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 .all_offers_delivered_event = COMPLETION_INITIALIZER(
38 vmbus_connection.all_offers_delivered_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_V6_0,
55 VERSION_WIN10_V5_3,
56 VERSION_WIN10_V5_2,
57 VERSION_WIN10_V5_1,
58 VERSION_WIN10_V5,
59 VERSION_WIN10_V4_1,
60 VERSION_WIN10,
61 VERSION_WIN8_1,
62 VERSION_WIN8
63 };
64
65 /*
66 * Maximal VMBus protocol version guests can negotiate. Useful to cap the
67 * VMBus version for testing and debugging purpose.
68 */
69 static uint max_version = VERSION_WIN10_V6_0;
70
71 module_param(max_version, uint, S_IRUGO);
72 MODULE_PARM_DESC(max_version,
73 "Maximal VMBus protocol version which can be negotiated");
74
vmbus_negotiate_version(struct vmbus_channel_msginfo * msginfo,u32 version)75 int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo, u32 version)
76 {
77 int ret = 0;
78 struct vmbus_channel_initiate_contact *msg;
79 unsigned long flags;
80
81 init_completion(&msginfo->waitevent);
82
83 msg = (struct vmbus_channel_initiate_contact *)msginfo->msg;
84
85 memset(msg, 0, sizeof(*msg));
86 msg->header.msgtype = CHANNELMSG_INITIATE_CONTACT;
87 msg->vmbus_version_requested = version;
88
89 /*
90 * VMBus protocol 5.0 (VERSION_WIN10_V5) and higher require that we must
91 * use VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate Contact Message,
92 * and for subsequent messages, we must use the Message Connection ID
93 * field in the host-returned Version Response Message. And, with
94 * VERSION_WIN10_V5 and higher, we don't use msg->interrupt_page, but we
95 * tell the host explicitly that we still use VMBUS_MESSAGE_SINT(2) for
96 * compatibility.
97 *
98 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1).
99 */
100 if (version >= VERSION_WIN10_V5) {
101 msg->msg_sint = VMBUS_MESSAGE_SINT;
102 msg->msg_vtl = ms_hyperv.vtl;
103 vmbus_connection.msg_conn_id = VMBUS_MESSAGE_CONNECTION_ID_4;
104 } else {
105 msg->interrupt_page = virt_to_phys(vmbus_connection.int_page);
106 vmbus_connection.msg_conn_id = VMBUS_MESSAGE_CONNECTION_ID;
107 }
108
109 if (vmbus_is_confidential() && version >= VERSION_WIN10_V6_0)
110 msg->feature_flags = VMBUS_FEATURE_FLAG_CONFIDENTIAL_CHANNELS;
111
112 /*
113 * shared_gpa_boundary is zero in non-SNP VMs, so it's safe to always
114 * bitwise OR it
115 */
116 msg->monitor_page1 = virt_to_phys(vmbus_connection.monitor_pages[0]) |
117 ms_hyperv.shared_gpa_boundary;
118 msg->monitor_page2 = virt_to_phys(vmbus_connection.monitor_pages[1]) |
119 ms_hyperv.shared_gpa_boundary;
120
121 msg->target_vcpu = hv_cpu_number_to_vp_number(VMBUS_CONNECT_CPU);
122
123 /*
124 * Add to list before we send the request since we may
125 * receive the response before returning from this routine
126 */
127 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
128 list_add_tail(&msginfo->msglistentry,
129 &vmbus_connection.chn_msg_list);
130
131 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
132
133 ret = vmbus_post_msg(msg,
134 sizeof(struct vmbus_channel_initiate_contact),
135 true);
136
137 trace_vmbus_negotiate_version(msg, ret);
138
139 if (ret != 0) {
140 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
141 list_del(&msginfo->msglistentry);
142 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock,
143 flags);
144 return ret;
145 }
146
147 /* Wait for the connection response */
148 wait_for_completion(&msginfo->waitevent);
149
150 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
151 list_del(&msginfo->msglistentry);
152 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
153
154 /* Check if successful */
155 if (msginfo->response.version_response.version_supported) {
156 vmbus_connection.conn_state = CONNECTED;
157
158 if (version >= VERSION_WIN10_V5)
159 vmbus_connection.msg_conn_id =
160 msginfo->response.version_response.msg_conn_id;
161 } else {
162 return -ECONNREFUSED;
163 }
164
165 return ret;
166 }
167
168 /*
169 * vmbus_connect - Sends a connect request on the partition service connection
170 */
vmbus_connect(void)171 int vmbus_connect(void)
172 {
173 struct vmbus_channel_msginfo *msginfo = NULL;
174 int i, ret = 0;
175 __u32 version;
176
177 /* Initialize the vmbus connection */
178 vmbus_connection.conn_state = CONNECTING;
179 vmbus_connection.work_queue = create_workqueue("hv_vmbus_con");
180 if (!vmbus_connection.work_queue) {
181 ret = -ENOMEM;
182 goto cleanup;
183 }
184
185 vmbus_connection.rescind_work_queue =
186 create_workqueue("hv_vmbus_rescind");
187 if (!vmbus_connection.rescind_work_queue) {
188 ret = -ENOMEM;
189 goto cleanup;
190 }
191 vmbus_connection.ignore_any_offer_msg = false;
192
193 vmbus_connection.handle_primary_chan_wq =
194 create_workqueue("hv_pri_chan");
195 if (!vmbus_connection.handle_primary_chan_wq) {
196 ret = -ENOMEM;
197 goto cleanup;
198 }
199
200 vmbus_connection.handle_sub_chan_wq =
201 create_workqueue("hv_sub_chan");
202 if (!vmbus_connection.handle_sub_chan_wq) {
203 ret = -ENOMEM;
204 goto cleanup;
205 }
206
207 INIT_LIST_HEAD(&vmbus_connection.chn_msg_list);
208 spin_lock_init(&vmbus_connection.channelmsg_lock);
209
210 INIT_LIST_HEAD(&vmbus_connection.chn_list);
211 mutex_init(&vmbus_connection.channel_mutex);
212
213 /*
214 * The following Hyper-V interrupt and monitor pages can be used by
215 * UIO for mapping to user-space, so they should always be allocated on
216 * system page boundaries. The system page size must be >= the Hyper-V
217 * page size.
218 */
219 BUILD_BUG_ON(PAGE_SIZE < HV_HYP_PAGE_SIZE);
220
221 /*
222 * Setup the vmbus event connection for channel interrupt
223 * abstraction stuff
224 */
225 vmbus_connection.int_page =
226 (void *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
227 if (vmbus_connection.int_page == NULL) {
228 ret = -ENOMEM;
229 goto cleanup;
230 }
231
232 vmbus_connection.recv_int_page = vmbus_connection.int_page;
233 vmbus_connection.send_int_page =
234 (void *)((unsigned long)vmbus_connection.int_page +
235 (HV_HYP_PAGE_SIZE >> 1));
236
237 /*
238 * Setup the monitor notification facility. The 1st page for
239 * parent->child and the 2nd page for child->parent
240 */
241 vmbus_connection.monitor_pages[0] = (void *)__get_free_page(GFP_KERNEL);
242 vmbus_connection.monitor_pages[1] = (void *)__get_free_page(GFP_KERNEL);
243 if ((vmbus_connection.monitor_pages[0] == NULL) ||
244 (vmbus_connection.monitor_pages[1] == NULL)) {
245 ret = -ENOMEM;
246 goto cleanup;
247 }
248
249 ret = set_memory_decrypted((unsigned long)
250 vmbus_connection.monitor_pages[0], 1);
251 ret |= set_memory_decrypted((unsigned long)
252 vmbus_connection.monitor_pages[1], 1);
253 if (ret) {
254 /*
255 * If set_memory_decrypted() fails, the encryption state
256 * of the memory is unknown. So leak the memory instead
257 * of risking returning decrypted memory to the free list.
258 * For simplicity, always handle both pages the same.
259 */
260 vmbus_connection.monitor_pages[0] = NULL;
261 vmbus_connection.monitor_pages[1] = NULL;
262 goto cleanup;
263 }
264
265 /*
266 * Set_memory_decrypted() will change the memory contents if
267 * decryption occurs, so zero monitor pages here.
268 */
269 memset(vmbus_connection.monitor_pages[0], 0x00, HV_HYP_PAGE_SIZE);
270 memset(vmbus_connection.monitor_pages[1], 0x00, HV_HYP_PAGE_SIZE);
271
272 msginfo = kzalloc(sizeof(*msginfo) +
273 sizeof(struct vmbus_channel_initiate_contact),
274 GFP_KERNEL);
275 if (msginfo == NULL) {
276 ret = -ENOMEM;
277 goto cleanup;
278 }
279
280 /*
281 * Negotiate a compatible VMBUS version number with the
282 * host. We start with the highest number we can support
283 * and work our way down until we negotiate a compatible
284 * version.
285 */
286
287 for (i = 0; ; i++) {
288 if (i == ARRAY_SIZE(vmbus_versions)) {
289 ret = -EDOM;
290 goto cleanup;
291 }
292
293 version = vmbus_versions[i];
294 if (version > max_version)
295 continue;
296
297 ret = vmbus_negotiate_version(msginfo, version);
298 if (ret == -ETIMEDOUT)
299 goto cleanup;
300
301 if (vmbus_connection.conn_state == CONNECTED)
302 break;
303 }
304
305 if (hv_is_isolation_supported() && version < VERSION_WIN10_V5_2) {
306 pr_err("Invalid VMBus version %d.%d (expected >= %d.%d) from the host supporting isolation\n",
307 version >> 16, version & 0xFFFF, VERSION_WIN10_V5_2 >> 16, VERSION_WIN10_V5_2 & 0xFFFF);
308 ret = -EINVAL;
309 goto cleanup;
310 }
311
312 vmbus_proto_version = version;
313 pr_info("Vmbus version:%d.%d\n",
314 version >> 16, version & 0xFFFF);
315
316 vmbus_connection.channels = kzalloc_objs(struct vmbus_channel *,
317 MAX_CHANNEL_RELIDS);
318 if (vmbus_connection.channels == NULL) {
319 ret = -ENOMEM;
320 goto cleanup;
321 }
322
323 kfree(msginfo);
324 return 0;
325
326 cleanup:
327 pr_err("Unable to connect to host\n");
328
329 vmbus_connection.conn_state = DISCONNECTED;
330 vmbus_disconnect();
331
332 kfree(msginfo);
333
334 return ret;
335 }
336
vmbus_disconnect(void)337 void vmbus_disconnect(void)
338 {
339 /*
340 * First send the unload request to the host.
341 */
342 vmbus_initiate_unload(false);
343
344 if (vmbus_connection.handle_sub_chan_wq)
345 destroy_workqueue(vmbus_connection.handle_sub_chan_wq);
346
347 if (vmbus_connection.handle_primary_chan_wq)
348 destroy_workqueue(vmbus_connection.handle_primary_chan_wq);
349
350 if (vmbus_connection.rescind_work_queue)
351 destroy_workqueue(vmbus_connection.rescind_work_queue);
352
353 if (vmbus_connection.work_queue)
354 destroy_workqueue(vmbus_connection.work_queue);
355
356 if (vmbus_connection.int_page) {
357 free_page((unsigned long)vmbus_connection.int_page);
358 vmbus_connection.int_page = NULL;
359 }
360
361 if (vmbus_connection.monitor_pages[0]) {
362 if (!set_memory_encrypted(
363 (unsigned long)vmbus_connection.monitor_pages[0], 1))
364 free_page((unsigned long)
365 vmbus_connection.monitor_pages[0]);
366 vmbus_connection.monitor_pages[0] = NULL;
367 }
368
369 if (vmbus_connection.monitor_pages[1]) {
370 if (!set_memory_encrypted(
371 (unsigned long)vmbus_connection.monitor_pages[1], 1))
372 free_page((unsigned long)
373 vmbus_connection.monitor_pages[1]);
374 vmbus_connection.monitor_pages[1] = NULL;
375 }
376 }
377
378 /*
379 * relid2channel - Get the channel object given its
380 * child relative id (ie channel id)
381 */
relid2channel(u32 relid)382 struct vmbus_channel *relid2channel(u32 relid)
383 {
384 if (vmbus_connection.channels == NULL) {
385 pr_warn_once("relid2channel: relid=%d: No channels mapped!\n", relid);
386 return NULL;
387 }
388 if (WARN_ON(relid >= MAX_CHANNEL_RELIDS))
389 return NULL;
390 return READ_ONCE(vmbus_connection.channels[relid]);
391 }
392
393 /*
394 * vmbus_on_event - Process a channel event notification
395 *
396 * For batched channels (default) optimize host to guest signaling
397 * by ensuring:
398 * 1. While reading the channel, we disable interrupts from host.
399 * 2. Ensure that we process all posted messages from the host
400 * before returning from this callback.
401 * 3. Once we return, enable signaling from the host. Once this
402 * state is set we check to see if additional packets are
403 * available to read. In this case we repeat the process.
404 * If this tasklet has been running for a long time
405 * then reschedule ourselves.
406 */
vmbus_on_event(unsigned long data)407 void vmbus_on_event(unsigned long data)
408 {
409 struct vmbus_channel *channel = (void *) data;
410 void (*callback_fn)(void *context);
411
412 trace_vmbus_on_event(channel);
413
414 hv_debug_delay_test(channel, INTERRUPT_DELAY);
415
416 /* A channel once created is persistent even when
417 * there is no driver handling the device. An
418 * unloading driver sets the onchannel_callback to NULL.
419 */
420 callback_fn = READ_ONCE(channel->onchannel_callback);
421 if (unlikely(!callback_fn))
422 return;
423
424 (*callback_fn)(channel->channel_callback_context);
425
426 if (channel->callback_mode != HV_CALL_BATCHED)
427 return;
428
429 if (likely(hv_end_read(&channel->inbound) == 0))
430 return;
431
432 hv_begin_read(&channel->inbound);
433 tasklet_schedule(&channel->callback_event);
434 }
435
436 /*
437 * vmbus_post_msg - Send a msg on the vmbus's message connection
438 */
vmbus_post_msg(void * buffer,size_t buflen,bool can_sleep)439 int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep)
440 {
441 struct vmbus_channel_message_header *hdr;
442 union hv_connection_id conn_id;
443 int ret = 0;
444 int retries = 0;
445 u32 usec = 1;
446
447 conn_id.asu32 = 0;
448 conn_id.u.id = vmbus_connection.msg_conn_id;
449
450 /*
451 * hv_post_message() can have transient failures because of
452 * insufficient resources. Retry the operation a couple of
453 * times before giving up.
454 */
455 while (retries < 100) {
456 ret = hv_post_message(conn_id, 1, buffer, buflen);
457
458 switch (ret) {
459 case HV_STATUS_INVALID_CONNECTION_ID:
460 /*
461 * See vmbus_negotiate_version(): VMBus protocol 5.0
462 * and higher require that we must use
463 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate
464 * Contact message, but on old hosts that only
465 * support VMBus protocol 4.0 or lower, here we get
466 * HV_STATUS_INVALID_CONNECTION_ID and we should
467 * return an error immediately without retrying.
468 */
469 hdr = buffer;
470 if (hdr->msgtype == CHANNELMSG_INITIATE_CONTACT)
471 return -EINVAL;
472 /*
473 * We could get this if we send messages too
474 * frequently.
475 */
476 ret = -EAGAIN;
477 break;
478 case HV_STATUS_INSUFFICIENT_MEMORY:
479 case HV_STATUS_INSUFFICIENT_BUFFERS:
480 ret = -ENOBUFS;
481 break;
482 case HV_STATUS_SUCCESS:
483 return ret;
484 default:
485 pr_err("hv_post_msg() failed; error code:%d\n", ret);
486 return -EINVAL;
487 }
488
489 retries++;
490 if (can_sleep && usec > 1000)
491 msleep(usec / 1000);
492 else if (usec < MAX_UDELAY_MS * 1000)
493 udelay(usec);
494 else
495 mdelay(usec / 1000);
496
497 if (retries < 22)
498 usec *= 2;
499 }
500 return ret;
501 }
502
503 /*
504 * vmbus_set_event - Send an event notification to the parent
505 */
vmbus_set_event(struct vmbus_channel * channel)506 void vmbus_set_event(struct vmbus_channel *channel)
507 {
508 u32 child_relid = channel->offermsg.child_relid;
509
510 if (!channel->is_dedicated_interrupt)
511 vmbus_send_interrupt(child_relid);
512
513 ++channel->sig_events;
514
515 if (ms_hyperv.paravisor_present) {
516 if (hv_isolation_type_snp())
517 hv_ghcb_hypercall(HVCALL_SIGNAL_EVENT, &channel->sig_event,
518 NULL, sizeof(channel->sig_event));
519 else if (hv_isolation_type_tdx())
520 hv_tdx_hypercall(HVCALL_SIGNAL_EVENT | HV_HYPERCALL_FAST_BIT,
521 channel->sig_event, 0);
522 else
523 WARN_ON_ONCE(1);
524 } else {
525 u64 control = HVCALL_SIGNAL_EVENT;
526
527 control |= hv_nested ? HV_HYPERCALL_NESTED : 0;
528 hv_do_fast_hypercall8(control, channel->sig_event);
529 }
530 }
531 EXPORT_SYMBOL_GPL(vmbus_set_event);
532