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 75 static int vmbus_try_connection_id(struct vmbus_channel_msginfo *msginfo, 76 u32 version, u32 connection_id) 77 { 78 int ret = 0; 79 struct vmbus_channel_initiate_contact *msg; 80 unsigned long flags; 81 82 init_completion(&msginfo->waitevent); 83 84 msg = (struct vmbus_channel_initiate_contact *)msginfo->msg; 85 86 memset(msg, 0, sizeof(*msg)); 87 msg->header.msgtype = CHANNELMSG_INITIATE_CONTACT; 88 msg->vmbus_version_requested = version; 89 90 /* 91 * For VMBus protocol 5.0 (VERSION_WIN10_V5) and higher, use the 92 * caller-supplied connection_id for the Initiate Contact message so 93 * the caller can implement the required retry scheme. For subsequent 94 * messages, use the Message Connection ID field in the host-returned 95 * Version Response message. With VERSION_WIN10_V5 and higher, we don't 96 * use msg->interrupt_page, but tell the host explicitly that we still 97 * use VMBUS_MESSAGE_SINT(2) for compatibility. 98 * 99 * On old hosts, we should always use VMBUS_MESSAGE_CONNECTION_ID (1). 100 */ 101 if (version >= VERSION_WIN10_V5) { 102 msg->msg_sint = VMBUS_MESSAGE_SINT; 103 msg->msg_vtl = ms_hyperv.vtl; 104 vmbus_connection.msg_conn_id = connection_id; 105 } else { 106 msg->interrupt_page = virt_to_phys(vmbus_connection.int_page); 107 vmbus_connection.msg_conn_id = VMBUS_MESSAGE_CONNECTION_ID; 108 } 109 110 if (vmbus_is_confidential() && version >= VERSION_WIN10_V6_0) 111 msg->feature_flags = VMBUS_FEATURE_FLAG_CONFIDENTIAL_CHANNELS; 112 113 /* 114 * shared_gpa_boundary is zero in non-SNP VMs, so it's safe to always 115 * bitwise OR it 116 */ 117 msg->monitor_page1 = virt_to_phys(vmbus_connection.monitor_pages[0]) | 118 ms_hyperv.shared_gpa_boundary; 119 msg->monitor_page2 = virt_to_phys(vmbus_connection.monitor_pages[1]) | 120 ms_hyperv.shared_gpa_boundary; 121 122 msg->target_vcpu = hv_cpu_number_to_vp_number(VMBUS_CONNECT_CPU); 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 int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo, u32 version) 170 { 171 int ret; 172 173 /* Try the redirect ID first for VTL2 with VMBus protocol 5.0+. */ 174 if (version >= VERSION_WIN10_V5 && ms_hyperv.vtl == 2) { 175 ret = vmbus_try_connection_id(msginfo, version, 176 VMBUS_MESSAGE_CONNECTION_ID_REDIRECT); 177 if (ret != -ENXIO) 178 return ret; 179 } 180 181 return vmbus_try_connection_id(msginfo, version, 182 VMBUS_MESSAGE_CONNECTION_ID_4); 183 } 184 185 /* 186 * vmbus_connect - Sends a connect request on the partition service connection 187 */ 188 int vmbus_connect(void) 189 { 190 struct vmbus_channel_msginfo *msginfo = NULL; 191 int i, ret = 0; 192 __u32 version; 193 194 /* Initialize the vmbus connection */ 195 vmbus_connection.conn_state = CONNECTING; 196 vmbus_connection.work_queue = create_workqueue("hv_vmbus_con"); 197 if (!vmbus_connection.work_queue) { 198 ret = -ENOMEM; 199 goto cleanup; 200 } 201 202 vmbus_connection.rescind_work_queue = 203 create_workqueue("hv_vmbus_rescind"); 204 if (!vmbus_connection.rescind_work_queue) { 205 ret = -ENOMEM; 206 goto cleanup; 207 } 208 vmbus_connection.ignore_any_offer_msg = false; 209 210 vmbus_connection.handle_primary_chan_wq = 211 create_workqueue("hv_pri_chan"); 212 if (!vmbus_connection.handle_primary_chan_wq) { 213 ret = -ENOMEM; 214 goto cleanup; 215 } 216 217 vmbus_connection.handle_sub_chan_wq = 218 create_workqueue("hv_sub_chan"); 219 if (!vmbus_connection.handle_sub_chan_wq) { 220 ret = -ENOMEM; 221 goto cleanup; 222 } 223 224 INIT_LIST_HEAD(&vmbus_connection.chn_msg_list); 225 spin_lock_init(&vmbus_connection.channelmsg_lock); 226 227 INIT_LIST_HEAD(&vmbus_connection.chn_list); 228 mutex_init(&vmbus_connection.channel_mutex); 229 230 /* 231 * The following Hyper-V interrupt and monitor pages can be used by 232 * UIO for mapping to user-space, so they should always be allocated on 233 * system page boundaries. The system page size must be >= the Hyper-V 234 * page size. 235 */ 236 BUILD_BUG_ON(PAGE_SIZE < HV_HYP_PAGE_SIZE); 237 238 /* 239 * Setup the vmbus event connection for channel interrupt 240 * abstraction stuff 241 */ 242 vmbus_connection.int_page = 243 (void *)__get_free_page(GFP_KERNEL | __GFP_ZERO); 244 if (vmbus_connection.int_page == NULL) { 245 ret = -ENOMEM; 246 goto cleanup; 247 } 248 249 vmbus_connection.recv_int_page = vmbus_connection.int_page; 250 vmbus_connection.send_int_page = 251 (void *)((unsigned long)vmbus_connection.int_page + 252 (HV_HYP_PAGE_SIZE >> 1)); 253 254 /* 255 * Setup the monitor notification facility. The 1st page for 256 * parent->child and the 2nd page for child->parent 257 */ 258 vmbus_connection.monitor_pages[0] = (void *)__get_free_page(GFP_KERNEL); 259 vmbus_connection.monitor_pages[1] = (void *)__get_free_page(GFP_KERNEL); 260 if ((vmbus_connection.monitor_pages[0] == NULL) || 261 (vmbus_connection.monitor_pages[1] == NULL)) { 262 ret = -ENOMEM; 263 goto cleanup; 264 } 265 266 ret = set_memory_decrypted((unsigned long) 267 vmbus_connection.monitor_pages[0], 1); 268 ret |= set_memory_decrypted((unsigned long) 269 vmbus_connection.monitor_pages[1], 1); 270 if (ret) { 271 /* 272 * If set_memory_decrypted() fails, the encryption state 273 * of the memory is unknown. So leak the memory instead 274 * of risking returning decrypted memory to the free list. 275 * For simplicity, always handle both pages the same. 276 */ 277 vmbus_connection.monitor_pages[0] = NULL; 278 vmbus_connection.monitor_pages[1] = NULL; 279 goto cleanup; 280 } 281 282 /* 283 * Set_memory_decrypted() will change the memory contents if 284 * decryption occurs, so zero monitor pages here. 285 */ 286 memset(vmbus_connection.monitor_pages[0], 0x00, HV_HYP_PAGE_SIZE); 287 memset(vmbus_connection.monitor_pages[1], 0x00, HV_HYP_PAGE_SIZE); 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 = kzalloc_objs(struct vmbus_channel *, 334 MAX_CHANNEL_RELIDS); 335 if (vmbus_connection.channels == NULL) { 336 ret = -ENOMEM; 337 goto cleanup; 338 } 339 340 kfree(msginfo); 341 return 0; 342 343 cleanup: 344 pr_err("Unable to connect to host\n"); 345 346 vmbus_connection.conn_state = DISCONNECTED; 347 vmbus_disconnect(); 348 349 kfree(msginfo); 350 351 return ret; 352 } 353 354 void vmbus_disconnect(void) 355 { 356 /* 357 * First send the unload request to the host. 358 */ 359 vmbus_initiate_unload(false); 360 361 if (vmbus_connection.handle_sub_chan_wq) 362 destroy_workqueue(vmbus_connection.handle_sub_chan_wq); 363 364 if (vmbus_connection.handle_primary_chan_wq) 365 destroy_workqueue(vmbus_connection.handle_primary_chan_wq); 366 367 if (vmbus_connection.rescind_work_queue) 368 destroy_workqueue(vmbus_connection.rescind_work_queue); 369 370 if (vmbus_connection.work_queue) 371 destroy_workqueue(vmbus_connection.work_queue); 372 373 if (vmbus_connection.int_page) { 374 free_page((unsigned long)vmbus_connection.int_page); 375 vmbus_connection.int_page = NULL; 376 } 377 378 if (vmbus_connection.monitor_pages[0]) { 379 if (!set_memory_encrypted( 380 (unsigned long)vmbus_connection.monitor_pages[0], 1)) 381 free_page((unsigned long) 382 vmbus_connection.monitor_pages[0]); 383 vmbus_connection.monitor_pages[0] = NULL; 384 } 385 386 if (vmbus_connection.monitor_pages[1]) { 387 if (!set_memory_encrypted( 388 (unsigned long)vmbus_connection.monitor_pages[1], 1)) 389 free_page((unsigned long) 390 vmbus_connection.monitor_pages[1]); 391 vmbus_connection.monitor_pages[1] = NULL; 392 } 393 } 394 395 /* 396 * relid2channel - Get the channel object given its 397 * child relative id (ie channel id) 398 */ 399 struct vmbus_channel *relid2channel(u32 relid) 400 { 401 if (vmbus_connection.channels == NULL) { 402 pr_warn_once("relid2channel: relid=%d: No channels mapped!\n", relid); 403 return NULL; 404 } 405 if (WARN_ON(relid >= MAX_CHANNEL_RELIDS)) 406 return NULL; 407 return READ_ONCE(vmbus_connection.channels[relid]); 408 } 409 410 /* 411 * vmbus_on_event - Process a channel event notification 412 * 413 * For batched channels (default) optimize host to guest signaling 414 * by ensuring: 415 * 1. While reading the channel, we disable interrupts from host. 416 * 2. Ensure that we process all posted messages from the host 417 * before returning from this callback. 418 * 3. Once we return, enable signaling from the host. Once this 419 * state is set we check to see if additional packets are 420 * available to read. In this case we repeat the process. 421 * If this tasklet has been running for a long time 422 * then reschedule ourselves. 423 */ 424 void vmbus_on_event(unsigned long data) 425 { 426 struct vmbus_channel *channel = (void *) data; 427 void (*callback_fn)(void *context); 428 429 trace_vmbus_on_event(channel); 430 431 hv_debug_delay_test(channel, INTERRUPT_DELAY); 432 433 /* A channel once created is persistent even when 434 * there is no driver handling the device. An 435 * unloading driver sets the onchannel_callback to NULL. 436 */ 437 callback_fn = READ_ONCE(channel->onchannel_callback); 438 if (unlikely(!callback_fn)) 439 return; 440 441 (*callback_fn)(channel->channel_callback_context); 442 443 if (channel->callback_mode != HV_CALL_BATCHED) 444 return; 445 446 if (likely(hv_end_read(&channel->inbound) == 0)) 447 return; 448 449 hv_begin_read(&channel->inbound); 450 tasklet_schedule(&channel->callback_event); 451 } 452 453 /* 454 * vmbus_post_msg - Send a msg on the vmbus's message connection 455 */ 456 int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep) 457 { 458 struct vmbus_channel_message_header *hdr; 459 union hv_connection_id conn_id; 460 int ret = 0; 461 int retries = 0; 462 u32 usec = 1; 463 464 conn_id.asu32 = 0; 465 conn_id.u.id = vmbus_connection.msg_conn_id; 466 467 /* 468 * hv_post_message() can have transient failures because of 469 * insufficient resources. Retry the operation a couple of 470 * times before giving up. 471 */ 472 while (retries < 100) { 473 ret = hv_post_message(conn_id, 1, buffer, buflen); 474 475 switch (ret) { 476 case HV_STATUS_INVALID_CONNECTION_ID: 477 /* Allow INITIATE_CONTACT to try another connection ID. */ 478 hdr = buffer; 479 if (hdr->msgtype == CHANNELMSG_INITIATE_CONTACT) 480 return -ENXIO; 481 /* 482 * We could get this if we send messages too 483 * frequently. 484 */ 485 ret = -EAGAIN; 486 break; 487 case HV_STATUS_INSUFFICIENT_MEMORY: 488 case HV_STATUS_INSUFFICIENT_BUFFERS: 489 ret = -ENOBUFS; 490 break; 491 case HV_STATUS_SUCCESS: 492 return ret; 493 default: 494 pr_err("hv_post_msg() failed; error code:%d\n", ret); 495 return -EINVAL; 496 } 497 498 retries++; 499 if (can_sleep && usec > 1000) 500 msleep(usec / 1000); 501 else if (usec < MAX_UDELAY_MS * 1000) 502 udelay(usec); 503 else 504 mdelay(usec / 1000); 505 506 if (retries < 22) 507 usec *= 2; 508 } 509 return ret; 510 } 511 512 /* 513 * vmbus_set_event - Send an event notification to the parent 514 */ 515 void vmbus_set_event(struct vmbus_channel *channel) 516 { 517 u32 child_relid = channel->offermsg.child_relid; 518 519 if (!channel->is_dedicated_interrupt) 520 vmbus_send_interrupt(child_relid); 521 522 ++channel->sig_events; 523 524 if (ms_hyperv.paravisor_present) { 525 if (hv_isolation_type_snp()) 526 hv_ghcb_hypercall(HVCALL_SIGNAL_EVENT, &channel->sig_event, 527 NULL, sizeof(channel->sig_event)); 528 else if (hv_isolation_type_tdx()) 529 hv_tdx_hypercall(HVCALL_SIGNAL_EVENT | HV_HYPERCALL_FAST_BIT, 530 channel->sig_event, 0); 531 else 532 WARN_ON_ONCE(1); 533 } else { 534 u64 control = HVCALL_SIGNAL_EVENT; 535 536 control |= hv_nested ? HV_HYPERCALL_NESTED : 0; 537 hv_do_fast_hypercall8(control, channel->sig_event); 538 } 539 } 540 EXPORT_SYMBOL_GPL(vmbus_set_event); 541