]>
Commit | Line | Data |
---|---|---|
1 | // SPDX-License-Identifier: GPL-2.0-only | |
2 | /* | |
3 | * | |
4 | * Copyright (c) 2009, Microsoft Corporation. | |
5 | * | |
6 | * Authors: | |
7 | * Haiyang Zhang <haiyangz@microsoft.com> | |
8 | * Hank Janssen <hjanssen@microsoft.com> | |
9 | */ | |
10 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
11 | ||
12 | #include <linux/kernel.h> | |
13 | #include <linux/sched.h> | |
14 | #include <linux/wait.h> | |
15 | #include <linux/delay.h> | |
16 | #include <linux/mm.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> | |
23 | ||
24 | #include "hyperv_vmbus.h" | |
25 | ||
26 | ||
27 | struct vmbus_connection vmbus_connection = { | |
28 | .conn_state = DISCONNECTED, | |
29 | .next_gpadl_handle = ATOMIC_INIT(0xE1E10), | |
30 | ||
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), | |
35 | }; | |
36 | EXPORT_SYMBOL_GPL(vmbus_connection); | |
37 | ||
38 | /* | |
39 | * Negotiated protocol version with the host. | |
40 | */ | |
41 | __u32 vmbus_proto_version; | |
42 | EXPORT_SYMBOL_GPL(vmbus_proto_version); | |
43 | ||
44 | /* | |
45 | * Table of VMBus versions listed from newest to oldest. | |
46 | */ | |
47 | static __u32 vmbus_versions[] = { | |
48 | VERSION_WIN10_V5_2, | |
49 | VERSION_WIN10_V5_1, | |
50 | VERSION_WIN10_V5, | |
51 | VERSION_WIN10_V4_1, | |
52 | VERSION_WIN10, | |
53 | VERSION_WIN8_1, | |
54 | VERSION_WIN8, | |
55 | VERSION_WIN7, | |
56 | VERSION_WS2008 | |
57 | }; | |
58 | ||
59 | /* | |
60 | * Maximal VMBus protocol version guests can negotiate. Useful to cap the | |
61 | * VMBus version for testing and debugging purpose. | |
62 | */ | |
63 | static uint max_version = VERSION_WIN10_V5_2; | |
64 | ||
65 | module_param(max_version, uint, S_IRUGO); | |
66 | MODULE_PARM_DESC(max_version, | |
67 | "Maximal VMBus protocol version which can be negotiated"); | |
68 | ||
69 | int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo, u32 version) | |
70 | { | |
71 | int ret = 0; | |
72 | unsigned int cur_cpu; | |
73 | struct vmbus_channel_initiate_contact *msg; | |
74 | unsigned long flags; | |
75 | ||
76 | init_completion(&msginfo->waitevent); | |
77 | ||
78 | msg = (struct vmbus_channel_initiate_contact *)msginfo->msg; | |
79 | ||
80 | memset(msg, 0, sizeof(*msg)); | |
81 | msg->header.msgtype = CHANNELMSG_INITIATE_CONTACT; | |
82 | msg->vmbus_version_requested = version; | |
83 | ||
84 | /* | |
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 | |
91 | * compatibility. | |
92 | * | |
93 | * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1). | |
94 | */ | |
95 | if (version >= VERSION_WIN10_V5) { | |
96 | msg->msg_sint = VMBUS_MESSAGE_SINT; | |
97 | vmbus_connection.msg_conn_id = VMBUS_MESSAGE_CONNECTION_ID_4; | |
98 | } else { | |
99 | msg->interrupt_page = virt_to_phys(vmbus_connection.int_page); | |
100 | vmbus_connection.msg_conn_id = VMBUS_MESSAGE_CONNECTION_ID; | |
101 | } | |
102 | ||
103 | msg->monitor_page1 = virt_to_phys(vmbus_connection.monitor_pages[0]); | |
104 | msg->monitor_page2 = virt_to_phys(vmbus_connection.monitor_pages[1]); | |
105 | /* | |
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 | |
109 | * would be ok. | |
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. | |
113 | */ | |
114 | if (version >= VERSION_WIN8_1) { | |
115 | cur_cpu = get_cpu(); | |
116 | msg->target_vcpu = hv_cpu_number_to_vp_number(cur_cpu); | |
117 | vmbus_connection.connect_cpu = cur_cpu; | |
118 | put_cpu(); | |
119 | } else { | |
120 | msg->target_vcpu = 0; | |
121 | vmbus_connection.connect_cpu = 0; | |
122 | } | |
123 | ||
124 | /* | |
125 | * Add to list before we send the request since we may | |
126 | * receive the response before returning from this routine | |
127 | */ | |
128 | spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); | |
129 | list_add_tail(&msginfo->msglistentry, | |
130 | &vmbus_connection.chn_msg_list); | |
131 | ||
132 | spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); | |
133 | ||
134 | ret = vmbus_post_msg(msg, | |
135 | sizeof(struct vmbus_channel_initiate_contact), | |
136 | true); | |
137 | ||
138 | trace_vmbus_negotiate_version(msg, ret); | |
139 | ||
140 | if (ret != 0) { | |
141 | spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); | |
142 | list_del(&msginfo->msglistentry); | |
143 | spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, | |
144 | flags); | |
145 | return ret; | |
146 | } | |
147 | ||
148 | /* Wait for the connection response */ | |
149 | wait_for_completion(&msginfo->waitevent); | |
150 | ||
151 | spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); | |
152 | list_del(&msginfo->msglistentry); | |
153 | spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); | |
154 | ||
155 | /* Check if successful */ | |
156 | if (msginfo->response.version_response.version_supported) { | |
157 | vmbus_connection.conn_state = CONNECTED; | |
158 | ||
159 | if (version >= VERSION_WIN10_V5) | |
160 | vmbus_connection.msg_conn_id = | |
161 | msginfo->response.version_response.msg_conn_id; | |
162 | } else { | |
163 | return -ECONNREFUSED; | |
164 | } | |
165 | ||
166 | return ret; | |
167 | } | |
168 | ||
169 | /* | |
170 | * vmbus_connect - Sends a connect request on the partition service connection | |
171 | */ | |
172 | int vmbus_connect(void) | |
173 | { | |
174 | struct vmbus_channel_msginfo *msginfo = NULL; | |
175 | int i, ret = 0; | |
176 | __u32 version; | |
177 | ||
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) { | |
182 | ret = -ENOMEM; | |
183 | goto cleanup; | |
184 | } | |
185 | ||
186 | vmbus_connection.handle_primary_chan_wq = | |
187 | create_workqueue("hv_pri_chan"); | |
188 | if (!vmbus_connection.handle_primary_chan_wq) { | |
189 | ret = -ENOMEM; | |
190 | goto cleanup; | |
191 | } | |
192 | ||
193 | vmbus_connection.handle_sub_chan_wq = | |
194 | create_workqueue("hv_sub_chan"); | |
195 | if (!vmbus_connection.handle_sub_chan_wq) { | |
196 | ret = -ENOMEM; | |
197 | goto cleanup; | |
198 | } | |
199 | ||
200 | INIT_LIST_HEAD(&vmbus_connection.chn_msg_list); | |
201 | spin_lock_init(&vmbus_connection.channelmsg_lock); | |
202 | ||
203 | INIT_LIST_HEAD(&vmbus_connection.chn_list); | |
204 | mutex_init(&vmbus_connection.channel_mutex); | |
205 | ||
206 | /* | |
207 | * Setup the vmbus event connection for channel interrupt | |
208 | * abstraction stuff | |
209 | */ | |
210 | vmbus_connection.int_page = | |
211 | (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, 0); | |
212 | if (vmbus_connection.int_page == NULL) { | |
213 | ret = -ENOMEM; | |
214 | goto cleanup; | |
215 | } | |
216 | ||
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 | (PAGE_SIZE >> 1)); | |
221 | ||
222 | /* | |
223 | * Setup the monitor notification facility. The 1st page for | |
224 | * parent->child and the 2nd page for child->parent | |
225 | */ | |
226 | vmbus_connection.monitor_pages[0] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0); | |
227 | vmbus_connection.monitor_pages[1] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0); | |
228 | if ((vmbus_connection.monitor_pages[0] == NULL) || | |
229 | (vmbus_connection.monitor_pages[1] == NULL)) { | |
230 | ret = -ENOMEM; | |
231 | goto cleanup; | |
232 | } | |
233 | ||
234 | msginfo = kzalloc(sizeof(*msginfo) + | |
235 | sizeof(struct vmbus_channel_initiate_contact), | |
236 | GFP_KERNEL); | |
237 | if (msginfo == NULL) { | |
238 | ret = -ENOMEM; | |
239 | goto cleanup; | |
240 | } | |
241 | ||
242 | /* | |
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 | |
246 | * version. | |
247 | */ | |
248 | ||
249 | for (i = 0; ; i++) { | |
250 | if (i == ARRAY_SIZE(vmbus_versions)) | |
251 | goto cleanup; | |
252 | ||
253 | version = vmbus_versions[i]; | |
254 | if (version > max_version) | |
255 | continue; | |
256 | ||
257 | ret = vmbus_negotiate_version(msginfo, version); | |
258 | if (ret == -ETIMEDOUT) | |
259 | goto cleanup; | |
260 | ||
261 | if (vmbus_connection.conn_state == CONNECTED) | |
262 | break; | |
263 | } | |
264 | ||
265 | vmbus_proto_version = version; | |
266 | pr_info("Vmbus version:%d.%d\n", | |
267 | version >> 16, version & 0xFFFF); | |
268 | ||
269 | kfree(msginfo); | |
270 | return 0; | |
271 | ||
272 | cleanup: | |
273 | pr_err("Unable to connect to host\n"); | |
274 | ||
275 | vmbus_connection.conn_state = DISCONNECTED; | |
276 | vmbus_disconnect(); | |
277 | ||
278 | kfree(msginfo); | |
279 | ||
280 | return ret; | |
281 | } | |
282 | ||
283 | void vmbus_disconnect(void) | |
284 | { | |
285 | /* | |
286 | * First send the unload request to the host. | |
287 | */ | |
288 | vmbus_initiate_unload(false); | |
289 | ||
290 | if (vmbus_connection.handle_sub_chan_wq) | |
291 | destroy_workqueue(vmbus_connection.handle_sub_chan_wq); | |
292 | ||
293 | if (vmbus_connection.handle_primary_chan_wq) | |
294 | destroy_workqueue(vmbus_connection.handle_primary_chan_wq); | |
295 | ||
296 | if (vmbus_connection.work_queue) | |
297 | destroy_workqueue(vmbus_connection.work_queue); | |
298 | ||
299 | if (vmbus_connection.int_page) { | |
300 | free_pages((unsigned long)vmbus_connection.int_page, 0); | |
301 | vmbus_connection.int_page = NULL; | |
302 | } | |
303 | ||
304 | free_pages((unsigned long)vmbus_connection.monitor_pages[0], 0); | |
305 | free_pages((unsigned long)vmbus_connection.monitor_pages[1], 0); | |
306 | vmbus_connection.monitor_pages[0] = NULL; | |
307 | vmbus_connection.monitor_pages[1] = NULL; | |
308 | } | |
309 | ||
310 | /* | |
311 | * relid2channel - Get the channel object given its | |
312 | * child relative id (ie channel id) | |
313 | */ | |
314 | struct vmbus_channel *relid2channel(u32 relid) | |
315 | { | |
316 | struct vmbus_channel *channel; | |
317 | struct vmbus_channel *found_channel = NULL; | |
318 | struct list_head *cur, *tmp; | |
319 | struct vmbus_channel *cur_sc; | |
320 | ||
321 | BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex)); | |
322 | ||
323 | list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) { | |
324 | if (channel->offermsg.child_relid == relid) { | |
325 | found_channel = channel; | |
326 | break; | |
327 | } else if (!list_empty(&channel->sc_list)) { | |
328 | /* | |
329 | * Deal with sub-channels. | |
330 | */ | |
331 | list_for_each_safe(cur, tmp, &channel->sc_list) { | |
332 | cur_sc = list_entry(cur, struct vmbus_channel, | |
333 | sc_list); | |
334 | if (cur_sc->offermsg.child_relid == relid) { | |
335 | found_channel = cur_sc; | |
336 | break; | |
337 | } | |
338 | } | |
339 | } | |
340 | } | |
341 | ||
342 | return found_channel; | |
343 | } | |
344 | ||
345 | /* | |
346 | * vmbus_on_event - Process a channel event notification | |
347 | * | |
348 | * For batched channels (default) optimize host to guest signaling | |
349 | * by ensuring: | |
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. | |
358 | */ | |
359 | void vmbus_on_event(unsigned long data) | |
360 | { | |
361 | struct vmbus_channel *channel = (void *) data; | |
362 | unsigned long time_limit = jiffies + 2; | |
363 | ||
364 | trace_vmbus_on_event(channel); | |
365 | ||
366 | do { | |
367 | void (*callback_fn)(void *); | |
368 | ||
369 | /* A channel once created is persistent even when | |
370 | * there is no driver handling the device. An | |
371 | * unloading driver sets the onchannel_callback to NULL. | |
372 | */ | |
373 | callback_fn = READ_ONCE(channel->onchannel_callback); | |
374 | if (unlikely(callback_fn == NULL)) | |
375 | return; | |
376 | ||
377 | (*callback_fn)(channel->channel_callback_context); | |
378 | ||
379 | if (channel->callback_mode != HV_CALL_BATCHED) | |
380 | return; | |
381 | ||
382 | if (likely(hv_end_read(&channel->inbound) == 0)) | |
383 | return; | |
384 | ||
385 | hv_begin_read(&channel->inbound); | |
386 | } while (likely(time_before(jiffies, time_limit))); | |
387 | ||
388 | /* The time limit (2 jiffies) has been reached */ | |
389 | tasklet_schedule(&channel->callback_event); | |
390 | } | |
391 | ||
392 | /* | |
393 | * vmbus_post_msg - Send a msg on the vmbus's message connection | |
394 | */ | |
395 | int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep) | |
396 | { | |
397 | struct vmbus_channel_message_header *hdr; | |
398 | union hv_connection_id conn_id; | |
399 | int ret = 0; | |
400 | int retries = 0; | |
401 | u32 usec = 1; | |
402 | ||
403 | conn_id.asu32 = 0; | |
404 | conn_id.u.id = vmbus_connection.msg_conn_id; | |
405 | ||
406 | /* | |
407 | * hv_post_message() can have transient failures because of | |
408 | * insufficient resources. Retry the operation a couple of | |
409 | * times before giving up. | |
410 | */ | |
411 | while (retries < 100) { | |
412 | ret = hv_post_message(conn_id, 1, buffer, buflen); | |
413 | ||
414 | switch (ret) { | |
415 | case HV_STATUS_INVALID_CONNECTION_ID: | |
416 | /* | |
417 | * See vmbus_negotiate_version(): VMBus protocol 5.0 | |
418 | * and higher require that we must use | |
419 | * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate | |
420 | * Contact message, but on old hosts that only | |
421 | * support VMBus protocol 4.0 or lower, here we get | |
422 | * HV_STATUS_INVALID_CONNECTION_ID and we should | |
423 | * return an error immediately without retrying. | |
424 | */ | |
425 | hdr = buffer; | |
426 | if (hdr->msgtype == CHANNELMSG_INITIATE_CONTACT) | |
427 | return -EINVAL; | |
428 | /* | |
429 | * We could get this if we send messages too | |
430 | * frequently. | |
431 | */ | |
432 | ret = -EAGAIN; | |
433 | break; | |
434 | case HV_STATUS_INSUFFICIENT_MEMORY: | |
435 | case HV_STATUS_INSUFFICIENT_BUFFERS: | |
436 | ret = -ENOBUFS; | |
437 | break; | |
438 | case HV_STATUS_SUCCESS: | |
439 | return ret; | |
440 | default: | |
441 | pr_err("hv_post_msg() failed; error code:%d\n", ret); | |
442 | return -EINVAL; | |
443 | } | |
444 | ||
445 | retries++; | |
446 | if (can_sleep && usec > 1000) | |
447 | msleep(usec / 1000); | |
448 | else if (usec < MAX_UDELAY_MS * 1000) | |
449 | udelay(usec); | |
450 | else | |
451 | mdelay(usec / 1000); | |
452 | ||
453 | if (retries < 22) | |
454 | usec *= 2; | |
455 | } | |
456 | return ret; | |
457 | } | |
458 | ||
459 | /* | |
460 | * vmbus_set_event - Send an event notification to the parent | |
461 | */ | |
462 | void vmbus_set_event(struct vmbus_channel *channel) | |
463 | { | |
464 | u32 child_relid = channel->offermsg.child_relid; | |
465 | ||
466 | if (!channel->is_dedicated_interrupt) | |
467 | vmbus_send_interrupt(child_relid); | |
468 | ||
469 | ++channel->sig_events; | |
470 | ||
471 | hv_do_fast_hypercall8(HVCALL_SIGNAL_EVENT, channel->sig_event); | |
472 | } | |
473 | EXPORT_SYMBOL_GPL(vmbus_set_event); |