2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
18 * Haiyang Zhang <haiyangz@microsoft.com>
19 * Hank Janssen <hjanssen@microsoft.com>
20 * K. Y. Srinivasan <kys@microsoft.com>
23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25 #include <linux/init.h>
26 #include <linux/module.h>
27 #include <linux/device.h>
28 #include <linux/interrupt.h>
29 #include <linux/sysctl.h>
30 #include <linux/slab.h>
31 #include <linux/acpi.h>
32 #include <linux/completion.h>
33 #include <linux/hyperv.h>
34 #include <linux/kernel_stat.h>
35 #include <linux/clockchips.h>
36 #include <linux/cpu.h>
37 #include <linux/sched/task_stack.h>
39 #include <asm/mshyperv.h>
40 #include <linux/notifier.h>
41 #include <linux/ptrace.h>
42 #include <linux/screen_info.h>
43 #include <linux/kdebug.h>
44 #include <linux/efi.h>
45 #include <linux/random.h>
46 #include <linux/kernel.h>
47 #include "hyperv_vmbus.h"
50 struct list_head node;
51 struct hv_vmbus_device_id id;
54 static struct acpi_device *hv_acpi_dev;
56 static struct completion probe_event;
58 static int hyperv_cpuhp_online;
60 static void *hv_panic_page;
63 * Boolean to control whether to report panic messages over Hyper-V.
65 * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
67 static int sysctl_record_panic_msg = 1;
69 static int hyperv_report_reg(void)
71 return !sysctl_record_panic_msg || !hv_panic_page;
74 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
79 vmbus_initiate_unload(true);
82 * Hyper-V should be notified only once about a panic. If we will be
83 * doing hyperv_report_panic_msg() later with kmsg data, don't do
84 * the notification here.
86 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE
87 && hyperv_report_reg()) {
88 regs = current_pt_regs();
89 hyperv_report_panic(regs, val, false);
94 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
97 struct die_args *die = (struct die_args *)args;
98 struct pt_regs *regs = die->regs;
101 * Hyper-V should be notified only once about a panic. If we will be
102 * doing hyperv_report_panic_msg() later with kmsg data, don't do
103 * the notification here.
105 if (hyperv_report_reg())
106 hyperv_report_panic(regs, val, true);
110 static struct notifier_block hyperv_die_block = {
111 .notifier_call = hyperv_die_event,
113 static struct notifier_block hyperv_panic_block = {
114 .notifier_call = hyperv_panic_event,
117 static const char *fb_mmio_name = "fb_range";
118 static struct resource *fb_mmio;
119 static struct resource *hyperv_mmio;
120 static DEFINE_SEMAPHORE(hyperv_mmio_lock);
122 static int vmbus_exists(void)
124 if (hv_acpi_dev == NULL)
130 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
131 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
134 for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
135 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
138 static u8 channel_monitor_group(const struct vmbus_channel *channel)
140 return (u8)channel->offermsg.monitorid / 32;
143 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
145 return (u8)channel->offermsg.monitorid % 32;
148 static u32 channel_pending(const struct vmbus_channel *channel,
149 const struct hv_monitor_page *monitor_page)
151 u8 monitor_group = channel_monitor_group(channel);
153 return monitor_page->trigger_group[monitor_group].pending;
156 static u32 channel_latency(const struct vmbus_channel *channel,
157 const struct hv_monitor_page *monitor_page)
159 u8 monitor_group = channel_monitor_group(channel);
160 u8 monitor_offset = channel_monitor_offset(channel);
162 return monitor_page->latency[monitor_group][monitor_offset];
165 static u32 channel_conn_id(struct vmbus_channel *channel,
166 struct hv_monitor_page *monitor_page)
168 u8 monitor_group = channel_monitor_group(channel);
169 u8 monitor_offset = channel_monitor_offset(channel);
170 return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
173 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
176 struct hv_device *hv_dev = device_to_hv_device(dev);
178 if (!hv_dev->channel)
180 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
182 static DEVICE_ATTR_RO(id);
184 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
187 struct hv_device *hv_dev = device_to_hv_device(dev);
189 if (!hv_dev->channel)
191 return sprintf(buf, "%d\n", hv_dev->channel->state);
193 static DEVICE_ATTR_RO(state);
195 static ssize_t monitor_id_show(struct device *dev,
196 struct device_attribute *dev_attr, char *buf)
198 struct hv_device *hv_dev = device_to_hv_device(dev);
200 if (!hv_dev->channel)
202 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
204 static DEVICE_ATTR_RO(monitor_id);
206 static ssize_t class_id_show(struct device *dev,
207 struct device_attribute *dev_attr, char *buf)
209 struct hv_device *hv_dev = device_to_hv_device(dev);
211 if (!hv_dev->channel)
213 return sprintf(buf, "{%pUl}\n",
214 hv_dev->channel->offermsg.offer.if_type.b);
216 static DEVICE_ATTR_RO(class_id);
218 static ssize_t device_id_show(struct device *dev,
219 struct device_attribute *dev_attr, char *buf)
221 struct hv_device *hv_dev = device_to_hv_device(dev);
223 if (!hv_dev->channel)
225 return sprintf(buf, "{%pUl}\n",
226 hv_dev->channel->offermsg.offer.if_instance.b);
228 static DEVICE_ATTR_RO(device_id);
230 static ssize_t modalias_show(struct device *dev,
231 struct device_attribute *dev_attr, char *buf)
233 struct hv_device *hv_dev = device_to_hv_device(dev);
234 char alias_name[VMBUS_ALIAS_LEN + 1];
236 print_alias_name(hv_dev, alias_name);
237 return sprintf(buf, "vmbus:%s\n", alias_name);
239 static DEVICE_ATTR_RO(modalias);
242 static ssize_t numa_node_show(struct device *dev,
243 struct device_attribute *attr, char *buf)
245 struct hv_device *hv_dev = device_to_hv_device(dev);
247 if (!hv_dev->channel)
250 return sprintf(buf, "%d\n", hv_dev->channel->numa_node);
252 static DEVICE_ATTR_RO(numa_node);
255 static ssize_t server_monitor_pending_show(struct device *dev,
256 struct device_attribute *dev_attr,
259 struct hv_device *hv_dev = device_to_hv_device(dev);
261 if (!hv_dev->channel)
263 return sprintf(buf, "%d\n",
264 channel_pending(hv_dev->channel,
265 vmbus_connection.monitor_pages[1]));
267 static DEVICE_ATTR_RO(server_monitor_pending);
269 static ssize_t client_monitor_pending_show(struct device *dev,
270 struct device_attribute *dev_attr,
273 struct hv_device *hv_dev = device_to_hv_device(dev);
275 if (!hv_dev->channel)
277 return sprintf(buf, "%d\n",
278 channel_pending(hv_dev->channel,
279 vmbus_connection.monitor_pages[1]));
281 static DEVICE_ATTR_RO(client_monitor_pending);
283 static ssize_t server_monitor_latency_show(struct device *dev,
284 struct device_attribute *dev_attr,
287 struct hv_device *hv_dev = device_to_hv_device(dev);
289 if (!hv_dev->channel)
291 return sprintf(buf, "%d\n",
292 channel_latency(hv_dev->channel,
293 vmbus_connection.monitor_pages[0]));
295 static DEVICE_ATTR_RO(server_monitor_latency);
297 static ssize_t client_monitor_latency_show(struct device *dev,
298 struct device_attribute *dev_attr,
301 struct hv_device *hv_dev = device_to_hv_device(dev);
303 if (!hv_dev->channel)
305 return sprintf(buf, "%d\n",
306 channel_latency(hv_dev->channel,
307 vmbus_connection.monitor_pages[1]));
309 static DEVICE_ATTR_RO(client_monitor_latency);
311 static ssize_t server_monitor_conn_id_show(struct device *dev,
312 struct device_attribute *dev_attr,
315 struct hv_device *hv_dev = device_to_hv_device(dev);
317 if (!hv_dev->channel)
319 return sprintf(buf, "%d\n",
320 channel_conn_id(hv_dev->channel,
321 vmbus_connection.monitor_pages[0]));
323 static DEVICE_ATTR_RO(server_monitor_conn_id);
325 static ssize_t client_monitor_conn_id_show(struct device *dev,
326 struct device_attribute *dev_attr,
329 struct hv_device *hv_dev = device_to_hv_device(dev);
331 if (!hv_dev->channel)
333 return sprintf(buf, "%d\n",
334 channel_conn_id(hv_dev->channel,
335 vmbus_connection.monitor_pages[1]));
337 static DEVICE_ATTR_RO(client_monitor_conn_id);
339 static ssize_t out_intr_mask_show(struct device *dev,
340 struct device_attribute *dev_attr, char *buf)
342 struct hv_device *hv_dev = device_to_hv_device(dev);
343 struct hv_ring_buffer_debug_info outbound;
346 if (!hv_dev->channel)
349 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
354 return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
356 static DEVICE_ATTR_RO(out_intr_mask);
358 static ssize_t out_read_index_show(struct device *dev,
359 struct device_attribute *dev_attr, char *buf)
361 struct hv_device *hv_dev = device_to_hv_device(dev);
362 struct hv_ring_buffer_debug_info outbound;
365 if (!hv_dev->channel)
368 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
372 return sprintf(buf, "%d\n", outbound.current_read_index);
374 static DEVICE_ATTR_RO(out_read_index);
376 static ssize_t out_write_index_show(struct device *dev,
377 struct device_attribute *dev_attr,
380 struct hv_device *hv_dev = device_to_hv_device(dev);
381 struct hv_ring_buffer_debug_info outbound;
384 if (!hv_dev->channel)
387 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
391 return sprintf(buf, "%d\n", outbound.current_write_index);
393 static DEVICE_ATTR_RO(out_write_index);
395 static ssize_t out_read_bytes_avail_show(struct device *dev,
396 struct device_attribute *dev_attr,
399 struct hv_device *hv_dev = device_to_hv_device(dev);
400 struct hv_ring_buffer_debug_info outbound;
403 if (!hv_dev->channel)
406 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
410 return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
412 static DEVICE_ATTR_RO(out_read_bytes_avail);
414 static ssize_t out_write_bytes_avail_show(struct device *dev,
415 struct device_attribute *dev_attr,
418 struct hv_device *hv_dev = device_to_hv_device(dev);
419 struct hv_ring_buffer_debug_info outbound;
422 if (!hv_dev->channel)
425 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
429 return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
431 static DEVICE_ATTR_RO(out_write_bytes_avail);
433 static ssize_t in_intr_mask_show(struct device *dev,
434 struct device_attribute *dev_attr, char *buf)
436 struct hv_device *hv_dev = device_to_hv_device(dev);
437 struct hv_ring_buffer_debug_info inbound;
440 if (!hv_dev->channel)
443 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
447 return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
449 static DEVICE_ATTR_RO(in_intr_mask);
451 static ssize_t in_read_index_show(struct device *dev,
452 struct device_attribute *dev_attr, char *buf)
454 struct hv_device *hv_dev = device_to_hv_device(dev);
455 struct hv_ring_buffer_debug_info inbound;
458 if (!hv_dev->channel)
461 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
465 return sprintf(buf, "%d\n", inbound.current_read_index);
467 static DEVICE_ATTR_RO(in_read_index);
469 static ssize_t in_write_index_show(struct device *dev,
470 struct device_attribute *dev_attr, char *buf)
472 struct hv_device *hv_dev = device_to_hv_device(dev);
473 struct hv_ring_buffer_debug_info inbound;
476 if (!hv_dev->channel)
479 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
483 return sprintf(buf, "%d\n", inbound.current_write_index);
485 static DEVICE_ATTR_RO(in_write_index);
487 static ssize_t in_read_bytes_avail_show(struct device *dev,
488 struct device_attribute *dev_attr,
491 struct hv_device *hv_dev = device_to_hv_device(dev);
492 struct hv_ring_buffer_debug_info inbound;
495 if (!hv_dev->channel)
498 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
502 return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
504 static DEVICE_ATTR_RO(in_read_bytes_avail);
506 static ssize_t in_write_bytes_avail_show(struct device *dev,
507 struct device_attribute *dev_attr,
510 struct hv_device *hv_dev = device_to_hv_device(dev);
511 struct hv_ring_buffer_debug_info inbound;
514 if (!hv_dev->channel)
517 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
521 return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
523 static DEVICE_ATTR_RO(in_write_bytes_avail);
525 static ssize_t channel_vp_mapping_show(struct device *dev,
526 struct device_attribute *dev_attr,
529 struct hv_device *hv_dev = device_to_hv_device(dev);
530 struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
532 int buf_size = PAGE_SIZE, n_written, tot_written;
533 struct list_head *cur;
538 tot_written = snprintf(buf, buf_size, "%u:%u\n",
539 channel->offermsg.child_relid, channel->target_cpu);
541 spin_lock_irqsave(&channel->lock, flags);
543 list_for_each(cur, &channel->sc_list) {
544 if (tot_written >= buf_size - 1)
547 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
548 n_written = scnprintf(buf + tot_written,
549 buf_size - tot_written,
551 cur_sc->offermsg.child_relid,
553 tot_written += n_written;
556 spin_unlock_irqrestore(&channel->lock, flags);
560 static DEVICE_ATTR_RO(channel_vp_mapping);
562 static ssize_t vendor_show(struct device *dev,
563 struct device_attribute *dev_attr,
566 struct hv_device *hv_dev = device_to_hv_device(dev);
567 return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
569 static DEVICE_ATTR_RO(vendor);
571 static ssize_t device_show(struct device *dev,
572 struct device_attribute *dev_attr,
575 struct hv_device *hv_dev = device_to_hv_device(dev);
576 return sprintf(buf, "0x%x\n", hv_dev->device_id);
578 static DEVICE_ATTR_RO(device);
580 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
581 static struct attribute *vmbus_dev_attrs[] = {
583 &dev_attr_state.attr,
584 &dev_attr_monitor_id.attr,
585 &dev_attr_class_id.attr,
586 &dev_attr_device_id.attr,
587 &dev_attr_modalias.attr,
589 &dev_attr_numa_node.attr,
591 &dev_attr_server_monitor_pending.attr,
592 &dev_attr_client_monitor_pending.attr,
593 &dev_attr_server_monitor_latency.attr,
594 &dev_attr_client_monitor_latency.attr,
595 &dev_attr_server_monitor_conn_id.attr,
596 &dev_attr_client_monitor_conn_id.attr,
597 &dev_attr_out_intr_mask.attr,
598 &dev_attr_out_read_index.attr,
599 &dev_attr_out_write_index.attr,
600 &dev_attr_out_read_bytes_avail.attr,
601 &dev_attr_out_write_bytes_avail.attr,
602 &dev_attr_in_intr_mask.attr,
603 &dev_attr_in_read_index.attr,
604 &dev_attr_in_write_index.attr,
605 &dev_attr_in_read_bytes_avail.attr,
606 &dev_attr_in_write_bytes_avail.attr,
607 &dev_attr_channel_vp_mapping.attr,
608 &dev_attr_vendor.attr,
609 &dev_attr_device.attr,
614 * Device-level attribute_group callback function. Returns the permission for
615 * each attribute, and returns 0 if an attribute is not visible.
617 static umode_t vmbus_dev_attr_is_visible(struct kobject *kobj,
618 struct attribute *attr, int idx)
620 struct device *dev = kobj_to_dev(kobj);
621 const struct hv_device *hv_dev = device_to_hv_device(dev);
623 /* Hide the monitor attributes if the monitor mechanism is not used. */
624 if (!hv_dev->channel->offermsg.monitor_allocated &&
625 (attr == &dev_attr_monitor_id.attr ||
626 attr == &dev_attr_server_monitor_pending.attr ||
627 attr == &dev_attr_client_monitor_pending.attr ||
628 attr == &dev_attr_server_monitor_latency.attr ||
629 attr == &dev_attr_client_monitor_latency.attr ||
630 attr == &dev_attr_server_monitor_conn_id.attr ||
631 attr == &dev_attr_client_monitor_conn_id.attr))
637 static const struct attribute_group vmbus_dev_group = {
638 .attrs = vmbus_dev_attrs,
639 .is_visible = vmbus_dev_attr_is_visible
641 __ATTRIBUTE_GROUPS(vmbus_dev);
644 * vmbus_uevent - add uevent for our device
646 * This routine is invoked when a device is added or removed on the vmbus to
647 * generate a uevent to udev in the userspace. The udev will then look at its
648 * rule and the uevent generated here to load the appropriate driver
650 * The alias string will be of the form vmbus:guid where guid is the string
651 * representation of the device guid (each byte of the guid will be
652 * represented with two hex characters.
654 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
656 struct hv_device *dev = device_to_hv_device(device);
658 char alias_name[VMBUS_ALIAS_LEN + 1];
660 print_alias_name(dev, alias_name);
661 ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
665 static const uuid_le null_guid;
667 static inline bool is_null_guid(const uuid_le *guid)
669 if (uuid_le_cmp(*guid, null_guid))
675 * Return a matching hv_vmbus_device_id pointer.
676 * If there is no match, return NULL.
678 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
681 const struct hv_vmbus_device_id *id = NULL;
682 struct vmbus_dynid *dynid;
684 /* Look at the dynamic ids first, before the static ones */
685 spin_lock(&drv->dynids.lock);
686 list_for_each_entry(dynid, &drv->dynids.list, node) {
687 if (!uuid_le_cmp(dynid->id.guid, *guid)) {
692 spin_unlock(&drv->dynids.lock);
699 return NULL; /* empty device table */
701 for (; !is_null_guid(&id->guid); id++)
702 if (!uuid_le_cmp(id->guid, *guid))
708 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
709 static int vmbus_add_dynid(struct hv_driver *drv, uuid_le *guid)
711 struct vmbus_dynid *dynid;
713 dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
717 dynid->id.guid = *guid;
719 spin_lock(&drv->dynids.lock);
720 list_add_tail(&dynid->node, &drv->dynids.list);
721 spin_unlock(&drv->dynids.lock);
723 return driver_attach(&drv->driver);
726 static void vmbus_free_dynids(struct hv_driver *drv)
728 struct vmbus_dynid *dynid, *n;
730 spin_lock(&drv->dynids.lock);
731 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
732 list_del(&dynid->node);
735 spin_unlock(&drv->dynids.lock);
739 * store_new_id - sysfs frontend to vmbus_add_dynid()
741 * Allow GUIDs to be added to an existing driver via sysfs.
743 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
746 struct hv_driver *drv = drv_to_hv_drv(driver);
750 retval = uuid_le_to_bin(buf, &guid);
754 if (hv_vmbus_get_id(drv, &guid))
757 retval = vmbus_add_dynid(drv, &guid);
762 static DRIVER_ATTR_WO(new_id);
765 * store_remove_id - remove a PCI device ID from this driver
767 * Removes a dynamic pci device ID to this driver.
769 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
772 struct hv_driver *drv = drv_to_hv_drv(driver);
773 struct vmbus_dynid *dynid, *n;
777 retval = uuid_le_to_bin(buf, &guid);
782 spin_lock(&drv->dynids.lock);
783 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
784 struct hv_vmbus_device_id *id = &dynid->id;
786 if (!uuid_le_cmp(id->guid, guid)) {
787 list_del(&dynid->node);
793 spin_unlock(&drv->dynids.lock);
797 static DRIVER_ATTR_WO(remove_id);
799 static struct attribute *vmbus_drv_attrs[] = {
800 &driver_attr_new_id.attr,
801 &driver_attr_remove_id.attr,
804 ATTRIBUTE_GROUPS(vmbus_drv);
808 * vmbus_match - Attempt to match the specified device to the specified driver
810 static int vmbus_match(struct device *device, struct device_driver *driver)
812 struct hv_driver *drv = drv_to_hv_drv(driver);
813 struct hv_device *hv_dev = device_to_hv_device(device);
815 /* The hv_sock driver handles all hv_sock offers. */
816 if (is_hvsock_channel(hv_dev->channel))
819 if (hv_vmbus_get_id(drv, &hv_dev->dev_type))
826 * vmbus_probe - Add the new vmbus's child device
828 static int vmbus_probe(struct device *child_device)
831 struct hv_driver *drv =
832 drv_to_hv_drv(child_device->driver);
833 struct hv_device *dev = device_to_hv_device(child_device);
834 const struct hv_vmbus_device_id *dev_id;
836 dev_id = hv_vmbus_get_id(drv, &dev->dev_type);
838 ret = drv->probe(dev, dev_id);
840 pr_err("probe failed for device %s (%d)\n",
841 dev_name(child_device), ret);
844 pr_err("probe not set for driver %s\n",
845 dev_name(child_device));
852 * vmbus_remove - Remove a vmbus device
854 static int vmbus_remove(struct device *child_device)
856 struct hv_driver *drv;
857 struct hv_device *dev = device_to_hv_device(child_device);
859 if (child_device->driver) {
860 drv = drv_to_hv_drv(child_device->driver);
870 * vmbus_shutdown - Shutdown a vmbus device
872 static void vmbus_shutdown(struct device *child_device)
874 struct hv_driver *drv;
875 struct hv_device *dev = device_to_hv_device(child_device);
878 /* The device may not be attached yet */
879 if (!child_device->driver)
882 drv = drv_to_hv_drv(child_device->driver);
890 * vmbus_device_release - Final callback release of the vmbus child device
892 static void vmbus_device_release(struct device *device)
894 struct hv_device *hv_dev = device_to_hv_device(device);
895 struct vmbus_channel *channel = hv_dev->channel;
897 mutex_lock(&vmbus_connection.channel_mutex);
898 hv_process_channel_removal(channel->offermsg.child_relid);
899 mutex_unlock(&vmbus_connection.channel_mutex);
904 /* The one and only one */
905 static struct bus_type hv_bus = {
907 .match = vmbus_match,
908 .shutdown = vmbus_shutdown,
909 .remove = vmbus_remove,
910 .probe = vmbus_probe,
911 .uevent = vmbus_uevent,
912 .dev_groups = vmbus_dev_groups,
913 .drv_groups = vmbus_drv_groups,
916 struct onmessage_work_context {
917 struct work_struct work;
918 struct hv_message msg;
921 static void vmbus_onmessage_work(struct work_struct *work)
923 struct onmessage_work_context *ctx;
925 /* Do not process messages if we're in DISCONNECTED state */
926 if (vmbus_connection.conn_state == DISCONNECTED)
929 ctx = container_of(work, struct onmessage_work_context,
931 vmbus_onmessage(&ctx->msg);
935 static void hv_process_timer_expiration(struct hv_message *msg,
936 struct hv_per_cpu_context *hv_cpu)
938 struct clock_event_device *dev = hv_cpu->clk_evt;
940 if (dev->event_handler)
941 dev->event_handler(dev);
943 vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
946 void vmbus_on_msg_dpc(unsigned long data)
948 struct hv_per_cpu_context *hv_cpu = (void *)data;
949 void *page_addr = hv_cpu->synic_message_page;
950 struct hv_message *msg = (struct hv_message *)page_addr +
952 struct vmbus_channel_message_header *hdr;
953 const struct vmbus_channel_message_table_entry *entry;
954 struct onmessage_work_context *ctx;
955 u32 message_type = msg->header.message_type;
957 if (message_type == HVMSG_NONE)
961 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
963 trace_vmbus_on_msg_dpc(hdr);
965 if (hdr->msgtype >= CHANNELMSG_COUNT) {
966 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
970 entry = &channel_message_table[hdr->msgtype];
972 if (!entry->message_handler)
975 if (entry->handler_type == VMHT_BLOCKING) {
976 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
980 INIT_WORK(&ctx->work, vmbus_onmessage_work);
981 memcpy(&ctx->msg, msg, sizeof(*msg));
984 * The host can generate a rescind message while we
985 * may still be handling the original offer. We deal with
986 * this condition by ensuring the processing is done on the
989 switch (hdr->msgtype) {
990 case CHANNELMSG_RESCIND_CHANNELOFFER:
992 * If we are handling the rescind message;
993 * schedule the work on the global work queue.
995 schedule_work_on(vmbus_connection.connect_cpu,
999 case CHANNELMSG_OFFERCHANNEL:
1000 atomic_inc(&vmbus_connection.offer_in_progress);
1001 queue_work_on(vmbus_connection.connect_cpu,
1002 vmbus_connection.work_queue,
1007 queue_work(vmbus_connection.work_queue, &ctx->work);
1010 entry->message_handler(hdr);
1013 vmbus_signal_eom(msg, message_type);
1018 * Direct callback for channels using other deferred processing
1020 static void vmbus_channel_isr(struct vmbus_channel *channel)
1022 void (*callback_fn)(void *);
1024 callback_fn = READ_ONCE(channel->onchannel_callback);
1025 if (likely(callback_fn != NULL))
1026 (*callback_fn)(channel->channel_callback_context);
1030 * Schedule all channels with events pending
1032 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1034 unsigned long *recv_int_page;
1037 if (vmbus_proto_version < VERSION_WIN8) {
1038 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1039 recv_int_page = vmbus_connection.recv_int_page;
1042 * When the host is win8 and beyond, the event page
1043 * can be directly checked to get the id of the channel
1044 * that has the interrupt pending.
1046 void *page_addr = hv_cpu->synic_event_page;
1047 union hv_synic_event_flags *event
1048 = (union hv_synic_event_flags *)page_addr +
1051 maxbits = HV_EVENT_FLAGS_COUNT;
1052 recv_int_page = event->flags;
1055 if (unlikely(!recv_int_page))
1058 for_each_set_bit(relid, recv_int_page, maxbits) {
1059 struct vmbus_channel *channel;
1061 if (!sync_test_and_clear_bit(relid, recv_int_page))
1064 /* Special case - vmbus channel protocol msg */
1070 /* Find channel based on relid */
1071 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
1072 if (channel->offermsg.child_relid != relid)
1075 if (channel->rescind)
1078 trace_vmbus_chan_sched(channel);
1080 ++channel->interrupts;
1082 switch (channel->callback_mode) {
1084 vmbus_channel_isr(channel);
1087 case HV_CALL_BATCHED:
1088 hv_begin_read(&channel->inbound);
1090 case HV_CALL_DIRECT:
1091 tasklet_schedule(&channel->callback_event);
1099 static void vmbus_isr(void)
1101 struct hv_per_cpu_context *hv_cpu
1102 = this_cpu_ptr(hv_context.cpu_context);
1103 void *page_addr = hv_cpu->synic_event_page;
1104 struct hv_message *msg;
1105 union hv_synic_event_flags *event;
1106 bool handled = false;
1108 if (unlikely(page_addr == NULL))
1111 event = (union hv_synic_event_flags *)page_addr +
1114 * Check for events before checking for messages. This is the order
1115 * in which events and messages are checked in Windows guests on
1116 * Hyper-V, and the Windows team suggested we do the same.
1119 if ((vmbus_proto_version == VERSION_WS2008) ||
1120 (vmbus_proto_version == VERSION_WIN7)) {
1122 /* Since we are a child, we only need to check bit 0 */
1123 if (sync_test_and_clear_bit(0, event->flags))
1127 * Our host is win8 or above. The signaling mechanism
1128 * has changed and we can directly look at the event page.
1129 * If bit n is set then we have an interrup on the channel
1136 vmbus_chan_sched(hv_cpu);
1138 page_addr = hv_cpu->synic_message_page;
1139 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1141 /* Check if there are actual msgs to be processed */
1142 if (msg->header.message_type != HVMSG_NONE) {
1143 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
1144 hv_process_timer_expiration(msg, hv_cpu);
1146 tasklet_schedule(&hv_cpu->msg_dpc);
1149 add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR);
1153 * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1154 * buffer and call into Hyper-V to transfer the data.
1156 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1157 enum kmsg_dump_reason reason)
1159 size_t bytes_written;
1160 phys_addr_t panic_pa;
1162 /* We are only interested in panics. */
1163 if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1166 panic_pa = virt_to_phys(hv_panic_page);
1169 * Write dump contents to the page. No need to synchronize; panic should
1170 * be single-threaded.
1172 kmsg_dump_get_buffer(dumper, true, hv_panic_page, PAGE_SIZE,
1175 hyperv_report_panic_msg(panic_pa, bytes_written);
1178 static struct kmsg_dumper hv_kmsg_dumper = {
1179 .dump = hv_kmsg_dump,
1182 static struct ctl_table_header *hv_ctl_table_hdr;
1187 * sysctl option to allow the user to control whether kmsg data should be
1188 * reported to Hyper-V on panic.
1190 static struct ctl_table hv_ctl_table[] = {
1192 .procname = "hyperv_record_panic_msg",
1193 .data = &sysctl_record_panic_msg,
1194 .maxlen = sizeof(int),
1196 .proc_handler = proc_dointvec_minmax,
1203 static struct ctl_table hv_root_table[] = {
1205 .procname = "kernel",
1207 .child = hv_ctl_table
1213 * vmbus_bus_init -Main vmbus driver initialization routine.
1216 * - initialize the vmbus driver context
1217 * - invoke the vmbus hv main init routine
1218 * - retrieve the channel offers
1220 static int vmbus_bus_init(void)
1224 /* Hypervisor initialization...setup hypercall page..etc */
1227 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1231 ret = bus_register(&hv_bus);
1235 hv_setup_vmbus_irq(vmbus_isr);
1237 ret = hv_synic_alloc();
1241 * Initialize the per-cpu interrupt state and
1242 * connect to the host.
1244 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1245 hv_synic_init, hv_synic_cleanup);
1248 hyperv_cpuhp_online = ret;
1250 ret = vmbus_connect();
1255 * Only register if the crash MSRs are available
1257 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1258 u64 hyperv_crash_ctl;
1260 * Sysctl registration is not fatal, since by default
1261 * reporting is enabled.
1263 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1264 if (!hv_ctl_table_hdr)
1265 pr_err("Hyper-V: sysctl table register error");
1268 * Register for panic kmsg callback only if the right
1269 * capability is supported by the hypervisor.
1271 hv_get_crash_ctl(hyperv_crash_ctl);
1272 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1273 hv_panic_page = (void *)get_zeroed_page(GFP_KERNEL);
1274 if (hv_panic_page) {
1275 ret = kmsg_dump_register(&hv_kmsg_dumper);
1277 pr_err("Hyper-V: kmsg dump register "
1278 "error 0x%x\n", ret);
1280 (unsigned long)hv_panic_page);
1281 hv_panic_page = NULL;
1284 pr_err("Hyper-V: panic message page memory "
1285 "allocation failed");
1288 register_die_notifier(&hyperv_die_block);
1292 * Always register the panic notifier because we need to unload
1293 * the VMbus channel connection to prevent any VMbus
1294 * activity after the VM panics.
1296 atomic_notifier_chain_register(&panic_notifier_list,
1297 &hyperv_panic_block);
1299 vmbus_request_offers();
1304 cpuhp_remove_state(hyperv_cpuhp_online);
1307 hv_remove_vmbus_irq();
1309 bus_unregister(&hv_bus);
1310 unregister_sysctl_table(hv_ctl_table_hdr);
1311 hv_ctl_table_hdr = NULL;
1316 * __vmbus_child_driver_register() - Register a vmbus's driver
1317 * @hv_driver: Pointer to driver structure you want to register
1318 * @owner: owner module of the drv
1319 * @mod_name: module name string
1321 * Registers the given driver with Linux through the 'driver_register()' call
1322 * and sets up the hyper-v vmbus handling for this driver.
1323 * It will return the state of the 'driver_register()' call.
1326 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1330 pr_info("registering driver %s\n", hv_driver->name);
1332 ret = vmbus_exists();
1336 hv_driver->driver.name = hv_driver->name;
1337 hv_driver->driver.owner = owner;
1338 hv_driver->driver.mod_name = mod_name;
1339 hv_driver->driver.bus = &hv_bus;
1341 spin_lock_init(&hv_driver->dynids.lock);
1342 INIT_LIST_HEAD(&hv_driver->dynids.list);
1344 ret = driver_register(&hv_driver->driver);
1348 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1351 * vmbus_driver_unregister() - Unregister a vmbus's driver
1352 * @hv_driver: Pointer to driver structure you want to
1355 * Un-register the given driver that was previous registered with a call to
1356 * vmbus_driver_register()
1358 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1360 pr_info("unregistering driver %s\n", hv_driver->name);
1362 if (!vmbus_exists()) {
1363 driver_unregister(&hv_driver->driver);
1364 vmbus_free_dynids(hv_driver);
1367 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1371 * Called when last reference to channel is gone.
1373 static void vmbus_chan_release(struct kobject *kobj)
1375 struct vmbus_channel *channel
1376 = container_of(kobj, struct vmbus_channel, kobj);
1378 kfree_rcu(channel, rcu);
1381 struct vmbus_chan_attribute {
1382 struct attribute attr;
1383 ssize_t (*show)(const struct vmbus_channel *chan, char *buf);
1384 ssize_t (*store)(struct vmbus_channel *chan,
1385 const char *buf, size_t count);
1387 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1388 struct vmbus_chan_attribute chan_attr_##_name \
1389 = __ATTR(_name, _mode, _show, _store)
1390 #define VMBUS_CHAN_ATTR_RW(_name) \
1391 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1392 #define VMBUS_CHAN_ATTR_RO(_name) \
1393 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1394 #define VMBUS_CHAN_ATTR_WO(_name) \
1395 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1397 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1398 struct attribute *attr, char *buf)
1400 const struct vmbus_chan_attribute *attribute
1401 = container_of(attr, struct vmbus_chan_attribute, attr);
1402 const struct vmbus_channel *chan
1403 = container_of(kobj, struct vmbus_channel, kobj);
1405 if (!attribute->show)
1408 if (chan->state != CHANNEL_OPENED_STATE)
1411 return attribute->show(chan, buf);
1414 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1415 .show = vmbus_chan_attr_show,
1418 static ssize_t out_mask_show(const struct vmbus_channel *channel, char *buf)
1420 const struct hv_ring_buffer_info *rbi = &channel->outbound;
1422 return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1424 static VMBUS_CHAN_ATTR_RO(out_mask);
1426 static ssize_t in_mask_show(const struct vmbus_channel *channel, char *buf)
1428 const struct hv_ring_buffer_info *rbi = &channel->inbound;
1430 return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1432 static VMBUS_CHAN_ATTR_RO(in_mask);
1434 static ssize_t read_avail_show(const struct vmbus_channel *channel, char *buf)
1436 const struct hv_ring_buffer_info *rbi = &channel->inbound;
1438 return sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1440 static VMBUS_CHAN_ATTR_RO(read_avail);
1442 static ssize_t write_avail_show(const struct vmbus_channel *channel, char *buf)
1444 const struct hv_ring_buffer_info *rbi = &channel->outbound;
1446 return sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1448 static VMBUS_CHAN_ATTR_RO(write_avail);
1450 static ssize_t show_target_cpu(const struct vmbus_channel *channel, char *buf)
1452 return sprintf(buf, "%u\n", channel->target_cpu);
1454 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1456 static ssize_t channel_pending_show(const struct vmbus_channel *channel,
1459 return sprintf(buf, "%d\n",
1460 channel_pending(channel,
1461 vmbus_connection.monitor_pages[1]));
1463 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1465 static ssize_t channel_latency_show(const struct vmbus_channel *channel,
1468 return sprintf(buf, "%d\n",
1469 channel_latency(channel,
1470 vmbus_connection.monitor_pages[1]));
1472 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1474 static ssize_t channel_interrupts_show(const struct vmbus_channel *channel, char *buf)
1476 return sprintf(buf, "%llu\n", channel->interrupts);
1478 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1480 static ssize_t channel_events_show(const struct vmbus_channel *channel, char *buf)
1482 return sprintf(buf, "%llu\n", channel->sig_events);
1484 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1486 static ssize_t subchannel_monitor_id_show(const struct vmbus_channel *channel,
1489 return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1491 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1493 static ssize_t subchannel_id_show(const struct vmbus_channel *channel,
1496 return sprintf(buf, "%u\n",
1497 channel->offermsg.offer.sub_channel_index);
1499 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1501 static struct attribute *vmbus_chan_attrs[] = {
1502 &chan_attr_out_mask.attr,
1503 &chan_attr_in_mask.attr,
1504 &chan_attr_read_avail.attr,
1505 &chan_attr_write_avail.attr,
1506 &chan_attr_cpu.attr,
1507 &chan_attr_pending.attr,
1508 &chan_attr_latency.attr,
1509 &chan_attr_interrupts.attr,
1510 &chan_attr_events.attr,
1511 &chan_attr_monitor_id.attr,
1512 &chan_attr_subchannel_id.attr,
1517 * Channel-level attribute_group callback function. Returns the permission for
1518 * each attribute, and returns 0 if an attribute is not visible.
1520 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1521 struct attribute *attr, int idx)
1523 const struct vmbus_channel *channel =
1524 container_of(kobj, struct vmbus_channel, kobj);
1526 /* Hide the monitor attributes if the monitor mechanism is not used. */
1527 if (!channel->offermsg.monitor_allocated &&
1528 (attr == &chan_attr_pending.attr ||
1529 attr == &chan_attr_latency.attr ||
1530 attr == &chan_attr_monitor_id.attr))
1536 static struct attribute_group vmbus_chan_group = {
1537 .attrs = vmbus_chan_attrs,
1538 .is_visible = vmbus_chan_attr_is_visible
1541 static struct kobj_type vmbus_chan_ktype = {
1542 .sysfs_ops = &vmbus_chan_sysfs_ops,
1543 .release = vmbus_chan_release,
1547 * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1549 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1551 const struct device *device = &dev->device;
1552 struct kobject *kobj = &channel->kobj;
1553 u32 relid = channel->offermsg.child_relid;
1556 kobj->kset = dev->channels_kset;
1557 ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1564 ret = sysfs_create_group(kobj, &vmbus_chan_group);
1568 * The calling functions' error handling paths will cleanup the
1569 * empty channel directory.
1572 dev_err(device, "Unable to set up channel sysfs files\n");
1576 kobject_uevent(kobj, KOBJ_ADD);
1582 * vmbus_remove_channel_attr_group - remove the channel's attribute group
1584 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
1586 sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
1590 * vmbus_device_create - Creates and registers a new child device
1593 struct hv_device *vmbus_device_create(const uuid_le *type,
1594 const uuid_le *instance,
1595 struct vmbus_channel *channel)
1597 struct hv_device *child_device_obj;
1599 child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1600 if (!child_device_obj) {
1601 pr_err("Unable to allocate device object for child device\n");
1605 child_device_obj->channel = channel;
1606 memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
1607 memcpy(&child_device_obj->dev_instance, instance,
1609 child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1612 return child_device_obj;
1616 * vmbus_device_register - Register the child device
1618 int vmbus_device_register(struct hv_device *child_device_obj)
1620 struct kobject *kobj = &child_device_obj->device.kobj;
1623 dev_set_name(&child_device_obj->device, "%pUl",
1624 child_device_obj->channel->offermsg.offer.if_instance.b);
1626 child_device_obj->device.bus = &hv_bus;
1627 child_device_obj->device.parent = &hv_acpi_dev->dev;
1628 child_device_obj->device.release = vmbus_device_release;
1631 * Register with the LDM. This will kick off the driver/device
1632 * binding...which will eventually call vmbus_match() and vmbus_probe()
1634 ret = device_register(&child_device_obj->device);
1636 pr_err("Unable to register child device\n");
1637 put_device(&child_device_obj->device);
1641 child_device_obj->channels_kset = kset_create_and_add("channels",
1643 if (!child_device_obj->channels_kset) {
1645 goto err_dev_unregister;
1648 ret = vmbus_add_channel_kobj(child_device_obj,
1649 child_device_obj->channel);
1651 pr_err("Unable to register primary channeln");
1652 goto err_kset_unregister;
1657 err_kset_unregister:
1658 kset_unregister(child_device_obj->channels_kset);
1661 device_unregister(&child_device_obj->device);
1666 * vmbus_device_unregister - Remove the specified child device
1669 void vmbus_device_unregister(struct hv_device *device_obj)
1671 pr_debug("child device %s unregistered\n",
1672 dev_name(&device_obj->device));
1674 kset_unregister(device_obj->channels_kset);
1677 * Kick off the process of unregistering the device.
1678 * This will call vmbus_remove() and eventually vmbus_device_release()
1680 device_unregister(&device_obj->device);
1685 * VMBUS is an acpi enumerated device. Get the information we
1688 #define VTPM_BASE_ADDRESS 0xfed40000
1689 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1691 resource_size_t start = 0;
1692 resource_size_t end = 0;
1693 struct resource *new_res;
1694 struct resource **old_res = &hyperv_mmio;
1695 struct resource **prev_res = NULL;
1697 switch (res->type) {
1700 * "Address" descriptors are for bus windows. Ignore
1701 * "memory" descriptors, which are for registers on
1704 case ACPI_RESOURCE_TYPE_ADDRESS32:
1705 start = res->data.address32.address.minimum;
1706 end = res->data.address32.address.maximum;
1709 case ACPI_RESOURCE_TYPE_ADDRESS64:
1710 start = res->data.address64.address.minimum;
1711 end = res->data.address64.address.maximum;
1715 /* Unused resource type */
1720 * Ignore ranges that are below 1MB, as they're not
1721 * necessary or useful here.
1726 new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1728 return AE_NO_MEMORY;
1730 /* If this range overlaps the virtual TPM, truncate it. */
1731 if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1732 end = VTPM_BASE_ADDRESS;
1734 new_res->name = "hyperv mmio";
1735 new_res->flags = IORESOURCE_MEM;
1736 new_res->start = start;
1740 * If two ranges are adjacent, merge them.
1748 if (((*old_res)->end + 1) == new_res->start) {
1749 (*old_res)->end = new_res->end;
1754 if ((*old_res)->start == new_res->end + 1) {
1755 (*old_res)->start = new_res->start;
1760 if ((*old_res)->start > new_res->end) {
1761 new_res->sibling = *old_res;
1763 (*prev_res)->sibling = new_res;
1769 old_res = &(*old_res)->sibling;
1776 static int vmbus_acpi_remove(struct acpi_device *device)
1778 struct resource *cur_res;
1779 struct resource *next_res;
1783 __release_region(hyperv_mmio, fb_mmio->start,
1784 resource_size(fb_mmio));
1788 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1789 next_res = cur_res->sibling;
1797 static void vmbus_reserve_fb(void)
1801 * Make a claim for the frame buffer in the resource tree under the
1802 * first node, which will be the one below 4GB. The length seems to
1803 * be underreported, particularly in a Generation 1 VM. So start out
1804 * reserving a larger area and make it smaller until it succeeds.
1807 if (screen_info.lfb_base) {
1808 if (efi_enabled(EFI_BOOT))
1809 size = max_t(__u32, screen_info.lfb_size, 0x800000);
1811 size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1813 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1814 fb_mmio = __request_region(hyperv_mmio,
1815 screen_info.lfb_base, size,
1822 * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1823 * @new: If successful, supplied a pointer to the
1824 * allocated MMIO space.
1825 * @device_obj: Identifies the caller
1826 * @min: Minimum guest physical address of the
1828 * @max: Maximum guest physical address
1829 * @size: Size of the range to be allocated
1830 * @align: Alignment of the range to be allocated
1831 * @fb_overlap_ok: Whether this allocation can be allowed
1832 * to overlap the video frame buffer.
1834 * This function walks the resources granted to VMBus by the
1835 * _CRS object in the ACPI namespace underneath the parent
1836 * "bridge" whether that's a root PCI bus in the Generation 1
1837 * case or a Module Device in the Generation 2 case. It then
1838 * attempts to allocate from the global MMIO pool in a way that
1839 * matches the constraints supplied in these parameters and by
1842 * Return: 0 on success, -errno on failure
1844 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1845 resource_size_t min, resource_size_t max,
1846 resource_size_t size, resource_size_t align,
1849 struct resource *iter, *shadow;
1850 resource_size_t range_min, range_max, start, end;
1851 const char *dev_n = dev_name(&device_obj->device);
1855 down(&hyperv_mmio_lock);
1858 * If overlaps with frame buffers are allowed, then first attempt to
1859 * make the allocation from within the reserved region. Because it
1860 * is already reserved, no shadow allocation is necessary.
1862 if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1863 !(max < fb_mmio->start)) {
1865 range_min = fb_mmio->start;
1866 range_max = fb_mmio->end;
1867 start = (range_min + align - 1) & ~(align - 1);
1868 for (; start + size - 1 <= range_max; start += align) {
1869 *new = request_mem_region_exclusive(start, size, dev_n);
1877 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1878 if ((iter->start >= max) || (iter->end <= min))
1881 range_min = iter->start;
1882 range_max = iter->end;
1883 start = (range_min + align - 1) & ~(align - 1);
1884 for (; start + size - 1 <= range_max; start += align) {
1885 end = start + size - 1;
1887 /* Skip the whole fb_mmio region if not fb_overlap_ok */
1888 if (!fb_overlap_ok && fb_mmio &&
1889 (((start >= fb_mmio->start) && (start <= fb_mmio->end)) ||
1890 ((end >= fb_mmio->start) && (end <= fb_mmio->end))))
1893 shadow = __request_region(iter, start, size, NULL,
1898 *new = request_mem_region_exclusive(start, size, dev_n);
1900 shadow->name = (char *)*new;
1905 __release_region(iter, start, size);
1910 up(&hyperv_mmio_lock);
1913 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1916 * vmbus_free_mmio() - Free a memory-mapped I/O range.
1917 * @start: Base address of region to release.
1918 * @size: Size of the range to be allocated
1920 * This function releases anything requested by
1921 * vmbus_mmio_allocate().
1923 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
1925 struct resource *iter;
1927 down(&hyperv_mmio_lock);
1928 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1929 if ((iter->start >= start + size) || (iter->end <= start))
1932 __release_region(iter, start, size);
1934 release_mem_region(start, size);
1935 up(&hyperv_mmio_lock);
1938 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
1940 static int vmbus_acpi_add(struct acpi_device *device)
1943 int ret_val = -ENODEV;
1944 struct acpi_device *ancestor;
1946 hv_acpi_dev = device;
1948 result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1949 vmbus_walk_resources, NULL);
1951 if (ACPI_FAILURE(result))
1954 * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1955 * firmware) is the VMOD that has the mmio ranges. Get that.
1957 for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1958 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1959 vmbus_walk_resources, NULL);
1961 if (ACPI_FAILURE(result))
1971 complete(&probe_event);
1973 vmbus_acpi_remove(device);
1977 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1982 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1984 static struct acpi_driver vmbus_acpi_driver = {
1986 .ids = vmbus_acpi_device_ids,
1988 .add = vmbus_acpi_add,
1989 .remove = vmbus_acpi_remove,
1993 static void hv_kexec_handler(void)
1995 hv_synic_clockevents_cleanup();
1996 vmbus_initiate_unload(false);
1997 /* Make sure conn_state is set as hv_synic_cleanup checks for it */
1999 cpuhp_remove_state(hyperv_cpuhp_online);
2003 static void hv_crash_handler(struct pt_regs *regs)
2005 vmbus_initiate_unload(true);
2007 * In crash handler we can't schedule synic cleanup for all CPUs,
2008 * doing the cleanup for current CPU only. This should be sufficient
2011 hv_synic_cleanup(smp_processor_id());
2015 static int __init hv_acpi_init(void)
2019 if (!hv_is_hyperv_initialized())
2022 init_completion(&probe_event);
2025 * Get ACPI resources first.
2027 ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2032 t = wait_for_completion_timeout(&probe_event, 5*HZ);
2038 ret = vmbus_bus_init();
2042 hv_setup_kexec_handler(hv_kexec_handler);
2043 hv_setup_crash_handler(hv_crash_handler);
2048 acpi_bus_unregister_driver(&vmbus_acpi_driver);
2053 static void __exit vmbus_exit(void)
2057 hv_remove_kexec_handler();
2058 hv_remove_crash_handler();
2059 vmbus_connection.conn_state = DISCONNECTED;
2060 hv_synic_clockevents_cleanup();
2062 hv_remove_vmbus_irq();
2063 for_each_online_cpu(cpu) {
2064 struct hv_per_cpu_context *hv_cpu
2065 = per_cpu_ptr(hv_context.cpu_context, cpu);
2067 tasklet_kill(&hv_cpu->msg_dpc);
2069 vmbus_free_channels();
2071 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2072 kmsg_dump_unregister(&hv_kmsg_dumper);
2073 unregister_die_notifier(&hyperv_die_block);
2077 * The panic notifier is always registered, hence we should
2078 * also unconditionally unregister it here as well.
2080 atomic_notifier_chain_unregister(&panic_notifier_list,
2081 &hyperv_panic_block);
2083 free_page((unsigned long)hv_panic_page);
2084 unregister_sysctl_table(hv_ctl_table_hdr);
2085 hv_ctl_table_hdr = NULL;
2086 bus_unregister(&hv_bus);
2088 cpuhp_remove_state(hyperv_cpuhp_online);
2090 acpi_bus_unregister_driver(&vmbus_acpi_driver);
2094 MODULE_LICENSE("GPL");
2096 subsys_initcall(hv_acpi_init);
2097 module_exit(vmbus_exit);