;
2457 tasklet_disable(&hv_cpu->msg_dpc);
2458 vmbus_connection.ignore_any_offer_msg = true;
2459 /* The tasklet_enable() takes care of providing a memory barrier */
2460 tasklet_enable(&hv_cpu->msg_dpc);
2462 /* Drain all the workqueues as we are in suspend */
2463 drain_workqueue(vmbus_connection.rescind_work_queue);
2464 drain_workqueue(vmbus_connection.work_queue);
2465 drain_workqueue(vmbus_connection.handle_primary_chan_wq);
2466 drain_workqueue(vmbus_connection.handle_sub_chan_wq);
2468 mutex_lock(&vmbus_connection.channel_mutex);
2469 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2470 if (!is_hvsock_channel(channel))
2473 vmbus_force_channel_rescinded(channel);
2475 mutex_unlock(&vmbus_connection.channel_mutex);
2478 * Wait until all the sub-channels and hv_sock channels have been
2479 * cleaned up. Sub-channels should be destroyed upon suspend, otherwise
2480 * they would conflict with the new sub-channels that will be created
2481 * in the resume path. hv_sock channels should also be destroyed, but
2482 * a hv_sock channel of an established hv_sock connection can not be
2483 * really destroyed since it may still be referenced by the userspace
2484 * application, so we just force the hv_sock channel to be rescinded
2485 * by vmbus_force_channel_rescinded(), and the userspace application
2486 * will thoroughly destroy the channel after hibernation.
2488 * Note: the counter nr_chan_close_on_suspend may never go above 0 if
2489 * the VM has no sub-channel and hv_sock channel, e.g. a 1-vCPU VM.
2491 if (atomic_read(&vmbus_connection.nr_chan_close_on_suspend) > 0)
2492 wait_for_completion(&vmbus_connection.ready_for_suspend_event);
2494 if (atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) != 0) {
2495 pr_err("Can not suspend due to a previous failed resuming\n");
2499 mutex_lock(&vmbus_connection.channel_mutex);
2501 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2503 * Remove the channel from the array of channels and invalidate
2504 * the channel's relid. Upon resume, vmbus_onoffer() will fix
2505 * up the relid (and other fields, if necessary) and add the
2506 * channel back to the array.
2508 vmbus_channel_unmap_relid(channel);
2509 channel->offermsg.child_relid = INVALID_RELID;
2511 if (is_hvsock_channel(channel)) {
2512 if (!channel->rescind) {
2513 pr_err("hv_sock channel not rescinded!\n");
2519 list_for_each_entry(sc, &channel->sc_list, sc_list) {
2520 pr_err("Sub-channel not deleted!\n");
2524 atomic_inc(&vmbus_connection.nr_chan_fixup_on_resume);
2527 mutex_unlock(&vmbus_connection.channel_mutex);
2529 vmbus_initiate_unload(false);
2531 /* Reset the event for the next resume. */
2532 reinit_completion(&vmbus_connection.ready_for_resume_event);
2537 static int vmbus_bus_resume(struct device *dev)
2539 struct vmbus_channel_msginfo *msginfo;
2543 vmbus_connection.ignore_any_offer_msg = false;
2546 * We only use the 'vmbus_proto_version', which was in use before
2547 * hibernation, to re-negotiate with the host.
2549 if (!vmbus_proto_version) {
2550 pr_err("Invalid proto version = 0x%x\n", vmbus_proto_version);
2554 msgsize = sizeof(*msginfo) +
2555 sizeof(struct vmbus_channel_initiate_contact);
2557 msginfo = kzalloc(msgsize, GFP_KERNEL);
2559 if (msginfo == NULL)
2562 ret = vmbus_negotiate_version(msginfo, vmbus_proto_version);
2569 WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) == 0);
2571 vmbus_request_offers();
2573 if (wait_for_completion_timeout(
2574 &vmbus_connection.ready_for_resume_event, 10 * HZ) == 0)
2575 pr_err("Some vmbus device is missing after suspending?\n");
2577 /* Reset the event for the next suspend. */
2578 reinit_completion(&vmbus_connection.ready_for_suspend_event);
2583 #define vmbus_bus_suspend NULL
2584 #define vmbus_bus_resume NULL
2585 #endif /* CONFIG_PM_SLEEP */
2587 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
2592 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2595 * Note: we must use the "no_irq" ops, otherwise hibernation can not work with
2596 * PCI device assignment, because "pci_dev_pm_ops" uses the "noirq" ops: in
2597 * the resume path, the pci "noirq" restore op runs before "non-noirq" op (see
2598 * resume_target_kernel() -> dpm_resume_start(), and hibernation_restore() ->
2599 * dpm_resume_end()). This means vmbus_bus_resume() and the pci-hyperv's
2600 * resume callback must also run via the "noirq" ops.
2602 * Set suspend_noirq/resume_noirq to NULL for Suspend-to-Idle: see the comment
2603 * earlier in this file before vmbus_pm.
2606 static const struct dev_pm_ops vmbus_bus_pm = {
2607 .suspend_noirq = NULL,
2608 .resume_noirq = NULL,
2609 .freeze_noirq = vmbus_bus_suspend,
2610 .thaw_noirq = vmbus_bus_resume,
2611 .poweroff_noirq = vmbus_bus_suspend,
2612 .restore_noirq = vmbus_bus_resume
2615 static struct acpi_driver vmbus_acpi_driver = {
2617 .ids = vmbus_acpi_device_ids,
2619 .add = vmbus_acpi_add,
2620 .remove = vmbus_acpi_remove,
2622 .drv.pm = &vmbus_bus_pm,
2625 static void hv_kexec_handler(void)
2627 hv_stimer_global_cleanup();
2628 vmbus_initiate_unload(false);
2629 /* Make sure conn_state is set as hv_synic_cleanup checks for it */
2631 cpuhp_remove_state(hyperv_cpuhp_online);
2634 static void hv_crash_handler(struct pt_regs *regs)
2638 vmbus_initiate_unload(true);
2640 * In crash handler we can't schedule synic cleanup for all CPUs,
2641 * doing the cleanup for current CPU only. This should be sufficient
2644 cpu = smp_processor_id();
2645 hv_stimer_cleanup(cpu);
2646 hv_synic_disable_regs(cpu);
2649 static int hv_synic_suspend(void)
2652 * When we reach here, all the non-boot CPUs have been offlined.
2653 * If we're in a legacy configuration where stimer Direct Mode is
2654 * not enabled, the stimers on the non-boot CPUs have been unbound
2655 * in hv_synic_cleanup() -> hv_stimer_legacy_cleanup() ->
2656 * hv_stimer_cleanup() -> clockevents_unbind_device().
2658 * hv_synic_suspend() only runs on CPU0 with interrupts disabled.
2659 * Here we do not call hv_stimer_legacy_cleanup() on CPU0 because:
2660 * 1) it's unnecessary as interrupts remain disabled between
2661 * syscore_suspend() and syscore_resume(): see create_image() and
2662 * resume_target_kernel()
2663 * 2) the stimer on CPU0 is automatically disabled later by
2664 * syscore_suspend() -> timekeeping_suspend() -> tick_suspend() -> ...
2665 * -> clockevents_shutdown() -> ... -> hv_ce_shutdown()
2666 * 3) a warning would be triggered if we call
2667 * clockevents_unbind_device(), which may sleep, in an
2668 * interrupts-disabled context.
2671 hv_synic_disable_regs(0);
2676 static void hv_synic_resume(void)
2678 hv_synic_enable_regs(0);
2681 * Note: we don't need to call hv_stimer_init(0), because the timer
2682 * on CPU0 is not unbound in hv_synic_suspend(), and the timer is
2683 * automatically re-enabled in timekeeping_resume().
2687 /* The callbacks run only on CPU0, with irqs_disabled. */
2688 static struct syscore_ops hv_synic_syscore_ops = {
2689 .suspend = hv_synic_suspend,
2690 .resume = hv_synic_resume,
2693 static int __init hv_acpi_init(void)
2697 if (!hv_is_hyperv_initialized())
2700 if (hv_root_partition)
2703 init_completion(&probe_event);
2706 * Get ACPI resources first.
2708 ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2713 t = wait_for_completion_timeout(&probe_event, 5*HZ);
2720 * If we're on an architecture with a hardcoded hypervisor
2721 * vector (i.e. x86/x64), override the VMbus interrupt found
2722 * in the ACPI tables. Ensure vmbus_irq is not set since the
2723 * normal Linux IRQ mechanism is not used in this case.
2725 #ifdef HYPERVISOR_CALLBACK_VECTOR
2726 vmbus_interrupt = HYPERVISOR_CALLBACK_VECTOR;
2732 ret = vmbus_bus_init();
2736 hv_setup_kexec_handler(hv_kexec_handler);
2737 hv_setup_crash_handler(hv_crash_handler);
2739 register_syscore_ops(&hv_synic_syscore_ops);
2744 acpi_bus_unregister_driver(&vmbus_acpi_driver);
2749 static void __exit vmbus_exit(void)
2753 unregister_syscore_ops(&hv_synic_syscore_ops);
2755 hv_remove_kexec_handler();
2756 hv_remove_crash_handler();
2757 vmbus_connection.conn_state = DISCONNECTED;
2758 hv_stimer_global_cleanup();
2760 if (vmbus_irq == -1) {
2761 hv_remove_vmbus_handler();
2763 free_percpu_irq(vmbus_irq, vmbus_evt);
2764 free_percpu(vmbus_evt);
2766 for_each_online_cpu(cpu) {
2767 struct hv_per_cpu_context *hv_cpu
2768 = per_cpu_ptr(hv_context.cpu_context, cpu);
2770 tasklet_kill(&hv_cpu->msg_dpc);
2772 hv_debug_rm_all_dir();
2774 vmbus_free_channels();
2775 kfree(vmbus_connection.channels);
2777 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2778 kmsg_dump_unregister(&hv_kmsg_dumper);
2779 unregister_die_notifier(&hyperv_die_block);
2783 * The panic notifier is always registered, hence we should
2784 * also unconditionally unregister it here as well.
2786 atomic_notifier_chain_unregister(&panic_notifier_list,
2787 &hyperv_panic_block);
2789 free_page((unsigned long)hv_panic_page);
2790 unregister_sysctl_table(hv_ctl_table_hdr);
2791 hv_ctl_table_hdr = NULL;
2792 bus_unregister(&hv_bus);
2794 cpuhp_remove_state(hyperv_cpuhp_online);
2796 acpi_bus_unregister_driver(&vmbus_acpi_driver);
2800 MODULE_LICENSE("GPL");
2801 MODULE_DESCRIPTION("Microsoft Hyper-V VMBus Driver");
2803 subsys_initcall(hv_acpi_init);
2804 module_exit(vmbus_exit);