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
);
612 * store_new_id - sysfs frontend to vmbus_add_dynid()
614 * Allow GUIDs to be added to an existing driver via sysfs.
616 static ssize_t
new_id_store(struct device_driver
*driver
, const char *buf
,
619 struct hv_driver
*drv
= drv_to_hv_drv(driver
);
623 retval
= uuid_le_to_bin(buf
, &guid
);
627 if (hv_vmbus_get_id(drv
, &guid
))
630 retval
= vmbus_add_dynid(drv
, &guid
);
635 static DRIVER_ATTR_WO(new_id
);
638 * store_remove_id - remove a PCI device ID from this driver
640 * Removes a dynamic pci device ID to this driver.
642 static ssize_t
remove_id_store(struct device_driver
*driver
, const char *buf
,
645 struct hv_driver
*drv
= drv_to_hv_drv(driver
);
646 struct vmbus_dynid
*dynid
, *n
;
650 retval
= uuid_le_to_bin(buf
, &guid
);
655 spin_lock(&drv
->dynids
.lock
);
656 list_for_each_entry_safe(dynid
, n
, &drv
->dynids
.list
, node
) {
657 struct hv_vmbus_device_id
*id
= &dynid
->id
;
659 if (!uuid_le_cmp(id
->guid
, guid
)) {
660 list_del(&dynid
->node
);
666 spin_unlock(&drv
->dynids
.lock
);
670 static DRIVER_ATTR_WO(remove_id
);
672 static struct attribute
*vmbus_drv_attrs
[] = {
673 &driver_attr_new_id
.attr
,
674 &driver_attr_remove_id
.attr
,
677 ATTRIBUTE_GROUPS(vmbus_drv
);
681 * vmbus_match - Attempt to match the specified device to the specified driver
683 static int vmbus_match(struct device
*device
, struct device_driver
*driver
)
685 struct hv_driver
*drv
= drv_to_hv_drv(driver
);
686 struct hv_device
*hv_dev
= device_to_hv_device(device
);
688 /* The hv_sock driver handles all hv_sock offers. */
689 if (is_hvsock_channel(hv_dev
->channel
))
692 if (hv_vmbus_get_id(drv
, &hv_dev
->dev_type
))
699 * vmbus_probe - Add the new vmbus's child device
701 static int vmbus_probe(struct device
*child_device
)
704 struct hv_driver
*drv
=
705 drv_to_hv_drv(child_device
->driver
);
706 struct hv_device
*dev
= device_to_hv_device(child_device
);
707 const struct hv_vmbus_device_id
*dev_id
;
709 dev_id
= hv_vmbus_get_id(drv
, &dev
->dev_type
);
711 ret
= drv
->probe(dev
, dev_id
);
713 pr_err("probe failed for device %s (%d)\n",
714 dev_name(child_device
), ret
);
717 pr_err("probe not set for driver %s\n",
718 dev_name(child_device
));
725 * vmbus_remove - Remove a vmbus device
727 static int vmbus_remove(struct device
*child_device
)
729 struct hv_driver
*drv
;
730 struct hv_device
*dev
= device_to_hv_device(child_device
);
732 if (child_device
->driver
) {
733 drv
= drv_to_hv_drv(child_device
->driver
);
743 * vmbus_shutdown - Shutdown a vmbus device
745 static void vmbus_shutdown(struct device
*child_device
)
747 struct hv_driver
*drv
;
748 struct hv_device
*dev
= device_to_hv_device(child_device
);
751 /* The device may not be attached yet */
752 if (!child_device
->driver
)
755 drv
= drv_to_hv_drv(child_device
->driver
);
763 * vmbus_device_release - Final callback release of the vmbus child device
765 static void vmbus_device_release(struct device
*device
)
767 struct hv_device
*hv_dev
= device_to_hv_device(device
);
768 struct vmbus_channel
*channel
= hv_dev
->channel
;
770 mutex_lock(&vmbus_connection
.channel_mutex
);
771 hv_process_channel_removal(channel
,
772 channel
->offermsg
.child_relid
);
773 mutex_unlock(&vmbus_connection
.channel_mutex
);
778 /* The one and only one */
779 static struct bus_type hv_bus
= {
781 .match
= vmbus_match
,
782 .shutdown
= vmbus_shutdown
,
783 .remove
= vmbus_remove
,
784 .probe
= vmbus_probe
,
785 .uevent
= vmbus_uevent
,
786 .dev_groups
= vmbus_dev_groups
,
787 .drv_groups
= vmbus_drv_groups
,
790 struct onmessage_work_context
{
791 struct work_struct work
;
792 struct hv_message msg
;
795 static void vmbus_onmessage_work(struct work_struct
*work
)
797 struct onmessage_work_context
*ctx
;
799 /* Do not process messages if we're in DISCONNECTED state */
800 if (vmbus_connection
.conn_state
== DISCONNECTED
)
803 ctx
= container_of(work
, struct onmessage_work_context
,
805 vmbus_onmessage(&ctx
->msg
);
809 static void hv_process_timer_expiration(struct hv_message
*msg
,
810 struct hv_per_cpu_context
*hv_cpu
)
812 struct clock_event_device
*dev
= hv_cpu
->clk_evt
;
814 if (dev
->event_handler
)
815 dev
->event_handler(dev
);
817 vmbus_signal_eom(msg
, HVMSG_TIMER_EXPIRED
);
820 void vmbus_on_msg_dpc(unsigned long data
)
822 struct hv_per_cpu_context
*hv_cpu
= (void *)data
;
823 void *page_addr
= hv_cpu
->synic_message_page
;
824 struct hv_message
*msg
= (struct hv_message
*)page_addr
+
826 struct vmbus_channel_message_header
*hdr
;
827 const struct vmbus_channel_message_table_entry
*entry
;
828 struct onmessage_work_context
*ctx
;
829 u32 message_type
= msg
->header
.message_type
;
831 if (message_type
== HVMSG_NONE
)
835 hdr
= (struct vmbus_channel_message_header
*)msg
->u
.payload
;
837 if (hdr
->msgtype
>= CHANNELMSG_COUNT
) {
838 WARN_ONCE(1, "unknown msgtype=%d\n", hdr
->msgtype
);
842 entry
= &channel_message_table
[hdr
->msgtype
];
843 if (entry
->handler_type
== VMHT_BLOCKING
) {
844 ctx
= kmalloc(sizeof(*ctx
), GFP_ATOMIC
);
848 INIT_WORK(&ctx
->work
, vmbus_onmessage_work
);
849 memcpy(&ctx
->msg
, msg
, sizeof(*msg
));
852 * The host can generate a rescind message while we
853 * may still be handling the original offer. We deal with
854 * this condition by ensuring the processing is done on the
857 switch (hdr
->msgtype
) {
858 case CHANNELMSG_RESCIND_CHANNELOFFER
:
860 * If we are handling the rescind message;
861 * schedule the work on the global work queue.
863 schedule_work_on(vmbus_connection
.connect_cpu
,
867 case CHANNELMSG_OFFERCHANNEL
:
868 atomic_inc(&vmbus_connection
.offer_in_progress
);
869 queue_work_on(vmbus_connection
.connect_cpu
,
870 vmbus_connection
.work_queue
,
875 queue_work(vmbus_connection
.work_queue
, &ctx
->work
);
878 entry
->message_handler(hdr
);
881 vmbus_signal_eom(msg
, message_type
);
886 * Direct callback for channels using other deferred processing
888 static void vmbus_channel_isr(struct vmbus_channel
*channel
)
890 void (*callback_fn
)(void *);
892 callback_fn
= READ_ONCE(channel
->onchannel_callback
);
893 if (likely(callback_fn
!= NULL
))
894 (*callback_fn
)(channel
->channel_callback_context
);
898 * Schedule all channels with events pending
900 static void vmbus_chan_sched(struct hv_per_cpu_context
*hv_cpu
)
902 unsigned long *recv_int_page
;
905 if (vmbus_proto_version
< VERSION_WIN8
) {
906 maxbits
= MAX_NUM_CHANNELS_SUPPORTED
;
907 recv_int_page
= vmbus_connection
.recv_int_page
;
910 * When the host is win8 and beyond, the event page
911 * can be directly checked to get the id of the channel
912 * that has the interrupt pending.
914 void *page_addr
= hv_cpu
->synic_event_page
;
915 union hv_synic_event_flags
*event
916 = (union hv_synic_event_flags
*)page_addr
+
919 maxbits
= HV_EVENT_FLAGS_COUNT
;
920 recv_int_page
= event
->flags
;
923 if (unlikely(!recv_int_page
))
926 for_each_set_bit(relid
, recv_int_page
, maxbits
) {
927 struct vmbus_channel
*channel
;
929 if (!sync_test_and_clear_bit(relid
, recv_int_page
))
932 /* Special case - vmbus channel protocol msg */
938 /* Find channel based on relid */
939 list_for_each_entry_rcu(channel
, &hv_cpu
->chan_list
, percpu_list
) {
940 if (channel
->offermsg
.child_relid
!= relid
)
943 switch (channel
->callback_mode
) {
945 vmbus_channel_isr(channel
);
948 case HV_CALL_BATCHED
:
949 hv_begin_read(&channel
->inbound
);
952 tasklet_schedule(&channel
->callback_event
);
960 static void vmbus_isr(void)
962 struct hv_per_cpu_context
*hv_cpu
963 = this_cpu_ptr(hv_context
.cpu_context
);
964 void *page_addr
= hv_cpu
->synic_event_page
;
965 struct hv_message
*msg
;
966 union hv_synic_event_flags
*event
;
967 bool handled
= false;
969 if (unlikely(page_addr
== NULL
))
972 event
= (union hv_synic_event_flags
*)page_addr
+
975 * Check for events before checking for messages. This is the order
976 * in which events and messages are checked in Windows guests on
977 * Hyper-V, and the Windows team suggested we do the same.
980 if ((vmbus_proto_version
== VERSION_WS2008
) ||
981 (vmbus_proto_version
== VERSION_WIN7
)) {
983 /* Since we are a child, we only need to check bit 0 */
984 if (sync_test_and_clear_bit(0, event
->flags
))
988 * Our host is win8 or above. The signaling mechanism
989 * has changed and we can directly look at the event page.
990 * If bit n is set then we have an interrup on the channel
997 vmbus_chan_sched(hv_cpu
);
999 page_addr
= hv_cpu
->synic_message_page
;
1000 msg
= (struct hv_message
*)page_addr
+ VMBUS_MESSAGE_SINT
;
1002 /* Check if there are actual msgs to be processed */
1003 if (msg
->header
.message_type
!= HVMSG_NONE
) {
1004 if (msg
->header
.message_type
== HVMSG_TIMER_EXPIRED
)
1005 hv_process_timer_expiration(msg
, hv_cpu
);
1007 tasklet_schedule(&hv_cpu
->msg_dpc
);
1010 add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR
, 0);
1015 * vmbus_bus_init -Main vmbus driver initialization routine.
1018 * - initialize the vmbus driver context
1019 * - invoke the vmbus hv main init routine
1020 * - retrieve the channel offers
1022 static int vmbus_bus_init(void)
1026 /* Hypervisor initialization...setup hypercall page..etc */
1029 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret
);
1033 ret
= bus_register(&hv_bus
);
1037 hv_setup_vmbus_irq(vmbus_isr
);
1039 ret
= hv_synic_alloc();
1043 * Initialize the per-cpu interrupt state and
1044 * connect to the host.
1046 ret
= cpuhp_setup_state(CPUHP_AP_ONLINE_DYN
, "x86/hyperv:online",
1047 hv_synic_init
, hv_synic_cleanup
);
1050 hyperv_cpuhp_online
= ret
;
1052 ret
= vmbus_connect();
1057 * Only register if the crash MSRs are available
1059 if (ms_hyperv
.misc_features
& HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE
) {
1060 register_die_notifier(&hyperv_die_block
);
1061 atomic_notifier_chain_register(&panic_notifier_list
,
1062 &hyperv_panic_block
);
1065 vmbus_request_offers();
1070 cpuhp_remove_state(hyperv_cpuhp_online
);
1073 hv_remove_vmbus_irq();
1075 bus_unregister(&hv_bus
);
1081 * __vmbus_child_driver_register() - Register a vmbus's driver
1082 * @hv_driver: Pointer to driver structure you want to register
1083 * @owner: owner module of the drv
1084 * @mod_name: module name string
1086 * Registers the given driver with Linux through the 'driver_register()' call
1087 * and sets up the hyper-v vmbus handling for this driver.
1088 * It will return the state of the 'driver_register()' call.
1091 int __vmbus_driver_register(struct hv_driver
*hv_driver
, struct module
*owner
, const char *mod_name
)
1095 pr_info("registering driver %s\n", hv_driver
->name
);
1097 ret
= vmbus_exists();
1101 hv_driver
->driver
.name
= hv_driver
->name
;
1102 hv_driver
->driver
.owner
= owner
;
1103 hv_driver
->driver
.mod_name
= mod_name
;
1104 hv_driver
->driver
.bus
= &hv_bus
;
1106 spin_lock_init(&hv_driver
->dynids
.lock
);
1107 INIT_LIST_HEAD(&hv_driver
->dynids
.list
);
1109 ret
= driver_register(&hv_driver
->driver
);
1113 EXPORT_SYMBOL_GPL(__vmbus_driver_register
);
1116 * vmbus_driver_unregister() - Unregister a vmbus's driver
1117 * @hv_driver: Pointer to driver structure you want to
1120 * Un-register the given driver that was previous registered with a call to
1121 * vmbus_driver_register()
1123 void vmbus_driver_unregister(struct hv_driver
*hv_driver
)
1125 pr_info("unregistering driver %s\n", hv_driver
->name
);
1127 if (!vmbus_exists()) {
1128 driver_unregister(&hv_driver
->driver
);
1129 vmbus_free_dynids(hv_driver
);
1132 EXPORT_SYMBOL_GPL(vmbus_driver_unregister
);
1135 * vmbus_device_create - Creates and registers a new child device
1138 struct hv_device
*vmbus_device_create(const uuid_le
*type
,
1139 const uuid_le
*instance
,
1140 struct vmbus_channel
*channel
)
1142 struct hv_device
*child_device_obj
;
1144 child_device_obj
= kzalloc(sizeof(struct hv_device
), GFP_KERNEL
);
1145 if (!child_device_obj
) {
1146 pr_err("Unable to allocate device object for child device\n");
1150 child_device_obj
->channel
= channel
;
1151 memcpy(&child_device_obj
->dev_type
, type
, sizeof(uuid_le
));
1152 memcpy(&child_device_obj
->dev_instance
, instance
,
1154 child_device_obj
->vendor_id
= 0x1414; /* MSFT vendor ID */
1157 return child_device_obj
;
1161 * vmbus_device_register - Register the child device
1163 int vmbus_device_register(struct hv_device
*child_device_obj
)
1167 dev_set_name(&child_device_obj
->device
, "%pUl",
1168 child_device_obj
->channel
->offermsg
.offer
.if_instance
.b
);
1170 child_device_obj
->device
.bus
= &hv_bus
;
1171 child_device_obj
->device
.parent
= &hv_acpi_dev
->dev
;
1172 child_device_obj
->device
.release
= vmbus_device_release
;
1175 * Register with the LDM. This will kick off the driver/device
1176 * binding...which will eventually call vmbus_match() and vmbus_probe()
1178 ret
= device_register(&child_device_obj
->device
);
1181 pr_err("Unable to register child device\n");
1183 pr_debug("child device %s registered\n",
1184 dev_name(&child_device_obj
->device
));
1190 * vmbus_device_unregister - Remove the specified child device
1193 void vmbus_device_unregister(struct hv_device
*device_obj
)
1195 pr_debug("child device %s unregistered\n",
1196 dev_name(&device_obj
->device
));
1199 * Kick off the process of unregistering the device.
1200 * This will call vmbus_remove() and eventually vmbus_device_release()
1202 device_unregister(&device_obj
->device
);
1207 * VMBUS is an acpi enumerated device. Get the information we
1210 #define VTPM_BASE_ADDRESS 0xfed40000
1211 static acpi_status
vmbus_walk_resources(struct acpi_resource
*res
, void *ctx
)
1213 resource_size_t start
= 0;
1214 resource_size_t end
= 0;
1215 struct resource
*new_res
;
1216 struct resource
**old_res
= &hyperv_mmio
;
1217 struct resource
**prev_res
= NULL
;
1219 switch (res
->type
) {
1222 * "Address" descriptors are for bus windows. Ignore
1223 * "memory" descriptors, which are for registers on
1226 case ACPI_RESOURCE_TYPE_ADDRESS32
:
1227 start
= res
->data
.address32
.address
.minimum
;
1228 end
= res
->data
.address32
.address
.maximum
;
1231 case ACPI_RESOURCE_TYPE_ADDRESS64
:
1232 start
= res
->data
.address64
.address
.minimum
;
1233 end
= res
->data
.address64
.address
.maximum
;
1237 /* Unused resource type */
1242 * Ignore ranges that are below 1MB, as they're not
1243 * necessary or useful here.
1248 new_res
= kzalloc(sizeof(*new_res
), GFP_ATOMIC
);
1250 return AE_NO_MEMORY
;
1252 /* If this range overlaps the virtual TPM, truncate it. */
1253 if (end
> VTPM_BASE_ADDRESS
&& start
< VTPM_BASE_ADDRESS
)
1254 end
= VTPM_BASE_ADDRESS
;
1256 new_res
->name
= "hyperv mmio";
1257 new_res
->flags
= IORESOURCE_MEM
;
1258 new_res
->start
= start
;
1262 * If two ranges are adjacent, merge them.
1270 if (((*old_res
)->end
+ 1) == new_res
->start
) {
1271 (*old_res
)->end
= new_res
->end
;
1276 if ((*old_res
)->start
== new_res
->end
+ 1) {
1277 (*old_res
)->start
= new_res
->start
;
1282 if ((*old_res
)->start
> new_res
->end
) {
1283 new_res
->sibling
= *old_res
;
1285 (*prev_res
)->sibling
= new_res
;
1291 old_res
= &(*old_res
)->sibling
;
1298 static int vmbus_acpi_remove(struct acpi_device
*device
)
1300 struct resource
*cur_res
;
1301 struct resource
*next_res
;
1305 __release_region(hyperv_mmio
, fb_mmio
->start
,
1306 resource_size(fb_mmio
));
1310 for (cur_res
= hyperv_mmio
; cur_res
; cur_res
= next_res
) {
1311 next_res
= cur_res
->sibling
;
1319 static void vmbus_reserve_fb(void)
1323 * Make a claim for the frame buffer in the resource tree under the
1324 * first node, which will be the one below 4GB. The length seems to
1325 * be underreported, particularly in a Generation 1 VM. So start out
1326 * reserving a larger area and make it smaller until it succeeds.
1329 if (screen_info
.lfb_base
) {
1330 if (efi_enabled(EFI_BOOT
))
1331 size
= max_t(__u32
, screen_info
.lfb_size
, 0x800000);
1333 size
= max_t(__u32
, screen_info
.lfb_size
, 0x4000000);
1335 for (; !fb_mmio
&& (size
>= 0x100000); size
>>= 1) {
1336 fb_mmio
= __request_region(hyperv_mmio
,
1337 screen_info
.lfb_base
, size
,
1344 * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1345 * @new: If successful, supplied a pointer to the
1346 * allocated MMIO space.
1347 * @device_obj: Identifies the caller
1348 * @min: Minimum guest physical address of the
1350 * @max: Maximum guest physical address
1351 * @size: Size of the range to be allocated
1352 * @align: Alignment of the range to be allocated
1353 * @fb_overlap_ok: Whether this allocation can be allowed
1354 * to overlap the video frame buffer.
1356 * This function walks the resources granted to VMBus by the
1357 * _CRS object in the ACPI namespace underneath the parent
1358 * "bridge" whether that's a root PCI bus in the Generation 1
1359 * case or a Module Device in the Generation 2 case. It then
1360 * attempts to allocate from the global MMIO pool in a way that
1361 * matches the constraints supplied in these parameters and by
1364 * Return: 0 on success, -errno on failure
1366 int vmbus_allocate_mmio(struct resource
**new, struct hv_device
*device_obj
,
1367 resource_size_t min
, resource_size_t max
,
1368 resource_size_t size
, resource_size_t align
,
1371 struct resource
*iter
, *shadow
;
1372 resource_size_t range_min
, range_max
, start
;
1373 const char *dev_n
= dev_name(&device_obj
->device
);
1377 down(&hyperv_mmio_lock
);
1380 * If overlaps with frame buffers are allowed, then first attempt to
1381 * make the allocation from within the reserved region. Because it
1382 * is already reserved, no shadow allocation is necessary.
1384 if (fb_overlap_ok
&& fb_mmio
&& !(min
> fb_mmio
->end
) &&
1385 !(max
< fb_mmio
->start
)) {
1387 range_min
= fb_mmio
->start
;
1388 range_max
= fb_mmio
->end
;
1389 start
= (range_min
+ align
- 1) & ~(align
- 1);
1390 for (; start
+ size
- 1 <= range_max
; start
+= align
) {
1391 *new = request_mem_region_exclusive(start
, size
, dev_n
);
1399 for (iter
= hyperv_mmio
; iter
; iter
= iter
->sibling
) {
1400 if ((iter
->start
>= max
) || (iter
->end
<= min
))
1403 range_min
= iter
->start
;
1404 range_max
= iter
->end
;
1405 start
= (range_min
+ align
- 1) & ~(align
- 1);
1406 for (; start
+ size
- 1 <= range_max
; start
+= align
) {
1407 shadow
= __request_region(iter
, start
, size
, NULL
,
1412 *new = request_mem_region_exclusive(start
, size
, dev_n
);
1414 shadow
->name
= (char *)*new;
1419 __release_region(iter
, start
, size
);
1424 up(&hyperv_mmio_lock
);
1427 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio
);
1430 * vmbus_free_mmio() - Free a memory-mapped I/O range.
1431 * @start: Base address of region to release.
1432 * @size: Size of the range to be allocated
1434 * This function releases anything requested by
1435 * vmbus_mmio_allocate().
1437 void vmbus_free_mmio(resource_size_t start
, resource_size_t size
)
1439 struct resource
*iter
;
1441 down(&hyperv_mmio_lock
);
1442 for (iter
= hyperv_mmio
; iter
; iter
= iter
->sibling
) {
1443 if ((iter
->start
>= start
+ size
) || (iter
->end
<= start
))
1446 __release_region(iter
, start
, size
);
1448 release_mem_region(start
, size
);
1449 up(&hyperv_mmio_lock
);
1452 EXPORT_SYMBOL_GPL(vmbus_free_mmio
);
1455 * vmbus_cpu_number_to_vp_number() - Map CPU to VP.
1456 * @cpu_number: CPU number in Linux terms
1458 * This function returns the mapping between the Linux processor
1459 * number and the hypervisor's virtual processor number, useful
1460 * in making hypercalls and such that talk about specific
1463 * Return: Virtual processor number in Hyper-V terms
1465 int vmbus_cpu_number_to_vp_number(int cpu_number
)
1467 return hv_context
.vp_index
[cpu_number
];
1469 EXPORT_SYMBOL_GPL(vmbus_cpu_number_to_vp_number
);
1471 static int vmbus_acpi_add(struct acpi_device
*device
)
1474 int ret_val
= -ENODEV
;
1475 struct acpi_device
*ancestor
;
1477 hv_acpi_dev
= device
;
1479 result
= acpi_walk_resources(device
->handle
, METHOD_NAME__CRS
,
1480 vmbus_walk_resources
, NULL
);
1482 if (ACPI_FAILURE(result
))
1485 * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1486 * firmware) is the VMOD that has the mmio ranges. Get that.
1488 for (ancestor
= device
->parent
; ancestor
; ancestor
= ancestor
->parent
) {
1489 result
= acpi_walk_resources(ancestor
->handle
, METHOD_NAME__CRS
,
1490 vmbus_walk_resources
, NULL
);
1492 if (ACPI_FAILURE(result
))
1502 complete(&probe_event
);
1504 vmbus_acpi_remove(device
);
1508 static const struct acpi_device_id vmbus_acpi_device_ids
[] = {
1513 MODULE_DEVICE_TABLE(acpi
, vmbus_acpi_device_ids
);
1515 static struct acpi_driver vmbus_acpi_driver
= {
1517 .ids
= vmbus_acpi_device_ids
,
1519 .add
= vmbus_acpi_add
,
1520 .remove
= vmbus_acpi_remove
,
1524 static void hv_kexec_handler(void)
1526 hv_synic_clockevents_cleanup();
1527 vmbus_initiate_unload(false);
1528 vmbus_connection
.conn_state
= DISCONNECTED
;
1529 /* Make sure conn_state is set as hv_synic_cleanup checks for it */
1531 cpuhp_remove_state(hyperv_cpuhp_online
);
1535 static void hv_crash_handler(struct pt_regs
*regs
)
1537 vmbus_initiate_unload(true);
1539 * In crash handler we can't schedule synic cleanup for all CPUs,
1540 * doing the cleanup for current CPU only. This should be sufficient
1543 vmbus_connection
.conn_state
= DISCONNECTED
;
1544 hv_synic_cleanup(smp_processor_id());
1548 static int __init
hv_acpi_init(void)
1552 if (x86_hyper
!= &x86_hyper_ms_hyperv
)
1555 init_completion(&probe_event
);
1558 * Get ACPI resources first.
1560 ret
= acpi_bus_register_driver(&vmbus_acpi_driver
);
1565 t
= wait_for_completion_timeout(&probe_event
, 5*HZ
);
1571 ret
= vmbus_bus_init();
1575 hv_setup_kexec_handler(hv_kexec_handler
);
1576 hv_setup_crash_handler(hv_crash_handler
);
1581 acpi_bus_unregister_driver(&vmbus_acpi_driver
);
1586 static void __exit
vmbus_exit(void)
1590 hv_remove_kexec_handler();
1591 hv_remove_crash_handler();
1592 vmbus_connection
.conn_state
= DISCONNECTED
;
1593 hv_synic_clockevents_cleanup();
1595 hv_remove_vmbus_irq();
1596 for_each_online_cpu(cpu
) {
1597 struct hv_per_cpu_context
*hv_cpu
1598 = per_cpu_ptr(hv_context
.cpu_context
, cpu
);
1600 tasklet_kill(&hv_cpu
->msg_dpc
);
1602 vmbus_free_channels();
1604 if (ms_hyperv
.misc_features
& HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE
) {
1605 unregister_die_notifier(&hyperv_die_block
);
1606 atomic_notifier_chain_unregister(&panic_notifier_list
,
1607 &hyperv_panic_block
);
1609 bus_unregister(&hv_bus
);
1611 cpuhp_remove_state(hyperv_cpuhp_online
);
1613 acpi_bus_unregister_driver(&vmbus_acpi_driver
);
1617 MODULE_LICENSE("GPL");
1619 subsys_initcall(hv_acpi_init
);
1620 module_exit(vmbus_exit
);