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 .unload_event = COMPLETION_INITIALIZER(
30 vmbus_connection.unload_event),
31 .next_gpadl_handle = ATOMIC_INIT(0xE1E10),
33 .ready_for_suspend_event = COMPLETION_INITIALIZER(
34 vmbus_connection.ready_for_suspend_event),
35 .ready_for_resume_event = COMPLETION_INITIALIZER(
36 vmbus_connection.ready_for_resume_event),
38 EXPORT_SYMBOL_GPL(vmbus_connection);
41 * Negotiated protocol version with the host.
43 __u32 vmbus_proto_version;
44 EXPORT_SYMBOL_GPL(vmbus_proto_version);
47 * Table of VMBus versions listed from newest to oldest.
49 static __u32 vmbus_versions[] = {
63 * Maximal VMBus protocol version guests can negotiate. Useful to cap the
64 * VMBus version for testing and debugging purpose.
66 static uint max_version = VERSION_WIN10_V5_3;
68 module_param(max_version, uint, S_IRUGO);
69 MODULE_PARM_DESC(max_version,
70 "Maximal VMBus protocol version which can be negotiated");
72 int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo, u32 version)
75 struct vmbus_channel_initiate_contact *msg;
78 init_completion(&msginfo->waitevent);
80 msg = (struct vmbus_channel_initiate_contact *)msginfo->msg;
82 memset(msg, 0, sizeof(*msg));
83 msg->header.msgtype = CHANNELMSG_INITIATE_CONTACT;
84 msg->vmbus_version_requested = version;
87 * VMBus protocol 5.0 (VERSION_WIN10_V5) and higher require that we must
88 * use VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
89 * and for subsequent messages, we must use the Message Connection ID
90 * field in the host-returned Version Response Message. And, with
91 * VERSION_WIN10_V5 and higher, we don't use msg->interrupt_page, but we
92 * tell the host explicitly that we still use VMBUS_MESSAGE_SINT(2) for
95 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
97 if (version >= VERSION_WIN10_V5) {
98 msg->msg_sint = VMBUS_MESSAGE_SINT;
99 vmbus_connection.msg_conn_id = VMBUS_MESSAGE_CONNECTION_ID_4;
101 msg->interrupt_page = virt_to_phys(vmbus_connection.int_page);
102 vmbus_connection.msg_conn_id = VMBUS_MESSAGE_CONNECTION_ID;
105 msg->monitor_page1 = virt_to_phys(vmbus_connection.monitor_pages[0]);
106 msg->monitor_page2 = virt_to_phys(vmbus_connection.monitor_pages[1]);
107 msg->target_vcpu = hv_cpu_number_to_vp_number(VMBUS_CONNECT_CPU);
110 * Add to list before we send the request since we may
111 * receive the response before returning from this routine
113 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
114 list_add_tail(&msginfo->msglistentry,
115 &vmbus_connection.chn_msg_list);
117 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
119 ret = vmbus_post_msg(msg,
120 sizeof(struct vmbus_channel_initiate_contact),
123 trace_vmbus_negotiate_version(msg, ret);
126 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
127 list_del(&msginfo->msglistentry);
128 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock,
133 /* Wait for the connection response */
134 wait_for_completion(&msginfo->waitevent);
136 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
137 list_del(&msginfo->msglistentry);
138 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
140 /* Check if successful */
141 if (msginfo->response.version_response.version_supported) {
142 vmbus_connection.conn_state = CONNECTED;
144 if (version >= VERSION_WIN10_V5)
145 vmbus_connection.msg_conn_id =
146 msginfo->response.version_response.msg_conn_id;
148 return -ECONNREFUSED;
155 * vmbus_connect - Sends a connect request on the partition service connection
157 int vmbus_connect(void)
159 struct vmbus_channel_msginfo *msginfo = NULL;
163 /* Initialize the vmbus connection */
164 vmbus_connection.conn_state = CONNECTING;
165 vmbus_connection.work_queue = create_workqueue("hv_vmbus_con");
166 if (!vmbus_connection.work_queue) {
171 vmbus_connection.handle_primary_chan_wq =
172 create_workqueue("hv_pri_chan");
173 if (!vmbus_connection.handle_primary_chan_wq) {
178 vmbus_connection.handle_sub_chan_wq =
179 create_workqueue("hv_sub_chan");
180 if (!vmbus_connection.handle_sub_chan_wq) {
185 INIT_LIST_HEAD(&vmbus_connection.chn_msg_list);
186 spin_lock_init(&vmbus_connection.channelmsg_lock);
188 INIT_LIST_HEAD(&vmbus_connection.chn_list);
189 mutex_init(&vmbus_connection.channel_mutex);
192 * Setup the vmbus event connection for channel interrupt
195 vmbus_connection.int_page =
196 (void *)hv_alloc_hyperv_zeroed_page();
197 if (vmbus_connection.int_page == NULL) {
202 vmbus_connection.recv_int_page = vmbus_connection.int_page;
203 vmbus_connection.send_int_page =
204 (void *)((unsigned long)vmbus_connection.int_page +
205 (HV_HYP_PAGE_SIZE >> 1));
208 * Setup the monitor notification facility. The 1st page for
209 * parent->child and the 2nd page for child->parent
211 vmbus_connection.monitor_pages[0] = (void *)hv_alloc_hyperv_zeroed_page();
212 vmbus_connection.monitor_pages[1] = (void *)hv_alloc_hyperv_zeroed_page();
213 if ((vmbus_connection.monitor_pages[0] == NULL) ||
214 (vmbus_connection.monitor_pages[1] == NULL)) {
219 msginfo = kzalloc(sizeof(*msginfo) +
220 sizeof(struct vmbus_channel_initiate_contact),
222 if (msginfo == NULL) {
228 * Negotiate a compatible VMBUS version number with the
229 * host. We start with the highest number we can support
230 * and work our way down until we negotiate a compatible
235 if (i == ARRAY_SIZE(vmbus_versions)) {
240 version = vmbus_versions[i];
241 if (version > max_version)
244 ret = vmbus_negotiate_version(msginfo, version);
245 if (ret == -ETIMEDOUT)
248 if (vmbus_connection.conn_state == CONNECTED)
252 if (hv_is_isolation_supported() && version < VERSION_WIN10_V5_2) {
253 pr_err("Invalid VMBus version %d.%d (expected >= %d.%d) from the host supporting isolation\n",
254 version >> 16, version & 0xFFFF, VERSION_WIN10_V5_2 >> 16, VERSION_WIN10_V5_2 & 0xFFFF);
259 vmbus_proto_version = version;
260 pr_info("Vmbus version:%d.%d\n",
261 version >> 16, version & 0xFFFF);
263 vmbus_connection.channels = kcalloc(MAX_CHANNEL_RELIDS,
264 sizeof(struct vmbus_channel *),
266 if (vmbus_connection.channels == NULL) {
275 pr_err("Unable to connect to host\n");
277 vmbus_connection.conn_state = DISCONNECTED;
285 void vmbus_disconnect(void)
288 * First send the unload request to the host.
290 vmbus_initiate_unload(false);
292 if (vmbus_connection.handle_sub_chan_wq)
293 destroy_workqueue(vmbus_connection.handle_sub_chan_wq);
295 if (vmbus_connection.handle_primary_chan_wq)
296 destroy_workqueue(vmbus_connection.handle_primary_chan_wq);
298 if (vmbus_connection.work_queue)
299 destroy_workqueue(vmbus_connection.work_queue);
301 if (vmbus_connection.int_page) {
302 hv_free_hyperv_page((unsigned long)vmbus_connection.int_page);
303 vmbus_connection.int_page = NULL;
306 hv_free_hyperv_page((unsigned long)vmbus_connection.monitor_pages[0]);
307 hv_free_hyperv_page((unsigned long)vmbus_connection.monitor_pages[1]);
308 vmbus_connection.monitor_pages[0] = NULL;
309 vmbus_connection.monitor_pages[1] = NULL;
313 * relid2channel - Get the channel object given its
314 * child relative id (ie channel id)
316 struct vmbus_channel *relid2channel(u32 relid)
318 if (WARN_ON(relid >= MAX_CHANNEL_RELIDS))
320 return READ_ONCE(vmbus_connection.channels[relid]);
324 * vmbus_on_event - Process a channel event notification
326 * For batched channels (default) optimize host to guest signaling
328 * 1. While reading the channel, we disable interrupts from host.
329 * 2. Ensure that we process all posted messages from the host
330 * before returning from this callback.
331 * 3. Once we return, enable signaling from the host. Once this
332 * state is set we check to see if additional packets are
333 * available to read. In this case we repeat the process.
334 * If this tasklet has been running for a long time
335 * then reschedule ourselves.
337 void vmbus_on_event(unsigned long data)
339 struct vmbus_channel *channel = (void *) data;
340 unsigned long time_limit = jiffies + 2;
342 trace_vmbus_on_event(channel);
344 hv_debug_delay_test(channel, INTERRUPT_DELAY);
346 void (*callback_fn)(void *);
348 /* A channel once created is persistent even when
349 * there is no driver handling the device. An
350 * unloading driver sets the onchannel_callback to NULL.
352 callback_fn = READ_ONCE(channel->onchannel_callback);
353 if (unlikely(callback_fn == NULL))
356 (*callback_fn)(channel->channel_callback_context);
358 if (channel->callback_mode != HV_CALL_BATCHED)
361 if (likely(hv_end_read(&channel->inbound) == 0))
364 hv_begin_read(&channel->inbound);
365 } while (likely(time_before(jiffies, time_limit)));
367 /* The time limit (2 jiffies) has been reached */
368 tasklet_schedule(&channel->callback_event);
372 * vmbus_post_msg - Send a msg on the vmbus's message connection
374 int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep)
376 struct vmbus_channel_message_header *hdr;
377 union hv_connection_id conn_id;
383 conn_id.u.id = vmbus_connection.msg_conn_id;
386 * hv_post_message() can have transient failures because of
387 * insufficient resources. Retry the operation a couple of
388 * times before giving up.
390 while (retries < 100) {
391 ret = hv_post_message(conn_id, 1, buffer, buflen);
394 case HV_STATUS_INVALID_CONNECTION_ID:
396 * See vmbus_negotiate_version(): VMBus protocol 5.0
397 * and higher require that we must use
398 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate
399 * Contact message, but on old hosts that only
400 * support VMBus protocol 4.0 or lower, here we get
401 * HV_STATUS_INVALID_CONNECTION_ID and we should
402 * return an error immediately without retrying.
405 if (hdr->msgtype == CHANNELMSG_INITIATE_CONTACT)
408 * We could get this if we send messages too
413 case HV_STATUS_INSUFFICIENT_MEMORY:
414 case HV_STATUS_INSUFFICIENT_BUFFERS:
417 case HV_STATUS_SUCCESS:
420 pr_err("hv_post_msg() failed; error code:%d\n", ret);
425 if (can_sleep && usec > 1000)
427 else if (usec < MAX_UDELAY_MS * 1000)
439 * vmbus_set_event - Send an event notification to the parent
441 void vmbus_set_event(struct vmbus_channel *channel)
443 u32 child_relid = channel->offermsg.child_relid;
445 if (!channel->is_dedicated_interrupt)
446 vmbus_send_interrupt(child_relid);
448 ++channel->sig_events;
450 hv_do_fast_hypercall8(HVCALL_SIGNAL_EVENT, channel->sig_event);
452 EXPORT_SYMBOL_GPL(vmbus_set_event);