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/task_stack.h>
26 #include <asm/mshyperv.h>
27 #include <linux/delay.h>
28 #include <linux/notifier.h>
29 #include <linux/ptrace.h>
30 #include <linux/screen_info.h>
31 #include <linux/kdebug.h>
32 #include <linux/efi.h>
33 #include <linux/random.h>
34 #include <linux/kernel.h>
35 #include <linux/syscore_ops.h>
36 #include <clocksource/hyperv_timer.h>
37 #include "hyperv_vmbus.h"
40 struct list_head node;
41 struct hv_vmbus_device_id id;
44 static struct acpi_device *hv_acpi_dev;
46 static struct completion probe_event;
48 static int hyperv_cpuhp_online;
50 static void *hv_panic_page;
53 * Boolean to control whether to report panic messages over Hyper-V.
55 * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
57 static int sysctl_record_panic_msg = 1;
59 static int hyperv_report_reg(void)
61 return !sysctl_record_panic_msg || !hv_panic_page;
64 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
69 vmbus_initiate_unload(true);
72 * Hyper-V should be notified only once about a panic. If we will be
73 * doing hyperv_report_panic_msg() later with kmsg data, don't do
74 * the notification here.
76 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE
77 && hyperv_report_reg()) {
78 regs = current_pt_regs();
79 hyperv_report_panic(regs, val, false);
84 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
87 struct die_args *die = (struct die_args *)args;
88 struct pt_regs *regs = die->regs;
91 * Hyper-V should be notified only once about a panic. If we will be
92 * doing hyperv_report_panic_msg() later with kmsg data, don't do
93 * the notification here.
95 if (hyperv_report_reg())
96 hyperv_report_panic(regs, val, true);
100 static struct notifier_block hyperv_die_block = {
101 .notifier_call = hyperv_die_event,
103 static struct notifier_block hyperv_panic_block = {
104 .notifier_call = hyperv_panic_event,
107 static const char *fb_mmio_name = "fb_range";
108 static struct resource *fb_mmio;
109 static struct resource *hyperv_mmio;
110 static DEFINE_MUTEX(hyperv_mmio_lock);
112 static int vmbus_exists(void)
114 if (hv_acpi_dev == NULL)
120 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
121 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
124 for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
125 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
128 static u8 channel_monitor_group(const struct vmbus_channel *channel)
130 return (u8)channel->offermsg.monitorid / 32;
133 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
135 return (u8)channel->offermsg.monitorid % 32;
138 static u32 channel_pending(const struct vmbus_channel *channel,
139 const struct hv_monitor_page *monitor_page)
141 u8 monitor_group = channel_monitor_group(channel);
143 return monitor_page->trigger_group[monitor_group].pending;
146 static u32 channel_latency(const struct vmbus_channel *channel,
147 const struct hv_monitor_page *monitor_page)
149 u8 monitor_group = channel_monitor_group(channel);
150 u8 monitor_offset = channel_monitor_offset(channel);
152 return monitor_page->latency[monitor_group][monitor_offset];
155 static u32 channel_conn_id(struct vmbus_channel *channel,
156 struct hv_monitor_page *monitor_page)
158 u8 monitor_group = channel_monitor_group(channel);
159 u8 monitor_offset = channel_monitor_offset(channel);
160 return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
163 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
166 struct hv_device *hv_dev = device_to_hv_device(dev);
168 if (!hv_dev->channel)
170 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
172 static DEVICE_ATTR_RO(id);
174 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
177 struct hv_device *hv_dev = device_to_hv_device(dev);
179 if (!hv_dev->channel)
181 return sprintf(buf, "%d\n", hv_dev->channel->state);
183 static DEVICE_ATTR_RO(state);
185 static ssize_t monitor_id_show(struct device *dev,
186 struct device_attribute *dev_attr, char *buf)
188 struct hv_device *hv_dev = device_to_hv_device(dev);
190 if (!hv_dev->channel)
192 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
194 static DEVICE_ATTR_RO(monitor_id);
196 static ssize_t class_id_show(struct device *dev,
197 struct device_attribute *dev_attr, char *buf)
199 struct hv_device *hv_dev = device_to_hv_device(dev);
201 if (!hv_dev->channel)
203 return sprintf(buf, "{%pUl}\n",
204 hv_dev->channel->offermsg.offer.if_type.b);
206 static DEVICE_ATTR_RO(class_id);
208 static ssize_t device_id_show(struct device *dev,
209 struct device_attribute *dev_attr, char *buf)
211 struct hv_device *hv_dev = device_to_hv_device(dev);
213 if (!hv_dev->channel)
215 return sprintf(buf, "{%pUl}\n",
216 hv_dev->channel->offermsg.offer.if_instance.b);
218 static DEVICE_ATTR_RO(device_id);
220 static ssize_t modalias_show(struct device *dev,
221 struct device_attribute *dev_attr, char *buf)
223 struct hv_device *hv_dev = device_to_hv_device(dev);
224 char alias_name[VMBUS_ALIAS_LEN + 1];
226 print_alias_name(hv_dev, alias_name);
227 return sprintf(buf, "vmbus:%s\n", alias_name);
229 static DEVICE_ATTR_RO(modalias);
232 static ssize_t numa_node_show(struct device *dev,
233 struct device_attribute *attr, char *buf)
235 struct hv_device *hv_dev = device_to_hv_device(dev);
237 if (!hv_dev->channel)
240 return sprintf(buf, "%d\n", hv_dev->channel->numa_node);
242 static DEVICE_ATTR_RO(numa_node);
245 static ssize_t server_monitor_pending_show(struct device *dev,
246 struct device_attribute *dev_attr,
249 struct hv_device *hv_dev = device_to_hv_device(dev);
251 if (!hv_dev->channel)
253 return sprintf(buf, "%d\n",
254 channel_pending(hv_dev->channel,
255 vmbus_connection.monitor_pages[0]));
257 static DEVICE_ATTR_RO(server_monitor_pending);
259 static ssize_t client_monitor_pending_show(struct device *dev,
260 struct device_attribute *dev_attr,
263 struct hv_device *hv_dev = device_to_hv_device(dev);
265 if (!hv_dev->channel)
267 return sprintf(buf, "%d\n",
268 channel_pending(hv_dev->channel,
269 vmbus_connection.monitor_pages[1]));
271 static DEVICE_ATTR_RO(client_monitor_pending);
273 static ssize_t server_monitor_latency_show(struct device *dev,
274 struct device_attribute *dev_attr,
277 struct hv_device *hv_dev = device_to_hv_device(dev);
279 if (!hv_dev->channel)
281 return sprintf(buf, "%d\n",
282 channel_latency(hv_dev->channel,
283 vmbus_connection.monitor_pages[0]));
285 static DEVICE_ATTR_RO(server_monitor_latency);
287 static ssize_t client_monitor_latency_show(struct device *dev,
288 struct device_attribute *dev_attr,
291 struct hv_device *hv_dev = device_to_hv_device(dev);
293 if (!hv_dev->channel)
295 return sprintf(buf, "%d\n",
296 channel_latency(hv_dev->channel,
297 vmbus_connection.monitor_pages[1]));
299 static DEVICE_ATTR_RO(client_monitor_latency);
301 static ssize_t server_monitor_conn_id_show(struct device *dev,
302 struct device_attribute *dev_attr,
305 struct hv_device *hv_dev = device_to_hv_device(dev);
307 if (!hv_dev->channel)
309 return sprintf(buf, "%d\n",
310 channel_conn_id(hv_dev->channel,
311 vmbus_connection.monitor_pages[0]));
313 static DEVICE_ATTR_RO(server_monitor_conn_id);
315 static ssize_t client_monitor_conn_id_show(struct device *dev,
316 struct device_attribute *dev_attr,
319 struct hv_device *hv_dev = device_to_hv_device(dev);
321 if (!hv_dev->channel)
323 return sprintf(buf, "%d\n",
324 channel_conn_id(hv_dev->channel,
325 vmbus_connection.monitor_pages[1]));
327 static DEVICE_ATTR_RO(client_monitor_conn_id);
329 static ssize_t out_intr_mask_show(struct device *dev,
330 struct device_attribute *dev_attr, char *buf)
332 struct hv_device *hv_dev = device_to_hv_device(dev);
333 struct hv_ring_buffer_debug_info outbound;
336 if (!hv_dev->channel)
339 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
344 return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
346 static DEVICE_ATTR_RO(out_intr_mask);
348 static ssize_t out_read_index_show(struct device *dev,
349 struct device_attribute *dev_attr, char *buf)
351 struct hv_device *hv_dev = device_to_hv_device(dev);
352 struct hv_ring_buffer_debug_info outbound;
355 if (!hv_dev->channel)
358 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
362 return sprintf(buf, "%d\n", outbound.current_read_index);
364 static DEVICE_ATTR_RO(out_read_index);
366 static ssize_t out_write_index_show(struct device *dev,
367 struct device_attribute *dev_attr,
370 struct hv_device *hv_dev = device_to_hv_device(dev);
371 struct hv_ring_buffer_debug_info outbound;
374 if (!hv_dev->channel)
377 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
381 return sprintf(buf, "%d\n", outbound.current_write_index);
383 static DEVICE_ATTR_RO(out_write_index);
385 static ssize_t out_read_bytes_avail_show(struct device *dev,
386 struct device_attribute *dev_attr,
389 struct hv_device *hv_dev = device_to_hv_device(dev);
390 struct hv_ring_buffer_debug_info outbound;
393 if (!hv_dev->channel)
396 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
400 return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
402 static DEVICE_ATTR_RO(out_read_bytes_avail);
404 static ssize_t out_write_bytes_avail_show(struct device *dev,
405 struct device_attribute *dev_attr,
408 struct hv_device *hv_dev = device_to_hv_device(dev);
409 struct hv_ring_buffer_debug_info outbound;
412 if (!hv_dev->channel)
415 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
419 return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
421 static DEVICE_ATTR_RO(out_write_bytes_avail);
423 static ssize_t in_intr_mask_show(struct device *dev,
424 struct device_attribute *dev_attr, char *buf)
426 struct hv_device *hv_dev = device_to_hv_device(dev);
427 struct hv_ring_buffer_debug_info inbound;
430 if (!hv_dev->channel)
433 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
437 return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
439 static DEVICE_ATTR_RO(in_intr_mask);
441 static ssize_t in_read_index_show(struct device *dev,
442 struct device_attribute *dev_attr, char *buf)
444 struct hv_device *hv_dev = device_to_hv_device(dev);
445 struct hv_ring_buffer_debug_info inbound;
448 if (!hv_dev->channel)
451 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
455 return sprintf(buf, "%d\n", inbound.current_read_index);
457 static DEVICE_ATTR_RO(in_read_index);
459 static ssize_t in_write_index_show(struct device *dev,
460 struct device_attribute *dev_attr, char *buf)
462 struct hv_device *hv_dev = device_to_hv_device(dev);
463 struct hv_ring_buffer_debug_info inbound;
466 if (!hv_dev->channel)
469 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
473 return sprintf(buf, "%d\n", inbound.current_write_index);
475 static DEVICE_ATTR_RO(in_write_index);
477 static ssize_t in_read_bytes_avail_show(struct device *dev,
478 struct device_attribute *dev_attr,
481 struct hv_device *hv_dev = device_to_hv_device(dev);
482 struct hv_ring_buffer_debug_info inbound;
485 if (!hv_dev->channel)
488 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
492 return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
494 static DEVICE_ATTR_RO(in_read_bytes_avail);
496 static ssize_t in_write_bytes_avail_show(struct device *dev,
497 struct device_attribute *dev_attr,
500 struct hv_device *hv_dev = device_to_hv_device(dev);
501 struct hv_ring_buffer_debug_info inbound;
504 if (!hv_dev->channel)
507 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
511 return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
513 static DEVICE_ATTR_RO(in_write_bytes_avail);
515 static ssize_t channel_vp_mapping_show(struct device *dev,
516 struct device_attribute *dev_attr,
519 struct hv_device *hv_dev = device_to_hv_device(dev);
520 struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
522 int buf_size = PAGE_SIZE, n_written, tot_written;
523 struct list_head *cur;
528 tot_written = snprintf(buf, buf_size, "%u:%u\n",
529 channel->offermsg.child_relid, channel->target_cpu);
531 spin_lock_irqsave(&channel->lock, flags);
533 list_for_each(cur, &channel->sc_list) {
534 if (tot_written >= buf_size - 1)
537 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
538 n_written = scnprintf(buf + tot_written,
539 buf_size - tot_written,
541 cur_sc->offermsg.child_relid,
543 tot_written += n_written;
546 spin_unlock_irqrestore(&channel->lock, flags);
550 static DEVICE_ATTR_RO(channel_vp_mapping);
552 static ssize_t vendor_show(struct device *dev,
553 struct device_attribute *dev_attr,
556 struct hv_device *hv_dev = device_to_hv_device(dev);
557 return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
559 static DEVICE_ATTR_RO(vendor);
561 static ssize_t device_show(struct device *dev,
562 struct device_attribute *dev_attr,
565 struct hv_device *hv_dev = device_to_hv_device(dev);
566 return sprintf(buf, "0x%x\n", hv_dev->device_id);
568 static DEVICE_ATTR_RO(device);
570 static ssize_t driver_override_store(struct device *dev,
571 struct device_attribute *attr,
572 const char *buf, size_t count)
574 struct hv_device *hv_dev = device_to_hv_device(dev);
575 char *driver_override, *old, *cp;
577 /* We need to keep extra room for a newline */
578 if (count >= (PAGE_SIZE - 1))
581 driver_override = kstrndup(buf, count, GFP_KERNEL);
582 if (!driver_override)
585 cp = strchr(driver_override, '\n');
590 old = hv_dev->driver_override;
591 if (strlen(driver_override)) {
592 hv_dev->driver_override = driver_override;
594 kfree(driver_override);
595 hv_dev->driver_override = NULL;
604 static ssize_t driver_override_show(struct device *dev,
605 struct device_attribute *attr, char *buf)
607 struct hv_device *hv_dev = device_to_hv_device(dev);
611 len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
616 static DEVICE_ATTR_RW(driver_override);
618 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
619 static struct attribute *vmbus_dev_attrs[] = {
621 &dev_attr_state.attr,
622 &dev_attr_monitor_id.attr,
623 &dev_attr_class_id.attr,
624 &dev_attr_device_id.attr,
625 &dev_attr_modalias.attr,
627 &dev_attr_numa_node.attr,
629 &dev_attr_server_monitor_pending.attr,
630 &dev_attr_client_monitor_pending.attr,
631 &dev_attr_server_monitor_latency.attr,
632 &dev_attr_client_monitor_latency.attr,
633 &dev_attr_server_monitor_conn_id.attr,
634 &dev_attr_client_monitor_conn_id.attr,
635 &dev_attr_out_intr_mask.attr,
636 &dev_attr_out_read_index.attr,
637 &dev_attr_out_write_index.attr,
638 &dev_attr_out_read_bytes_avail.attr,
639 &dev_attr_out_write_bytes_avail.attr,
640 &dev_attr_in_intr_mask.attr,
641 &dev_attr_in_read_index.attr,
642 &dev_attr_in_write_index.attr,
643 &dev_attr_in_read_bytes_avail.attr,
644 &dev_attr_in_write_bytes_avail.attr,
645 &dev_attr_channel_vp_mapping.attr,
646 &dev_attr_vendor.attr,
647 &dev_attr_device.attr,
648 &dev_attr_driver_override.attr,
653 * Device-level attribute_group callback function. Returns the permission for
654 * each attribute, and returns 0 if an attribute is not visible.
656 static umode_t vmbus_dev_attr_is_visible(struct kobject *kobj,
657 struct attribute *attr, int idx)
659 struct device *dev = kobj_to_dev(kobj);
660 const struct hv_device *hv_dev = device_to_hv_device(dev);
662 /* Hide the monitor attributes if the monitor mechanism is not used. */
663 if (!hv_dev->channel->offermsg.monitor_allocated &&
664 (attr == &dev_attr_monitor_id.attr ||
665 attr == &dev_attr_server_monitor_pending.attr ||
666 attr == &dev_attr_client_monitor_pending.attr ||
667 attr == &dev_attr_server_monitor_latency.attr ||
668 attr == &dev_attr_client_monitor_latency.attr ||
669 attr == &dev_attr_server_monitor_conn_id.attr ||
670 attr == &dev_attr_client_monitor_conn_id.attr))
676 static const struct attribute_group vmbus_dev_group = {
677 .attrs = vmbus_dev_attrs,
678 .is_visible = vmbus_dev_attr_is_visible
680 __ATTRIBUTE_GROUPS(vmbus_dev);
683 * vmbus_uevent - add uevent for our device
685 * This routine is invoked when a device is added or removed on the vmbus to
686 * generate a uevent to udev in the userspace. The udev will then look at its
687 * rule and the uevent generated here to load the appropriate driver
689 * The alias string will be of the form vmbus:guid where guid is the string
690 * representation of the device guid (each byte of the guid will be
691 * represented with two hex characters.
693 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
695 struct hv_device *dev = device_to_hv_device(device);
697 char alias_name[VMBUS_ALIAS_LEN + 1];
699 print_alias_name(dev, alias_name);
700 ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
704 static const struct hv_vmbus_device_id *
705 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const guid_t *guid)
708 return NULL; /* empty device table */
710 for (; !guid_is_null(&id->guid); id++)
711 if (guid_equal(&id->guid, guid))
717 static const struct hv_vmbus_device_id *
718 hv_vmbus_dynid_match(struct hv_driver *drv, const guid_t *guid)
720 const struct hv_vmbus_device_id *id = NULL;
721 struct vmbus_dynid *dynid;
723 spin_lock(&drv->dynids.lock);
724 list_for_each_entry(dynid, &drv->dynids.list, node) {
725 if (guid_equal(&dynid->id.guid, guid)) {
730 spin_unlock(&drv->dynids.lock);
735 static const struct hv_vmbus_device_id vmbus_device_null;
738 * Return a matching hv_vmbus_device_id pointer.
739 * If there is no match, return NULL.
741 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
742 struct hv_device *dev)
744 const guid_t *guid = &dev->dev_type;
745 const struct hv_vmbus_device_id *id;
747 /* When driver_override is set, only bind to the matching driver */
748 if (dev->driver_override && strcmp(dev->driver_override, drv->name))
751 /* Look at the dynamic ids first, before the static ones */
752 id = hv_vmbus_dynid_match(drv, guid);
754 id = hv_vmbus_dev_match(drv->id_table, guid);
756 /* driver_override will always match, send a dummy id */
757 if (!id && dev->driver_override)
758 id = &vmbus_device_null;
763 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
764 static int vmbus_add_dynid(struct hv_driver *drv, guid_t *guid)
766 struct vmbus_dynid *dynid;
768 dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
772 dynid->id.guid = *guid;
774 spin_lock(&drv->dynids.lock);
775 list_add_tail(&dynid->node, &drv->dynids.list);
776 spin_unlock(&drv->dynids.lock);
778 return driver_attach(&drv->driver);
781 static void vmbus_free_dynids(struct hv_driver *drv)
783 struct vmbus_dynid *dynid, *n;
785 spin_lock(&drv->dynids.lock);
786 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
787 list_del(&dynid->node);
790 spin_unlock(&drv->dynids.lock);
794 * store_new_id - sysfs frontend to vmbus_add_dynid()
796 * Allow GUIDs to be added to an existing driver via sysfs.
798 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
801 struct hv_driver *drv = drv_to_hv_drv(driver);
805 retval = guid_parse(buf, &guid);
809 if (hv_vmbus_dynid_match(drv, &guid))
812 retval = vmbus_add_dynid(drv, &guid);
817 static DRIVER_ATTR_WO(new_id);
820 * store_remove_id - remove a PCI device ID from this driver
822 * Removes a dynamic pci device ID to this driver.
824 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
827 struct hv_driver *drv = drv_to_hv_drv(driver);
828 struct vmbus_dynid *dynid, *n;
832 retval = guid_parse(buf, &guid);
837 spin_lock(&drv->dynids.lock);
838 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
839 struct hv_vmbus_device_id *id = &dynid->id;
841 if (guid_equal(&id->guid, &guid)) {
842 list_del(&dynid->node);
848 spin_unlock(&drv->dynids.lock);
852 static DRIVER_ATTR_WO(remove_id);
854 static struct attribute *vmbus_drv_attrs[] = {
855 &driver_attr_new_id.attr,
856 &driver_attr_remove_id.attr,
859 ATTRIBUTE_GROUPS(vmbus_drv);
863 * vmbus_match - Attempt to match the specified device to the specified driver
865 static int vmbus_match(struct device *device, struct device_driver *driver)
867 struct hv_driver *drv = drv_to_hv_drv(driver);
868 struct hv_device *hv_dev = device_to_hv_device(device);
870 /* The hv_sock driver handles all hv_sock offers. */
871 if (is_hvsock_channel(hv_dev->channel))
874 if (hv_vmbus_get_id(drv, hv_dev))
881 * vmbus_probe - Add the new vmbus's child device
883 static int vmbus_probe(struct device *child_device)
886 struct hv_driver *drv =
887 drv_to_hv_drv(child_device->driver);
888 struct hv_device *dev = device_to_hv_device(child_device);
889 const struct hv_vmbus_device_id *dev_id;
891 dev_id = hv_vmbus_get_id(drv, dev);
893 ret = drv->probe(dev, dev_id);
895 pr_err("probe failed for device %s (%d)\n",
896 dev_name(child_device), ret);
899 pr_err("probe not set for driver %s\n",
900 dev_name(child_device));
907 * vmbus_remove - Remove a vmbus device
909 static int vmbus_remove(struct device *child_device)
911 struct hv_driver *drv;
912 struct hv_device *dev = device_to_hv_device(child_device);
914 if (child_device->driver) {
915 drv = drv_to_hv_drv(child_device->driver);
925 * vmbus_shutdown - Shutdown a vmbus device
927 static void vmbus_shutdown(struct device *child_device)
929 struct hv_driver *drv;
930 struct hv_device *dev = device_to_hv_device(child_device);
933 /* The device may not be attached yet */
934 if (!child_device->driver)
937 drv = drv_to_hv_drv(child_device->driver);
943 #ifdef CONFIG_PM_SLEEP
945 * vmbus_suspend - Suspend a vmbus device
947 static int vmbus_suspend(struct device *child_device)
949 struct hv_driver *drv;
950 struct hv_device *dev = device_to_hv_device(child_device);
952 /* The device may not be attached yet */
953 if (!child_device->driver)
956 drv = drv_to_hv_drv(child_device->driver);
960 return drv->suspend(dev);
964 * vmbus_resume - Resume a vmbus device
966 static int vmbus_resume(struct device *child_device)
968 struct hv_driver *drv;
969 struct hv_device *dev = device_to_hv_device(child_device);
971 /* The device may not be attached yet */
972 if (!child_device->driver)
975 drv = drv_to_hv_drv(child_device->driver);
979 return drv->resume(dev);
982 #define vmbus_suspend NULL
983 #define vmbus_resume NULL
984 #endif /* CONFIG_PM_SLEEP */
987 * vmbus_device_release - Final callback release of the vmbus child device
989 static void vmbus_device_release(struct device *device)
991 struct hv_device *hv_dev = device_to_hv_device(device);
992 struct vmbus_channel *channel = hv_dev->channel;
994 hv_debug_rm_dev_dir(hv_dev);
996 mutex_lock(&vmbus_connection.channel_mutex);
997 hv_process_channel_removal(channel);
998 mutex_unlock(&vmbus_connection.channel_mutex);
1003 * Note: we must use the "noirq" ops: see the comment before vmbus_bus_pm.
1005 * suspend_noirq/resume_noirq are set to NULL to support Suspend-to-Idle: we
1006 * shouldn't suspend the vmbus devices upon Suspend-to-Idle, otherwise there
1007 * is no way to wake up a Generation-2 VM.
1009 * The other 4 ops are for hibernation.
1012 static const struct dev_pm_ops vmbus_pm = {
1013 .suspend_noirq = NULL,
1014 .resume_noirq = NULL,
1015 .freeze_noirq = vmbus_suspend,
1016 .thaw_noirq = vmbus_resume,
1017 .poweroff_noirq = vmbus_suspend,
1018 .restore_noirq = vmbus_resume,
1021 /* The one and only one */
1022 static struct bus_type hv_bus = {
1024 .match = vmbus_match,
1025 .shutdown = vmbus_shutdown,
1026 .remove = vmbus_remove,
1027 .probe = vmbus_probe,
1028 .uevent = vmbus_uevent,
1029 .dev_groups = vmbus_dev_groups,
1030 .drv_groups = vmbus_drv_groups,
1034 struct onmessage_work_context {
1035 struct work_struct work;
1036 struct hv_message msg;
1039 static void vmbus_onmessage_work(struct work_struct *work)
1041 struct onmessage_work_context *ctx;
1043 /* Do not process messages if we're in DISCONNECTED state */
1044 if (vmbus_connection.conn_state == DISCONNECTED)
1047 ctx = container_of(work, struct onmessage_work_context,
1049 vmbus_onmessage(&ctx->msg);
1053 void vmbus_on_msg_dpc(unsigned long data)
1055 struct hv_per_cpu_context *hv_cpu = (void *)data;
1056 void *page_addr = hv_cpu->synic_message_page;
1057 struct hv_message *msg = (struct hv_message *)page_addr +
1059 struct vmbus_channel_message_header *hdr;
1060 const struct vmbus_channel_message_table_entry *entry;
1061 struct onmessage_work_context *ctx;
1062 u32 message_type = msg->header.message_type;
1064 if (message_type == HVMSG_NONE)
1068 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1070 trace_vmbus_on_msg_dpc(hdr);
1072 if (hdr->msgtype >= CHANNELMSG_COUNT) {
1073 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
1077 entry = &channel_message_table[hdr->msgtype];
1079 if (!entry->message_handler)
1082 if (entry->handler_type == VMHT_BLOCKING) {
1083 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
1087 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1088 memcpy(&ctx->msg, msg, sizeof(*msg));
1091 * The host can generate a rescind message while we
1092 * may still be handling the original offer. We deal with
1093 * this condition by ensuring the processing is done on the
1096 switch (hdr->msgtype) {
1097 case CHANNELMSG_RESCIND_CHANNELOFFER:
1099 * If we are handling the rescind message;
1100 * schedule the work on the global work queue.
1102 schedule_work_on(vmbus_connection.connect_cpu,
1106 case CHANNELMSG_OFFERCHANNEL:
1107 atomic_inc(&vmbus_connection.offer_in_progress);
1108 queue_work_on(vmbus_connection.connect_cpu,
1109 vmbus_connection.work_queue,
1114 queue_work(vmbus_connection.work_queue, &ctx->work);
1117 entry->message_handler(hdr);
1120 vmbus_signal_eom(msg, message_type);
1123 #ifdef CONFIG_PM_SLEEP
1125 * Fake RESCIND_CHANNEL messages to clean up hv_sock channels by force for
1126 * hibernation, because hv_sock connections can not persist across hibernation.
1128 static void vmbus_force_channel_rescinded(struct vmbus_channel *channel)
1130 struct onmessage_work_context *ctx;
1131 struct vmbus_channel_rescind_offer *rescind;
1133 WARN_ON(!is_hvsock_channel(channel));
1136 * sizeof(*ctx) is small and the allocation should really not fail,
1137 * otherwise the state of the hv_sock connections ends up in limbo.
1139 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL | __GFP_NOFAIL);
1142 * So far, these are not really used by Linux. Just set them to the
1143 * reasonable values conforming to the definitions of the fields.
1145 ctx->msg.header.message_type = 1;
1146 ctx->msg.header.payload_size = sizeof(*rescind);
1148 /* These values are actually used by Linux. */
1149 rescind = (struct vmbus_channel_rescind_offer *)ctx->msg.u.payload;
1150 rescind->header.msgtype = CHANNELMSG_RESCIND_CHANNELOFFER;
1151 rescind->child_relid = channel->offermsg.child_relid;
1153 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1155 queue_work_on(vmbus_connection.connect_cpu,
1156 vmbus_connection.work_queue,
1159 #endif /* CONFIG_PM_SLEEP */
1162 * Direct callback for channels using other deferred processing
1164 static void vmbus_channel_isr(struct vmbus_channel *channel)
1166 void (*callback_fn)(void *);
1168 callback_fn = READ_ONCE(channel->onchannel_callback);
1169 if (likely(callback_fn != NULL))
1170 (*callback_fn)(channel->channel_callback_context);
1174 * Schedule all channels with events pending
1176 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1178 unsigned long *recv_int_page;
1181 if (vmbus_proto_version < VERSION_WIN8) {
1182 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1183 recv_int_page = vmbus_connection.recv_int_page;
1186 * When the host is win8 and beyond, the event page
1187 * can be directly checked to get the id of the channel
1188 * that has the interrupt pending.
1190 void *page_addr = hv_cpu->synic_event_page;
1191 union hv_synic_event_flags *event
1192 = (union hv_synic_event_flags *)page_addr +
1195 maxbits = HV_EVENT_FLAGS_COUNT;
1196 recv_int_page = event->flags;
1199 if (unlikely(!recv_int_page))
1202 for_each_set_bit(relid, recv_int_page, maxbits) {
1203 struct vmbus_channel *channel;
1205 if (!sync_test_and_clear_bit(relid, recv_int_page))
1208 /* Special case - vmbus channel protocol msg */
1214 /* Find channel based on relid */
1215 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
1216 if (channel->offermsg.child_relid != relid)
1219 if (channel->rescind)
1222 trace_vmbus_chan_sched(channel);
1224 ++channel->interrupts;
1226 switch (channel->callback_mode) {
1228 vmbus_channel_isr(channel);
1231 case HV_CALL_BATCHED:
1232 hv_begin_read(&channel->inbound);
1234 case HV_CALL_DIRECT:
1235 tasklet_schedule(&channel->callback_event);
1243 static void vmbus_isr(void)
1245 struct hv_per_cpu_context *hv_cpu
1246 = this_cpu_ptr(hv_context.cpu_context);
1247 void *page_addr = hv_cpu->synic_event_page;
1248 struct hv_message *msg;
1249 union hv_synic_event_flags *event;
1250 bool handled = false;
1252 if (unlikely(page_addr == NULL))
1255 event = (union hv_synic_event_flags *)page_addr +
1258 * Check for events before checking for messages. This is the order
1259 * in which events and messages are checked in Windows guests on
1260 * Hyper-V, and the Windows team suggested we do the same.
1263 if ((vmbus_proto_version == VERSION_WS2008) ||
1264 (vmbus_proto_version == VERSION_WIN7)) {
1266 /* Since we are a child, we only need to check bit 0 */
1267 if (sync_test_and_clear_bit(0, event->flags))
1271 * Our host is win8 or above. The signaling mechanism
1272 * has changed and we can directly look at the event page.
1273 * If bit n is set then we have an interrup on the channel
1280 vmbus_chan_sched(hv_cpu);
1282 page_addr = hv_cpu->synic_message_page;
1283 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1285 /* Check if there are actual msgs to be processed */
1286 if (msg->header.message_type != HVMSG_NONE) {
1287 if (msg->header.message_type == HVMSG_TIMER_EXPIRED) {
1289 vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
1291 tasklet_schedule(&hv_cpu->msg_dpc);
1294 add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1298 * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1299 * buffer and call into Hyper-V to transfer the data.
1301 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1302 enum kmsg_dump_reason reason)
1304 size_t bytes_written;
1305 phys_addr_t panic_pa;
1307 /* We are only interested in panics. */
1308 if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1311 panic_pa = virt_to_phys(hv_panic_page);
1314 * Write dump contents to the page. No need to synchronize; panic should
1315 * be single-threaded.
1317 kmsg_dump_get_buffer(dumper, true, hv_panic_page, HV_HYP_PAGE_SIZE,
1320 hyperv_report_panic_msg(panic_pa, bytes_written);
1323 static struct kmsg_dumper hv_kmsg_dumper = {
1324 .dump = hv_kmsg_dump,
1327 static struct ctl_table_header *hv_ctl_table_hdr;
1330 * sysctl option to allow the user to control whether kmsg data should be
1331 * reported to Hyper-V on panic.
1333 static struct ctl_table hv_ctl_table[] = {
1335 .procname = "hyperv_record_panic_msg",
1336 .data = &sysctl_record_panic_msg,
1337 .maxlen = sizeof(int),
1339 .proc_handler = proc_dointvec_minmax,
1340 .extra1 = SYSCTL_ZERO,
1341 .extra2 = SYSCTL_ONE
1346 static struct ctl_table hv_root_table[] = {
1348 .procname = "kernel",
1350 .child = hv_ctl_table
1356 * vmbus_bus_init -Main vmbus driver initialization routine.
1359 * - initialize the vmbus driver context
1360 * - invoke the vmbus hv main init routine
1361 * - retrieve the channel offers
1363 static int vmbus_bus_init(void)
1367 /* Hypervisor initialization...setup hypercall page..etc */
1370 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1374 ret = bus_register(&hv_bus);
1378 hv_setup_vmbus_irq(vmbus_isr);
1380 ret = hv_synic_alloc();
1385 * Initialize the per-cpu interrupt state and stimer state.
1386 * Then connect to the host.
1388 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1389 hv_synic_init, hv_synic_cleanup);
1392 hyperv_cpuhp_online = ret;
1394 ret = vmbus_connect();
1399 * Only register if the crash MSRs are available
1401 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1402 u64 hyperv_crash_ctl;
1404 * Sysctl registration is not fatal, since by default
1405 * reporting is enabled.
1407 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1408 if (!hv_ctl_table_hdr)
1409 pr_err("Hyper-V: sysctl table register error");
1412 * Register for panic kmsg callback only if the right
1413 * capability is supported by the hypervisor.
1415 hv_get_crash_ctl(hyperv_crash_ctl);
1416 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1417 hv_panic_page = (void *)hv_alloc_hyperv_zeroed_page();
1418 if (hv_panic_page) {
1419 ret = kmsg_dump_register(&hv_kmsg_dumper);
1421 pr_err("Hyper-V: kmsg dump register "
1422 "error 0x%x\n", ret);
1423 hv_free_hyperv_page(
1424 (unsigned long)hv_panic_page);
1425 hv_panic_page = NULL;
1428 pr_err("Hyper-V: panic message page memory "
1429 "allocation failed");
1432 register_die_notifier(&hyperv_die_block);
1436 * Always register the panic notifier because we need to unload
1437 * the VMbus channel connection to prevent any VMbus
1438 * activity after the VM panics.
1440 atomic_notifier_chain_register(&panic_notifier_list,
1441 &hyperv_panic_block);
1443 vmbus_request_offers();
1448 cpuhp_remove_state(hyperv_cpuhp_online);
1452 hv_remove_vmbus_irq();
1454 bus_unregister(&hv_bus);
1455 unregister_sysctl_table(hv_ctl_table_hdr);
1456 hv_ctl_table_hdr = NULL;
1461 * __vmbus_child_driver_register() - Register a vmbus's driver
1462 * @hv_driver: Pointer to driver structure you want to register
1463 * @owner: owner module of the drv
1464 * @mod_name: module name string
1466 * Registers the given driver with Linux through the 'driver_register()' call
1467 * and sets up the hyper-v vmbus handling for this driver.
1468 * It will return the state of the 'driver_register()' call.
1471 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1475 pr_info("registering driver %s\n", hv_driver->name);
1477 ret = vmbus_exists();
1481 hv_driver->driver.name = hv_driver->name;
1482 hv_driver->driver.owner = owner;
1483 hv_driver->driver.mod_name = mod_name;
1484 hv_driver->driver.bus = &hv_bus;
1486 spin_lock_init(&hv_driver->dynids.lock);
1487 INIT_LIST_HEAD(&hv_driver->dynids.list);
1489 ret = driver_register(&hv_driver->driver);
1493 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1496 * vmbus_driver_unregister() - Unregister a vmbus's driver
1497 * @hv_driver: Pointer to driver structure you want to
1500 * Un-register the given driver that was previous registered with a call to
1501 * vmbus_driver_register()
1503 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1505 pr_info("unregistering driver %s\n", hv_driver->name);
1507 if (!vmbus_exists()) {
1508 driver_unregister(&hv_driver->driver);
1509 vmbus_free_dynids(hv_driver);
1512 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1516 * Called when last reference to channel is gone.
1518 static void vmbus_chan_release(struct kobject *kobj)
1520 struct vmbus_channel *channel
1521 = container_of(kobj, struct vmbus_channel, kobj);
1523 kfree_rcu(channel, rcu);
1526 struct vmbus_chan_attribute {
1527 struct attribute attr;
1528 ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1529 ssize_t (*store)(struct vmbus_channel *chan,
1530 const char *buf, size_t count);
1532 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1533 struct vmbus_chan_attribute chan_attr_##_name \
1534 = __ATTR(_name, _mode, _show, _store)
1535 #define VMBUS_CHAN_ATTR_RW(_name) \
1536 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1537 #define VMBUS_CHAN_ATTR_RO(_name) \
1538 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1539 #define VMBUS_CHAN_ATTR_WO(_name) \
1540 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1542 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1543 struct attribute *attr, char *buf)
1545 const struct vmbus_chan_attribute *attribute
1546 = container_of(attr, struct vmbus_chan_attribute, attr);
1547 struct vmbus_channel *chan
1548 = container_of(kobj, struct vmbus_channel, kobj);
1550 if (!attribute->show)
1553 return attribute->show(chan, buf);
1556 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1557 .show = vmbus_chan_attr_show,
1560 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1562 struct hv_ring_buffer_info *rbi = &channel->outbound;
1565 mutex_lock(&rbi->ring_buffer_mutex);
1566 if (!rbi->ring_buffer) {
1567 mutex_unlock(&rbi->ring_buffer_mutex);
1571 ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1572 mutex_unlock(&rbi->ring_buffer_mutex);
1575 static VMBUS_CHAN_ATTR_RO(out_mask);
1577 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1579 struct hv_ring_buffer_info *rbi = &channel->inbound;
1582 mutex_lock(&rbi->ring_buffer_mutex);
1583 if (!rbi->ring_buffer) {
1584 mutex_unlock(&rbi->ring_buffer_mutex);
1588 ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1589 mutex_unlock(&rbi->ring_buffer_mutex);
1592 static VMBUS_CHAN_ATTR_RO(in_mask);
1594 static ssize_t read_avail_show(struct vmbus_channel *channel, char *buf)
1596 struct hv_ring_buffer_info *rbi = &channel->inbound;
1599 mutex_lock(&rbi->ring_buffer_mutex);
1600 if (!rbi->ring_buffer) {
1601 mutex_unlock(&rbi->ring_buffer_mutex);
1605 ret = sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1606 mutex_unlock(&rbi->ring_buffer_mutex);
1609 static VMBUS_CHAN_ATTR_RO(read_avail);
1611 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1613 struct hv_ring_buffer_info *rbi = &channel->outbound;
1616 mutex_lock(&rbi->ring_buffer_mutex);
1617 if (!rbi->ring_buffer) {
1618 mutex_unlock(&rbi->ring_buffer_mutex);
1622 ret = sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1623 mutex_unlock(&rbi->ring_buffer_mutex);
1626 static VMBUS_CHAN_ATTR_RO(write_avail);
1628 static ssize_t show_target_cpu(struct vmbus_channel *channel, char *buf)
1630 return sprintf(buf, "%u\n", channel->target_cpu);
1632 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1634 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1637 return sprintf(buf, "%d\n",
1638 channel_pending(channel,
1639 vmbus_connection.monitor_pages[1]));
1641 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1643 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1646 return sprintf(buf, "%d\n",
1647 channel_latency(channel,
1648 vmbus_connection.monitor_pages[1]));
1650 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1652 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1654 return sprintf(buf, "%llu\n", channel->interrupts);
1656 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1658 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1660 return sprintf(buf, "%llu\n", channel->sig_events);
1662 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1664 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1667 return sprintf(buf, "%llu\n",
1668 (unsigned long long)channel->intr_in_full);
1670 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1672 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1675 return sprintf(buf, "%llu\n",
1676 (unsigned long long)channel->intr_out_empty);
1678 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1680 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1683 return sprintf(buf, "%llu\n",
1684 (unsigned long long)channel->out_full_first);
1686 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1688 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1691 return sprintf(buf, "%llu\n",
1692 (unsigned long long)channel->out_full_total);
1694 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1696 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1699 return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1701 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1703 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1706 return sprintf(buf, "%u\n",
1707 channel->offermsg.offer.sub_channel_index);
1709 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1711 static struct attribute *vmbus_chan_attrs[] = {
1712 &chan_attr_out_mask.attr,
1713 &chan_attr_in_mask.attr,
1714 &chan_attr_read_avail.attr,
1715 &chan_attr_write_avail.attr,
1716 &chan_attr_cpu.attr,
1717 &chan_attr_pending.attr,
1718 &chan_attr_latency.attr,
1719 &chan_attr_interrupts.attr,
1720 &chan_attr_events.attr,
1721 &chan_attr_intr_in_full.attr,
1722 &chan_attr_intr_out_empty.attr,
1723 &chan_attr_out_full_first.attr,
1724 &chan_attr_out_full_total.attr,
1725 &chan_attr_monitor_id.attr,
1726 &chan_attr_subchannel_id.attr,
1731 * Channel-level attribute_group callback function. Returns the permission for
1732 * each attribute, and returns 0 if an attribute is not visible.
1734 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1735 struct attribute *attr, int idx)
1737 const struct vmbus_channel *channel =
1738 container_of(kobj, struct vmbus_channel, kobj);
1740 /* Hide the monitor attributes if the monitor mechanism is not used. */
1741 if (!channel->offermsg.monitor_allocated &&
1742 (attr == &chan_attr_pending.attr ||
1743 attr == &chan_attr_latency.attr ||
1744 attr == &chan_attr_monitor_id.attr))
1750 static struct attribute_group vmbus_chan_group = {
1751 .attrs = vmbus_chan_attrs,
1752 .is_visible = vmbus_chan_attr_is_visible
1755 static struct kobj_type vmbus_chan_ktype = {
1756 .sysfs_ops = &vmbus_chan_sysfs_ops,
1757 .release = vmbus_chan_release,
1761 * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1763 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1765 const struct device *device = &dev->device;
1766 struct kobject *kobj = &channel->kobj;
1767 u32 relid = channel->offermsg.child_relid;
1770 kobj->kset = dev->channels_kset;
1771 ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1776 ret = sysfs_create_group(kobj, &vmbus_chan_group);
1780 * The calling functions' error handling paths will cleanup the
1781 * empty channel directory.
1783 dev_err(device, "Unable to set up channel sysfs files\n");
1787 kobject_uevent(kobj, KOBJ_ADD);
1793 * vmbus_remove_channel_attr_group - remove the channel's attribute group
1795 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
1797 sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
1801 * vmbus_device_create - Creates and registers a new child device
1804 struct hv_device *vmbus_device_create(const guid_t *type,
1805 const guid_t *instance,
1806 struct vmbus_channel *channel)
1808 struct hv_device *child_device_obj;
1810 child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1811 if (!child_device_obj) {
1812 pr_err("Unable to allocate device object for child device\n");
1816 child_device_obj->channel = channel;
1817 guid_copy(&child_device_obj->dev_type, type);
1818 guid_copy(&child_device_obj->dev_instance, instance);
1819 child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1821 return child_device_obj;
1825 * vmbus_device_register - Register the child device
1827 int vmbus_device_register(struct hv_device *child_device_obj)
1829 struct kobject *kobj = &child_device_obj->device.kobj;
1832 dev_set_name(&child_device_obj->device, "%pUl",
1833 child_device_obj->channel->offermsg.offer.if_instance.b);
1835 child_device_obj->device.bus = &hv_bus;
1836 child_device_obj->device.parent = &hv_acpi_dev->dev;
1837 child_device_obj->device.release = vmbus_device_release;
1840 * Register with the LDM. This will kick off the driver/device
1841 * binding...which will eventually call vmbus_match() and vmbus_probe()
1843 ret = device_register(&child_device_obj->device);
1845 pr_err("Unable to register child device\n");
1849 child_device_obj->channels_kset = kset_create_and_add("channels",
1851 if (!child_device_obj->channels_kset) {
1853 goto err_dev_unregister;
1856 ret = vmbus_add_channel_kobj(child_device_obj,
1857 child_device_obj->channel);
1859 pr_err("Unable to register primary channeln");
1860 goto err_kset_unregister;
1862 hv_debug_add_dev_dir(child_device_obj);
1866 err_kset_unregister:
1867 kset_unregister(child_device_obj->channels_kset);
1870 device_unregister(&child_device_obj->device);
1875 * vmbus_device_unregister - Remove the specified child device
1878 void vmbus_device_unregister(struct hv_device *device_obj)
1880 pr_debug("child device %s unregistered\n",
1881 dev_name(&device_obj->device));
1883 kset_unregister(device_obj->channels_kset);
1886 * Kick off the process of unregistering the device.
1887 * This will call vmbus_remove() and eventually vmbus_device_release()
1889 device_unregister(&device_obj->device);
1894 * VMBUS is an acpi enumerated device. Get the information we
1897 #define VTPM_BASE_ADDRESS 0xfed40000
1898 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1900 resource_size_t start = 0;
1901 resource_size_t end = 0;
1902 struct resource *new_res;
1903 struct resource **old_res = &hyperv_mmio;
1904 struct resource **prev_res = NULL;
1906 switch (res->type) {
1909 * "Address" descriptors are for bus windows. Ignore
1910 * "memory" descriptors, which are for registers on
1913 case ACPI_RESOURCE_TYPE_ADDRESS32:
1914 start = res->data.address32.address.minimum;
1915 end = res->data.address32.address.maximum;
1918 case ACPI_RESOURCE_TYPE_ADDRESS64:
1919 start = res->data.address64.address.minimum;
1920 end = res->data.address64.address.maximum;
1924 /* Unused resource type */
1929 * Ignore ranges that are below 1MB, as they're not
1930 * necessary or useful here.
1935 new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1937 return AE_NO_MEMORY;
1939 /* If this range overlaps the virtual TPM, truncate it. */
1940 if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1941 end = VTPM_BASE_ADDRESS;
1943 new_res->name = "hyperv mmio";
1944 new_res->flags = IORESOURCE_MEM;
1945 new_res->start = start;
1949 * If two ranges are adjacent, merge them.
1957 if (((*old_res)->end + 1) == new_res->start) {
1958 (*old_res)->end = new_res->end;
1963 if ((*old_res)->start == new_res->end + 1) {
1964 (*old_res)->start = new_res->start;
1969 if ((*old_res)->start > new_res->end) {
1970 new_res->sibling = *old_res;
1972 (*prev_res)->sibling = new_res;
1978 old_res = &(*old_res)->sibling;
1985 static int vmbus_acpi_remove(struct acpi_device *device)
1987 struct resource *cur_res;
1988 struct resource *next_res;
1992 __release_region(hyperv_mmio, fb_mmio->start,
1993 resource_size(fb_mmio));
1997 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1998 next_res = cur_res->sibling;
2006 static void vmbus_reserve_fb(void)
2010 * Make a claim for the frame buffer in the resource tree under the
2011 * first node, which will be the one below 4GB. The length seems to
2012 * be underreported, particularly in a Generation 1 VM. So start out
2013 * reserving a larger area and make it smaller until it succeeds.
2016 if (screen_info.lfb_base) {
2017 if (efi_enabled(EFI_BOOT))
2018 size = max_t(__u32, screen_info.lfb_size, 0x800000);
2020 size = max_t(__u32, screen_info.lfb_size, 0x4000000);
2022 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
2023 fb_mmio = __request_region(hyperv_mmio,
2024 screen_info.lfb_base, size,
2031 * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
2032 * @new: If successful, supplied a pointer to the
2033 * allocated MMIO space.
2034 * @device_obj: Identifies the caller
2035 * @min: Minimum guest physical address of the
2037 * @max: Maximum guest physical address
2038 * @size: Size of the range to be allocated
2039 * @align: Alignment of the range to be allocated
2040 * @fb_overlap_ok: Whether this allocation can be allowed
2041 * to overlap the video frame buffer.
2043 * This function walks the resources granted to VMBus by the
2044 * _CRS object in the ACPI namespace underneath the parent
2045 * "bridge" whether that's a root PCI bus in the Generation 1
2046 * case or a Module Device in the Generation 2 case. It then
2047 * attempts to allocate from the global MMIO pool in a way that
2048 * matches the constraints supplied in these parameters and by
2051 * Return: 0 on success, -errno on failure
2053 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
2054 resource_size_t min, resource_size_t max,
2055 resource_size_t size, resource_size_t align,
2058 struct resource *iter, *shadow;
2059 resource_size_t range_min, range_max, start;
2060 const char *dev_n = dev_name(&device_obj->device);
2064 mutex_lock(&hyperv_mmio_lock);
2067 * If overlaps with frame buffers are allowed, then first attempt to
2068 * make the allocation from within the reserved region. Because it
2069 * is already reserved, no shadow allocation is necessary.
2071 if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
2072 !(max < fb_mmio->start)) {
2074 range_min = fb_mmio->start;
2075 range_max = fb_mmio->end;
2076 start = (range_min + align - 1) & ~(align - 1);
2077 for (; start + size - 1 <= range_max; start += align) {
2078 *new = request_mem_region_exclusive(start, size, dev_n);
2086 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2087 if ((iter->start >= max) || (iter->end <= min))
2090 range_min = iter->start;
2091 range_max = iter->end;
2092 start = (range_min + align - 1) & ~(align - 1);
2093 for (; start + size - 1 <= range_max; start += align) {
2094 shadow = __request_region(iter, start, size, NULL,
2099 *new = request_mem_region_exclusive(start, size, dev_n);
2101 shadow->name = (char *)*new;
2106 __release_region(iter, start, size);
2111 mutex_unlock(&hyperv_mmio_lock);
2114 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
2117 * vmbus_free_mmio() - Free a memory-mapped I/O range.
2118 * @start: Base address of region to release.
2119 * @size: Size of the range to be allocated
2121 * This function releases anything requested by
2122 * vmbus_mmio_allocate().
2124 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
2126 struct resource *iter;
2128 mutex_lock(&hyperv_mmio_lock);
2129 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2130 if ((iter->start >= start + size) || (iter->end <= start))
2133 __release_region(iter, start, size);
2135 release_mem_region(start, size);
2136 mutex_unlock(&hyperv_mmio_lock);
2139 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2141 static int vmbus_acpi_add(struct acpi_device *device)
2144 int ret_val = -ENODEV;
2145 struct acpi_device *ancestor;
2147 hv_acpi_dev = device;
2149 result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2150 vmbus_walk_resources, NULL);
2152 if (ACPI_FAILURE(result))
2155 * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2156 * firmware) is the VMOD that has the mmio ranges. Get that.
2158 for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
2159 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2160 vmbus_walk_resources, NULL);
2162 if (ACPI_FAILURE(result))
2172 complete(&probe_event);
2174 vmbus_acpi_remove(device);
2178 #ifdef CONFIG_PM_SLEEP
2179 static int vmbus_bus_suspend(struct device *dev)
2181 struct vmbus_channel *channel, *sc;
2182 unsigned long flags;
2184 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
2186 * We wait here until the completion of any channel
2187 * offers that are currently in progress.
2192 mutex_lock(&vmbus_connection.channel_mutex);
2193 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2194 if (!is_hvsock_channel(channel))
2197 vmbus_force_channel_rescinded(channel);
2199 mutex_unlock(&vmbus_connection.channel_mutex);
2202 * Wait until all the sub-channels and hv_sock channels have been
2203 * cleaned up. Sub-channels should be destroyed upon suspend, otherwise
2204 * they would conflict with the new sub-channels that will be created
2205 * in the resume path. hv_sock channels should also be destroyed, but
2206 * a hv_sock channel of an established hv_sock connection can not be
2207 * really destroyed since it may still be referenced by the userspace
2208 * application, so we just force the hv_sock channel to be rescinded
2209 * by vmbus_force_channel_rescinded(), and the userspace application
2210 * will thoroughly destroy the channel after hibernation.
2212 * Note: the counter nr_chan_close_on_suspend may never go above 0 if
2213 * the VM has no sub-channel and hv_sock channel, e.g. a 1-vCPU VM.
2215 if (atomic_read(&vmbus_connection.nr_chan_close_on_suspend) > 0)
2216 wait_for_completion(&vmbus_connection.ready_for_suspend_event);
2218 WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) != 0);
2220 mutex_lock(&vmbus_connection.channel_mutex);
2222 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2224 * Invalidate the field. Upon resume, vmbus_onoffer() will fix
2225 * up the field, and the other fields (if necessary).
2227 channel->offermsg.child_relid = INVALID_RELID;
2229 if (is_hvsock_channel(channel)) {
2230 if (!channel->rescind) {
2231 pr_err("hv_sock channel not rescinded!\n");
2237 spin_lock_irqsave(&channel->lock, flags);
2238 list_for_each_entry(sc, &channel->sc_list, sc_list) {
2239 pr_err("Sub-channel not deleted!\n");
2242 spin_unlock_irqrestore(&channel->lock, flags);
2244 atomic_inc(&vmbus_connection.nr_chan_fixup_on_resume);
2247 mutex_unlock(&vmbus_connection.channel_mutex);
2249 vmbus_initiate_unload(false);
2251 /* Reset the event for the next resume. */
2252 reinit_completion(&vmbus_connection.ready_for_resume_event);
2257 static int vmbus_bus_resume(struct device *dev)
2259 struct vmbus_channel_msginfo *msginfo;
2264 * We only use the 'vmbus_proto_version', which was in use before
2265 * hibernation, to re-negotiate with the host.
2267 if (!vmbus_proto_version) {
2268 pr_err("Invalid proto version = 0x%x\n", vmbus_proto_version);
2272 msgsize = sizeof(*msginfo) +
2273 sizeof(struct vmbus_channel_initiate_contact);
2275 msginfo = kzalloc(msgsize, GFP_KERNEL);
2277 if (msginfo == NULL)
2280 ret = vmbus_negotiate_version(msginfo, vmbus_proto_version);
2287 WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) == 0);
2289 vmbus_request_offers();
2291 wait_for_completion(&vmbus_connection.ready_for_resume_event);
2293 /* Reset the event for the next suspend. */
2294 reinit_completion(&vmbus_connection.ready_for_suspend_event);
2299 #define vmbus_bus_suspend NULL
2300 #define vmbus_bus_resume NULL
2301 #endif /* CONFIG_PM_SLEEP */
2303 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
2308 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2311 * Note: we must use the "no_irq" ops, otherwise hibernation can not work with
2312 * PCI device assignment, because "pci_dev_pm_ops" uses the "noirq" ops: in
2313 * the resume path, the pci "noirq" restore op runs before "non-noirq" op (see
2314 * resume_target_kernel() -> dpm_resume_start(), and hibernation_restore() ->
2315 * dpm_resume_end()). This means vmbus_bus_resume() and the pci-hyperv's
2316 * resume callback must also run via the "noirq" ops.
2318 * Set suspend_noirq/resume_noirq to NULL for Suspend-to-Idle: see the comment
2319 * earlier in this file before vmbus_pm.
2322 static const struct dev_pm_ops vmbus_bus_pm = {
2323 .suspend_noirq = NULL,
2324 .resume_noirq = NULL,
2325 .freeze_noirq = vmbus_bus_suspend,
2326 .thaw_noirq = vmbus_bus_resume,
2327 .poweroff_noirq = vmbus_bus_suspend,
2328 .restore_noirq = vmbus_bus_resume
2331 static struct acpi_driver vmbus_acpi_driver = {
2333 .ids = vmbus_acpi_device_ids,
2335 .add = vmbus_acpi_add,
2336 .remove = vmbus_acpi_remove,
2338 .drv.pm = &vmbus_bus_pm,
2341 static void hv_kexec_handler(void)
2343 hv_stimer_global_cleanup();
2344 vmbus_initiate_unload(false);
2345 /* Make sure conn_state is set as hv_synic_cleanup checks for it */
2347 cpuhp_remove_state(hyperv_cpuhp_online);
2351 static void hv_crash_handler(struct pt_regs *regs)
2355 vmbus_initiate_unload(true);
2357 * In crash handler we can't schedule synic cleanup for all CPUs,
2358 * doing the cleanup for current CPU only. This should be sufficient
2361 cpu = smp_processor_id();
2362 hv_stimer_cleanup(cpu);
2363 hv_synic_disable_regs(cpu);
2367 static int hv_synic_suspend(void)
2370 * When we reach here, all the non-boot CPUs have been offlined.
2371 * If we're in a legacy configuration where stimer Direct Mode is
2372 * not enabled, the stimers on the non-boot CPUs have been unbound
2373 * in hv_synic_cleanup() -> hv_stimer_legacy_cleanup() ->
2374 * hv_stimer_cleanup() -> clockevents_unbind_device().
2376 * hv_synic_suspend() only runs on CPU0 with interrupts disabled.
2377 * Here we do not call hv_stimer_legacy_cleanup() on CPU0 because:
2378 * 1) it's unnecessary as interrupts remain disabled between
2379 * syscore_suspend() and syscore_resume(): see create_image() and
2380 * resume_target_kernel()
2381 * 2) the stimer on CPU0 is automatically disabled later by
2382 * syscore_suspend() -> timekeeping_suspend() -> tick_suspend() -> ...
2383 * -> clockevents_shutdown() -> ... -> hv_ce_shutdown()
2384 * 3) a warning would be triggered if we call
2385 * clockevents_unbind_device(), which may sleep, in an
2386 * interrupts-disabled context.
2389 hv_synic_disable_regs(0);
2394 static void hv_synic_resume(void)
2396 hv_synic_enable_regs(0);
2399 * Note: we don't need to call hv_stimer_init(0), because the timer
2400 * on CPU0 is not unbound in hv_synic_suspend(), and the timer is
2401 * automatically re-enabled in timekeeping_resume().
2405 /* The callbacks run only on CPU0, with irqs_disabled. */
2406 static struct syscore_ops hv_synic_syscore_ops = {
2407 .suspend = hv_synic_suspend,
2408 .resume = hv_synic_resume,
2411 static int __init hv_acpi_init(void)
2415 if (!hv_is_hyperv_initialized())
2418 init_completion(&probe_event);
2421 * Get ACPI resources first.
2423 ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2428 t = wait_for_completion_timeout(&probe_event, 5*HZ);
2435 ret = vmbus_bus_init();
2439 hv_setup_kexec_handler(hv_kexec_handler);
2440 hv_setup_crash_handler(hv_crash_handler);
2442 register_syscore_ops(&hv_synic_syscore_ops);
2447 acpi_bus_unregister_driver(&vmbus_acpi_driver);
2452 static void __exit vmbus_exit(void)
2456 unregister_syscore_ops(&hv_synic_syscore_ops);
2458 hv_remove_kexec_handler();
2459 hv_remove_crash_handler();
2460 vmbus_connection.conn_state = DISCONNECTED;
2461 hv_stimer_global_cleanup();
2463 hv_remove_vmbus_irq();
2464 for_each_online_cpu(cpu) {
2465 struct hv_per_cpu_context *hv_cpu
2466 = per_cpu_ptr(hv_context.cpu_context, cpu);
2468 tasklet_kill(&hv_cpu->msg_dpc);
2470 hv_debug_rm_all_dir();
2472 vmbus_free_channels();
2474 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2475 kmsg_dump_unregister(&hv_kmsg_dumper);
2476 unregister_die_notifier(&hyperv_die_block);
2477 atomic_notifier_chain_unregister(&panic_notifier_list,
2478 &hyperv_panic_block);
2481 free_page((unsigned long)hv_panic_page);
2482 unregister_sysctl_table(hv_ctl_table_hdr);
2483 hv_ctl_table_hdr = NULL;
2484 bus_unregister(&hv_bus);
2486 cpuhp_remove_state(hyperv_cpuhp_online);
2488 acpi_bus_unregister_driver(&vmbus_acpi_driver);
2492 MODULE_LICENSE("GPL");
2493 MODULE_DESCRIPTION("Microsoft Hyper-V VMBus Driver");
2495 subsys_initcall(hv_acpi_init);
2496 module_exit(vmbus_exit);