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/hyperv.h>
40 #include <asm/hypervisor.h>
41 #include <asm/mshyperv.h>
42 #include <linux/notifier.h>
43 #include <linux/ptrace.h>
44 #include <linux/screen_info.h>
45 #include <linux/kdebug.h>
46 #include <linux/efi.h>
47 #include <linux/random.h>
48 #include "hyperv_vmbus.h"
51 struct list_head node;
52 struct hv_vmbus_device_id id;
55 static struct acpi_device *hv_acpi_dev;
57 static struct completion probe_event;
59 static int hyperv_cpuhp_online;
61 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
66 regs = current_pt_regs();
68 hyperv_report_panic(regs);
72 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
75 struct die_args *die = (struct die_args *)args;
76 struct pt_regs *regs = die->regs;
78 hyperv_report_panic(regs);
82 static struct notifier_block hyperv_die_block = {
83 .notifier_call = hyperv_die_event,
85 static struct notifier_block hyperv_panic_block = {
86 .notifier_call = hyperv_panic_event,
89 static const char *fb_mmio_name = "fb_range";
90 static struct resource *fb_mmio;
91 static struct resource *hyperv_mmio;
92 static DEFINE_SEMAPHORE(hyperv_mmio_lock);
94 static int vmbus_exists(void)
96 if (hv_acpi_dev == NULL)
102 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
103 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
106 for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
107 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
110 static u8 channel_monitor_group(struct vmbus_channel *channel)
112 return (u8)channel->offermsg.monitorid / 32;
115 static u8 channel_monitor_offset(struct vmbus_channel *channel)
117 return (u8)channel->offermsg.monitorid % 32;
120 static u32 channel_pending(struct vmbus_channel *channel,
121 struct hv_monitor_page *monitor_page)
123 u8 monitor_group = channel_monitor_group(channel);
124 return monitor_page->trigger_group[monitor_group].pending;
127 static u32 channel_latency(struct vmbus_channel *channel,
128 struct hv_monitor_page *monitor_page)
130 u8 monitor_group = channel_monitor_group(channel);
131 u8 monitor_offset = channel_monitor_offset(channel);
132 return monitor_page->latency[monitor_group][monitor_offset];
135 static u32 channel_conn_id(struct vmbus_channel *channel,
136 struct hv_monitor_page *monitor_page)
138 u8 monitor_group = channel_monitor_group(channel);
139 u8 monitor_offset = channel_monitor_offset(channel);
140 return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
143 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
146 struct hv_device *hv_dev = device_to_hv_device(dev);
148 if (!hv_dev->channel)
150 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
152 static DEVICE_ATTR_RO(id);
154 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
157 struct hv_device *hv_dev = device_to_hv_device(dev);
159 if (!hv_dev->channel)
161 return sprintf(buf, "%d\n", hv_dev->channel->state);
163 static DEVICE_ATTR_RO(state);
165 static ssize_t monitor_id_show(struct device *dev,
166 struct device_attribute *dev_attr, char *buf)
168 struct hv_device *hv_dev = device_to_hv_device(dev);
170 if (!hv_dev->channel)
172 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
174 static DEVICE_ATTR_RO(monitor_id);
176 static ssize_t class_id_show(struct device *dev,
177 struct device_attribute *dev_attr, char *buf)
179 struct hv_device *hv_dev = device_to_hv_device(dev);
181 if (!hv_dev->channel)
183 return sprintf(buf, "{%pUl}\n",
184 hv_dev->channel->offermsg.offer.if_type.b);
186 static DEVICE_ATTR_RO(class_id);
188 static ssize_t device_id_show(struct device *dev,
189 struct device_attribute *dev_attr, char *buf)
191 struct hv_device *hv_dev = device_to_hv_device(dev);
193 if (!hv_dev->channel)
195 return sprintf(buf, "{%pUl}\n",
196 hv_dev->channel->offermsg.offer.if_instance.b);
198 static DEVICE_ATTR_RO(device_id);
200 static ssize_t modalias_show(struct device *dev,
201 struct device_attribute *dev_attr, char *buf)
203 struct hv_device *hv_dev = device_to_hv_device(dev);
204 char alias_name[VMBUS_ALIAS_LEN + 1];
206 print_alias_name(hv_dev, alias_name);
207 return sprintf(buf, "vmbus:%s\n", alias_name);
209 static DEVICE_ATTR_RO(modalias);
211 static ssize_t server_monitor_pending_show(struct device *dev,
212 struct device_attribute *dev_attr,
215 struct hv_device *hv_dev = device_to_hv_device(dev);
217 if (!hv_dev->channel)
219 return sprintf(buf, "%d\n",
220 channel_pending(hv_dev->channel,
221 vmbus_connection.monitor_pages[1]));
223 static DEVICE_ATTR_RO(server_monitor_pending);
225 static ssize_t client_monitor_pending_show(struct device *dev,
226 struct device_attribute *dev_attr,
229 struct hv_device *hv_dev = device_to_hv_device(dev);
231 if (!hv_dev->channel)
233 return sprintf(buf, "%d\n",
234 channel_pending(hv_dev->channel,
235 vmbus_connection.monitor_pages[1]));
237 static DEVICE_ATTR_RO(client_monitor_pending);
239 static ssize_t server_monitor_latency_show(struct device *dev,
240 struct device_attribute *dev_attr,
243 struct hv_device *hv_dev = device_to_hv_device(dev);
245 if (!hv_dev->channel)
247 return sprintf(buf, "%d\n",
248 channel_latency(hv_dev->channel,
249 vmbus_connection.monitor_pages[0]));
251 static DEVICE_ATTR_RO(server_monitor_latency);
253 static ssize_t client_monitor_latency_show(struct device *dev,
254 struct device_attribute *dev_attr,
257 struct hv_device *hv_dev = device_to_hv_device(dev);
259 if (!hv_dev->channel)
261 return sprintf(buf, "%d\n",
262 channel_latency(hv_dev->channel,
263 vmbus_connection.monitor_pages[1]));
265 static DEVICE_ATTR_RO(client_monitor_latency);
267 static ssize_t server_monitor_conn_id_show(struct device *dev,
268 struct device_attribute *dev_attr,
271 struct hv_device *hv_dev = device_to_hv_device(dev);
273 if (!hv_dev->channel)
275 return sprintf(buf, "%d\n",
276 channel_conn_id(hv_dev->channel,
277 vmbus_connection.monitor_pages[0]));
279 static DEVICE_ATTR_RO(server_monitor_conn_id);
281 static ssize_t client_monitor_conn_id_show(struct device *dev,
282 struct device_attribute *dev_attr,
285 struct hv_device *hv_dev = device_to_hv_device(dev);
287 if (!hv_dev->channel)
289 return sprintf(buf, "%d\n",
290 channel_conn_id(hv_dev->channel,
291 vmbus_connection.monitor_pages[1]));
293 static DEVICE_ATTR_RO(client_monitor_conn_id);
295 static ssize_t out_intr_mask_show(struct device *dev,
296 struct device_attribute *dev_attr, char *buf)
298 struct hv_device *hv_dev = device_to_hv_device(dev);
299 struct hv_ring_buffer_debug_info outbound;
302 if (!hv_dev->channel)
305 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
310 return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
312 static DEVICE_ATTR_RO(out_intr_mask);
314 static ssize_t out_read_index_show(struct device *dev,
315 struct device_attribute *dev_attr, char *buf)
317 struct hv_device *hv_dev = device_to_hv_device(dev);
318 struct hv_ring_buffer_debug_info outbound;
321 if (!hv_dev->channel)
324 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
328 return sprintf(buf, "%d\n", outbound.current_read_index);
330 static DEVICE_ATTR_RO(out_read_index);
332 static ssize_t out_write_index_show(struct device *dev,
333 struct device_attribute *dev_attr,
336 struct hv_device *hv_dev = device_to_hv_device(dev);
337 struct hv_ring_buffer_debug_info outbound;
340 if (!hv_dev->channel)
343 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
347 return sprintf(buf, "%d\n", outbound.current_write_index);
349 static DEVICE_ATTR_RO(out_write_index);
351 static ssize_t out_read_bytes_avail_show(struct device *dev,
352 struct device_attribute *dev_attr,
355 struct hv_device *hv_dev = device_to_hv_device(dev);
356 struct hv_ring_buffer_debug_info outbound;
359 if (!hv_dev->channel)
362 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
366 return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
368 static DEVICE_ATTR_RO(out_read_bytes_avail);
370 static ssize_t out_write_bytes_avail_show(struct device *dev,
371 struct device_attribute *dev_attr,
374 struct hv_device *hv_dev = device_to_hv_device(dev);
375 struct hv_ring_buffer_debug_info outbound;
378 if (!hv_dev->channel)
381 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
385 return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
387 static DEVICE_ATTR_RO(out_write_bytes_avail);
389 static ssize_t in_intr_mask_show(struct device *dev,
390 struct device_attribute *dev_attr, char *buf)
392 struct hv_device *hv_dev = device_to_hv_device(dev);
393 struct hv_ring_buffer_debug_info inbound;
396 if (!hv_dev->channel)
399 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
403 return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
405 static DEVICE_ATTR_RO(in_intr_mask);
407 static ssize_t in_read_index_show(struct device *dev,
408 struct device_attribute *dev_attr, char *buf)
410 struct hv_device *hv_dev = device_to_hv_device(dev);
411 struct hv_ring_buffer_debug_info inbound;
414 if (!hv_dev->channel)
417 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
421 return sprintf(buf, "%d\n", inbound.current_read_index);
423 static DEVICE_ATTR_RO(in_read_index);
425 static ssize_t in_write_index_show(struct device *dev,
426 struct device_attribute *dev_attr, char *buf)
428 struct hv_device *hv_dev = device_to_hv_device(dev);
429 struct hv_ring_buffer_debug_info inbound;
432 if (!hv_dev->channel)
435 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
439 return sprintf(buf, "%d\n", inbound.current_write_index);
441 static DEVICE_ATTR_RO(in_write_index);
443 static ssize_t in_read_bytes_avail_show(struct device *dev,
444 struct device_attribute *dev_attr,
447 struct hv_device *hv_dev = device_to_hv_device(dev);
448 struct hv_ring_buffer_debug_info inbound;
451 if (!hv_dev->channel)
454 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
458 return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
460 static DEVICE_ATTR_RO(in_read_bytes_avail);
462 static ssize_t in_write_bytes_avail_show(struct device *dev,
463 struct device_attribute *dev_attr,
466 struct hv_device *hv_dev = device_to_hv_device(dev);
467 struct hv_ring_buffer_debug_info inbound;
470 if (!hv_dev->channel)
473 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
477 return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
479 static DEVICE_ATTR_RO(in_write_bytes_avail);
481 static ssize_t channel_vp_mapping_show(struct device *dev,
482 struct device_attribute *dev_attr,
485 struct hv_device *hv_dev = device_to_hv_device(dev);
486 struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
488 int buf_size = PAGE_SIZE, n_written, tot_written;
489 struct list_head *cur;
494 tot_written = snprintf(buf, buf_size, "%u:%u\n",
495 channel->offermsg.child_relid, channel->target_cpu);
497 spin_lock_irqsave(&channel->lock, flags);
499 list_for_each(cur, &channel->sc_list) {
500 if (tot_written >= buf_size - 1)
503 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
504 n_written = scnprintf(buf + tot_written,
505 buf_size - tot_written,
507 cur_sc->offermsg.child_relid,
509 tot_written += n_written;
512 spin_unlock_irqrestore(&channel->lock, flags);
516 static DEVICE_ATTR_RO(channel_vp_mapping);
518 static ssize_t vendor_show(struct device *dev,
519 struct device_attribute *dev_attr,
522 struct hv_device *hv_dev = device_to_hv_device(dev);
523 return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
525 static DEVICE_ATTR_RO(vendor);
527 static ssize_t device_show(struct device *dev,
528 struct device_attribute *dev_attr,
531 struct hv_device *hv_dev = device_to_hv_device(dev);
532 return sprintf(buf, "0x%x\n", hv_dev->device_id);
534 static DEVICE_ATTR_RO(device);
536 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
537 static struct attribute *vmbus_dev_attrs[] = {
539 &dev_attr_state.attr,
540 &dev_attr_monitor_id.attr,
541 &dev_attr_class_id.attr,
542 &dev_attr_device_id.attr,
543 &dev_attr_modalias.attr,
544 &dev_attr_server_monitor_pending.attr,
545 &dev_attr_client_monitor_pending.attr,
546 &dev_attr_server_monitor_latency.attr,
547 &dev_attr_client_monitor_latency.attr,
548 &dev_attr_server_monitor_conn_id.attr,
549 &dev_attr_client_monitor_conn_id.attr,
550 &dev_attr_out_intr_mask.attr,
551 &dev_attr_out_read_index.attr,
552 &dev_attr_out_write_index.attr,
553 &dev_attr_out_read_bytes_avail.attr,
554 &dev_attr_out_write_bytes_avail.attr,
555 &dev_attr_in_intr_mask.attr,
556 &dev_attr_in_read_index.attr,
557 &dev_attr_in_write_index.attr,
558 &dev_attr_in_read_bytes_avail.attr,
559 &dev_attr_in_write_bytes_avail.attr,
560 &dev_attr_channel_vp_mapping.attr,
561 &dev_attr_vendor.attr,
562 &dev_attr_device.attr,
565 ATTRIBUTE_GROUPS(vmbus_dev);
568 * vmbus_uevent - add uevent for our device
570 * This routine is invoked when a device is added or removed on the vmbus to
571 * generate a uevent to udev in the userspace. The udev will then look at its
572 * rule and the uevent generated here to load the appropriate driver
574 * The alias string will be of the form vmbus:guid where guid is the string
575 * representation of the device guid (each byte of the guid will be
576 * represented with two hex characters.
578 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
580 struct hv_device *dev = device_to_hv_device(device);
582 char alias_name[VMBUS_ALIAS_LEN + 1];
584 print_alias_name(dev, alias_name);
585 ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
589 static const uuid_le null_guid;
591 static inline bool is_null_guid(const uuid_le *guid)
593 if (uuid_le_cmp(*guid, null_guid))
599 * Return a matching hv_vmbus_device_id pointer.
600 * If there is no match, return NULL.
602 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
605 const struct hv_vmbus_device_id *id = NULL;
606 struct vmbus_dynid *dynid;
608 /* Look at the dynamic ids first, before the static ones */
609 spin_lock(&drv->dynids.lock);
610 list_for_each_entry(dynid, &drv->dynids.list, node) {
611 if (!uuid_le_cmp(dynid->id.guid, *guid)) {
616 spin_unlock(&drv->dynids.lock);
623 return NULL; /* empty device table */
625 for (; !is_null_guid(&id->guid); id++)
626 if (!uuid_le_cmp(id->guid, *guid))
632 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
633 static int vmbus_add_dynid(struct hv_driver *drv, uuid_le *guid)
635 struct vmbus_dynid *dynid;
637 dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
641 dynid->id.guid = *guid;
643 spin_lock(&drv->dynids.lock);
644 list_add_tail(&dynid->node, &drv->dynids.list);
645 spin_unlock(&drv->dynids.lock);
647 return driver_attach(&drv->driver);
650 static void vmbus_free_dynids(struct hv_driver *drv)
652 struct vmbus_dynid *dynid, *n;
654 spin_lock(&drv->dynids.lock);
655 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
656 list_del(&dynid->node);
659 spin_unlock(&drv->dynids.lock);
663 * store_new_id - sysfs frontend to vmbus_add_dynid()
665 * Allow GUIDs to be added to an existing driver via sysfs.
667 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
670 struct hv_driver *drv = drv_to_hv_drv(driver);
674 retval = uuid_le_to_bin(buf, &guid);
678 if (hv_vmbus_get_id(drv, &guid))
681 retval = vmbus_add_dynid(drv, &guid);
686 static DRIVER_ATTR_WO(new_id);
689 * store_remove_id - remove a PCI device ID from this driver
691 * Removes a dynamic pci device ID to this driver.
693 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
696 struct hv_driver *drv = drv_to_hv_drv(driver);
697 struct vmbus_dynid *dynid, *n;
701 retval = uuid_le_to_bin(buf, &guid);
706 spin_lock(&drv->dynids.lock);
707 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
708 struct hv_vmbus_device_id *id = &dynid->id;
710 if (!uuid_le_cmp(id->guid, guid)) {
711 list_del(&dynid->node);
717 spin_unlock(&drv->dynids.lock);
721 static DRIVER_ATTR_WO(remove_id);
723 static struct attribute *vmbus_drv_attrs[] = {
724 &driver_attr_new_id.attr,
725 &driver_attr_remove_id.attr,
728 ATTRIBUTE_GROUPS(vmbus_drv);
732 * vmbus_match - Attempt to match the specified device to the specified driver
734 static int vmbus_match(struct device *device, struct device_driver *driver)
736 struct hv_driver *drv = drv_to_hv_drv(driver);
737 struct hv_device *hv_dev = device_to_hv_device(device);
739 /* The hv_sock driver handles all hv_sock offers. */
740 if (is_hvsock_channel(hv_dev->channel))
743 if (hv_vmbus_get_id(drv, &hv_dev->dev_type))
750 * vmbus_probe - Add the new vmbus's child device
752 static int vmbus_probe(struct device *child_device)
755 struct hv_driver *drv =
756 drv_to_hv_drv(child_device->driver);
757 struct hv_device *dev = device_to_hv_device(child_device);
758 const struct hv_vmbus_device_id *dev_id;
760 dev_id = hv_vmbus_get_id(drv, &dev->dev_type);
762 ret = drv->probe(dev, dev_id);
764 pr_err("probe failed for device %s (%d)\n",
765 dev_name(child_device), ret);
768 pr_err("probe not set for driver %s\n",
769 dev_name(child_device));
776 * vmbus_remove - Remove a vmbus device
778 static int vmbus_remove(struct device *child_device)
780 struct hv_driver *drv;
781 struct hv_device *dev = device_to_hv_device(child_device);
783 if (child_device->driver) {
784 drv = drv_to_hv_drv(child_device->driver);
794 * vmbus_shutdown - Shutdown a vmbus device
796 static void vmbus_shutdown(struct device *child_device)
798 struct hv_driver *drv;
799 struct hv_device *dev = device_to_hv_device(child_device);
802 /* The device may not be attached yet */
803 if (!child_device->driver)
806 drv = drv_to_hv_drv(child_device->driver);
814 * vmbus_device_release - Final callback release of the vmbus child device
816 static void vmbus_device_release(struct device *device)
818 struct hv_device *hv_dev = device_to_hv_device(device);
819 struct vmbus_channel *channel = hv_dev->channel;
821 mutex_lock(&vmbus_connection.channel_mutex);
822 hv_process_channel_removal(channel->offermsg.child_relid);
823 mutex_unlock(&vmbus_connection.channel_mutex);
828 /* The one and only one */
829 static struct bus_type hv_bus = {
831 .match = vmbus_match,
832 .shutdown = vmbus_shutdown,
833 .remove = vmbus_remove,
834 .probe = vmbus_probe,
835 .uevent = vmbus_uevent,
836 .dev_groups = vmbus_dev_groups,
837 .drv_groups = vmbus_drv_groups,
840 struct onmessage_work_context {
841 struct work_struct work;
842 struct hv_message msg;
845 static void vmbus_onmessage_work(struct work_struct *work)
847 struct onmessage_work_context *ctx;
849 /* Do not process messages if we're in DISCONNECTED state */
850 if (vmbus_connection.conn_state == DISCONNECTED)
853 ctx = container_of(work, struct onmessage_work_context,
855 vmbus_onmessage(&ctx->msg);
859 static void hv_process_timer_expiration(struct hv_message *msg,
860 struct hv_per_cpu_context *hv_cpu)
862 struct clock_event_device *dev = hv_cpu->clk_evt;
864 if (dev->event_handler)
865 dev->event_handler(dev);
867 vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
870 void vmbus_on_msg_dpc(unsigned long data)
872 struct hv_per_cpu_context *hv_cpu = (void *)data;
873 void *page_addr = hv_cpu->synic_message_page;
874 struct hv_message *msg = (struct hv_message *)page_addr +
876 struct vmbus_channel_message_header *hdr;
877 const struct vmbus_channel_message_table_entry *entry;
878 struct onmessage_work_context *ctx;
879 u32 message_type = msg->header.message_type;
881 if (message_type == HVMSG_NONE)
885 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
887 if (hdr->msgtype >= CHANNELMSG_COUNT) {
888 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
892 entry = &channel_message_table[hdr->msgtype];
893 if (entry->handler_type == VMHT_BLOCKING) {
894 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
898 INIT_WORK(&ctx->work, vmbus_onmessage_work);
899 memcpy(&ctx->msg, msg, sizeof(*msg));
902 * The host can generate a rescind message while we
903 * may still be handling the original offer. We deal with
904 * this condition by ensuring the processing is done on the
907 switch (hdr->msgtype) {
908 case CHANNELMSG_RESCIND_CHANNELOFFER:
910 * If we are handling the rescind message;
911 * schedule the work on the global work queue.
913 schedule_work_on(vmbus_connection.connect_cpu,
917 case CHANNELMSG_OFFERCHANNEL:
918 atomic_inc(&vmbus_connection.offer_in_progress);
919 queue_work_on(vmbus_connection.connect_cpu,
920 vmbus_connection.work_queue,
925 queue_work(vmbus_connection.work_queue, &ctx->work);
928 entry->message_handler(hdr);
931 vmbus_signal_eom(msg, message_type);
936 * Direct callback for channels using other deferred processing
938 static void vmbus_channel_isr(struct vmbus_channel *channel)
940 void (*callback_fn)(void *);
942 callback_fn = READ_ONCE(channel->onchannel_callback);
943 if (likely(callback_fn != NULL))
944 (*callback_fn)(channel->channel_callback_context);
948 * Schedule all channels with events pending
950 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
952 unsigned long *recv_int_page;
955 if (vmbus_proto_version < VERSION_WIN8) {
956 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
957 recv_int_page = vmbus_connection.recv_int_page;
960 * When the host is win8 and beyond, the event page
961 * can be directly checked to get the id of the channel
962 * that has the interrupt pending.
964 void *page_addr = hv_cpu->synic_event_page;
965 union hv_synic_event_flags *event
966 = (union hv_synic_event_flags *)page_addr +
969 maxbits = HV_EVENT_FLAGS_COUNT;
970 recv_int_page = event->flags;
973 if (unlikely(!recv_int_page))
976 for_each_set_bit(relid, recv_int_page, maxbits) {
977 struct vmbus_channel *channel;
979 if (!sync_test_and_clear_bit(relid, recv_int_page))
982 /* Special case - vmbus channel protocol msg */
988 /* Find channel based on relid */
989 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
990 if (channel->offermsg.child_relid != relid)
993 if (channel->rescind)
996 switch (channel->callback_mode) {
998 vmbus_channel_isr(channel);
1001 case HV_CALL_BATCHED:
1002 hv_begin_read(&channel->inbound);
1004 case HV_CALL_DIRECT:
1005 tasklet_schedule(&channel->callback_event);
1013 static void vmbus_isr(void)
1015 struct hv_per_cpu_context *hv_cpu
1016 = this_cpu_ptr(hv_context.cpu_context);
1017 void *page_addr = hv_cpu->synic_event_page;
1018 struct hv_message *msg;
1019 union hv_synic_event_flags *event;
1020 bool handled = false;
1022 if (unlikely(page_addr == NULL))
1025 event = (union hv_synic_event_flags *)page_addr +
1028 * Check for events before checking for messages. This is the order
1029 * in which events and messages are checked in Windows guests on
1030 * Hyper-V, and the Windows team suggested we do the same.
1033 if ((vmbus_proto_version == VERSION_WS2008) ||
1034 (vmbus_proto_version == VERSION_WIN7)) {
1036 /* Since we are a child, we only need to check bit 0 */
1037 if (sync_test_and_clear_bit(0, event->flags))
1041 * Our host is win8 or above. The signaling mechanism
1042 * has changed and we can directly look at the event page.
1043 * If bit n is set then we have an interrup on the channel
1050 vmbus_chan_sched(hv_cpu);
1052 page_addr = hv_cpu->synic_message_page;
1053 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1055 /* Check if there are actual msgs to be processed */
1056 if (msg->header.message_type != HVMSG_NONE) {
1057 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
1058 hv_process_timer_expiration(msg, hv_cpu);
1060 tasklet_schedule(&hv_cpu->msg_dpc);
1063 add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1068 * vmbus_bus_init -Main vmbus driver initialization routine.
1071 * - initialize the vmbus driver context
1072 * - invoke the vmbus hv main init routine
1073 * - retrieve the channel offers
1075 static int vmbus_bus_init(void)
1079 /* Hypervisor initialization...setup hypercall page..etc */
1082 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1086 ret = bus_register(&hv_bus);
1090 hv_setup_vmbus_irq(vmbus_isr);
1092 ret = hv_synic_alloc();
1096 * Initialize the per-cpu interrupt state and
1097 * connect to the host.
1099 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv:online",
1100 hv_synic_init, hv_synic_cleanup);
1103 hyperv_cpuhp_online = ret;
1105 ret = vmbus_connect();
1110 * Only register if the crash MSRs are available
1112 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1113 register_die_notifier(&hyperv_die_block);
1114 atomic_notifier_chain_register(&panic_notifier_list,
1115 &hyperv_panic_block);
1118 vmbus_request_offers();
1123 cpuhp_remove_state(hyperv_cpuhp_online);
1126 hv_remove_vmbus_irq();
1128 bus_unregister(&hv_bus);
1134 * __vmbus_child_driver_register() - Register a vmbus's driver
1135 * @hv_driver: Pointer to driver structure you want to register
1136 * @owner: owner module of the drv
1137 * @mod_name: module name string
1139 * Registers the given driver with Linux through the 'driver_register()' call
1140 * and sets up the hyper-v vmbus handling for this driver.
1141 * It will return the state of the 'driver_register()' call.
1144 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1148 pr_info("registering driver %s\n", hv_driver->name);
1150 ret = vmbus_exists();
1154 hv_driver->driver.name = hv_driver->name;
1155 hv_driver->driver.owner = owner;
1156 hv_driver->driver.mod_name = mod_name;
1157 hv_driver->driver.bus = &hv_bus;
1159 spin_lock_init(&hv_driver->dynids.lock);
1160 INIT_LIST_HEAD(&hv_driver->dynids.list);
1162 ret = driver_register(&hv_driver->driver);
1166 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1169 * vmbus_driver_unregister() - Unregister a vmbus's driver
1170 * @hv_driver: Pointer to driver structure you want to
1173 * Un-register the given driver that was previous registered with a call to
1174 * vmbus_driver_register()
1176 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1178 pr_info("unregistering driver %s\n", hv_driver->name);
1180 if (!vmbus_exists()) {
1181 driver_unregister(&hv_driver->driver);
1182 vmbus_free_dynids(hv_driver);
1185 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1188 * vmbus_device_create - Creates and registers a new child device
1191 struct hv_device *vmbus_device_create(const uuid_le *type,
1192 const uuid_le *instance,
1193 struct vmbus_channel *channel)
1195 struct hv_device *child_device_obj;
1197 child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1198 if (!child_device_obj) {
1199 pr_err("Unable to allocate device object for child device\n");
1203 child_device_obj->channel = channel;
1204 memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
1205 memcpy(&child_device_obj->dev_instance, instance,
1207 child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1210 return child_device_obj;
1214 * vmbus_device_register - Register the child device
1216 int vmbus_device_register(struct hv_device *child_device_obj)
1220 dev_set_name(&child_device_obj->device, "%pUl",
1221 child_device_obj->channel->offermsg.offer.if_instance.b);
1223 child_device_obj->device.bus = &hv_bus;
1224 child_device_obj->device.parent = &hv_acpi_dev->dev;
1225 child_device_obj->device.release = vmbus_device_release;
1228 * Register with the LDM. This will kick off the driver/device
1229 * binding...which will eventually call vmbus_match() and vmbus_probe()
1231 ret = device_register(&child_device_obj->device);
1234 pr_err("Unable to register child device\n");
1236 pr_debug("child device %s registered\n",
1237 dev_name(&child_device_obj->device));
1243 * vmbus_device_unregister - Remove the specified child device
1246 void vmbus_device_unregister(struct hv_device *device_obj)
1248 pr_debug("child device %s unregistered\n",
1249 dev_name(&device_obj->device));
1252 * Kick off the process of unregistering the device.
1253 * This will call vmbus_remove() and eventually vmbus_device_release()
1255 device_unregister(&device_obj->device);
1260 * VMBUS is an acpi enumerated device. Get the information we
1263 #define VTPM_BASE_ADDRESS 0xfed40000
1264 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1266 resource_size_t start = 0;
1267 resource_size_t end = 0;
1268 struct resource *new_res;
1269 struct resource **old_res = &hyperv_mmio;
1270 struct resource **prev_res = NULL;
1272 switch (res->type) {
1275 * "Address" descriptors are for bus windows. Ignore
1276 * "memory" descriptors, which are for registers on
1279 case ACPI_RESOURCE_TYPE_ADDRESS32:
1280 start = res->data.address32.address.minimum;
1281 end = res->data.address32.address.maximum;
1284 case ACPI_RESOURCE_TYPE_ADDRESS64:
1285 start = res->data.address64.address.minimum;
1286 end = res->data.address64.address.maximum;
1290 /* Unused resource type */
1295 * Ignore ranges that are below 1MB, as they're not
1296 * necessary or useful here.
1301 new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1303 return AE_NO_MEMORY;
1305 /* If this range overlaps the virtual TPM, truncate it. */
1306 if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1307 end = VTPM_BASE_ADDRESS;
1309 new_res->name = "hyperv mmio";
1310 new_res->flags = IORESOURCE_MEM;
1311 new_res->start = start;
1315 * If two ranges are adjacent, merge them.
1323 if (((*old_res)->end + 1) == new_res->start) {
1324 (*old_res)->end = new_res->end;
1329 if ((*old_res)->start == new_res->end + 1) {
1330 (*old_res)->start = new_res->start;
1335 if ((*old_res)->start > new_res->end) {
1336 new_res->sibling = *old_res;
1338 (*prev_res)->sibling = new_res;
1344 old_res = &(*old_res)->sibling;
1351 static int vmbus_acpi_remove(struct acpi_device *device)
1353 struct resource *cur_res;
1354 struct resource *next_res;
1358 __release_region(hyperv_mmio, fb_mmio->start,
1359 resource_size(fb_mmio));
1363 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1364 next_res = cur_res->sibling;
1372 static void vmbus_reserve_fb(void)
1376 * Make a claim for the frame buffer in the resource tree under the
1377 * first node, which will be the one below 4GB. The length seems to
1378 * be underreported, particularly in a Generation 1 VM. So start out
1379 * reserving a larger area and make it smaller until it succeeds.
1382 if (screen_info.lfb_base) {
1383 if (efi_enabled(EFI_BOOT))
1384 size = max_t(__u32, screen_info.lfb_size, 0x800000);
1386 size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1388 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1389 fb_mmio = __request_region(hyperv_mmio,
1390 screen_info.lfb_base, size,
1397 * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1398 * @new: If successful, supplied a pointer to the
1399 * allocated MMIO space.
1400 * @device_obj: Identifies the caller
1401 * @min: Minimum guest physical address of the
1403 * @max: Maximum guest physical address
1404 * @size: Size of the range to be allocated
1405 * @align: Alignment of the range to be allocated
1406 * @fb_overlap_ok: Whether this allocation can be allowed
1407 * to overlap the video frame buffer.
1409 * This function walks the resources granted to VMBus by the
1410 * _CRS object in the ACPI namespace underneath the parent
1411 * "bridge" whether that's a root PCI bus in the Generation 1
1412 * case or a Module Device in the Generation 2 case. It then
1413 * attempts to allocate from the global MMIO pool in a way that
1414 * matches the constraints supplied in these parameters and by
1417 * Return: 0 on success, -errno on failure
1419 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1420 resource_size_t min, resource_size_t max,
1421 resource_size_t size, resource_size_t align,
1424 struct resource *iter, *shadow;
1425 resource_size_t range_min, range_max, start;
1426 const char *dev_n = dev_name(&device_obj->device);
1430 down(&hyperv_mmio_lock);
1433 * If overlaps with frame buffers are allowed, then first attempt to
1434 * make the allocation from within the reserved region. Because it
1435 * is already reserved, no shadow allocation is necessary.
1437 if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1438 !(max < fb_mmio->start)) {
1440 range_min = fb_mmio->start;
1441 range_max = fb_mmio->end;
1442 start = (range_min + align - 1) & ~(align - 1);
1443 for (; start + size - 1 <= range_max; start += align) {
1444 *new = request_mem_region_exclusive(start, size, dev_n);
1452 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1453 if ((iter->start >= max) || (iter->end <= min))
1456 range_min = iter->start;
1457 range_max = iter->end;
1458 start = (range_min + align - 1) & ~(align - 1);
1459 for (; start + size - 1 <= range_max; start += align) {
1460 shadow = __request_region(iter, start, size, NULL,
1465 *new = request_mem_region_exclusive(start, size, dev_n);
1467 shadow->name = (char *)*new;
1472 __release_region(iter, start, size);
1477 up(&hyperv_mmio_lock);
1480 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1483 * vmbus_free_mmio() - Free a memory-mapped I/O range.
1484 * @start: Base address of region to release.
1485 * @size: Size of the range to be allocated
1487 * This function releases anything requested by
1488 * vmbus_mmio_allocate().
1490 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
1492 struct resource *iter;
1494 down(&hyperv_mmio_lock);
1495 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1496 if ((iter->start >= start + size) || (iter->end <= start))
1499 __release_region(iter, start, size);
1501 release_mem_region(start, size);
1502 up(&hyperv_mmio_lock);
1505 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
1507 static int vmbus_acpi_add(struct acpi_device *device)
1510 int ret_val = -ENODEV;
1511 struct acpi_device *ancestor;
1513 hv_acpi_dev = device;
1515 result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1516 vmbus_walk_resources, NULL);
1518 if (ACPI_FAILURE(result))
1521 * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1522 * firmware) is the VMOD that has the mmio ranges. Get that.
1524 for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1525 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1526 vmbus_walk_resources, NULL);
1528 if (ACPI_FAILURE(result))
1538 complete(&probe_event);
1540 vmbus_acpi_remove(device);
1544 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1549 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1551 static struct acpi_driver vmbus_acpi_driver = {
1553 .ids = vmbus_acpi_device_ids,
1555 .add = vmbus_acpi_add,
1556 .remove = vmbus_acpi_remove,
1560 static void hv_kexec_handler(void)
1562 hv_synic_clockevents_cleanup();
1563 vmbus_initiate_unload(false);
1564 vmbus_connection.conn_state = DISCONNECTED;
1565 /* Make sure conn_state is set as hv_synic_cleanup checks for it */
1567 cpuhp_remove_state(hyperv_cpuhp_online);
1571 static void hv_crash_handler(struct pt_regs *regs)
1573 vmbus_initiate_unload(true);
1575 * In crash handler we can't schedule synic cleanup for all CPUs,
1576 * doing the cleanup for current CPU only. This should be sufficient
1579 vmbus_connection.conn_state = DISCONNECTED;
1580 hv_synic_cleanup(smp_processor_id());
1584 static int __init hv_acpi_init(void)
1588 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
1591 init_completion(&probe_event);
1594 * Get ACPI resources first.
1596 ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1601 t = wait_for_completion_timeout(&probe_event, 5*HZ);
1607 ret = vmbus_bus_init();
1611 hv_setup_kexec_handler(hv_kexec_handler);
1612 hv_setup_crash_handler(hv_crash_handler);
1617 acpi_bus_unregister_driver(&vmbus_acpi_driver);
1622 static void __exit vmbus_exit(void)
1626 hv_remove_kexec_handler();
1627 hv_remove_crash_handler();
1628 vmbus_connection.conn_state = DISCONNECTED;
1629 hv_synic_clockevents_cleanup();
1631 hv_remove_vmbus_irq();
1632 for_each_online_cpu(cpu) {
1633 struct hv_per_cpu_context *hv_cpu
1634 = per_cpu_ptr(hv_context.cpu_context, cpu);
1636 tasklet_kill(&hv_cpu->msg_dpc);
1638 vmbus_free_channels();
1640 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1641 unregister_die_notifier(&hyperv_die_block);
1642 atomic_notifier_chain_unregister(&panic_notifier_list,
1643 &hyperv_panic_block);
1645 bus_unregister(&hv_bus);
1647 cpuhp_remove_state(hyperv_cpuhp_online);
1649 acpi_bus_unregister_driver(&vmbus_acpi_driver);
1653 MODULE_LICENSE("GPL");
1655 subsys_initcall(hv_acpi_init);
1656 module_exit(vmbus_exit);