1 // SPDX-License-Identifier: GPL-2.0-only
4 * Copyright (c) 2009, Microsoft Corporation.
7 * Haiyang Zhang <haiyangz@microsoft.com>
8 * Hank Janssen <hjanssen@microsoft.com>
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/kernel.h>
13 #include <linux/sched.h>
14 #include <linux/wait.h>
15 #include <linux/delay.h>
17 #include <linux/module.h>
18 #include <linux/slab.h>
19 #include <linux/vmalloc.h>
20 #include <linux/hyperv.h>
21 #include <linux/export.h>
22 #include <asm/mshyperv.h>
24 #include "hyperv_vmbus.h"
27 struct vmbus_connection vmbus_connection
= {
28 .conn_state
= DISCONNECTED
,
29 .next_gpadl_handle
= ATOMIC_INIT(0xE1E10),
31 .ready_for_suspend_event
= COMPLETION_INITIALIZER(
32 vmbus_connection
.ready_for_suspend_event
),
33 .ready_for_resume_event
= COMPLETION_INITIALIZER(
34 vmbus_connection
.ready_for_resume_event
),
36 EXPORT_SYMBOL_GPL(vmbus_connection
);
39 * Negotiated protocol version with the host.
41 __u32 vmbus_proto_version
;
42 EXPORT_SYMBOL_GPL(vmbus_proto_version
);
45 * Table of VMBus versions listed from newest to oldest.
47 static __u32 vmbus_versions
[] = {
60 * Maximal VMBus protocol version guests can negotiate. Useful to cap the
61 * VMBus version for testing and debugging purpose.
63 static uint max_version
= VERSION_WIN10_V5_2
;
65 module_param(max_version
, uint
, S_IRUGO
);
66 MODULE_PARM_DESC(max_version
,
67 "Maximal VMBus protocol version which can be negotiated");
69 int vmbus_negotiate_version(struct vmbus_channel_msginfo
*msginfo
, u32 version
)
73 struct vmbus_channel_initiate_contact
*msg
;
76 init_completion(&msginfo
->waitevent
);
78 msg
= (struct vmbus_channel_initiate_contact
*)msginfo
->msg
;
80 memset(msg
, 0, sizeof(*msg
));
81 msg
->header
.msgtype
= CHANNELMSG_INITIATE_CONTACT
;
82 msg
->vmbus_version_requested
= version
;
85 * VMBus protocol 5.0 (VERSION_WIN10_V5) and higher require that we must
86 * use VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
87 * and for subsequent messages, we must use the Message Connection ID
88 * field in the host-returned Version Response Message. And, with
89 * VERSION_WIN10_V5 and higher, we don't use msg->interrupt_page, but we
90 * tell the host explicitly that we still use VMBUS_MESSAGE_SINT(2) for
93 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
95 if (version
>= VERSION_WIN10_V5
) {
96 msg
->msg_sint
= VMBUS_MESSAGE_SINT
;
97 vmbus_connection
.msg_conn_id
= VMBUS_MESSAGE_CONNECTION_ID_4
;
99 msg
->interrupt_page
= virt_to_phys(vmbus_connection
.int_page
);
100 vmbus_connection
.msg_conn_id
= VMBUS_MESSAGE_CONNECTION_ID
;
103 msg
->monitor_page1
= virt_to_phys(vmbus_connection
.monitor_pages
[0]);
104 msg
->monitor_page2
= virt_to_phys(vmbus_connection
.monitor_pages
[1]);
106 * We want all channel messages to be delivered on CPU 0.
107 * This has been the behavior pre-win8. This is not
108 * perf issue and having all channel messages delivered on CPU 0
110 * For post win8 hosts, we support receiving channel messagges on
111 * all the CPUs. This is needed for kexec to work correctly where
112 * the CPU attempting to connect may not be CPU 0.
114 if (version
>= VERSION_WIN8_1
) {
116 msg
->target_vcpu
= hv_cpu_number_to_vp_number(cur_cpu
);
117 vmbus_connection
.connect_cpu
= cur_cpu
;
120 msg
->target_vcpu
= 0;
121 vmbus_connection
.connect_cpu
= 0;
125 * Add to list before we send the request since we may
126 * receive the response before returning from this routine
128 spin_lock_irqsave(&vmbus_connection
.channelmsg_lock
, flags
);
129 list_add_tail(&msginfo
->msglistentry
,
130 &vmbus_connection
.chn_msg_list
);
132 spin_unlock_irqrestore(&vmbus_connection
.channelmsg_lock
, flags
);
134 ret
= vmbus_post_msg(msg
,
135 sizeof(struct vmbus_channel_initiate_contact
),
138 trace_vmbus_negotiate_version(msg
, ret
);
141 spin_lock_irqsave(&vmbus_connection
.channelmsg_lock
, flags
);
142 list_del(&msginfo
->msglistentry
);
143 spin_unlock_irqrestore(&vmbus_connection
.channelmsg_lock
,
148 /* Wait for the connection response */
149 wait_for_completion(&msginfo
->waitevent
);
151 spin_lock_irqsave(&vmbus_connection
.channelmsg_lock
, flags
);
152 list_del(&msginfo
->msglistentry
);
153 spin_unlock_irqrestore(&vmbus_connection
.channelmsg_lock
, flags
);
155 /* Check if successful */
156 if (msginfo
->response
.version_response
.version_supported
) {
157 vmbus_connection
.conn_state
= CONNECTED
;
159 if (version
>= VERSION_WIN10_V5
)
160 vmbus_connection
.msg_conn_id
=
161 msginfo
->response
.version_response
.msg_conn_id
;
163 return -ECONNREFUSED
;
170 * vmbus_connect - Sends a connect request on the partition service connection
172 int vmbus_connect(void)
174 struct vmbus_channel_msginfo
*msginfo
= NULL
;
178 /* Initialize the vmbus connection */
179 vmbus_connection
.conn_state
= CONNECTING
;
180 vmbus_connection
.work_queue
= create_workqueue("hv_vmbus_con");
181 if (!vmbus_connection
.work_queue
) {
186 vmbus_connection
.handle_primary_chan_wq
=
187 create_workqueue("hv_pri_chan");
188 if (!vmbus_connection
.handle_primary_chan_wq
) {
193 vmbus_connection
.handle_sub_chan_wq
=
194 create_workqueue("hv_sub_chan");
195 if (!vmbus_connection
.handle_sub_chan_wq
) {
200 INIT_LIST_HEAD(&vmbus_connection
.chn_msg_list
);
201 spin_lock_init(&vmbus_connection
.channelmsg_lock
);
203 INIT_LIST_HEAD(&vmbus_connection
.chn_list
);
204 mutex_init(&vmbus_connection
.channel_mutex
);
207 * Setup the vmbus event connection for channel interrupt
210 vmbus_connection
.int_page
=
211 (void *)hv_alloc_hyperv_zeroed_page();
212 if (vmbus_connection
.int_page
== NULL
) {
217 vmbus_connection
.recv_int_page
= vmbus_connection
.int_page
;
218 vmbus_connection
.send_int_page
=
219 (void *)((unsigned long)vmbus_connection
.int_page
+
220 (HV_HYP_PAGE_SIZE
>> 1));
223 * Setup the monitor notification facility. The 1st page for
224 * parent->child and the 2nd page for child->parent
226 vmbus_connection
.monitor_pages
[0] = (void *)hv_alloc_hyperv_zeroed_page();
227 vmbus_connection
.monitor_pages
[1] = (void *)hv_alloc_hyperv_zeroed_page();
228 if ((vmbus_connection
.monitor_pages
[0] == NULL
) ||
229 (vmbus_connection
.monitor_pages
[1] == NULL
)) {
234 msginfo
= kzalloc(sizeof(*msginfo
) +
235 sizeof(struct vmbus_channel_initiate_contact
),
237 if (msginfo
== NULL
) {
243 * Negotiate a compatible VMBUS version number with the
244 * host. We start with the highest number we can support
245 * and work our way down until we negotiate a compatible
250 if (i
== ARRAY_SIZE(vmbus_versions
))
253 version
= vmbus_versions
[i
];
254 if (version
> max_version
)
257 ret
= vmbus_negotiate_version(msginfo
, version
);
258 if (ret
== -ETIMEDOUT
)
261 if (vmbus_connection
.conn_state
== CONNECTED
)
265 vmbus_proto_version
= version
;
266 pr_info("Vmbus version:%d.%d\n",
267 version
>> 16, version
& 0xFFFF);
273 pr_err("Unable to connect to host\n");
275 vmbus_connection
.conn_state
= DISCONNECTED
;
283 void vmbus_disconnect(void)
286 * First send the unload request to the host.
288 vmbus_initiate_unload(false);
290 if (vmbus_connection
.handle_sub_chan_wq
)
291 destroy_workqueue(vmbus_connection
.handle_sub_chan_wq
);
293 if (vmbus_connection
.handle_primary_chan_wq
)
294 destroy_workqueue(vmbus_connection
.handle_primary_chan_wq
);
296 if (vmbus_connection
.work_queue
)
297 destroy_workqueue(vmbus_connection
.work_queue
);
299 if (vmbus_connection
.int_page
) {
300 hv_free_hyperv_page((unsigned long)vmbus_connection
.int_page
);
301 vmbus_connection
.int_page
= NULL
;
304 hv_free_hyperv_page((unsigned long)vmbus_connection
.monitor_pages
[0]);
305 hv_free_hyperv_page((unsigned long)vmbus_connection
.monitor_pages
[1]);
306 vmbus_connection
.monitor_pages
[0] = NULL
;
307 vmbus_connection
.monitor_pages
[1] = NULL
;
311 * relid2channel - Get the channel object given its
312 * child relative id (ie channel id)
314 struct vmbus_channel
*relid2channel(u32 relid
)
316 struct vmbus_channel
*channel
;
317 struct vmbus_channel
*found_channel
= NULL
;
318 struct list_head
*cur
, *tmp
;
319 struct vmbus_channel
*cur_sc
;
321 BUG_ON(!mutex_is_locked(&vmbus_connection
.channel_mutex
));
323 list_for_each_entry(channel
, &vmbus_connection
.chn_list
, listentry
) {
324 if (channel
->offermsg
.child_relid
== relid
) {
325 found_channel
= channel
;
327 } else if (!list_empty(&channel
->sc_list
)) {
329 * Deal with sub-channels.
331 list_for_each_safe(cur
, tmp
, &channel
->sc_list
) {
332 cur_sc
= list_entry(cur
, struct vmbus_channel
,
334 if (cur_sc
->offermsg
.child_relid
== relid
) {
335 found_channel
= cur_sc
;
342 return found_channel
;
346 * vmbus_on_event - Process a channel event notification
348 * For batched channels (default) optimize host to guest signaling
350 * 1. While reading the channel, we disable interrupts from host.
351 * 2. Ensure that we process all posted messages from the host
352 * before returning from this callback.
353 * 3. Once we return, enable signaling from the host. Once this
354 * state is set we check to see if additional packets are
355 * available to read. In this case we repeat the process.
356 * If this tasklet has been running for a long time
357 * then reschedule ourselves.
359 void vmbus_on_event(unsigned long data
)
361 struct vmbus_channel
*channel
= (void *) data
;
362 unsigned long time_limit
= jiffies
+ 2;
364 trace_vmbus_on_event(channel
);
366 hv_debug_delay_test(channel
, INTERRUPT_DELAY
);
368 void (*callback_fn
)(void *);
370 /* A channel once created is persistent even when
371 * there is no driver handling the device. An
372 * unloading driver sets the onchannel_callback to NULL.
374 callback_fn
= READ_ONCE(channel
->onchannel_callback
);
375 if (unlikely(callback_fn
== NULL
))
378 (*callback_fn
)(channel
->channel_callback_context
);
380 if (channel
->callback_mode
!= HV_CALL_BATCHED
)
383 if (likely(hv_end_read(&channel
->inbound
) == 0))
386 hv_begin_read(&channel
->inbound
);
387 } while (likely(time_before(jiffies
, time_limit
)));
389 /* The time limit (2 jiffies) has been reached */
390 tasklet_schedule(&channel
->callback_event
);
394 * vmbus_post_msg - Send a msg on the vmbus's message connection
396 int vmbus_post_msg(void *buffer
, size_t buflen
, bool can_sleep
)
398 struct vmbus_channel_message_header
*hdr
;
399 union hv_connection_id conn_id
;
405 conn_id
.u
.id
= vmbus_connection
.msg_conn_id
;
408 * hv_post_message() can have transient failures because of
409 * insufficient resources. Retry the operation a couple of
410 * times before giving up.
412 while (retries
< 100) {
413 ret
= hv_post_message(conn_id
, 1, buffer
, buflen
);
416 case HV_STATUS_INVALID_CONNECTION_ID
:
418 * See vmbus_negotiate_version(): VMBus protocol 5.0
419 * and higher require that we must use
420 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate
421 * Contact message, but on old hosts that only
422 * support VMBus protocol 4.0 or lower, here we get
423 * HV_STATUS_INVALID_CONNECTION_ID and we should
424 * return an error immediately without retrying.
427 if (hdr
->msgtype
== CHANNELMSG_INITIATE_CONTACT
)
430 * We could get this if we send messages too
435 case HV_STATUS_INSUFFICIENT_MEMORY
:
436 case HV_STATUS_INSUFFICIENT_BUFFERS
:
439 case HV_STATUS_SUCCESS
:
442 pr_err("hv_post_msg() failed; error code:%d\n", ret
);
447 if (can_sleep
&& usec
> 1000)
449 else if (usec
< MAX_UDELAY_MS
* 1000)
461 * vmbus_set_event - Send an event notification to the parent
463 void vmbus_set_event(struct vmbus_channel
*channel
)
465 u32 child_relid
= channel
->offermsg
.child_relid
;
467 if (!channel
->is_dedicated_interrupt
)
468 vmbus_send_interrupt(child_relid
);
470 ++channel
->sig_events
;
472 hv_do_fast_hypercall8(HVCALL_SIGNAL_EVENT
, channel
->sig_event
);
474 EXPORT_SYMBOL_GPL(vmbus_set_event
);