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 <asm/hyperv.h>
38 #include <asm/hypervisor.h>
39 #include <asm/mshyperv.h>
40 #include <linux/notifier.h>
41 #include <linux/ptrace.h>
42 #include <linux/screen_info.h>
43 #include <linux/kdebug.h>
44 #include <linux/efi.h>
45 #include "hyperv_vmbus.h"
47 static struct acpi_device
*hv_acpi_dev
;
49 static struct completion probe_event
;
52 static void hyperv_report_panic(struct pt_regs
*regs
)
54 static bool panic_reported
;
57 * We prefer to report panic on 'die' chain as we have proper
58 * registers to report, but if we miss it (e.g. on BUG()) we need
59 * to report it on 'panic'.
63 panic_reported
= true;
65 wrmsrl(HV_X64_MSR_CRASH_P0
, regs
->ip
);
66 wrmsrl(HV_X64_MSR_CRASH_P1
, regs
->ax
);
67 wrmsrl(HV_X64_MSR_CRASH_P2
, regs
->bx
);
68 wrmsrl(HV_X64_MSR_CRASH_P3
, regs
->cx
);
69 wrmsrl(HV_X64_MSR_CRASH_P4
, regs
->dx
);
72 * Let Hyper-V know there is crash data available
74 wrmsrl(HV_X64_MSR_CRASH_CTL
, HV_CRASH_CTL_CRASH_NOTIFY
);
77 static int hyperv_panic_event(struct notifier_block
*nb
, unsigned long val
,
82 regs
= current_pt_regs();
84 hyperv_report_panic(regs
);
88 static int hyperv_die_event(struct notifier_block
*nb
, unsigned long val
,
91 struct die_args
*die
= (struct die_args
*)args
;
92 struct pt_regs
*regs
= die
->regs
;
94 hyperv_report_panic(regs
);
98 static struct notifier_block hyperv_die_block
= {
99 .notifier_call
= hyperv_die_event
,
101 static struct notifier_block hyperv_panic_block
= {
102 .notifier_call
= hyperv_panic_event
,
105 static const char *fb_mmio_name
= "fb_range";
106 static struct resource
*fb_mmio
;
107 struct resource
*hyperv_mmio
;
108 DEFINE_SEMAPHORE(hyperv_mmio_lock
);
110 static int vmbus_exists(void)
112 if (hv_acpi_dev
== NULL
)
118 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
119 static void print_alias_name(struct hv_device
*hv_dev
, char *alias_name
)
122 for (i
= 0; i
< VMBUS_ALIAS_LEN
; i
+= 2)
123 sprintf(&alias_name
[i
], "%02x", hv_dev
->dev_type
.b
[i
/2]);
126 static u8
channel_monitor_group(struct vmbus_channel
*channel
)
128 return (u8
)channel
->offermsg
.monitorid
/ 32;
131 static u8
channel_monitor_offset(struct vmbus_channel
*channel
)
133 return (u8
)channel
->offermsg
.monitorid
% 32;
136 static u32
channel_pending(struct vmbus_channel
*channel
,
137 struct hv_monitor_page
*monitor_page
)
139 u8 monitor_group
= channel_monitor_group(channel
);
140 return monitor_page
->trigger_group
[monitor_group
].pending
;
143 static u32
channel_latency(struct vmbus_channel
*channel
,
144 struct hv_monitor_page
*monitor_page
)
146 u8 monitor_group
= channel_monitor_group(channel
);
147 u8 monitor_offset
= channel_monitor_offset(channel
);
148 return monitor_page
->latency
[monitor_group
][monitor_offset
];
151 static u32
channel_conn_id(struct vmbus_channel
*channel
,
152 struct hv_monitor_page
*monitor_page
)
154 u8 monitor_group
= channel_monitor_group(channel
);
155 u8 monitor_offset
= channel_monitor_offset(channel
);
156 return monitor_page
->parameter
[monitor_group
][monitor_offset
].connectionid
.u
.id
;
159 static ssize_t
id_show(struct device
*dev
, struct device_attribute
*dev_attr
,
162 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
164 if (!hv_dev
->channel
)
166 return sprintf(buf
, "%d\n", hv_dev
->channel
->offermsg
.child_relid
);
168 static DEVICE_ATTR_RO(id
);
170 static ssize_t
state_show(struct device
*dev
, struct device_attribute
*dev_attr
,
173 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
175 if (!hv_dev
->channel
)
177 return sprintf(buf
, "%d\n", hv_dev
->channel
->state
);
179 static DEVICE_ATTR_RO(state
);
181 static ssize_t
monitor_id_show(struct device
*dev
,
182 struct device_attribute
*dev_attr
, char *buf
)
184 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
186 if (!hv_dev
->channel
)
188 return sprintf(buf
, "%d\n", hv_dev
->channel
->offermsg
.monitorid
);
190 static DEVICE_ATTR_RO(monitor_id
);
192 static ssize_t
class_id_show(struct device
*dev
,
193 struct device_attribute
*dev_attr
, char *buf
)
195 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
197 if (!hv_dev
->channel
)
199 return sprintf(buf
, "{%pUl}\n",
200 hv_dev
->channel
->offermsg
.offer
.if_type
.b
);
202 static DEVICE_ATTR_RO(class_id
);
204 static ssize_t
device_id_show(struct device
*dev
,
205 struct device_attribute
*dev_attr
, char *buf
)
207 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
209 if (!hv_dev
->channel
)
211 return sprintf(buf
, "{%pUl}\n",
212 hv_dev
->channel
->offermsg
.offer
.if_instance
.b
);
214 static DEVICE_ATTR_RO(device_id
);
216 static ssize_t
modalias_show(struct device
*dev
,
217 struct device_attribute
*dev_attr
, char *buf
)
219 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
220 char alias_name
[VMBUS_ALIAS_LEN
+ 1];
222 print_alias_name(hv_dev
, alias_name
);
223 return sprintf(buf
, "vmbus:%s\n", alias_name
);
225 static DEVICE_ATTR_RO(modalias
);
227 static ssize_t
server_monitor_pending_show(struct device
*dev
,
228 struct device_attribute
*dev_attr
,
231 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
233 if (!hv_dev
->channel
)
235 return sprintf(buf
, "%d\n",
236 channel_pending(hv_dev
->channel
,
237 vmbus_connection
.monitor_pages
[1]));
239 static DEVICE_ATTR_RO(server_monitor_pending
);
241 static ssize_t
client_monitor_pending_show(struct device
*dev
,
242 struct device_attribute
*dev_attr
,
245 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
247 if (!hv_dev
->channel
)
249 return sprintf(buf
, "%d\n",
250 channel_pending(hv_dev
->channel
,
251 vmbus_connection
.monitor_pages
[1]));
253 static DEVICE_ATTR_RO(client_monitor_pending
);
255 static ssize_t
server_monitor_latency_show(struct device
*dev
,
256 struct device_attribute
*dev_attr
,
259 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
261 if (!hv_dev
->channel
)
263 return sprintf(buf
, "%d\n",
264 channel_latency(hv_dev
->channel
,
265 vmbus_connection
.monitor_pages
[0]));
267 static DEVICE_ATTR_RO(server_monitor_latency
);
269 static ssize_t
client_monitor_latency_show(struct device
*dev
,
270 struct device_attribute
*dev_attr
,
273 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
275 if (!hv_dev
->channel
)
277 return sprintf(buf
, "%d\n",
278 channel_latency(hv_dev
->channel
,
279 vmbus_connection
.monitor_pages
[1]));
281 static DEVICE_ATTR_RO(client_monitor_latency
);
283 static ssize_t
server_monitor_conn_id_show(struct device
*dev
,
284 struct device_attribute
*dev_attr
,
287 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
289 if (!hv_dev
->channel
)
291 return sprintf(buf
, "%d\n",
292 channel_conn_id(hv_dev
->channel
,
293 vmbus_connection
.monitor_pages
[0]));
295 static DEVICE_ATTR_RO(server_monitor_conn_id
);
297 static ssize_t
client_monitor_conn_id_show(struct device
*dev
,
298 struct device_attribute
*dev_attr
,
301 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
303 if (!hv_dev
->channel
)
305 return sprintf(buf
, "%d\n",
306 channel_conn_id(hv_dev
->channel
,
307 vmbus_connection
.monitor_pages
[1]));
309 static DEVICE_ATTR_RO(client_monitor_conn_id
);
311 static ssize_t
out_intr_mask_show(struct device
*dev
,
312 struct device_attribute
*dev_attr
, char *buf
)
314 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
315 struct hv_ring_buffer_debug_info outbound
;
317 if (!hv_dev
->channel
)
319 hv_ringbuffer_get_debuginfo(&hv_dev
->channel
->outbound
, &outbound
);
320 return sprintf(buf
, "%d\n", outbound
.current_interrupt_mask
);
322 static DEVICE_ATTR_RO(out_intr_mask
);
324 static ssize_t
out_read_index_show(struct device
*dev
,
325 struct device_attribute
*dev_attr
, char *buf
)
327 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
328 struct hv_ring_buffer_debug_info outbound
;
330 if (!hv_dev
->channel
)
332 hv_ringbuffer_get_debuginfo(&hv_dev
->channel
->outbound
, &outbound
);
333 return sprintf(buf
, "%d\n", outbound
.current_read_index
);
335 static DEVICE_ATTR_RO(out_read_index
);
337 static ssize_t
out_write_index_show(struct device
*dev
,
338 struct device_attribute
*dev_attr
,
341 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
342 struct hv_ring_buffer_debug_info outbound
;
344 if (!hv_dev
->channel
)
346 hv_ringbuffer_get_debuginfo(&hv_dev
->channel
->outbound
, &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
;
358 if (!hv_dev
->channel
)
360 hv_ringbuffer_get_debuginfo(&hv_dev
->channel
->outbound
, &outbound
);
361 return sprintf(buf
, "%d\n", outbound
.bytes_avail_toread
);
363 static DEVICE_ATTR_RO(out_read_bytes_avail
);
365 static ssize_t
out_write_bytes_avail_show(struct device
*dev
,
366 struct device_attribute
*dev_attr
,
369 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
370 struct hv_ring_buffer_debug_info outbound
;
372 if (!hv_dev
->channel
)
374 hv_ringbuffer_get_debuginfo(&hv_dev
->channel
->outbound
, &outbound
);
375 return sprintf(buf
, "%d\n", outbound
.bytes_avail_towrite
);
377 static DEVICE_ATTR_RO(out_write_bytes_avail
);
379 static ssize_t
in_intr_mask_show(struct device
*dev
,
380 struct device_attribute
*dev_attr
, char *buf
)
382 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
383 struct hv_ring_buffer_debug_info inbound
;
385 if (!hv_dev
->channel
)
387 hv_ringbuffer_get_debuginfo(&hv_dev
->channel
->inbound
, &inbound
);
388 return sprintf(buf
, "%d\n", inbound
.current_interrupt_mask
);
390 static DEVICE_ATTR_RO(in_intr_mask
);
392 static ssize_t
in_read_index_show(struct device
*dev
,
393 struct device_attribute
*dev_attr
, char *buf
)
395 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
396 struct hv_ring_buffer_debug_info inbound
;
398 if (!hv_dev
->channel
)
400 hv_ringbuffer_get_debuginfo(&hv_dev
->channel
->inbound
, &inbound
);
401 return sprintf(buf
, "%d\n", inbound
.current_read_index
);
403 static DEVICE_ATTR_RO(in_read_index
);
405 static ssize_t
in_write_index_show(struct device
*dev
,
406 struct device_attribute
*dev_attr
, char *buf
)
408 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
409 struct hv_ring_buffer_debug_info inbound
;
411 if (!hv_dev
->channel
)
413 hv_ringbuffer_get_debuginfo(&hv_dev
->channel
->inbound
, &inbound
);
414 return sprintf(buf
, "%d\n", inbound
.current_write_index
);
416 static DEVICE_ATTR_RO(in_write_index
);
418 static ssize_t
in_read_bytes_avail_show(struct device
*dev
,
419 struct device_attribute
*dev_attr
,
422 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
423 struct hv_ring_buffer_debug_info inbound
;
425 if (!hv_dev
->channel
)
427 hv_ringbuffer_get_debuginfo(&hv_dev
->channel
->inbound
, &inbound
);
428 return sprintf(buf
, "%d\n", inbound
.bytes_avail_toread
);
430 static DEVICE_ATTR_RO(in_read_bytes_avail
);
432 static ssize_t
in_write_bytes_avail_show(struct device
*dev
,
433 struct device_attribute
*dev_attr
,
436 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
437 struct hv_ring_buffer_debug_info inbound
;
439 if (!hv_dev
->channel
)
441 hv_ringbuffer_get_debuginfo(&hv_dev
->channel
->inbound
, &inbound
);
442 return sprintf(buf
, "%d\n", inbound
.bytes_avail_towrite
);
444 static DEVICE_ATTR_RO(in_write_bytes_avail
);
446 static ssize_t
channel_vp_mapping_show(struct device
*dev
,
447 struct device_attribute
*dev_attr
,
450 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
451 struct vmbus_channel
*channel
= hv_dev
->channel
, *cur_sc
;
453 int buf_size
= PAGE_SIZE
, n_written
, tot_written
;
454 struct list_head
*cur
;
459 tot_written
= snprintf(buf
, buf_size
, "%u:%u\n",
460 channel
->offermsg
.child_relid
, channel
->target_cpu
);
462 spin_lock_irqsave(&channel
->lock
, flags
);
464 list_for_each(cur
, &channel
->sc_list
) {
465 if (tot_written
>= buf_size
- 1)
468 cur_sc
= list_entry(cur
, struct vmbus_channel
, sc_list
);
469 n_written
= scnprintf(buf
+ tot_written
,
470 buf_size
- tot_written
,
472 cur_sc
->offermsg
.child_relid
,
474 tot_written
+= n_written
;
477 spin_unlock_irqrestore(&channel
->lock
, flags
);
481 static DEVICE_ATTR_RO(channel_vp_mapping
);
483 static ssize_t
vendor_show(struct device
*dev
,
484 struct device_attribute
*dev_attr
,
487 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
488 return sprintf(buf
, "0x%x\n", hv_dev
->vendor_id
);
490 static DEVICE_ATTR_RO(vendor
);
492 static ssize_t
device_show(struct device
*dev
,
493 struct device_attribute
*dev_attr
,
496 struct hv_device
*hv_dev
= device_to_hv_device(dev
);
497 return sprintf(buf
, "0x%x\n", hv_dev
->device_id
);
499 static DEVICE_ATTR_RO(device
);
501 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
502 static struct attribute
*vmbus_attrs
[] = {
504 &dev_attr_state
.attr
,
505 &dev_attr_monitor_id
.attr
,
506 &dev_attr_class_id
.attr
,
507 &dev_attr_device_id
.attr
,
508 &dev_attr_modalias
.attr
,
509 &dev_attr_server_monitor_pending
.attr
,
510 &dev_attr_client_monitor_pending
.attr
,
511 &dev_attr_server_monitor_latency
.attr
,
512 &dev_attr_client_monitor_latency
.attr
,
513 &dev_attr_server_monitor_conn_id
.attr
,
514 &dev_attr_client_monitor_conn_id
.attr
,
515 &dev_attr_out_intr_mask
.attr
,
516 &dev_attr_out_read_index
.attr
,
517 &dev_attr_out_write_index
.attr
,
518 &dev_attr_out_read_bytes_avail
.attr
,
519 &dev_attr_out_write_bytes_avail
.attr
,
520 &dev_attr_in_intr_mask
.attr
,
521 &dev_attr_in_read_index
.attr
,
522 &dev_attr_in_write_index
.attr
,
523 &dev_attr_in_read_bytes_avail
.attr
,
524 &dev_attr_in_write_bytes_avail
.attr
,
525 &dev_attr_channel_vp_mapping
.attr
,
526 &dev_attr_vendor
.attr
,
527 &dev_attr_device
.attr
,
530 ATTRIBUTE_GROUPS(vmbus
);
533 * vmbus_uevent - add uevent for our device
535 * This routine is invoked when a device is added or removed on the vmbus to
536 * generate a uevent to udev in the userspace. The udev will then look at its
537 * rule and the uevent generated here to load the appropriate driver
539 * The alias string will be of the form vmbus:guid where guid is the string
540 * representation of the device guid (each byte of the guid will be
541 * represented with two hex characters.
543 static int vmbus_uevent(struct device
*device
, struct kobj_uevent_env
*env
)
545 struct hv_device
*dev
= device_to_hv_device(device
);
547 char alias_name
[VMBUS_ALIAS_LEN
+ 1];
549 print_alias_name(dev
, alias_name
);
550 ret
= add_uevent_var(env
, "MODALIAS=vmbus:%s", alias_name
);
554 static const uuid_le null_guid
;
556 static inline bool is_null_guid(const uuid_le
*guid
)
558 if (uuid_le_cmp(*guid
, null_guid
))
564 * Return a matching hv_vmbus_device_id pointer.
565 * If there is no match, return NULL.
567 static const struct hv_vmbus_device_id
*hv_vmbus_get_id(
568 const struct hv_vmbus_device_id
*id
,
571 for (; !is_null_guid(&id
->guid
); id
++)
572 if (!uuid_le_cmp(id
->guid
, *guid
))
581 * vmbus_match - Attempt to match the specified device to the specified driver
583 static int vmbus_match(struct device
*device
, struct device_driver
*driver
)
585 struct hv_driver
*drv
= drv_to_hv_drv(driver
);
586 struct hv_device
*hv_dev
= device_to_hv_device(device
);
588 /* The hv_sock driver handles all hv_sock offers. */
589 if (is_hvsock_channel(hv_dev
->channel
))
592 if (hv_vmbus_get_id(drv
->id_table
, &hv_dev
->dev_type
))
599 * vmbus_probe - Add the new vmbus's child device
601 static int vmbus_probe(struct device
*child_device
)
604 struct hv_driver
*drv
=
605 drv_to_hv_drv(child_device
->driver
);
606 struct hv_device
*dev
= device_to_hv_device(child_device
);
607 const struct hv_vmbus_device_id
*dev_id
;
609 dev_id
= hv_vmbus_get_id(drv
->id_table
, &dev
->dev_type
);
611 ret
= drv
->probe(dev
, dev_id
);
613 pr_err("probe failed for device %s (%d)\n",
614 dev_name(child_device
), ret
);
617 pr_err("probe not set for driver %s\n",
618 dev_name(child_device
));
625 * vmbus_remove - Remove a vmbus device
627 static int vmbus_remove(struct device
*child_device
)
629 struct hv_driver
*drv
;
630 struct hv_device
*dev
= device_to_hv_device(child_device
);
632 if (child_device
->driver
) {
633 drv
= drv_to_hv_drv(child_device
->driver
);
643 * vmbus_shutdown - Shutdown a vmbus device
645 static void vmbus_shutdown(struct device
*child_device
)
647 struct hv_driver
*drv
;
648 struct hv_device
*dev
= device_to_hv_device(child_device
);
651 /* The device may not be attached yet */
652 if (!child_device
->driver
)
655 drv
= drv_to_hv_drv(child_device
->driver
);
665 * vmbus_device_release - Final callback release of the vmbus child device
667 static void vmbus_device_release(struct device
*device
)
669 struct hv_device
*hv_dev
= device_to_hv_device(device
);
670 struct vmbus_channel
*channel
= hv_dev
->channel
;
672 hv_process_channel_removal(channel
,
673 channel
->offermsg
.child_relid
);
678 /* The one and only one */
679 static struct bus_type hv_bus
= {
681 .match
= vmbus_match
,
682 .shutdown
= vmbus_shutdown
,
683 .remove
= vmbus_remove
,
684 .probe
= vmbus_probe
,
685 .uevent
= vmbus_uevent
,
686 .dev_groups
= vmbus_groups
,
689 struct onmessage_work_context
{
690 struct work_struct work
;
691 struct hv_message msg
;
694 static void vmbus_onmessage_work(struct work_struct
*work
)
696 struct onmessage_work_context
*ctx
;
698 /* Do not process messages if we're in DISCONNECTED state */
699 if (vmbus_connection
.conn_state
== DISCONNECTED
)
702 ctx
= container_of(work
, struct onmessage_work_context
,
704 vmbus_onmessage(&ctx
->msg
);
708 static void hv_process_timer_expiration(struct hv_message
*msg
, int cpu
)
710 struct clock_event_device
*dev
= hv_context
.clk_evt
[cpu
];
712 if (dev
->event_handler
)
713 dev
->event_handler(dev
);
715 vmbus_signal_eom(msg
, HVMSG_TIMER_EXPIRED
);
718 void vmbus_on_msg_dpc(unsigned long data
)
720 int cpu
= smp_processor_id();
721 void *page_addr
= hv_context
.synic_message_page
[cpu
];
722 struct hv_message
*msg
= (struct hv_message
*)page_addr
+
724 struct vmbus_channel_message_header
*hdr
;
725 struct vmbus_channel_message_table_entry
*entry
;
726 struct onmessage_work_context
*ctx
;
727 u32 message_type
= msg
->header
.message_type
;
729 if (message_type
== HVMSG_NONE
)
733 hdr
= (struct vmbus_channel_message_header
*)msg
->u
.payload
;
735 if (hdr
->msgtype
>= CHANNELMSG_COUNT
) {
736 WARN_ONCE(1, "unknown msgtype=%d\n", hdr
->msgtype
);
740 entry
= &channel_message_table
[hdr
->msgtype
];
741 if (entry
->handler_type
== VMHT_BLOCKING
) {
742 ctx
= kmalloc(sizeof(*ctx
), GFP_ATOMIC
);
746 INIT_WORK(&ctx
->work
, vmbus_onmessage_work
);
747 memcpy(&ctx
->msg
, msg
, sizeof(*msg
));
749 queue_work(vmbus_connection
.work_queue
, &ctx
->work
);
751 entry
->message_handler(hdr
);
754 vmbus_signal_eom(msg
, message_type
);
757 static void vmbus_isr(void)
759 int cpu
= smp_processor_id();
761 struct hv_message
*msg
;
762 union hv_synic_event_flags
*event
;
763 bool handled
= false;
765 page_addr
= hv_context
.synic_event_page
[cpu
];
766 if (page_addr
== NULL
)
769 event
= (union hv_synic_event_flags
*)page_addr
+
772 * Check for events before checking for messages. This is the order
773 * in which events and messages are checked in Windows guests on
774 * Hyper-V, and the Windows team suggested we do the same.
777 if ((vmbus_proto_version
== VERSION_WS2008
) ||
778 (vmbus_proto_version
== VERSION_WIN7
)) {
780 /* Since we are a child, we only need to check bit 0 */
781 if (sync_test_and_clear_bit(0,
782 (unsigned long *) &event
->flags32
[0])) {
787 * Our host is win8 or above. The signaling mechanism
788 * has changed and we can directly look at the event page.
789 * If bit n is set then we have an interrup on the channel
796 tasklet_schedule(hv_context
.event_dpc
[cpu
]);
799 page_addr
= hv_context
.synic_message_page
[cpu
];
800 msg
= (struct hv_message
*)page_addr
+ VMBUS_MESSAGE_SINT
;
802 /* Check if there are actual msgs to be processed */
803 if (msg
->header
.message_type
!= HVMSG_NONE
) {
804 if (msg
->header
.message_type
== HVMSG_TIMER_EXPIRED
)
805 hv_process_timer_expiration(msg
, cpu
);
807 tasklet_schedule(hv_context
.msg_dpc
[cpu
]);
813 * vmbus_bus_init -Main vmbus driver initialization routine.
816 * - initialize the vmbus driver context
817 * - invoke the vmbus hv main init routine
818 * - retrieve the channel offers
820 static int vmbus_bus_init(void)
824 /* Hypervisor initialization...setup hypercall page..etc */
827 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret
);
831 ret
= bus_register(&hv_bus
);
835 hv_setup_vmbus_irq(vmbus_isr
);
837 ret
= hv_synic_alloc();
841 * Initialize the per-cpu interrupt state and
842 * connect to the host.
844 on_each_cpu(hv_synic_init
, NULL
, 1);
845 ret
= vmbus_connect();
849 if (vmbus_proto_version
> VERSION_WIN7
)
850 cpu_hotplug_disable();
853 * Only register if the crash MSRs are available
855 if (ms_hyperv
.misc_features
& HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE
) {
856 register_die_notifier(&hyperv_die_block
);
857 atomic_notifier_chain_register(&panic_notifier_list
,
858 &hyperv_panic_block
);
861 vmbus_request_offers();
866 on_each_cpu(hv_synic_cleanup
, NULL
, 1);
869 hv_remove_vmbus_irq();
871 bus_unregister(&hv_bus
);
880 * __vmbus_child_driver_register() - Register a vmbus's driver
881 * @hv_driver: Pointer to driver structure you want to register
882 * @owner: owner module of the drv
883 * @mod_name: module name string
885 * Registers the given driver with Linux through the 'driver_register()' call
886 * and sets up the hyper-v vmbus handling for this driver.
887 * It will return the state of the 'driver_register()' call.
890 int __vmbus_driver_register(struct hv_driver
*hv_driver
, struct module
*owner
, const char *mod_name
)
894 pr_info("registering driver %s\n", hv_driver
->name
);
896 ret
= vmbus_exists();
900 hv_driver
->driver
.name
= hv_driver
->name
;
901 hv_driver
->driver
.owner
= owner
;
902 hv_driver
->driver
.mod_name
= mod_name
;
903 hv_driver
->driver
.bus
= &hv_bus
;
905 ret
= driver_register(&hv_driver
->driver
);
909 EXPORT_SYMBOL_GPL(__vmbus_driver_register
);
912 * vmbus_driver_unregister() - Unregister a vmbus's driver
913 * @hv_driver: Pointer to driver structure you want to
916 * Un-register the given driver that was previous registered with a call to
917 * vmbus_driver_register()
919 void vmbus_driver_unregister(struct hv_driver
*hv_driver
)
921 pr_info("unregistering driver %s\n", hv_driver
->name
);
924 driver_unregister(&hv_driver
->driver
);
926 EXPORT_SYMBOL_GPL(vmbus_driver_unregister
);
929 * vmbus_device_create - Creates and registers a new child device
932 struct hv_device
*vmbus_device_create(const uuid_le
*type
,
933 const uuid_le
*instance
,
934 struct vmbus_channel
*channel
)
936 struct hv_device
*child_device_obj
;
938 child_device_obj
= kzalloc(sizeof(struct hv_device
), GFP_KERNEL
);
939 if (!child_device_obj
) {
940 pr_err("Unable to allocate device object for child device\n");
944 child_device_obj
->channel
= channel
;
945 memcpy(&child_device_obj
->dev_type
, type
, sizeof(uuid_le
));
946 memcpy(&child_device_obj
->dev_instance
, instance
,
948 child_device_obj
->vendor_id
= 0x1414; /* MSFT vendor ID */
951 return child_device_obj
;
955 * vmbus_device_register - Register the child device
957 int vmbus_device_register(struct hv_device
*child_device_obj
)
961 dev_set_name(&child_device_obj
->device
, "vmbus_%d",
962 child_device_obj
->channel
->id
);
964 child_device_obj
->device
.bus
= &hv_bus
;
965 child_device_obj
->device
.parent
= &hv_acpi_dev
->dev
;
966 child_device_obj
->device
.release
= vmbus_device_release
;
969 * Register with the LDM. This will kick off the driver/device
970 * binding...which will eventually call vmbus_match() and vmbus_probe()
972 ret
= device_register(&child_device_obj
->device
);
975 pr_err("Unable to register child device\n");
977 pr_debug("child device %s registered\n",
978 dev_name(&child_device_obj
->device
));
984 * vmbus_device_unregister - Remove the specified child device
987 void vmbus_device_unregister(struct hv_device
*device_obj
)
989 pr_debug("child device %s unregistered\n",
990 dev_name(&device_obj
->device
));
993 * Kick off the process of unregistering the device.
994 * This will call vmbus_remove() and eventually vmbus_device_release()
996 device_unregister(&device_obj
->device
);
1001 * VMBUS is an acpi enumerated device. Get the information we
1004 #define VTPM_BASE_ADDRESS 0xfed40000
1005 static acpi_status
vmbus_walk_resources(struct acpi_resource
*res
, void *ctx
)
1007 resource_size_t start
= 0;
1008 resource_size_t end
= 0;
1009 struct resource
*new_res
;
1010 struct resource
**old_res
= &hyperv_mmio
;
1011 struct resource
**prev_res
= NULL
;
1013 switch (res
->type
) {
1016 * "Address" descriptors are for bus windows. Ignore
1017 * "memory" descriptors, which are for registers on
1020 case ACPI_RESOURCE_TYPE_ADDRESS32
:
1021 start
= res
->data
.address32
.address
.minimum
;
1022 end
= res
->data
.address32
.address
.maximum
;
1025 case ACPI_RESOURCE_TYPE_ADDRESS64
:
1026 start
= res
->data
.address64
.address
.minimum
;
1027 end
= res
->data
.address64
.address
.maximum
;
1031 /* Unused resource type */
1036 * Ignore ranges that are below 1MB, as they're not
1037 * necessary or useful here.
1042 new_res
= kzalloc(sizeof(*new_res
), GFP_ATOMIC
);
1044 return AE_NO_MEMORY
;
1046 /* If this range overlaps the virtual TPM, truncate it. */
1047 if (end
> VTPM_BASE_ADDRESS
&& start
< VTPM_BASE_ADDRESS
)
1048 end
= VTPM_BASE_ADDRESS
;
1050 new_res
->name
= "hyperv mmio";
1051 new_res
->flags
= IORESOURCE_MEM
;
1052 new_res
->start
= start
;
1056 * If two ranges are adjacent, merge them.
1064 if (((*old_res
)->end
+ 1) == new_res
->start
) {
1065 (*old_res
)->end
= new_res
->end
;
1070 if ((*old_res
)->start
== new_res
->end
+ 1) {
1071 (*old_res
)->start
= new_res
->start
;
1076 if ((*old_res
)->start
> new_res
->end
) {
1077 new_res
->sibling
= *old_res
;
1079 (*prev_res
)->sibling
= new_res
;
1085 old_res
= &(*old_res
)->sibling
;
1092 static int vmbus_acpi_remove(struct acpi_device
*device
)
1094 struct resource
*cur_res
;
1095 struct resource
*next_res
;
1099 __release_region(hyperv_mmio
, fb_mmio
->start
,
1100 resource_size(fb_mmio
));
1104 for (cur_res
= hyperv_mmio
; cur_res
; cur_res
= next_res
) {
1105 next_res
= cur_res
->sibling
;
1113 static void vmbus_reserve_fb(void)
1117 * Make a claim for the frame buffer in the resource tree under the
1118 * first node, which will be the one below 4GB. The length seems to
1119 * be underreported, particularly in a Generation 1 VM. So start out
1120 * reserving a larger area and make it smaller until it succeeds.
1123 if (screen_info
.lfb_base
) {
1124 if (efi_enabled(EFI_BOOT
))
1125 size
= max_t(__u32
, screen_info
.lfb_size
, 0x800000);
1127 size
= max_t(__u32
, screen_info
.lfb_size
, 0x4000000);
1129 for (; !fb_mmio
&& (size
>= 0x100000); size
>>= 1) {
1130 fb_mmio
= __request_region(hyperv_mmio
,
1131 screen_info
.lfb_base
, size
,
1138 * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1139 * @new: If successful, supplied a pointer to the
1140 * allocated MMIO space.
1141 * @device_obj: Identifies the caller
1142 * @min: Minimum guest physical address of the
1144 * @max: Maximum guest physical address
1145 * @size: Size of the range to be allocated
1146 * @align: Alignment of the range to be allocated
1147 * @fb_overlap_ok: Whether this allocation can be allowed
1148 * to overlap the video frame buffer.
1150 * This function walks the resources granted to VMBus by the
1151 * _CRS object in the ACPI namespace underneath the parent
1152 * "bridge" whether that's a root PCI bus in the Generation 1
1153 * case or a Module Device in the Generation 2 case. It then
1154 * attempts to allocate from the global MMIO pool in a way that
1155 * matches the constraints supplied in these parameters and by
1158 * Return: 0 on success, -errno on failure
1160 int vmbus_allocate_mmio(struct resource
**new, struct hv_device
*device_obj
,
1161 resource_size_t min
, resource_size_t max
,
1162 resource_size_t size
, resource_size_t align
,
1165 struct resource
*iter
, *shadow
;
1166 resource_size_t range_min
, range_max
, start
;
1167 const char *dev_n
= dev_name(&device_obj
->device
);
1171 down(&hyperv_mmio_lock
);
1174 * If overlaps with frame buffers are allowed, then first attempt to
1175 * make the allocation from within the reserved region. Because it
1176 * is already reserved, no shadow allocation is necessary.
1178 if (fb_overlap_ok
&& fb_mmio
&& !(min
> fb_mmio
->end
) &&
1179 !(max
< fb_mmio
->start
)) {
1181 range_min
= fb_mmio
->start
;
1182 range_max
= fb_mmio
->end
;
1183 start
= (range_min
+ align
- 1) & ~(align
- 1);
1184 for (; start
+ size
- 1 <= range_max
; start
+= align
) {
1185 *new = request_mem_region_exclusive(start
, size
, dev_n
);
1193 for (iter
= hyperv_mmio
; iter
; iter
= iter
->sibling
) {
1194 if ((iter
->start
>= max
) || (iter
->end
<= min
))
1197 range_min
= iter
->start
;
1198 range_max
= iter
->end
;
1199 start
= (range_min
+ align
- 1) & ~(align
- 1);
1200 for (; start
+ size
- 1 <= range_max
; start
+= align
) {
1201 shadow
= __request_region(iter
, start
, size
, NULL
,
1206 *new = request_mem_region_exclusive(start
, size
, dev_n
);
1208 shadow
->name
= (char *)*new;
1213 __release_region(iter
, start
, size
);
1218 up(&hyperv_mmio_lock
);
1221 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio
);
1224 * vmbus_free_mmio() - Free a memory-mapped I/O range.
1225 * @start: Base address of region to release.
1226 * @size: Size of the range to be allocated
1228 * This function releases anything requested by
1229 * vmbus_mmio_allocate().
1231 void vmbus_free_mmio(resource_size_t start
, resource_size_t size
)
1233 struct resource
*iter
;
1235 down(&hyperv_mmio_lock
);
1236 for (iter
= hyperv_mmio
; iter
; iter
= iter
->sibling
) {
1237 if ((iter
->start
>= start
+ size
) || (iter
->end
<= start
))
1240 __release_region(iter
, start
, size
);
1242 release_mem_region(start
, size
);
1243 up(&hyperv_mmio_lock
);
1246 EXPORT_SYMBOL_GPL(vmbus_free_mmio
);
1249 * vmbus_cpu_number_to_vp_number() - Map CPU to VP.
1250 * @cpu_number: CPU number in Linux terms
1252 * This function returns the mapping between the Linux processor
1253 * number and the hypervisor's virtual processor number, useful
1254 * in making hypercalls and such that talk about specific
1257 * Return: Virtual processor number in Hyper-V terms
1259 int vmbus_cpu_number_to_vp_number(int cpu_number
)
1261 return hv_context
.vp_index
[cpu_number
];
1263 EXPORT_SYMBOL_GPL(vmbus_cpu_number_to_vp_number
);
1265 static int vmbus_acpi_add(struct acpi_device
*device
)
1268 int ret_val
= -ENODEV
;
1269 struct acpi_device
*ancestor
;
1271 hv_acpi_dev
= device
;
1273 result
= acpi_walk_resources(device
->handle
, METHOD_NAME__CRS
,
1274 vmbus_walk_resources
, NULL
);
1276 if (ACPI_FAILURE(result
))
1279 * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1280 * firmware) is the VMOD that has the mmio ranges. Get that.
1282 for (ancestor
= device
->parent
; ancestor
; ancestor
= ancestor
->parent
) {
1283 result
= acpi_walk_resources(ancestor
->handle
, METHOD_NAME__CRS
,
1284 vmbus_walk_resources
, NULL
);
1286 if (ACPI_FAILURE(result
))
1296 complete(&probe_event
);
1298 vmbus_acpi_remove(device
);
1302 static const struct acpi_device_id vmbus_acpi_device_ids
[] = {
1307 MODULE_DEVICE_TABLE(acpi
, vmbus_acpi_device_ids
);
1309 static struct acpi_driver vmbus_acpi_driver
= {
1311 .ids
= vmbus_acpi_device_ids
,
1313 .add
= vmbus_acpi_add
,
1314 .remove
= vmbus_acpi_remove
,
1318 static void hv_kexec_handler(void)
1322 hv_synic_clockevents_cleanup();
1323 vmbus_initiate_unload(false);
1324 for_each_online_cpu(cpu
)
1325 smp_call_function_single(cpu
, hv_synic_cleanup
, NULL
, 1);
1329 static void hv_crash_handler(struct pt_regs
*regs
)
1331 vmbus_initiate_unload(true);
1333 * In crash handler we can't schedule synic cleanup for all CPUs,
1334 * doing the cleanup for current CPU only. This should be sufficient
1337 hv_synic_cleanup(NULL
);
1341 static int __init
hv_acpi_init(void)
1345 if (x86_hyper
!= &x86_hyper_ms_hyperv
)
1348 init_completion(&probe_event
);
1351 * Get ACPI resources first.
1353 ret
= acpi_bus_register_driver(&vmbus_acpi_driver
);
1358 t
= wait_for_completion_timeout(&probe_event
, 5*HZ
);
1364 ret
= vmbus_bus_init();
1368 hv_setup_kexec_handler(hv_kexec_handler
);
1369 hv_setup_crash_handler(hv_crash_handler
);
1374 acpi_bus_unregister_driver(&vmbus_acpi_driver
);
1379 static void __exit
vmbus_exit(void)
1383 hv_remove_kexec_handler();
1384 hv_remove_crash_handler();
1385 vmbus_connection
.conn_state
= DISCONNECTED
;
1386 hv_synic_clockevents_cleanup();
1388 hv_remove_vmbus_irq();
1389 for_each_online_cpu(cpu
)
1390 tasklet_kill(hv_context
.msg_dpc
[cpu
]);
1391 vmbus_free_channels();
1392 if (ms_hyperv
.misc_features
& HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE
) {
1393 unregister_die_notifier(&hyperv_die_block
);
1394 atomic_notifier_chain_unregister(&panic_notifier_list
,
1395 &hyperv_panic_block
);
1397 bus_unregister(&hv_bus
);
1399 for_each_online_cpu(cpu
) {
1400 tasklet_kill(hv_context
.event_dpc
[cpu
]);
1401 smp_call_function_single(cpu
, hv_synic_cleanup
, NULL
, 1);
1404 acpi_bus_unregister_driver(&vmbus_acpi_driver
);
1405 if (vmbus_proto_version
> VERSION_WIN7
)
1406 cpu_hotplug_enable();
1410 MODULE_LICENSE("GPL");
1412 subsys_initcall(hv_acpi_init
);
1413 module_exit(vmbus_exit
);