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 <linux/io.h>
23 #include <linux/set_memory.h>
24 #include <asm/mshyperv.h>
25
26 #include "hyperv_vmbus.h"
27
28
29 struct vmbus_connection vmbus_connection = {
30 .conn_state = DISCONNECTED,
31 .unload_event = COMPLETION_INITIALIZER(
32 vmbus_connection.unload_event),
33 .next_gpadl_handle = ATOMIC_INIT(0xE1E10),
34
35 .ready_for_suspend_event = COMPLETION_INITIALIZER(
36 vmbus_connection.ready_for_suspend_event),
37 .ready_for_resume_event = COMPLETION_INITIALIZER(
38 vmbus_connection.ready_for_resume_event),
39 };
40 EXPORT_SYMBOL_GPL(vmbus_connection);
41
42 /*
43 * Negotiated protocol version with the host.
44 */
45 __u32 vmbus_proto_version;
46 EXPORT_SYMBOL_GPL(vmbus_proto_version);
47
48 /*
49 * Table of VMBus versions listed from newest to oldest.
50 * VERSION_WIN7 and VERSION_WS2008 are no longer supported in
51 * Linux guests and are not listed.
52 */
53 static __u32 vmbus_versions[] = {
54 VERSION_WIN10_V5_3,
55 VERSION_WIN10_V5_2,
56 VERSION_WIN10_V5_1,
57 VERSION_WIN10_V5,
58 VERSION_WIN10_V4_1,
59 VERSION_WIN10,
60 VERSION_WIN8_1,
61 VERSION_WIN8
62 };
63
64 /*
65 * Maximal VMBus protocol version guests can negotiate. Useful to cap the
66 * VMBus version for testing and debugging purpose.
67 */
68 static uint max_version = VERSION_WIN10_V5_3;
69
70 module_param(max_version, uint, S_IRUGO);
71 MODULE_PARM_DESC(max_version,
72 "Maximal VMBus protocol version which can be negotiated");
73
vmbus_negotiate_version(struct vmbus_channel_msginfo * msginfo,u32 version)74 int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo, u32 version)
75 {
76 int ret = 0;
77 struct vmbus_channel_initiate_contact *msg;
78 unsigned long flags;
79
80 init_completion(&msginfo->waitevent);
81
82 msg = (struct vmbus_channel_initiate_contact *)msginfo->msg;
83
84 memset(msg, 0, sizeof(*msg));
85 msg->header.msgtype = CHANNELMSG_INITIATE_CONTACT;
86 msg->vmbus_version_requested = version;
87
88 /*
89 * VMBus protocol 5.0 (VERSION_WIN10_V5) and higher require that we must
90 * use VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
91 * and for subsequent messages, we must use the Message Connection ID
92 * field in the host-returned Version Response Message. And, with
93 * VERSION_WIN10_V5 and higher, we don't use msg->interrupt_page, but we
94 * tell the host explicitly that we still use VMBUS_MESSAGE_SINT(2) for
95 * compatibility.
96 *
97 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
98 */
99 if (version >= VERSION_WIN10_V5) {
100 msg->msg_sint = VMBUS_MESSAGE_SINT;
101 vmbus_connection.msg_conn_id = VMBUS_MESSAGE_CONNECTION_ID_4;
102 } else {
103 msg->interrupt_page = virt_to_phys(vmbus_connection.int_page);
104 vmbus_connection.msg_conn_id = VMBUS_MESSAGE_CONNECTION_ID;
105 }
106
107 msg->monitor_page1 = vmbus_connection.monitor_pages_pa[0];
108 msg->monitor_page2 = vmbus_connection.monitor_pages_pa[1];
109
110 msg->target_vcpu = hv_cpu_number_to_vp_number(VMBUS_CONNECT_CPU);
111
112 /*
113 * Add to list before we send the request since we may
114 * receive the response before returning from this routine
115 */
116 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
117 list_add_tail(&msginfo->msglistentry,
118 &vmbus_connection.chn_msg_list);
119
120 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
121
122 ret = vmbus_post_msg(msg,
123 sizeof(struct vmbus_channel_initiate_contact),
124 true);
125
126 trace_vmbus_negotiate_version(msg, ret);
127
128 if (ret != 0) {
129 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
130 list_del(&msginfo->msglistentry);
131 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock,
132 flags);
133 return ret;
134 }
135
136 /* Wait for the connection response */
137 wait_for_completion(&msginfo->waitevent);
138
139 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
140 list_del(&msginfo->msglistentry);
141 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
142
143 /* Check if successful */
144 if (msginfo->response.version_response.version_supported) {
145 vmbus_connection.conn_state = CONNECTED;
146
147 if (version >= VERSION_WIN10_V5)
148 vmbus_connection.msg_conn_id =
149 msginfo->response.version_response.msg_conn_id;
150 } else {
151 return -ECONNREFUSED;
152 }
153
154 return ret;
155 }
156
157 /*
158 * vmbus_connect - Sends a connect request on the partition service connection
159 */
vmbus_connect(void)160 int vmbus_connect(void)
161 {
162 struct vmbus_channel_msginfo *msginfo = NULL;
163 int i, ret = 0;
164 __u32 version;
165
166 /* Initialize the vmbus connection */
167 vmbus_connection.conn_state = CONNECTING;
168 vmbus_connection.work_queue = create_workqueue("hv_vmbus_con");
169 if (!vmbus_connection.work_queue) {
170 ret = -ENOMEM;
171 goto cleanup;
172 }
173
174 vmbus_connection.rescind_work_queue =
175 create_workqueue("hv_vmbus_rescind");
176 if (!vmbus_connection.rescind_work_queue) {
177 ret = -ENOMEM;
178 goto cleanup;
179 }
180 vmbus_connection.ignore_any_offer_msg = false;
181
182 vmbus_connection.handle_primary_chan_wq =
183 create_workqueue("hv_pri_chan");
184 if (!vmbus_connection.handle_primary_chan_wq) {
185 ret = -ENOMEM;
186 goto cleanup;
187 }
188
189 vmbus_connection.handle_sub_chan_wq =
190 create_workqueue("hv_sub_chan");
191 if (!vmbus_connection.handle_sub_chan_wq) {
192 ret = -ENOMEM;
193 goto cleanup;
194 }
195
196 INIT_LIST_HEAD(&vmbus_connection.chn_msg_list);
197 spin_lock_init(&vmbus_connection.channelmsg_lock);
198
199 INIT_LIST_HEAD(&vmbus_connection.chn_list);
200 mutex_init(&vmbus_connection.channel_mutex);
201
202 /*
203 * Setup the vmbus event connection for channel interrupt
204 * abstraction stuff
205 */
206 vmbus_connection.int_page =
207 (void *)hv_alloc_hyperv_zeroed_page();
208 if (vmbus_connection.int_page == NULL) {
209 ret = -ENOMEM;
210 goto cleanup;
211 }
212
213 vmbus_connection.recv_int_page = vmbus_connection.int_page;
214 vmbus_connection.send_int_page =
215 (void *)((unsigned long)vmbus_connection.int_page +
216 (HV_HYP_PAGE_SIZE >> 1));
217
218 /*
219 * Setup the monitor notification facility. The 1st page for
220 * parent->child and the 2nd page for child->parent
221 */
222 vmbus_connection.monitor_pages[0] = (void *)hv_alloc_hyperv_zeroed_page();
223 vmbus_connection.monitor_pages[1] = (void *)hv_alloc_hyperv_zeroed_page();
224 if ((vmbus_connection.monitor_pages[0] == NULL) ||
225 (vmbus_connection.monitor_pages[1] == NULL)) {
226 ret = -ENOMEM;
227 goto cleanup;
228 }
229
230 vmbus_connection.monitor_pages_original[0]
231 = vmbus_connection.monitor_pages[0];
232 vmbus_connection.monitor_pages_original[1]
233 = vmbus_connection.monitor_pages[1];
234 vmbus_connection.monitor_pages_pa[0]
235 = virt_to_phys(vmbus_connection.monitor_pages[0]);
236 vmbus_connection.monitor_pages_pa[1]
237 = virt_to_phys(vmbus_connection.monitor_pages[1]);
238
239 if (hv_is_isolation_supported()) {
240 ret = set_memory_decrypted((unsigned long)
241 vmbus_connection.monitor_pages[0],
242 1);
243 ret |= set_memory_decrypted((unsigned long)
244 vmbus_connection.monitor_pages[1],
245 1);
246 if (ret)
247 goto cleanup;
248
249 /*
250 * Isolation VM with AMD SNP needs to access monitor page via
251 * address space above shared gpa boundary.
252 */
253 if (hv_isolation_type_snp()) {
254 vmbus_connection.monitor_pages_pa[0] +=
255 ms_hyperv.shared_gpa_boundary;
256 vmbus_connection.monitor_pages_pa[1] +=
257 ms_hyperv.shared_gpa_boundary;
258
259 vmbus_connection.monitor_pages[0]
260 = memremap(vmbus_connection.monitor_pages_pa[0],
261 HV_HYP_PAGE_SIZE,
262 MEMREMAP_WB);
263 if (!vmbus_connection.monitor_pages[0]) {
264 ret = -ENOMEM;
265 goto cleanup;
266 }
267
268 vmbus_connection.monitor_pages[1]
269 = memremap(vmbus_connection.monitor_pages_pa[1],
270 HV_HYP_PAGE_SIZE,
271 MEMREMAP_WB);
272 if (!vmbus_connection.monitor_pages[1]) {
273 ret = -ENOMEM;
274 goto cleanup;
275 }
276 }
277
278 /*
279 * Set memory host visibility hvcall smears memory
280 * and so zero monitor pages here.
281 */
282 memset(vmbus_connection.monitor_pages[0], 0x00,
283 HV_HYP_PAGE_SIZE);
284 memset(vmbus_connection.monitor_pages[1], 0x00,
285 HV_HYP_PAGE_SIZE);
286
287 }
288
289 msginfo = kzalloc(sizeof(*msginfo) +
290 sizeof(struct vmbus_channel_initiate_contact),
291 GFP_KERNEL);
292 if (msginfo == NULL) {
293 ret = -ENOMEM;
294 goto cleanup;
295 }
296
297 /*
298 * Negotiate a compatible VMBUS version number with the
299 * host. We start with the highest number we can support
300 * and work our way down until we negotiate a compatible
301 * version.
302 */
303
304 for (i = 0; ; i++) {
305 if (i == ARRAY_SIZE(vmbus_versions)) {
306 ret = -EDOM;
307 goto cleanup;
308 }
309
310 version = vmbus_versions[i];
311 if (version > max_version)
312 continue;
313
314 ret = vmbus_negotiate_version(msginfo, version);
315 if (ret == -ETIMEDOUT)
316 goto cleanup;
317
318 if (vmbus_connection.conn_state == CONNECTED)
319 break;
320 }
321
322 if (hv_is_isolation_supported() && version < VERSION_WIN10_V5_2) {
323 pr_err("Invalid VMBus version %d.%d (expected >= %d.%d) from the host supporting isolation\n",
324 version >> 16, version & 0xFFFF, VERSION_WIN10_V5_2 >> 16, VERSION_WIN10_V5_2 & 0xFFFF);
325 ret = -EINVAL;
326 goto cleanup;
327 }
328
329 vmbus_proto_version = version;
330 pr_info("Vmbus version:%d.%d\n",
331 version >> 16, version & 0xFFFF);
332
333 vmbus_connection.channels = kcalloc(MAX_CHANNEL_RELIDS,
334 sizeof(struct vmbus_channel *),
335 GFP_KERNEL);
336 if (vmbus_connection.channels == NULL) {
337 ret = -ENOMEM;
338 goto cleanup;
339 }
340
341 kfree(msginfo);
342 return 0;
343
344 cleanup:
345 pr_err("Unable to connect to host\n");
346
347 vmbus_connection.conn_state = DISCONNECTED;
348 vmbus_disconnect();
349
350 kfree(msginfo);
351
352 return ret;
353 }
354
vmbus_disconnect(void)355 void vmbus_disconnect(void)
356 {
357 /*
358 * First send the unload request to the host.
359 */
360 vmbus_initiate_unload(false);
361
362 if (vmbus_connection.handle_sub_chan_wq)
363 destroy_workqueue(vmbus_connection.handle_sub_chan_wq);
364
365 if (vmbus_connection.handle_primary_chan_wq)
366 destroy_workqueue(vmbus_connection.handle_primary_chan_wq);
367
368 if (vmbus_connection.rescind_work_queue)
369 destroy_workqueue(vmbus_connection.rescind_work_queue);
370
371 if (vmbus_connection.work_queue)
372 destroy_workqueue(vmbus_connection.work_queue);
373
374 if (vmbus_connection.int_page) {
375 hv_free_hyperv_page((unsigned long)vmbus_connection.int_page);
376 vmbus_connection.int_page = NULL;
377 }
378
379 if (hv_is_isolation_supported()) {
380 /*
381 * memunmap() checks input address is ioremap address or not
382 * inside. It doesn't unmap any thing in the non-SNP CVM and
383 * so not check CVM type here.
384 */
385 memunmap(vmbus_connection.monitor_pages[0]);
386 memunmap(vmbus_connection.monitor_pages[1]);
387
388 set_memory_encrypted((unsigned long)
389 vmbus_connection.monitor_pages_original[0],
390 1);
391 set_memory_encrypted((unsigned long)
392 vmbus_connection.monitor_pages_original[1],
393 1);
394 }
395
396 hv_free_hyperv_page((unsigned long)
397 vmbus_connection.monitor_pages_original[0]);
398 hv_free_hyperv_page((unsigned long)
399 vmbus_connection.monitor_pages_original[1]);
400 vmbus_connection.monitor_pages_original[0] =
401 vmbus_connection.monitor_pages[0] = NULL;
402 vmbus_connection.monitor_pages_original[1] =
403 vmbus_connection.monitor_pages[1] = NULL;
404 }
405
406 /*
407 * relid2channel - Get the channel object given its
408 * child relative id (ie channel id)
409 */
relid2channel(u32 relid)410 struct vmbus_channel *relid2channel(u32 relid)
411 {
412 if (WARN_ON(relid >= MAX_CHANNEL_RELIDS))
413 return NULL;
414 return READ_ONCE(vmbus_connection.channels[relid]);
415 }
416
417 /*
418 * vmbus_on_event - Process a channel event notification
419 *
420 * For batched channels (default) optimize host to guest signaling
421 * by ensuring:
422 * 1. While reading the channel, we disable interrupts from host.
423 * 2. Ensure that we process all posted messages from the host
424 * before returning from this callback.
425 * 3. Once we return, enable signaling from the host. Once this
426 * state is set we check to see if additional packets are
427 * available to read. In this case we repeat the process.
428 * If this tasklet has been running for a long time
429 * then reschedule ourselves.
430 */
vmbus_on_event(unsigned long data)431 void vmbus_on_event(unsigned long data)
432 {
433 struct vmbus_channel *channel = (void *) data;
434 void (*callback_fn)(void *context);
435
436 trace_vmbus_on_event(channel);
437
438 hv_debug_delay_test(channel, INTERRUPT_DELAY);
439
440 /* A channel once created is persistent even when
441 * there is no driver handling the device. An
442 * unloading driver sets the onchannel_callback to NULL.
443 */
444 callback_fn = READ_ONCE(channel->onchannel_callback);
445 if (unlikely(!callback_fn))
446 return;
447
448 (*callback_fn)(channel->channel_callback_context);
449
450 if (channel->callback_mode != HV_CALL_BATCHED)
451 return;
452
453 if (likely(hv_end_read(&channel->inbound) == 0))
454 return;
455
456 hv_begin_read(&channel->inbound);
457 tasklet_schedule(&channel->callback_event);
458 }
459
460 /*
461 * vmbus_post_msg - Send a msg on the vmbus's message connection
462 */
vmbus_post_msg(void * buffer,size_t buflen,bool can_sleep)463 int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep)
464 {
465 struct vmbus_channel_message_header *hdr;
466 union hv_connection_id conn_id;
467 int ret = 0;
468 int retries = 0;
469 u32 usec = 1;
470
471 conn_id.asu32 = 0;
472 conn_id.u.id = vmbus_connection.msg_conn_id;
473
474 /*
475 * hv_post_message() can have transient failures because of
476 * insufficient resources. Retry the operation a couple of
477 * times before giving up.
478 */
479 while (retries < 100) {
480 ret = hv_post_message(conn_id, 1, buffer, buflen);
481
482 switch (ret) {
483 case HV_STATUS_INVALID_CONNECTION_ID:
484 /*
485 * See vmbus_negotiate_version(): VMBus protocol 5.0
486 * and higher require that we must use
487 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate
488 * Contact message, but on old hosts that only
489 * support VMBus protocol 4.0 or lower, here we get
490 * HV_STATUS_INVALID_CONNECTION_ID and we should
491 * return an error immediately without retrying.
492 */
493 hdr = buffer;
494 if (hdr->msgtype == CHANNELMSG_INITIATE_CONTACT)
495 return -EINVAL;
496 /*
497 * We could get this if we send messages too
498 * frequently.
499 */
500 ret = -EAGAIN;
501 break;
502 case HV_STATUS_INSUFFICIENT_MEMORY:
503 case HV_STATUS_INSUFFICIENT_BUFFERS:
504 ret = -ENOBUFS;
505 break;
506 case HV_STATUS_SUCCESS:
507 return ret;
508 default:
509 pr_err("hv_post_msg() failed; error code:%d\n", ret);
510 return -EINVAL;
511 }
512
513 retries++;
514 if (can_sleep && usec > 1000)
515 msleep(usec / 1000);
516 else if (usec < MAX_UDELAY_MS * 1000)
517 udelay(usec);
518 else
519 mdelay(usec / 1000);
520
521 if (retries < 22)
522 usec *= 2;
523 }
524 return ret;
525 }
526
527 /*
528 * vmbus_set_event - Send an event notification to the parent
529 */
vmbus_set_event(struct vmbus_channel * channel)530 void vmbus_set_event(struct vmbus_channel *channel)
531 {
532 u32 child_relid = channel->offermsg.child_relid;
533
534 if (!channel->is_dedicated_interrupt)
535 vmbus_send_interrupt(child_relid);
536
537 ++channel->sig_events;
538
539 if (hv_isolation_type_snp())
540 hv_ghcb_hypercall(HVCALL_SIGNAL_EVENT, &channel->sig_event,
541 NULL, sizeof(channel->sig_event));
542 else
543 hv_do_fast_hypercall8(HVCALL_SIGNAL_EVENT, channel->sig_event);
544 }
545 EXPORT_SYMBOL_GPL(vmbus_set_event);
546