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;
301 if (!hv_dev->channel)
303 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
304 return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
306 static DEVICE_ATTR_RO(out_intr_mask);
308 static ssize_t out_read_index_show(struct device *dev,
309 struct device_attribute *dev_attr, char *buf)
311 struct hv_device *hv_dev = device_to_hv_device(dev);
312 struct hv_ring_buffer_debug_info outbound;
314 if (!hv_dev->channel)
316 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
317 return sprintf(buf, "%d\n", outbound.current_read_index);
319 static DEVICE_ATTR_RO(out_read_index);
321 static ssize_t out_write_index_show(struct device *dev,
322 struct device_attribute *dev_attr,
325 struct hv_device *hv_dev = device_to_hv_device(dev);
326 struct hv_ring_buffer_debug_info outbound;
328 if (!hv_dev->channel)
330 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
331 return sprintf(buf, "%d\n", outbound.current_write_index);
333 static DEVICE_ATTR_RO(out_write_index);
335 static ssize_t out_read_bytes_avail_show(struct device *dev,
336 struct device_attribute *dev_attr,
339 struct hv_device *hv_dev = device_to_hv_device(dev);
340 struct hv_ring_buffer_debug_info outbound;
342 if (!hv_dev->channel)
344 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
345 return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
347 static DEVICE_ATTR_RO(out_read_bytes_avail);
349 static ssize_t out_write_bytes_avail_show(struct device *dev,
350 struct device_attribute *dev_attr,
353 struct hv_device *hv_dev = device_to_hv_device(dev);
354 struct hv_ring_buffer_debug_info outbound;
356 if (!hv_dev->channel)
358 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
359 return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
361 static DEVICE_ATTR_RO(out_write_bytes_avail);
363 static ssize_t in_intr_mask_show(struct device *dev,
364 struct device_attribute *dev_attr, char *buf)
366 struct hv_device *hv_dev = device_to_hv_device(dev);
367 struct hv_ring_buffer_debug_info inbound;
369 if (!hv_dev->channel)
371 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
372 return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
374 static DEVICE_ATTR_RO(in_intr_mask);
376 static ssize_t in_read_index_show(struct device *dev,
377 struct device_attribute *dev_attr, char *buf)
379 struct hv_device *hv_dev = device_to_hv_device(dev);
380 struct hv_ring_buffer_debug_info inbound;
382 if (!hv_dev->channel)
384 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
385 return sprintf(buf, "%d\n", inbound.current_read_index);
387 static DEVICE_ATTR_RO(in_read_index);
389 static ssize_t in_write_index_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;
395 if (!hv_dev->channel)
397 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
398 return sprintf(buf, "%d\n", inbound.current_write_index);
400 static DEVICE_ATTR_RO(in_write_index);
402 static ssize_t in_read_bytes_avail_show(struct device *dev,
403 struct device_attribute *dev_attr,
406 struct hv_device *hv_dev = device_to_hv_device(dev);
407 struct hv_ring_buffer_debug_info inbound;
409 if (!hv_dev->channel)
411 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
412 return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
414 static DEVICE_ATTR_RO(in_read_bytes_avail);
416 static ssize_t in_write_bytes_avail_show(struct device *dev,
417 struct device_attribute *dev_attr,
420 struct hv_device *hv_dev = device_to_hv_device(dev);
421 struct hv_ring_buffer_debug_info inbound;
423 if (!hv_dev->channel)
425 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
426 return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
428 static DEVICE_ATTR_RO(in_write_bytes_avail);
430 static ssize_t channel_vp_mapping_show(struct device *dev,
431 struct device_attribute *dev_attr,
434 struct hv_device *hv_dev = device_to_hv_device(dev);
435 struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
437 int buf_size = PAGE_SIZE, n_written, tot_written;
438 struct list_head *cur;
443 tot_written = snprintf(buf, buf_size, "%u:%u\n",
444 channel->offermsg.child_relid, channel->target_cpu);
446 spin_lock_irqsave(&channel->lock, flags);
448 list_for_each(cur, &channel->sc_list) {
449 if (tot_written >= buf_size - 1)
452 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
453 n_written = scnprintf(buf + tot_written,
454 buf_size - tot_written,
456 cur_sc->offermsg.child_relid,
458 tot_written += n_written;
461 spin_unlock_irqrestore(&channel->lock, flags);
465 static DEVICE_ATTR_RO(channel_vp_mapping);
467 static ssize_t vendor_show(struct device *dev,
468 struct device_attribute *dev_attr,
471 struct hv_device *hv_dev = device_to_hv_device(dev);
472 return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
474 static DEVICE_ATTR_RO(vendor);
476 static ssize_t device_show(struct device *dev,
477 struct device_attribute *dev_attr,
480 struct hv_device *hv_dev = device_to_hv_device(dev);
481 return sprintf(buf, "0x%x\n", hv_dev->device_id);
483 static DEVICE_ATTR_RO(device);
485 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
486 static struct attribute *vmbus_dev_attrs[] = {
488 &dev_attr_state.attr,
489 &dev_attr_monitor_id.attr,
490 &dev_attr_class_id.attr,
491 &dev_attr_device_id.attr,
492 &dev_attr_modalias.attr,
493 &dev_attr_server_monitor_pending.attr,
494 &dev_attr_client_monitor_pending.attr,
495 &dev_attr_server_monitor_latency.attr,
496 &dev_attr_client_monitor_latency.attr,
497 &dev_attr_server_monitor_conn_id.attr,
498 &dev_attr_client_monitor_conn_id.attr,
499 &dev_attr_out_intr_mask.attr,
500 &dev_attr_out_read_index.attr,
501 &dev_attr_out_write_index.attr,
502 &dev_attr_out_read_bytes_avail.attr,
503 &dev_attr_out_write_bytes_avail.attr,
504 &dev_attr_in_intr_mask.attr,
505 &dev_attr_in_read_index.attr,
506 &dev_attr_in_write_index.attr,
507 &dev_attr_in_read_bytes_avail.attr,
508 &dev_attr_in_write_bytes_avail.attr,
509 &dev_attr_channel_vp_mapping.attr,
510 &dev_attr_vendor.attr,
511 &dev_attr_device.attr,
514 ATTRIBUTE_GROUPS(vmbus_dev);
517 * vmbus_uevent - add uevent for our device
519 * This routine is invoked when a device is added or removed on the vmbus to
520 * generate a uevent to udev in the userspace. The udev will then look at its
521 * rule and the uevent generated here to load the appropriate driver
523 * The alias string will be of the form vmbus:guid where guid is the string
524 * representation of the device guid (each byte of the guid will be
525 * represented with two hex characters.
527 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
529 struct hv_device *dev = device_to_hv_device(device);
531 char alias_name[VMBUS_ALIAS_LEN + 1];
533 print_alias_name(dev, alias_name);
534 ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
538 static const uuid_le null_guid;
540 static inline bool is_null_guid(const uuid_le *guid)
542 if (uuid_le_cmp(*guid, null_guid))
548 * Return a matching hv_vmbus_device_id pointer.
549 * If there is no match, return NULL.
551 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
554 const struct hv_vmbus_device_id *id = NULL;
555 struct vmbus_dynid *dynid;
557 /* Look at the dynamic ids first, before the static ones */
558 spin_lock(&drv->dynids.lock);
559 list_for_each_entry(dynid, &drv->dynids.list, node) {
560 if (!uuid_le_cmp(dynid->id.guid, *guid)) {
565 spin_unlock(&drv->dynids.lock);
572 return NULL; /* empty device table */
574 for (; !is_null_guid(&id->guid); id++)
575 if (!uuid_le_cmp(id->guid, *guid))
581 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
582 static int vmbus_add_dynid(struct hv_driver *drv, uuid_le *guid)
584 struct vmbus_dynid *dynid;
586 dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
590 dynid->id.guid = *guid;
592 spin_lock(&drv->dynids.lock);
593 list_add_tail(&dynid->node, &drv->dynids.list);
594 spin_unlock(&drv->dynids.lock);
596 return driver_attach(&drv->driver);
599 static void vmbus_free_dynids(struct hv_driver *drv)
601 struct vmbus_dynid *dynid, *n;
603 spin_lock(&drv->dynids.lock);
604 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
605 list_del(&dynid->node);
608 spin_unlock(&drv->dynids.lock);
611 /* Parse string of form: 1b4e28ba-2fa1-11d2-883f-b9a761bde3f */
612 static int get_uuid_le(const char *str, uuid_le *uu)
617 if (strlen(str) < 37)
620 for (i = 0; i < 36; i++) {
622 case 8: case 13: case 18: case 23:
627 if (!isxdigit(str[i]))
632 /* unparse little endian output byte order */
634 "%2x%2x%2x%2x-%2x%2x-%2x%2x-%2x%2x-%2x%2x%2x%2x%2x%2x",
635 &b[3], &b[2], &b[1], &b[0],
636 &b[5], &b[4], &b[7], &b[6], &b[8], &b[9],
637 &b[10], &b[11], &b[12], &b[13], &b[14], &b[15]) != 16)
640 for (i = 0; i < 16; i++)
646 * store_new_id - sysfs frontend to vmbus_add_dynid()
648 * Allow GUIDs to be added to an existing driver via sysfs.
650 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
653 struct hv_driver *drv = drv_to_hv_drv(driver);
654 uuid_le guid = NULL_UUID_LE;
657 if (get_uuid_le(buf, &guid) != 0)
660 if (hv_vmbus_get_id(drv, &guid))
663 retval = vmbus_add_dynid(drv, &guid);
668 static DRIVER_ATTR_WO(new_id);
671 * store_remove_id - remove a PCI device ID from this driver
673 * Removes a dynamic pci device ID to this driver.
675 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
678 struct hv_driver *drv = drv_to_hv_drv(driver);
679 struct vmbus_dynid *dynid, *n;
680 uuid_le guid = NULL_UUID_LE;
681 size_t retval = -ENODEV;
683 if (get_uuid_le(buf, &guid))
686 spin_lock(&drv->dynids.lock);
687 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
688 struct hv_vmbus_device_id *id = &dynid->id;
690 if (!uuid_le_cmp(id->guid, guid)) {
691 list_del(&dynid->node);
697 spin_unlock(&drv->dynids.lock);
701 static DRIVER_ATTR_WO(remove_id);
703 static struct attribute *vmbus_drv_attrs[] = {
704 &driver_attr_new_id.attr,
705 &driver_attr_remove_id.attr,
708 ATTRIBUTE_GROUPS(vmbus_drv);
712 * vmbus_match - Attempt to match the specified device to the specified driver
714 static int vmbus_match(struct device *device, struct device_driver *driver)
716 struct hv_driver *drv = drv_to_hv_drv(driver);
717 struct hv_device *hv_dev = device_to_hv_device(device);
719 /* The hv_sock driver handles all hv_sock offers. */
720 if (is_hvsock_channel(hv_dev->channel))
723 if (hv_vmbus_get_id(drv, &hv_dev->dev_type))
730 * vmbus_probe - Add the new vmbus's child device
732 static int vmbus_probe(struct device *child_device)
735 struct hv_driver *drv =
736 drv_to_hv_drv(child_device->driver);
737 struct hv_device *dev = device_to_hv_device(child_device);
738 const struct hv_vmbus_device_id *dev_id;
740 dev_id = hv_vmbus_get_id(drv, &dev->dev_type);
742 ret = drv->probe(dev, dev_id);
744 pr_err("probe failed for device %s (%d)\n",
745 dev_name(child_device), ret);
748 pr_err("probe not set for driver %s\n",
749 dev_name(child_device));
756 * vmbus_remove - Remove a vmbus device
758 static int vmbus_remove(struct device *child_device)
760 struct hv_driver *drv;
761 struct hv_device *dev = device_to_hv_device(child_device);
763 if (child_device->driver) {
764 drv = drv_to_hv_drv(child_device->driver);
774 * vmbus_shutdown - Shutdown a vmbus device
776 static void vmbus_shutdown(struct device *child_device)
778 struct hv_driver *drv;
779 struct hv_device *dev = device_to_hv_device(child_device);
782 /* The device may not be attached yet */
783 if (!child_device->driver)
786 drv = drv_to_hv_drv(child_device->driver);
796 * vmbus_device_release - Final callback release of the vmbus child device
798 static void vmbus_device_release(struct device *device)
800 struct hv_device *hv_dev = device_to_hv_device(device);
801 struct vmbus_channel *channel = hv_dev->channel;
803 hv_process_channel_removal(channel,
804 channel->offermsg.child_relid);
809 /* The one and only one */
810 static struct bus_type hv_bus = {
812 .match = vmbus_match,
813 .shutdown = vmbus_shutdown,
814 .remove = vmbus_remove,
815 .probe = vmbus_probe,
816 .uevent = vmbus_uevent,
817 .dev_groups = vmbus_dev_groups,
818 .drv_groups = vmbus_drv_groups,
821 struct onmessage_work_context {
822 struct work_struct work;
823 struct hv_message msg;
826 static void vmbus_onmessage_work(struct work_struct *work)
828 struct onmessage_work_context *ctx;
830 /* Do not process messages if we're in DISCONNECTED state */
831 if (vmbus_connection.conn_state == DISCONNECTED)
834 ctx = container_of(work, struct onmessage_work_context,
836 vmbus_onmessage(&ctx->msg);
840 static void hv_process_timer_expiration(struct hv_message *msg,
841 struct hv_per_cpu_context *hv_cpu)
843 struct clock_event_device *dev = hv_cpu->clk_evt;
845 if (dev->event_handler)
846 dev->event_handler(dev);
848 vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
851 void vmbus_on_msg_dpc(unsigned long data)
853 struct hv_per_cpu_context *hv_cpu = (void *)data;
854 void *page_addr = hv_cpu->synic_message_page;
855 struct hv_message *msg = (struct hv_message *)page_addr +
857 struct vmbus_channel_message_header *hdr;
858 struct vmbus_channel_message_table_entry *entry;
859 struct onmessage_work_context *ctx;
860 u32 message_type = msg->header.message_type;
862 if (message_type == HVMSG_NONE)
866 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
868 if (hdr->msgtype >= CHANNELMSG_COUNT) {
869 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
873 entry = &channel_message_table[hdr->msgtype];
874 if (entry->handler_type == VMHT_BLOCKING) {
875 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
879 INIT_WORK(&ctx->work, vmbus_onmessage_work);
880 memcpy(&ctx->msg, msg, sizeof(*msg));
882 queue_work(vmbus_connection.work_queue, &ctx->work);
884 entry->message_handler(hdr);
887 vmbus_signal_eom(msg, message_type);
892 * Direct callback for channels using other deferred processing
894 static void vmbus_channel_isr(struct vmbus_channel *channel)
896 void (*callback_fn)(void *);
898 callback_fn = READ_ONCE(channel->onchannel_callback);
899 if (likely(callback_fn != NULL))
900 (*callback_fn)(channel->channel_callback_context);
904 * Schedule all channels with events pending
906 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
908 unsigned long *recv_int_page;
911 if (vmbus_proto_version < VERSION_WIN8) {
912 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
913 recv_int_page = vmbus_connection.recv_int_page;
916 * When the host is win8 and beyond, the event page
917 * can be directly checked to get the id of the channel
918 * that has the interrupt pending.
920 void *page_addr = hv_cpu->synic_event_page;
921 union hv_synic_event_flags *event
922 = (union hv_synic_event_flags *)page_addr +
925 maxbits = HV_EVENT_FLAGS_COUNT;
926 recv_int_page = event->flags;
929 if (unlikely(!recv_int_page))
932 for_each_set_bit(relid, recv_int_page, maxbits) {
933 struct vmbus_channel *channel;
935 if (!sync_test_and_clear_bit(relid, recv_int_page))
938 /* Special case - vmbus channel protocol msg */
944 /* Find channel based on relid */
945 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
946 if (channel->offermsg.child_relid != relid)
949 switch (channel->callback_mode) {
951 vmbus_channel_isr(channel);
954 case HV_CALL_BATCHED:
955 hv_begin_read(&channel->inbound);
958 tasklet_schedule(&channel->callback_event);
966 static void vmbus_isr(void)
968 struct hv_per_cpu_context *hv_cpu
969 = this_cpu_ptr(hv_context.cpu_context);
970 void *page_addr = hv_cpu->synic_event_page;
971 struct hv_message *msg;
972 union hv_synic_event_flags *event;
973 bool handled = false;
975 if (unlikely(page_addr == NULL))
978 event = (union hv_synic_event_flags *)page_addr +
981 * Check for events before checking for messages. This is the order
982 * in which events and messages are checked in Windows guests on
983 * Hyper-V, and the Windows team suggested we do the same.
986 if ((vmbus_proto_version == VERSION_WS2008) ||
987 (vmbus_proto_version == VERSION_WIN7)) {
989 /* Since we are a child, we only need to check bit 0 */
990 if (sync_test_and_clear_bit(0, event->flags))
994 * Our host is win8 or above. The signaling mechanism
995 * has changed and we can directly look at the event page.
996 * If bit n is set then we have an interrup on the channel
1003 vmbus_chan_sched(hv_cpu);
1005 page_addr = hv_cpu->synic_message_page;
1006 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1008 /* Check if there are actual msgs to be processed */
1009 if (msg->header.message_type != HVMSG_NONE) {
1010 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
1011 hv_process_timer_expiration(msg, hv_cpu);
1013 tasklet_schedule(&hv_cpu->msg_dpc);
1016 add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1021 * vmbus_bus_init -Main vmbus driver initialization routine.
1024 * - initialize the vmbus driver context
1025 * - invoke the vmbus hv main init routine
1026 * - retrieve the channel offers
1028 static int vmbus_bus_init(void)
1032 /* Hypervisor initialization...setup hypercall page..etc */
1035 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1039 ret = bus_register(&hv_bus);
1043 hv_setup_vmbus_irq(vmbus_isr);
1045 ret = hv_synic_alloc();
1049 * Initialize the per-cpu interrupt state and
1050 * connect to the host.
1052 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv:online",
1053 hv_synic_init, hv_synic_cleanup);
1056 hyperv_cpuhp_online = ret;
1058 ret = vmbus_connect();
1063 * Only register if the crash MSRs are available
1065 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1066 register_die_notifier(&hyperv_die_block);
1067 atomic_notifier_chain_register(&panic_notifier_list,
1068 &hyperv_panic_block);
1071 vmbus_request_offers();
1076 cpuhp_remove_state(hyperv_cpuhp_online);
1079 hv_remove_vmbus_irq();
1081 bus_unregister(&hv_bus);
1087 * __vmbus_child_driver_register() - Register a vmbus's driver
1088 * @hv_driver: Pointer to driver structure you want to register
1089 * @owner: owner module of the drv
1090 * @mod_name: module name string
1092 * Registers the given driver with Linux through the 'driver_register()' call
1093 * and sets up the hyper-v vmbus handling for this driver.
1094 * It will return the state of the 'driver_register()' call.
1097 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1101 pr_info("registering driver %s\n", hv_driver->name);
1103 ret = vmbus_exists();
1107 hv_driver->driver.name = hv_driver->name;
1108 hv_driver->driver.owner = owner;
1109 hv_driver->driver.mod_name = mod_name;
1110 hv_driver->driver.bus = &hv_bus;
1112 spin_lock_init(&hv_driver->dynids.lock);
1113 INIT_LIST_HEAD(&hv_driver->dynids.list);
1115 ret = driver_register(&hv_driver->driver);
1119 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1122 * vmbus_driver_unregister() - Unregister a vmbus's driver
1123 * @hv_driver: Pointer to driver structure you want to
1126 * Un-register the given driver that was previous registered with a call to
1127 * vmbus_driver_register()
1129 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1131 pr_info("unregistering driver %s\n", hv_driver->name);
1133 if (!vmbus_exists()) {
1134 driver_unregister(&hv_driver->driver);
1135 vmbus_free_dynids(hv_driver);
1138 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1141 * vmbus_device_create - Creates and registers a new child device
1144 struct hv_device *vmbus_device_create(const uuid_le *type,
1145 const uuid_le *instance,
1146 struct vmbus_channel *channel)
1148 struct hv_device *child_device_obj;
1150 child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1151 if (!child_device_obj) {
1152 pr_err("Unable to allocate device object for child device\n");
1156 child_device_obj->channel = channel;
1157 memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
1158 memcpy(&child_device_obj->dev_instance, instance,
1160 child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1163 return child_device_obj;
1167 * vmbus_device_register - Register the child device
1169 int vmbus_device_register(struct hv_device *child_device_obj)
1173 dev_set_name(&child_device_obj->device, "%pUl",
1174 child_device_obj->channel->offermsg.offer.if_instance.b);
1176 child_device_obj->device.bus = &hv_bus;
1177 child_device_obj->device.parent = &hv_acpi_dev->dev;
1178 child_device_obj->device.release = vmbus_device_release;
1181 * Register with the LDM. This will kick off the driver/device
1182 * binding...which will eventually call vmbus_match() and vmbus_probe()
1184 ret = device_register(&child_device_obj->device);
1187 pr_err("Unable to register child device\n");
1189 pr_debug("child device %s registered\n",
1190 dev_name(&child_device_obj->device));
1196 * vmbus_device_unregister - Remove the specified child device
1199 void vmbus_device_unregister(struct hv_device *device_obj)
1201 pr_debug("child device %s unregistered\n",
1202 dev_name(&device_obj->device));
1205 * Kick off the process of unregistering the device.
1206 * This will call vmbus_remove() and eventually vmbus_device_release()
1208 device_unregister(&device_obj->device);
1213 * VMBUS is an acpi enumerated device. Get the information we
1216 #define VTPM_BASE_ADDRESS 0xfed40000
1217 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1219 resource_size_t start = 0;
1220 resource_size_t end = 0;
1221 struct resource *new_res;
1222 struct resource **old_res = &hyperv_mmio;
1223 struct resource **prev_res = NULL;
1225 switch (res->type) {
1228 * "Address" descriptors are for bus windows. Ignore
1229 * "memory" descriptors, which are for registers on
1232 case ACPI_RESOURCE_TYPE_ADDRESS32:
1233 start = res->data.address32.address.minimum;
1234 end = res->data.address32.address.maximum;
1237 case ACPI_RESOURCE_TYPE_ADDRESS64:
1238 start = res->data.address64.address.minimum;
1239 end = res->data.address64.address.maximum;
1243 /* Unused resource type */
1248 * Ignore ranges that are below 1MB, as they're not
1249 * necessary or useful here.
1254 new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1256 return AE_NO_MEMORY;
1258 /* If this range overlaps the virtual TPM, truncate it. */
1259 if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1260 end = VTPM_BASE_ADDRESS;
1262 new_res->name = "hyperv mmio";
1263 new_res->flags = IORESOURCE_MEM;
1264 new_res->start = start;
1268 * If two ranges are adjacent, merge them.
1276 if (((*old_res)->end + 1) == new_res->start) {
1277 (*old_res)->end = new_res->end;
1282 if ((*old_res)->start == new_res->end + 1) {
1283 (*old_res)->start = new_res->start;
1288 if ((*old_res)->start > new_res->end) {
1289 new_res->sibling = *old_res;
1291 (*prev_res)->sibling = new_res;
1297 old_res = &(*old_res)->sibling;
1304 static int vmbus_acpi_remove(struct acpi_device *device)
1306 struct resource *cur_res;
1307 struct resource *next_res;
1311 __release_region(hyperv_mmio, fb_mmio->start,
1312 resource_size(fb_mmio));
1316 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1317 next_res = cur_res->sibling;
1325 static void vmbus_reserve_fb(void)
1329 * Make a claim for the frame buffer in the resource tree under the
1330 * first node, which will be the one below 4GB. The length seems to
1331 * be underreported, particularly in a Generation 1 VM. So start out
1332 * reserving a larger area and make it smaller until it succeeds.
1335 if (screen_info.lfb_base) {
1336 if (efi_enabled(EFI_BOOT))
1337 size = max_t(__u32, screen_info.lfb_size, 0x800000);
1339 size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1341 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1342 fb_mmio = __request_region(hyperv_mmio,
1343 screen_info.lfb_base, size,
1350 * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1351 * @new: If successful, supplied a pointer to the
1352 * allocated MMIO space.
1353 * @device_obj: Identifies the caller
1354 * @min: Minimum guest physical address of the
1356 * @max: Maximum guest physical address
1357 * @size: Size of the range to be allocated
1358 * @align: Alignment of the range to be allocated
1359 * @fb_overlap_ok: Whether this allocation can be allowed
1360 * to overlap the video frame buffer.
1362 * This function walks the resources granted to VMBus by the
1363 * _CRS object in the ACPI namespace underneath the parent
1364 * "bridge" whether that's a root PCI bus in the Generation 1
1365 * case or a Module Device in the Generation 2 case. It then
1366 * attempts to allocate from the global MMIO pool in a way that
1367 * matches the constraints supplied in these parameters and by
1370 * Return: 0 on success, -errno on failure
1372 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1373 resource_size_t min, resource_size_t max,
1374 resource_size_t size, resource_size_t align,
1377 struct resource *iter, *shadow;
1378 resource_size_t range_min, range_max, start;
1379 const char *dev_n = dev_name(&device_obj->device);
1383 down(&hyperv_mmio_lock);
1386 * If overlaps with frame buffers are allowed, then first attempt to
1387 * make the allocation from within the reserved region. Because it
1388 * is already reserved, no shadow allocation is necessary.
1390 if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1391 !(max < fb_mmio->start)) {
1393 range_min = fb_mmio->start;
1394 range_max = fb_mmio->end;
1395 start = (range_min + align - 1) & ~(align - 1);
1396 for (; start + size - 1 <= range_max; start += align) {
1397 *new = request_mem_region_exclusive(start, size, dev_n);
1405 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1406 if ((iter->start >= max) || (iter->end <= min))
1409 range_min = iter->start;
1410 range_max = iter->end;
1411 start = (range_min + align - 1) & ~(align - 1);
1412 for (; start + size - 1 <= range_max; start += align) {
1413 shadow = __request_region(iter, start, size, NULL,
1418 *new = request_mem_region_exclusive(start, size, dev_n);
1420 shadow->name = (char *)*new;
1425 __release_region(iter, start, size);
1430 up(&hyperv_mmio_lock);
1433 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1436 * vmbus_free_mmio() - Free a memory-mapped I/O range.
1437 * @start: Base address of region to release.
1438 * @size: Size of the range to be allocated
1440 * This function releases anything requested by
1441 * vmbus_mmio_allocate().
1443 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
1445 struct resource *iter;
1447 down(&hyperv_mmio_lock);
1448 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1449 if ((iter->start >= start + size) || (iter->end <= start))
1452 __release_region(iter, start, size);
1454 release_mem_region(start, size);
1455 up(&hyperv_mmio_lock);
1458 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
1461 * vmbus_cpu_number_to_vp_number() - Map CPU to VP.
1462 * @cpu_number: CPU number in Linux terms
1464 * This function returns the mapping between the Linux processor
1465 * number and the hypervisor's virtual processor number, useful
1466 * in making hypercalls and such that talk about specific
1469 * Return: Virtual processor number in Hyper-V terms
1471 int vmbus_cpu_number_to_vp_number(int cpu_number)
1473 return hv_context.vp_index[cpu_number];
1475 EXPORT_SYMBOL_GPL(vmbus_cpu_number_to_vp_number);
1477 static int vmbus_acpi_add(struct acpi_device *device)
1480 int ret_val = -ENODEV;
1481 struct acpi_device *ancestor;
1483 hv_acpi_dev = device;
1485 result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1486 vmbus_walk_resources, NULL);
1488 if (ACPI_FAILURE(result))
1491 * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1492 * firmware) is the VMOD that has the mmio ranges. Get that.
1494 for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1495 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1496 vmbus_walk_resources, NULL);
1498 if (ACPI_FAILURE(result))
1508 complete(&probe_event);
1510 vmbus_acpi_remove(device);
1514 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1519 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1521 static struct acpi_driver vmbus_acpi_driver = {
1523 .ids = vmbus_acpi_device_ids,
1525 .add = vmbus_acpi_add,
1526 .remove = vmbus_acpi_remove,
1530 static void hv_kexec_handler(void)
1532 hv_synic_clockevents_cleanup();
1533 vmbus_initiate_unload(false);
1534 vmbus_connection.conn_state = DISCONNECTED;
1535 /* Make sure conn_state is set as hv_synic_cleanup checks for it */
1537 cpuhp_remove_state(hyperv_cpuhp_online);
1541 static void hv_crash_handler(struct pt_regs *regs)
1543 vmbus_initiate_unload(true);
1545 * In crash handler we can't schedule synic cleanup for all CPUs,
1546 * doing the cleanup for current CPU only. This should be sufficient
1549 vmbus_connection.conn_state = DISCONNECTED;
1550 hv_synic_cleanup(smp_processor_id());
1554 static int __init hv_acpi_init(void)
1558 if (x86_hyper != &x86_hyper_ms_hyperv)
1561 init_completion(&probe_event);
1564 * Get ACPI resources first.
1566 ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1571 t = wait_for_completion_timeout(&probe_event, 5*HZ);
1577 ret = vmbus_bus_init();
1581 hv_setup_kexec_handler(hv_kexec_handler);
1582 hv_setup_crash_handler(hv_crash_handler);
1587 acpi_bus_unregister_driver(&vmbus_acpi_driver);
1592 static void __exit vmbus_exit(void)
1596 hv_remove_kexec_handler();
1597 hv_remove_crash_handler();
1598 vmbus_connection.conn_state = DISCONNECTED;
1599 hv_synic_clockevents_cleanup();
1601 hv_remove_vmbus_irq();
1602 for_each_online_cpu(cpu) {
1603 struct hv_per_cpu_context *hv_cpu
1604 = per_cpu_ptr(hv_context.cpu_context, cpu);
1606 tasklet_kill(&hv_cpu->msg_dpc);
1608 vmbus_free_channels();
1610 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1611 unregister_die_notifier(&hyperv_die_block);
1612 atomic_notifier_chain_unregister(&panic_notifier_list,
1613 &hyperv_panic_block);
1615 bus_unregister(&hv_bus);
1617 cpuhp_remove_state(hyperv_cpuhp_online);
1619 acpi_bus_unregister_driver(&vmbus_acpi_driver);
1623 MODULE_LICENSE("GPL");
1625 subsys_initcall(hv_acpi_init);
1626 module_exit(vmbus_exit);