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)
72 struct vmbus_channel_initiate_contact *msg;
75 init_completion(&msginfo->waitevent);
77 msg = (struct vmbus_channel_initiate_contact *)msginfo->msg;
79 memset(msg, 0, sizeof(*msg));
80 msg->header.msgtype = CHANNELMSG_INITIATE_CONTACT;
81 msg->vmbus_version_requested = version;
84 * VMBus protocol 5.0 (VERSION_WIN10_V5) and higher require that we must
85 * use VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
86 * and for subsequent messages, we must use the Message Connection ID
87 * field in the host-returned Version Response Message. And, with
88 * VERSION_WIN10_V5 and higher, we don't use msg->interrupt_page, but we
89 * tell the host explicitly that we still use VMBUS_MESSAGE_SINT(2) for
92 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
94 if (version >= VERSION_WIN10_V5) {
95 msg->msg_sint = VMBUS_MESSAGE_SINT;
96 vmbus_connection.msg_conn_id = VMBUS_MESSAGE_CONNECTION_ID_4;
98 msg->interrupt_page = virt_to_phys(vmbus_connection.int_page);
99 vmbus_connection.msg_conn_id = VMBUS_MESSAGE_CONNECTION_ID;
102 msg->monitor_page1 = virt_to_phys(vmbus_connection.monitor_pages[0]);
103 msg->monitor_page2 = virt_to_phys(vmbus_connection.monitor_pages[1]);
104 msg->target_vcpu = hv_cpu_number_to_vp_number(VMBUS_CONNECT_CPU);
107 * Add to list before we send the request since we may
108 * receive the response before returning from this routine
110 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
111 list_add_tail(&msginfo->msglistentry,
112 &vmbus_connection.chn_msg_list);
114 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
116 ret = vmbus_post_msg(msg,
117 sizeof(struct vmbus_channel_initiate_contact),
120 trace_vmbus_negotiate_version(msg, ret);
123 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
124 list_del(&msginfo->msglistentry);
125 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock,
130 /* Wait for the connection response */
131 wait_for_completion(&msginfo->waitevent);
133 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
134 list_del(&msginfo->msglistentry);
135 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
137 /* Check if successful */
138 if (msginfo->response.version_response.version_supported) {
139 vmbus_connection.conn_state = CONNECTED;
141 if (version >= VERSION_WIN10_V5)
142 vmbus_connection.msg_conn_id =
143 msginfo->response.version_response.msg_conn_id;
145 return -ECONNREFUSED;
152 * vmbus_connect - Sends a connect request on the partition service connection
154 int vmbus_connect(void)
156 struct vmbus_channel_msginfo *msginfo = NULL;
160 /* Initialize the vmbus connection */
161 vmbus_connection.conn_state = CONNECTING;
162 vmbus_connection.work_queue = create_workqueue("hv_vmbus_con");
163 if (!vmbus_connection.work_queue) {
168 vmbus_connection.handle_primary_chan_wq =
169 create_workqueue("hv_pri_chan");
170 if (!vmbus_connection.handle_primary_chan_wq) {
175 vmbus_connection.handle_sub_chan_wq =
176 create_workqueue("hv_sub_chan");
177 if (!vmbus_connection.handle_sub_chan_wq) {
182 INIT_LIST_HEAD(&vmbus_connection.chn_msg_list);
183 spin_lock_init(&vmbus_connection.channelmsg_lock);
185 INIT_LIST_HEAD(&vmbus_connection.chn_list);
186 mutex_init(&vmbus_connection.channel_mutex);
189 * Setup the vmbus event connection for channel interrupt
192 vmbus_connection.int_page =
193 (void *)hv_alloc_hyperv_zeroed_page();
194 if (vmbus_connection.int_page == NULL) {
199 vmbus_connection.recv_int_page = vmbus_connection.int_page;
200 vmbus_connection.send_int_page =
201 (void *)((unsigned long)vmbus_connection.int_page +
202 (HV_HYP_PAGE_SIZE >> 1));
205 * Setup the monitor notification facility. The 1st page for
206 * parent->child and the 2nd page for child->parent
208 vmbus_connection.monitor_pages[0] = (void *)hv_alloc_hyperv_zeroed_page();
209 vmbus_connection.monitor_pages[1] = (void *)hv_alloc_hyperv_zeroed_page();
210 if ((vmbus_connection.monitor_pages[0] == NULL) ||
211 (vmbus_connection.monitor_pages[1] == NULL)) {
216 msginfo = kzalloc(sizeof(*msginfo) +
217 sizeof(struct vmbus_channel_initiate_contact),
219 if (msginfo == NULL) {
225 * Negotiate a compatible VMBUS version number with the
226 * host. We start with the highest number we can support
227 * and work our way down until we negotiate a compatible
232 if (i == ARRAY_SIZE(vmbus_versions)) {
237 version = vmbus_versions[i];
238 if (version > max_version)
241 ret = vmbus_negotiate_version(msginfo, version);
242 if (ret == -ETIMEDOUT)
245 if (vmbus_connection.conn_state == CONNECTED)
249 vmbus_proto_version = version;
250 pr_info("Vmbus version:%d.%d\n",
251 version >> 16, version & 0xFFFF);
253 vmbus_connection.channels = kcalloc(MAX_CHANNEL_RELIDS,
254 sizeof(struct vmbus_channel *),
256 if (vmbus_connection.channels == NULL) {
265 pr_err("Unable to connect to host\n");
267 vmbus_connection.conn_state = DISCONNECTED;
275 void vmbus_disconnect(void)
278 * First send the unload request to the host.
280 vmbus_initiate_unload(false);
282 if (vmbus_connection.handle_sub_chan_wq)
283 destroy_workqueue(vmbus_connection.handle_sub_chan_wq);
285 if (vmbus_connection.handle_primary_chan_wq)
286 destroy_workqueue(vmbus_connection.handle_primary_chan_wq);
288 if (vmbus_connection.work_queue)
289 destroy_workqueue(vmbus_connection.work_queue);
291 if (vmbus_connection.int_page) {
292 hv_free_hyperv_page((unsigned long)vmbus_connection.int_page);
293 vmbus_connection.int_page = NULL;
296 hv_free_hyperv_page((unsigned long)vmbus_connection.monitor_pages[0]);
297 hv_free_hyperv_page((unsigned long)vmbus_connection.monitor_pages[1]);
298 vmbus_connection.monitor_pages[0] = NULL;
299 vmbus_connection.monitor_pages[1] = NULL;
303 * relid2channel - Get the channel object given its
304 * child relative id (ie channel id)
306 struct vmbus_channel *relid2channel(u32 relid)
308 if (vmbus_connection.channels == NULL) {
309 pr_warn_once("relid2channel: relid=%d: No channels mapped!\n", relid);
312 if (WARN_ON(relid >= MAX_CHANNEL_RELIDS))
314 return READ_ONCE(vmbus_connection.channels[relid]);
318 * vmbus_on_event - Process a channel event notification
320 * For batched channels (default) optimize host to guest signaling
322 * 1. While reading the channel, we disable interrupts from host.
323 * 2. Ensure that we process all posted messages from the host
324 * before returning from this callback.
325 * 3. Once we return, enable signaling from the host. Once this
326 * state is set we check to see if additional packets are
327 * available to read. In this case we repeat the process.
328 * If this tasklet has been running for a long time
329 * then reschedule ourselves.
331 void vmbus_on_event(unsigned long data)
333 struct vmbus_channel *channel = (void *) data;
334 unsigned long time_limit = jiffies + 2;
336 trace_vmbus_on_event(channel);
338 hv_debug_delay_test(channel, INTERRUPT_DELAY);
340 void (*callback_fn)(void *);
342 /* A channel once created is persistent even when
343 * there is no driver handling the device. An
344 * unloading driver sets the onchannel_callback to NULL.
346 callback_fn = READ_ONCE(channel->onchannel_callback);
347 if (unlikely(callback_fn == NULL))
350 (*callback_fn)(channel->channel_callback_context);
352 if (channel->callback_mode != HV_CALL_BATCHED)
355 if (likely(hv_end_read(&channel->inbound) == 0))
358 hv_begin_read(&channel->inbound);
359 } while (likely(time_before(jiffies, time_limit)));
361 /* The time limit (2 jiffies) has been reached */
362 tasklet_schedule(&channel->callback_event);
366 * vmbus_post_msg - Send a msg on the vmbus's message connection
368 int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep)
370 struct vmbus_channel_message_header *hdr;
371 union hv_connection_id conn_id;
377 conn_id.u.id = vmbus_connection.msg_conn_id;
380 * hv_post_message() can have transient failures because of
381 * insufficient resources. Retry the operation a couple of
382 * times before giving up.
384 while (retries < 100) {
385 ret = hv_post_message(conn_id, 1, buffer, buflen);
388 case HV_STATUS_INVALID_CONNECTION_ID:
390 * See vmbus_negotiate_version(): VMBus protocol 5.0
391 * and higher require that we must use
392 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate
393 * Contact message, but on old hosts that only
394 * support VMBus protocol 4.0 or lower, here we get
395 * HV_STATUS_INVALID_CONNECTION_ID and we should
396 * return an error immediately without retrying.
399 if (hdr->msgtype == CHANNELMSG_INITIATE_CONTACT)
402 * We could get this if we send messages too
407 case HV_STATUS_INSUFFICIENT_MEMORY:
408 case HV_STATUS_INSUFFICIENT_BUFFERS:
411 case HV_STATUS_SUCCESS:
414 pr_err("hv_post_msg() failed; error code:%d\n", ret);
419 if (can_sleep && usec > 1000)
421 else if (usec < MAX_UDELAY_MS * 1000)
433 * vmbus_set_event - Send an event notification to the parent
435 void vmbus_set_event(struct vmbus_channel *channel)
437 u32 child_relid = channel->offermsg.child_relid;
439 if (!channel->is_dedicated_interrupt)
440 vmbus_send_interrupt(child_relid);
442 ++channel->sig_events;
444 hv_do_fast_hypercall8(HVCALL_SIGNAL_EVENT, channel->sig_event);
446 EXPORT_SYMBOL_GPL(vmbus_set_event);