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.handle_primary_chan_wq =
175 create_workqueue("hv_pri_chan");
176 if (!vmbus_connection.handle_primary_chan_wq) {
177 ret = -ENOMEM;
178 goto cleanup;
179 }
180
181 vmbus_connection.handle_sub_chan_wq =
182 create_workqueue("hv_sub_chan");
183 if (!vmbus_connection.handle_sub_chan_wq) {
184 ret = -ENOMEM;
185 goto cleanup;
186 }
187
188 INIT_LIST_HEAD(&vmbus_connection.chn_msg_list);
189 spin_lock_init(&vmbus_connection.channelmsg_lock);
190
191 INIT_LIST_HEAD(&vmbus_connection.chn_list);
192 mutex_init(&vmbus_connection.channel_mutex);
193
194 /*
195 * Setup the vmbus event connection for channel interrupt
196 * abstraction stuff
197 */
198 vmbus_connection.int_page =
199 (void *)hv_alloc_hyperv_zeroed_page();
200 if (vmbus_connection.int_page == NULL) {
201 ret = -ENOMEM;
202 goto cleanup;
203 }
204
205 vmbus_connection.recv_int_page = vmbus_connection.int_page;
206 vmbus_connection.send_int_page =
207 (void *)((unsigned long)vmbus_connection.int_page +
208 (HV_HYP_PAGE_SIZE >> 1));
209
210 /*
211 * Setup the monitor notification facility. The 1st page for
212 * parent->child and the 2nd page for child->parent
213 */
214 vmbus_connection.monitor_pages[0] = (void *)hv_alloc_hyperv_zeroed_page();
215 vmbus_connection.monitor_pages[1] = (void *)hv_alloc_hyperv_zeroed_page();
216 if ((vmbus_connection.monitor_pages[0] == NULL) ||
217 (vmbus_connection.monitor_pages[1] == NULL)) {
218 ret = -ENOMEM;
219 goto cleanup;
220 }
221
222 vmbus_connection.monitor_pages_original[0]
223 = vmbus_connection.monitor_pages[0];
224 vmbus_connection.monitor_pages_original[1]
225 = vmbus_connection.monitor_pages[1];
226 vmbus_connection.monitor_pages_pa[0]
227 = virt_to_phys(vmbus_connection.monitor_pages[0]);
228 vmbus_connection.monitor_pages_pa[1]
229 = virt_to_phys(vmbus_connection.monitor_pages[1]);
230
231 if (hv_is_isolation_supported()) {
232 ret = set_memory_decrypted((unsigned long)
233 vmbus_connection.monitor_pages[0],
234 1);
235 ret |= set_memory_decrypted((unsigned long)
236 vmbus_connection.monitor_pages[1],
237 1);
238 if (ret)
239 goto cleanup;
240
241 /*
242 * Isolation VM with AMD SNP needs to access monitor page via
243 * address space above shared gpa boundary.
244 */
245 if (hv_isolation_type_snp()) {
246 vmbus_connection.monitor_pages_pa[0] +=
247 ms_hyperv.shared_gpa_boundary;
248 vmbus_connection.monitor_pages_pa[1] +=
249 ms_hyperv.shared_gpa_boundary;
250
251 vmbus_connection.monitor_pages[0]
252 = memremap(vmbus_connection.monitor_pages_pa[0],
253 HV_HYP_PAGE_SIZE,
254 MEMREMAP_WB);
255 if (!vmbus_connection.monitor_pages[0]) {
256 ret = -ENOMEM;
257 goto cleanup;
258 }
259
260 vmbus_connection.monitor_pages[1]
261 = memremap(vmbus_connection.monitor_pages_pa[1],
262 HV_HYP_PAGE_SIZE,
263 MEMREMAP_WB);
264 if (!vmbus_connection.monitor_pages[1]) {
265 ret = -ENOMEM;
266 goto cleanup;
267 }
268 }
269
270 /*
271 * Set memory host visibility hvcall smears memory
272 * and so zero monitor pages here.
273 */
274 memset(vmbus_connection.monitor_pages[0], 0x00,
275 HV_HYP_PAGE_SIZE);
276 memset(vmbus_connection.monitor_pages[1], 0x00,
277 HV_HYP_PAGE_SIZE);
278
279 }
280
281 msginfo = kzalloc(sizeof(*msginfo) +
282 sizeof(struct vmbus_channel_initiate_contact),
283 GFP_KERNEL);
284 if (msginfo == NULL) {
285 ret = -ENOMEM;
286 goto cleanup;
287 }
288
289 /*
290 * Negotiate a compatible VMBUS version number with the
291 * host. We start with the highest number we can support
292 * and work our way down until we negotiate a compatible
293 * version.
294 */
295
296 for (i = 0; ; i++) {
297 if (i == ARRAY_SIZE(vmbus_versions)) {
298 ret = -EDOM;
299 goto cleanup;
300 }
301
302 version = vmbus_versions[i];
303 if (version > max_version)
304 continue;
305
306 ret = vmbus_negotiate_version(msginfo, version);
307 if (ret == -ETIMEDOUT)
308 goto cleanup;
309
310 if (vmbus_connection.conn_state == CONNECTED)
311 break;
312 }
313
314 if (hv_is_isolation_supported() && version < VERSION_WIN10_V5_2) {
315 pr_err("Invalid VMBus version %d.%d (expected >= %d.%d) from the host supporting isolation\n",
316 version >> 16, version & 0xFFFF, VERSION_WIN10_V5_2 >> 16, VERSION_WIN10_V5_2 & 0xFFFF);
317 ret = -EINVAL;
318 goto cleanup;
319 }
320
321 vmbus_proto_version = version;
322 pr_info("Vmbus version:%d.%d\n",
323 version >> 16, version & 0xFFFF);
324
325 vmbus_connection.channels = kcalloc(MAX_CHANNEL_RELIDS,
326 sizeof(struct vmbus_channel *),
327 GFP_KERNEL);
328 if (vmbus_connection.channels == NULL) {
329 ret = -ENOMEM;
330 goto cleanup;
331 }
332
333 kfree(msginfo);
334 return 0;
335
336 cleanup:
337 pr_err("Unable to connect to host\n");
338
339 vmbus_connection.conn_state = DISCONNECTED;
340 vmbus_disconnect();
341
342 kfree(msginfo);
343
344 return ret;
345 }
346
vmbus_disconnect(void)347 void vmbus_disconnect(void)
348 {
349 /*
350 * First send the unload request to the host.
351 */
352 vmbus_initiate_unload(false);
353
354 if (vmbus_connection.handle_sub_chan_wq)
355 destroy_workqueue(vmbus_connection.handle_sub_chan_wq);
356
357 if (vmbus_connection.handle_primary_chan_wq)
358 destroy_workqueue(vmbus_connection.handle_primary_chan_wq);
359
360 if (vmbus_connection.work_queue)
361 destroy_workqueue(vmbus_connection.work_queue);
362
363 if (vmbus_connection.int_page) {
364 hv_free_hyperv_page((unsigned long)vmbus_connection.int_page);
365 vmbus_connection.int_page = NULL;
366 }
367
368 if (hv_is_isolation_supported()) {
369 /*
370 * memunmap() checks input address is ioremap address or not
371 * inside. It doesn't unmap any thing in the non-SNP CVM and
372 * so not check CVM type here.
373 */
374 memunmap(vmbus_connection.monitor_pages[0]);
375 memunmap(vmbus_connection.monitor_pages[1]);
376
377 set_memory_encrypted((unsigned long)
378 vmbus_connection.monitor_pages_original[0],
379 1);
380 set_memory_encrypted((unsigned long)
381 vmbus_connection.monitor_pages_original[1],
382 1);
383 }
384
385 hv_free_hyperv_page((unsigned long)
386 vmbus_connection.monitor_pages_original[0]);
387 hv_free_hyperv_page((unsigned long)
388 vmbus_connection.monitor_pages_original[1]);
389 vmbus_connection.monitor_pages_original[0] =
390 vmbus_connection.monitor_pages[0] = NULL;
391 vmbus_connection.monitor_pages_original[1] =
392 vmbus_connection.monitor_pages[1] = NULL;
393 }
394
395 /*
396 * relid2channel - Get the channel object given its
397 * child relative id (ie channel id)
398 */
relid2channel(u32 relid)399 struct vmbus_channel *relid2channel(u32 relid)
400 {
401 if (WARN_ON(relid >= MAX_CHANNEL_RELIDS))
402 return NULL;
403 return READ_ONCE(vmbus_connection.channels[relid]);
404 }
405
406 /*
407 * vmbus_on_event - Process a channel event notification
408 *
409 * For batched channels (default) optimize host to guest signaling
410 * by ensuring:
411 * 1. While reading the channel, we disable interrupts from host.
412 * 2. Ensure that we process all posted messages from the host
413 * before returning from this callback.
414 * 3. Once we return, enable signaling from the host. Once this
415 * state is set we check to see if additional packets are
416 * available to read. In this case we repeat the process.
417 * If this tasklet has been running for a long time
418 * then reschedule ourselves.
419 */
vmbus_on_event(unsigned long data)420 void vmbus_on_event(unsigned long data)
421 {
422 struct vmbus_channel *channel = (void *) data;
423 unsigned long time_limit = jiffies + 2;
424
425 trace_vmbus_on_event(channel);
426
427 hv_debug_delay_test(channel, INTERRUPT_DELAY);
428 do {
429 void (*callback_fn)(void *);
430
431 /* A channel once created is persistent even when
432 * there is no driver handling the device. An
433 * unloading driver sets the onchannel_callback to NULL.
434 */
435 callback_fn = READ_ONCE(channel->onchannel_callback);
436 if (unlikely(callback_fn == NULL))
437 return;
438
439 (*callback_fn)(channel->channel_callback_context);
440
441 if (channel->callback_mode != HV_CALL_BATCHED)
442 return;
443
444 if (likely(hv_end_read(&channel->inbound) == 0))
445 return;
446
447 hv_begin_read(&channel->inbound);
448 } while (likely(time_before(jiffies, time_limit)));
449
450 /* The time limit (2 jiffies) has been reached */
451 tasklet_schedule(&channel->callback_event);
452 }
453
454 /*
455 * vmbus_post_msg - Send a msg on the vmbus's message connection
456 */
vmbus_post_msg(void * buffer,size_t buflen,bool can_sleep)457 int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep)
458 {
459 struct vmbus_channel_message_header *hdr;
460 union hv_connection_id conn_id;
461 int ret = 0;
462 int retries = 0;
463 u32 usec = 1;
464
465 conn_id.asu32 = 0;
466 conn_id.u.id = vmbus_connection.msg_conn_id;
467
468 /*
469 * hv_post_message() can have transient failures because of
470 * insufficient resources. Retry the operation a couple of
471 * times before giving up.
472 */
473 while (retries < 100) {
474 ret = hv_post_message(conn_id, 1, buffer, buflen);
475
476 switch (ret) {
477 case HV_STATUS_INVALID_CONNECTION_ID:
478 /*
479 * See vmbus_negotiate_version(): VMBus protocol 5.0
480 * and higher require that we must use
481 * VMBUS_MESSAGE_CONNECTION_ID_4 for the Initiate
482 * Contact message, but on old hosts that only
483 * support VMBus protocol 4.0 or lower, here we get
484 * HV_STATUS_INVALID_CONNECTION_ID and we should
485 * return an error immediately without retrying.
486 */
487 hdr = buffer;
488 if (hdr->msgtype == CHANNELMSG_INITIATE_CONTACT)
489 return -EINVAL;
490 /*
491 * We could get this if we send messages too
492 * frequently.
493 */
494 ret = -EAGAIN;
495 break;
496 case HV_STATUS_INSUFFICIENT_MEMORY:
497 case HV_STATUS_INSUFFICIENT_BUFFERS:
498 ret = -ENOBUFS;
499 break;
500 case HV_STATUS_SUCCESS:
501 return ret;
502 default:
503 pr_err("hv_post_msg() failed; error code:%d\n", ret);
504 return -EINVAL;
505 }
506
507 retries++;
508 if (can_sleep && usec > 1000)
509 msleep(usec / 1000);
510 else if (usec < MAX_UDELAY_MS * 1000)
511 udelay(usec);
512 else
513 mdelay(usec / 1000);
514
515 if (retries < 22)
516 usec *= 2;
517 }
518 return ret;
519 }
520
521 /*
522 * vmbus_set_event - Send an event notification to the parent
523 */
vmbus_set_event(struct vmbus_channel * channel)524 void vmbus_set_event(struct vmbus_channel *channel)
525 {
526 u32 child_relid = channel->offermsg.child_relid;
527
528 if (!channel->is_dedicated_interrupt)
529 vmbus_send_interrupt(child_relid);
530
531 ++channel->sig_events;
532
533 if (hv_isolation_type_snp())
534 hv_ghcb_hypercall(HVCALL_SIGNAL_EVENT, &channel->sig_event,
535 NULL, sizeof(channel->sig_event));
536 else
537 hv_do_fast_hypercall8(HVCALL_SIGNAL_EVENT, channel->sig_event);
538 }
539 EXPORT_SYMBOL_GPL(vmbus_set_event);
540