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/slab.h>
18 #include <linux/vmalloc.h>
19 #include <linux/hyperv.h>
20 #include <linux/export.h>
21 #include <asm/mshyperv.h>
23 #include "hyperv_vmbus.h"
26 struct vmbus_connection vmbus_connection = {
27 .conn_state = DISCONNECTED,
28 .next_gpadl_handle = ATOMIC_INIT(0xE1E10),
30 .ready_for_suspend_event= COMPLETION_INITIALIZER(
31 vmbus_connection.ready_for_suspend_event),
32 .ready_for_resume_event = COMPLETION_INITIALIZER(
33 vmbus_connection.ready_for_resume_event),
35 EXPORT_SYMBOL_GPL(vmbus_connection);
38 * Negotiated protocol version with the host.
40 __u32 vmbus_proto_version;
41 EXPORT_SYMBOL_GPL(vmbus_proto_version);
43 static __u32 vmbus_get_next_version(__u32 current_version)
45 switch (current_version) {
47 return VERSION_WS2008;
52 case (VERSION_WIN8_1):
56 return VERSION_WIN8_1;
58 case (VERSION_WIN10_V5):
61 case (VERSION_WS2008):
67 int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo, u32 version)
71 struct vmbus_channel_initiate_contact *msg;
74 init_completion(&msginfo->waitevent);
76 msg = (struct vmbus_channel_initiate_contact *)msginfo->msg;
78 memset(msg, 0, sizeof(*msg));
79 msg->header.msgtype = CHANNELMSG_INITIATE_CONTACT;
80 msg->vmbus_version_requested = version;
83 * VMBus protocol 5.0 (VERSION_WIN10_V5) requires that we must use
84 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
85 * and for subsequent messages, we must use the Message Connection ID
86 * field in the host-returned Version Response Message. And, with
87 * VERSION_WIN10_V5, we don't use msg->interrupt_page, but we tell
88 * the host explicitly that we still use VMBUS_MESSAGE_SINT(2) for
91 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
93 if (version >= VERSION_WIN10_V5) {
94 msg->msg_sint = VMBUS_MESSAGE_SINT;
95 vmbus_connection.msg_conn_id = VMBUS_MESSAGE_CONNECTION_ID_4;
97 msg->interrupt_page = virt_to_phys(vmbus_connection.int_page);
98 vmbus_connection.msg_conn_id = VMBUS_MESSAGE_CONNECTION_ID;
101 msg->monitor_page1 = virt_to_phys(vmbus_connection.monitor_pages[0]);
102 msg->monitor_page2 = virt_to_phys(vmbus_connection.monitor_pages[1]);
104 * We want all channel messages to be delivered on CPU 0.
105 * This has been the behavior pre-win8. This is not
106 * perf issue and having all channel messages delivered on CPU 0
108 * For post win8 hosts, we support receiving channel messagges on
109 * all the CPUs. This is needed for kexec to work correctly where
110 * the CPU attempting to connect may not be CPU 0.
112 if (version >= VERSION_WIN8_1) {
114 msg->target_vcpu = hv_cpu_number_to_vp_number(cur_cpu);
115 vmbus_connection.connect_cpu = cur_cpu;
118 msg->target_vcpu = 0;
119 vmbus_connection.connect_cpu = 0;
123 * Add to list before we send the request since we may
124 * receive the response before returning from this routine
126 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
127 list_add_tail(&msginfo->msglistentry,
128 &vmbus_connection.chn_msg_list);
130 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
132 ret = vmbus_post_msg(msg,
133 sizeof(struct vmbus_channel_initiate_contact),
136 trace_vmbus_negotiate_version(msg, ret);
139 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
140 list_del(&msginfo->msglistentry);
141 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock,
146 /* Wait for the connection response */
147 wait_for_completion(&msginfo->waitevent);
149 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
150 list_del(&msginfo->msglistentry);
151 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
153 /* Check if successful */
154 if (msginfo->response.version_response.version_supported) {
155 vmbus_connection.conn_state = CONNECTED;
157 if (version >= VERSION_WIN10_V5)
158 vmbus_connection.msg_conn_id =
159 msginfo->response.version_response.msg_conn_id;
161 return -ECONNREFUSED;
168 * vmbus_connect - Sends a connect request on the partition service connection
170 int vmbus_connect(void)
173 struct vmbus_channel_msginfo *msginfo = NULL;
176 /* Initialize the vmbus connection */
177 vmbus_connection.conn_state = CONNECTING;
178 vmbus_connection.work_queue = create_workqueue("hv_vmbus_con");
179 if (!vmbus_connection.work_queue) {
184 vmbus_connection.handle_primary_chan_wq =
185 create_workqueue("hv_pri_chan");
186 if (!vmbus_connection.handle_primary_chan_wq) {
191 vmbus_connection.handle_sub_chan_wq =
192 create_workqueue("hv_sub_chan");
193 if (!vmbus_connection.handle_sub_chan_wq) {
198 INIT_LIST_HEAD(&vmbus_connection.chn_msg_list);
199 spin_lock_init(&vmbus_connection.channelmsg_lock);
201 INIT_LIST_HEAD(&vmbus_connection.chn_list);
202 mutex_init(&vmbus_connection.channel_mutex);
205 * Setup the vmbus event connection for channel interrupt
208 vmbus_connection.int_page =
209 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, 0);
210 if (vmbus_connection.int_page == NULL) {
215 vmbus_connection.recv_int_page = vmbus_connection.int_page;
216 vmbus_connection.send_int_page =
217 (void *)((unsigned long)vmbus_connection.int_page +
221 * Setup the monitor notification facility. The 1st page for
222 * parent->child and the 2nd page for child->parent
224 vmbus_connection.monitor_pages[0] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0);
225 vmbus_connection.monitor_pages[1] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0);
226 if ((vmbus_connection.monitor_pages[0] == NULL) ||
227 (vmbus_connection.monitor_pages[1] == NULL)) {
232 msginfo = kzalloc(sizeof(*msginfo) +
233 sizeof(struct vmbus_channel_initiate_contact),
235 if (msginfo == NULL) {
241 * Negotiate a compatible VMBUS version number with the
242 * host. We start with the highest number we can support
243 * and work our way down until we negotiate a compatible
247 version = VERSION_CURRENT;
250 ret = vmbus_negotiate_version(msginfo, version);
251 if (ret == -ETIMEDOUT)
254 if (vmbus_connection.conn_state == CONNECTED)
257 version = vmbus_get_next_version(version);
258 } while (version != VERSION_INVAL);
260 if (version == VERSION_INVAL)
263 vmbus_proto_version = version;
264 pr_info("Vmbus version:%d.%d\n",
265 version >> 16, version & 0xFFFF);
271 pr_err("Unable to connect to host\n");
273 vmbus_connection.conn_state = DISCONNECTED;
281 void vmbus_disconnect(void)
284 * First send the unload request to the host.
286 vmbus_initiate_unload(false);
288 if (vmbus_connection.handle_sub_chan_wq)
289 destroy_workqueue(vmbus_connection.handle_sub_chan_wq);
291 if (vmbus_connection.handle_primary_chan_wq)
292 destroy_workqueue(vmbus_connection.handle_primary_chan_wq);
294 if (vmbus_connection.work_queue)
295 destroy_workqueue(vmbus_connection.work_queue);
297 if (vmbus_connection.int_page) {
298 free_pages((unsigned long)vmbus_connection.int_page, 0);
299 vmbus_connection.int_page = NULL;
302 free_pages((unsigned long)vmbus_connection.monitor_pages[0], 0);
303 free_pages((unsigned long)vmbus_connection.monitor_pages[1], 0);
304 vmbus_connection.monitor_pages[0] = NULL;
305 vmbus_connection.monitor_pages[1] = NULL;
309 * relid2channel - Get the channel object given its
310 * child relative id (ie channel id)
312 struct vmbus_channel *relid2channel(u32 relid)
314 struct vmbus_channel *channel;
315 struct vmbus_channel *found_channel = NULL;
316 struct list_head *cur, *tmp;
317 struct vmbus_channel *cur_sc;
319 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
321 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
322 if (channel->offermsg.child_relid == relid) {
323 found_channel = channel;
325 } else if (!list_empty(&channel->sc_list)) {
327 * Deal with sub-channels.
329 list_for_each_safe(cur, tmp, &channel->sc_list) {
330 cur_sc = list_entry(cur, struct vmbus_channel,
332 if (cur_sc->offermsg.child_relid == relid) {
333 found_channel = cur_sc;
340 return found_channel;
344 * vmbus_on_event - Process a channel event notification
346 * For batched channels (default) optimize host to guest signaling
348 * 1. While reading the channel, we disable interrupts from host.
349 * 2. Ensure that we process all posted messages from the host
350 * before returning from this callback.
351 * 3. Once we return, enable signaling from the host. Once this
352 * state is set we check to see if additional packets are
353 * available to read. In this case we repeat the process.
354 * If this tasklet has been running for a long time
355 * then reschedule ourselves.
357 void vmbus_on_event(unsigned long data)
359 struct vmbus_channel *channel = (void *) data;
360 unsigned long time_limit = jiffies + 2;
362 trace_vmbus_on_event(channel);
365 void (*callback_fn)(void *);
367 /* A channel once created is persistent even when
368 * there is no driver handling the device. An
369 * unloading driver sets the onchannel_callback to NULL.
371 callback_fn = READ_ONCE(channel->onchannel_callback);
372 if (unlikely(callback_fn == NULL))
375 (*callback_fn)(channel->channel_callback_context);
377 if (channel->callback_mode != HV_CALL_BATCHED)
380 if (likely(hv_end_read(&channel->inbound) == 0))
383 hv_begin_read(&channel->inbound);
384 } while (likely(time_before(jiffies, time_limit)));
386 /* The time limit (2 jiffies) has been reached */
387 tasklet_schedule(&channel->callback_event);
391 * vmbus_post_msg - Send a msg on the vmbus's message connection
393 int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep)
395 struct vmbus_channel_message_header *hdr;
396 union hv_connection_id conn_id;
402 conn_id.u.id = vmbus_connection.msg_conn_id;
405 * hv_post_message() can have transient failures because of
406 * insufficient resources. Retry the operation a couple of
407 * times before giving up.
409 while (retries < 100) {
410 ret = hv_post_message(conn_id, 1, buffer, buflen);
413 case HV_STATUS_INVALID_CONNECTION_ID:
415 * See vmbus_negotiate_version(): VMBus protocol 5.0
416 * requires that we must use
417 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate
418 * Contact message, but on old hosts that only
419 * support VMBus protocol 4.0 or lower, here we get
420 * HV_STATUS_INVALID_CONNECTION_ID and we should
421 * return an error immediately without retrying.
424 if (hdr->msgtype == CHANNELMSG_INITIATE_CONTACT)
427 * We could get this if we send messages too
432 case HV_STATUS_INSUFFICIENT_MEMORY:
433 case HV_STATUS_INSUFFICIENT_BUFFERS:
436 case HV_STATUS_SUCCESS:
439 pr_err("hv_post_msg() failed; error code:%d\n", ret);
444 if (can_sleep && usec > 1000)
446 else if (usec < MAX_UDELAY_MS * 1000)
458 * vmbus_set_event - Send an event notification to the parent
460 void vmbus_set_event(struct vmbus_channel *channel)
462 u32 child_relid = channel->offermsg.child_relid;
464 if (!channel->is_dedicated_interrupt)
465 vmbus_send_interrupt(child_relid);
467 ++channel->sig_events;
469 hv_do_fast_hypercall8(HVCALL_SIGNAL_EVENT, channel->sig_event);
471 EXPORT_SYMBOL_GPL(vmbus_set_event);