1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* 3 * 4 * Copyright (c) 2011, Microsoft Corporation. 5 * 6 * Authors: 7 * Haiyang Zhang <haiyangz@microsoft.com> 8 * Hank Janssen <hjanssen@microsoft.com> 9 * K. Y. Srinivasan <kys@microsoft.com> 10 */ 11 12 #ifndef _HYPERV_VMBUS_H 13 #define _HYPERV_VMBUS_H 14 15 #include <linux/list.h> 16 #include <linux/bitops.h> 17 #include <asm/sync_bitops.h> 18 #include <asm/mshyperv.h> 19 #include <linux/atomic.h> 20 #include <linux/hyperv.h> 21 #include <linux/interrupt.h> 22 #include <hyperv/hvhdk.h> 23 24 #include "hv_trace.h" 25 26 /* 27 * Timeout for services such as KVP and fcopy. 28 */ 29 #define HV_UTIL_TIMEOUT 30 30 31 /* 32 * Timeout for guest-host handshake for services. 33 */ 34 #define HV_UTIL_NEGO_TIMEOUT 55 35 36 void vmbus_isr(void); 37 38 /* Definitions for the monitored notification facility */ 39 union hv_monitor_trigger_group { 40 u64 as_uint64; 41 struct { 42 u32 pending; 43 u32 armed; 44 }; 45 }; 46 47 struct hv_monitor_parameter { 48 union hv_connection_id connectionid; 49 u16 flagnumber; 50 u16 rsvdz; 51 }; 52 53 union hv_monitor_trigger_state { 54 u32 asu32; 55 56 struct { 57 u32 group_enable:4; 58 u32 rsvdz:28; 59 }; 60 }; 61 62 /* struct hv_monitor_page Layout */ 63 /* ------------------------------------------------------ */ 64 /* | 0 | TriggerState (4 bytes) | Rsvd1 (4 bytes) | */ 65 /* | 8 | TriggerGroup[0] | */ 66 /* | 10 | TriggerGroup[1] | */ 67 /* | 18 | TriggerGroup[2] | */ 68 /* | 20 | TriggerGroup[3] | */ 69 /* | 28 | Rsvd2[0] | */ 70 /* | 30 | Rsvd2[1] | */ 71 /* | 38 | Rsvd2[2] | */ 72 /* | 40 | NextCheckTime[0][0] | NextCheckTime[0][1] | */ 73 /* | ... | */ 74 /* | 240 | Latency[0][0..3] | */ 75 /* | 340 | Rsvz3[0] | */ 76 /* | 440 | Parameter[0][0] | */ 77 /* | 448 | Parameter[0][1] | */ 78 /* | ... | */ 79 /* | 840 | Rsvd4[0] | */ 80 /* ------------------------------------------------------ */ 81 struct hv_monitor_page { 82 union hv_monitor_trigger_state trigger_state; 83 u32 rsvdz1; 84 85 union hv_monitor_trigger_group trigger_group[4]; 86 u64 rsvdz2[3]; 87 88 s32 next_checktime[4][32]; 89 90 u16 latency[4][32]; 91 u64 rsvdz3[32]; 92 93 struct hv_monitor_parameter parameter[4][32]; 94 95 u8 rsvdz4[1984]; 96 }; 97 98 #define HV_HYPERCALL_PARAM_ALIGN sizeof(u64) 99 100 /* Definition of the hv_post_message hypercall input structure. */ 101 struct hv_input_post_message { 102 union hv_connection_id connectionid; 103 u32 reserved; 104 u32 message_type; 105 u32 payload_size; 106 u64 payload[HV_MESSAGE_PAYLOAD_QWORD_COUNT]; 107 }; 108 109 110 enum { 111 VMBUS_MESSAGE_CONNECTION_ID = 1, 112 VMBUS_MESSAGE_CONNECTION_ID_4 = 4, 113 /* VTL2 redirect connection ID for INITIATE_CONTACT. */ 114 VMBUS_MESSAGE_CONNECTION_ID_REDIRECT = 0x800074, 115 VMBUS_MESSAGE_PORT_ID = 1, 116 VMBUS_EVENT_CONNECTION_ID = 2, 117 VMBUS_EVENT_PORT_ID = 2, 118 VMBUS_MONITOR_CONNECTION_ID = 3, 119 VMBUS_MONITOR_PORT_ID = 3, 120 VMBUS_MESSAGE_SINT = 2, 121 }; 122 123 /* 124 * Per cpu state for channel handling 125 */ 126 struct hv_per_cpu_context { 127 /* 128 * SynIC pages for communicating with the host. 129 * 130 * These pages are accessible to the host partition and the hypervisor. 131 * They may be used for exchanging data with the host partition and the 132 * hypervisor even when they aren't trusted yet the guest partition 133 * must be prepared to handle the malicious behavior. 134 */ 135 void *hyp_synic_message_page; 136 void *hyp_synic_event_page; 137 /* 138 * SynIC pages for communicating with the paravisor. 139 * 140 * These pages may be accessed from within the guest partition only in 141 * CoCo VMs. Neither the host partition nor the hypervisor can access 142 * these pages in that case; they are used for exchanging data with the 143 * paravisor. 144 */ 145 void *para_synic_message_page; 146 void *para_synic_event_page; 147 148 /* 149 * The page is only used in hv_post_message() for a TDX VM (with the 150 * paravisor) to post a messages to Hyper-V: when such a VM calls 151 * HVCALL_POST_MESSAGE, it can't use the hyperv_pcpu_input_arg (which 152 * is encrypted in such a VM) as the hypercall input page, because 153 * the input page for HVCALL_POST_MESSAGE must be decrypted in such a 154 * VM, so post_msg_page (which is decrypted in hv_synic_alloc()) is 155 * introduced for this purpose. See hyperv_init() for more comments. 156 */ 157 void *post_msg_page; 158 159 /* 160 * Starting with win8, we can take channel interrupts on any CPU; 161 * we will manage the tasklet that handles events messages on a per CPU 162 * basis. 163 */ 164 struct tasklet_struct msg_dpc; 165 }; 166 167 struct hv_context { 168 /* We only support running on top of Hyper-V 169 * So at this point this really can only contain the Hyper-V ID 170 */ 171 u64 guestid; 172 173 struct hv_per_cpu_context __percpu *cpu_context; 174 175 /* 176 * To manage allocations in a NUMA node. 177 * Array indexed by numa node ID. 178 */ 179 struct cpumask *hv_numa_map; 180 }; 181 182 extern struct hv_context hv_context; 183 184 /* Hv Interface */ 185 186 extern int hv_init(void); 187 188 extern int hv_post_message(union hv_connection_id connection_id, 189 enum hv_message_type message_type, 190 void *payload, size_t payload_size); 191 192 extern int hv_synic_alloc(void); 193 194 extern void hv_synic_free(void); 195 196 extern void hv_hyp_synic_enable_regs(unsigned int cpu); 197 extern int hv_synic_init(unsigned int cpu); 198 199 extern void hv_hyp_synic_disable_regs(unsigned int cpu); 200 extern int hv_synic_cleanup(unsigned int cpu); 201 202 /* Interface */ 203 204 void hv_ringbuffer_pre_init(struct vmbus_channel *channel); 205 206 int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info, 207 struct page *pages, u32 pagecnt, u32 max_pkt_size, 208 bool confidential); 209 210 void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info); 211 212 int hv_ringbuffer_write(struct vmbus_channel *channel, 213 const struct kvec *kv_list, u32 kv_count, 214 u64 requestid, u64 *trans_id); 215 216 int hv_ringbuffer_read(struct vmbus_channel *channel, 217 void *buffer, u32 buflen, u32 *buffer_actual_len, 218 u64 *requestid, bool raw); 219 220 /* 221 * The Maximum number of channels (16384) is determined by the size of the 222 * interrupt page, which is HV_HYP_PAGE_SIZE. 1/2 of HV_HYP_PAGE_SIZE is to 223 * send endpoint interrupts, and the other is to receive endpoint interrupts. 224 */ 225 #define MAX_NUM_CHANNELS ((HV_HYP_PAGE_SIZE >> 1) << 3) 226 227 /* The value here must be in multiple of 32 */ 228 #define MAX_NUM_CHANNELS_SUPPORTED 256 229 230 #define MAX_CHANNEL_RELIDS \ 231 max(MAX_NUM_CHANNELS_SUPPORTED, HV_EVENT_FLAGS_COUNT) 232 233 enum vmbus_connect_state { 234 DISCONNECTED, 235 CONNECTING, 236 CONNECTED, 237 DISCONNECTING 238 }; 239 240 #define MAX_SIZE_CHANNEL_MESSAGE HV_MESSAGE_PAYLOAD_BYTE_COUNT 241 242 /* 243 * The CPU that Hyper-V will interrupt for VMBUS messages, such as 244 * CHANNELMSG_OFFERCHANNEL and CHANNELMSG_RESCIND_CHANNELOFFER. 245 */ 246 #define VMBUS_CONNECT_CPU 0 247 248 struct vmbus_connection { 249 u32 msg_conn_id; 250 251 atomic_t offer_in_progress; 252 253 enum vmbus_connect_state conn_state; 254 255 atomic_t next_gpadl_handle; 256 257 struct completion unload_event; 258 /* 259 * Represents channel interrupts. Each bit position represents a 260 * channel. When a channel sends an interrupt via VMBUS, it finds its 261 * bit in the sendInterruptPage, set it and calls Hv to generate a port 262 * event. The other end receives the port event and parse the 263 * recvInterruptPage to see which bit is set 264 */ 265 void *int_page; 266 void *send_int_page; 267 void *recv_int_page; 268 269 /* 270 * 2 pages - 1st page for parent->child notification and 2nd 271 * is child->parent notification 272 */ 273 struct hv_monitor_page *monitor_pages[2]; 274 struct list_head chn_msg_list; 275 spinlock_t channelmsg_lock; 276 277 /* List of channels */ 278 struct list_head chn_list; 279 struct mutex channel_mutex; 280 281 /* Array of channel pointers, indexed by relid */ 282 struct vmbus_channel **channels; 283 u32 relid_hiwater; 284 285 /* 286 * An offer message is handled first on the work_queue, and then 287 * is further handled on handle_primary_chan_wq or 288 * handle_sub_chan_wq. 289 */ 290 struct workqueue_struct *work_queue; 291 struct workqueue_struct *handle_primary_chan_wq; 292 struct workqueue_struct *handle_sub_chan_wq; 293 struct workqueue_struct *rescind_work_queue; 294 295 /* 296 * On suspension of the vmbus, the accumulated offer messages 297 * must be dropped. 298 */ 299 bool ignore_any_offer_msg; 300 301 /* 302 * The number of sub-channels and hv_sock channels that should be 303 * cleaned up upon suspend: sub-channels will be re-created upon 304 * resume, and hv_sock channels should not survive suspend. 305 */ 306 atomic_t nr_chan_close_on_suspend; 307 /* 308 * vmbus_bus_suspend() waits for "nr_chan_close_on_suspend" to 309 * drop to zero. 310 */ 311 struct completion ready_for_suspend_event; 312 313 /* 314 * Completed once the host has offered all boot-time channels. 315 * Note that some channels may still be under process on a workqueue. 316 */ 317 struct completion all_offers_delivered_event; 318 }; 319 320 321 struct vmbus_msginfo { 322 /* Bookkeeping stuff */ 323 struct list_head msglist_entry; 324 325 /* The message itself */ 326 unsigned char msg[]; 327 }; 328 329 330 extern struct vmbus_connection vmbus_connection; 331 332 int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo, u32 version); 333 334 static inline void vmbus_send_interrupt(u32 relid) 335 { 336 sync_set_bit(relid, vmbus_connection.send_int_page); 337 } 338 339 enum vmbus_message_handler_type { 340 /* The related handler can sleep. */ 341 VMHT_BLOCKING = 0, 342 343 /* The related handler must NOT sleep. */ 344 VMHT_NON_BLOCKING = 1, 345 }; 346 347 struct vmbus_channel_message_table_entry { 348 enum vmbus_channel_message_type message_type; 349 enum vmbus_message_handler_type handler_type; 350 void (*message_handler)(struct vmbus_channel_message_header *msg); 351 u32 min_payload_len; 352 }; 353 354 extern const struct vmbus_channel_message_table_entry 355 channel_message_table[CHANNELMSG_COUNT]; 356 357 358 /* General vmbus interface */ 359 360 bool vmbus_is_confidential(void); 361 362 #if IS_ENABLED(CONFIG_HYPERV_VMBUS) 363 /* Free the message slot and signal end-of-message if required */ 364 static inline void vmbus_signal_eom(struct hv_message *msg, u32 old_msg_type) 365 { 366 /* 367 * On crash we're reading some other CPU's message page and we need 368 * to be careful: this other CPU may already had cleared the header 369 * and the host may already had delivered some other message there. 370 * In case we blindly write msg->header.message_type we're going 371 * to lose it. We can still lose a message of the same type but 372 * we count on the fact that there can only be one 373 * CHANNELMSG_UNLOAD_RESPONSE and we don't care about other messages 374 * on crash. 375 */ 376 if (!try_cmpxchg(&msg->header.message_type, 377 &old_msg_type, HVMSG_NONE)) 378 return; 379 380 /* 381 * The cmpxchg() above does an implicit memory barrier to 382 * ensure the write to MessageType (ie set to 383 * HVMSG_NONE) happens before we read the 384 * MessagePending and EOMing. Otherwise, the EOMing 385 * will not deliver any more messages since there is 386 * no empty slot 387 */ 388 if (msg->header.message_flags.msg_pending) { 389 /* 390 * This will cause message queue rescan to 391 * possibly deliver another msg from the 392 * hypervisor 393 */ 394 if (vmbus_is_confidential()) 395 hv_para_set_synic_register(HV_MSR_EOM, 0); 396 else 397 hv_set_msr(HV_MSR_EOM, 0); 398 } 399 } 400 401 extern int vmbus_interrupt; 402 extern int vmbus_irq; 403 #endif /* CONFIG_HYPERV_VMBUS */ 404 405 struct hv_device *vmbus_device_create(const guid_t *type, 406 const guid_t *instance, 407 struct vmbus_channel *channel); 408 409 int vmbus_device_register(struct hv_device *child_device_obj); 410 void vmbus_device_unregister(struct hv_device *device_obj); 411 int vmbus_add_channel_kobj(struct hv_device *device_obj, 412 struct vmbus_channel *channel); 413 414 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel); 415 416 void vmbus_channel_map_relid(struct vmbus_channel *channel); 417 void vmbus_channel_unmap_relid(struct vmbus_channel *channel); 418 419 struct vmbus_channel *relid2channel(u32 relid); 420 421 void vmbus_free_channels(void); 422 423 /* Connection interface */ 424 425 int vmbus_connect(void); 426 void vmbus_disconnect(void); 427 428 int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep); 429 430 void vmbus_on_event(unsigned long data); 431 void vmbus_on_msg_dpc(unsigned long data); 432 433 int hv_kvp_init(struct hv_util_service *srv); 434 int hv_kvp_init_transport(void); 435 void hv_kvp_deinit(void); 436 int hv_kvp_pre_suspend(void); 437 int hv_kvp_pre_resume(void); 438 void hv_kvp_onchannelcallback(void *context); 439 440 int hv_vss_init(struct hv_util_service *srv); 441 int hv_vss_init_transport(void); 442 void hv_vss_deinit(void); 443 int hv_vss_pre_suspend(void); 444 int hv_vss_pre_resume(void); 445 void hv_vss_onchannelcallback(void *context); 446 447 static inline void hv_poll_channel(struct vmbus_channel *channel, 448 void (*cb)(void *)) 449 { 450 if (!channel) 451 return; 452 cb(channel); 453 } 454 455 enum hvutil_device_state { 456 HVUTIL_DEVICE_INIT = 0, /* driver is loaded, waiting for userspace */ 457 HVUTIL_READY, /* userspace is registered */ 458 HVUTIL_HOSTMSG_RECEIVED, /* message from the host was received */ 459 HVUTIL_USERSPACE_REQ, /* request to userspace was sent */ 460 HVUTIL_USERSPACE_RECV, /* reply from userspace was received */ 461 HVUTIL_DEVICE_DYING, /* driver unload is in progress */ 462 }; 463 464 enum delay { 465 INTERRUPT_DELAY = 0, 466 MESSAGE_DELAY = 1, 467 }; 468 469 extern const struct vmbus_device vmbus_devs[]; 470 471 static inline bool hv_is_perf_channel(struct vmbus_channel *channel) 472 { 473 return vmbus_devs[channel->device_id].perf_device; 474 } 475 476 static inline size_t hv_dev_ring_size(struct vmbus_channel *channel) 477 { 478 return vmbus_devs[channel->device_id].pref_ring_size; 479 } 480 481 static inline bool hv_is_allocated_cpu(unsigned int cpu) 482 { 483 struct vmbus_channel *channel, *sc; 484 485 lockdep_assert_held(&vmbus_connection.channel_mutex); 486 /* 487 * List additions/deletions as well as updates of the target CPUs are 488 * protected by channel_mutex. 489 */ 490 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) { 491 if (!hv_is_perf_channel(channel)) 492 continue; 493 if (channel->target_cpu == cpu) 494 return true; 495 list_for_each_entry(sc, &channel->sc_list, sc_list) { 496 if (sc->target_cpu == cpu) 497 return true; 498 } 499 } 500 return false; 501 } 502 503 static inline void hv_set_allocated_cpu(unsigned int cpu) 504 { 505 cpumask_set_cpu(cpu, &hv_context.hv_numa_map[cpu_to_node(cpu)]); 506 } 507 508 static inline void hv_clear_allocated_cpu(unsigned int cpu) 509 { 510 if (hv_is_allocated_cpu(cpu)) 511 return; 512 cpumask_clear_cpu(cpu, &hv_context.hv_numa_map[cpu_to_node(cpu)]); 513 } 514 515 static inline void hv_update_allocated_cpus(unsigned int old_cpu, 516 unsigned int new_cpu) 517 { 518 hv_set_allocated_cpu(new_cpu); 519 hv_clear_allocated_cpu(old_cpu); 520 } 521 522 #ifdef CONFIG_HYPERV_TESTING 523 524 int hv_debug_add_dev_dir(struct hv_device *dev); 525 void hv_debug_rm_dev_dir(struct hv_device *dev); 526 void hv_debug_rm_all_dir(void); 527 int hv_debug_init(void); 528 void hv_debug_delay_test(struct vmbus_channel *channel, enum delay delay_type); 529 530 #else /* CONFIG_HYPERV_TESTING */ 531 532 static inline void hv_debug_rm_dev_dir(struct hv_device *dev) {}; 533 static inline void hv_debug_rm_all_dir(void) {}; 534 static inline void hv_debug_delay_test(struct vmbus_channel *channel, 535 enum delay delay_type) {}; 536 static inline int hv_debug_init(void) 537 { 538 return -1; 539 } 540 541 static inline int hv_debug_add_dev_dir(struct hv_device *dev) 542 { 543 return -1; 544 } 545 546 #endif /* CONFIG_HYPERV_TESTING */ 547 548 /* Create and remove sysfs entry for memory mapped ring buffers for a channel */ 549 int hv_create_ring_sysfs(struct vmbus_channel *channel, 550 int (*hv_mmap_prepare_ring_buffer)(struct vmbus_channel *channel, 551 struct vm_area_desc *desc)); 552 int hv_remove_ring_sysfs(struct vmbus_channel *channel); 553 554 #endif /* _HYPERV_VMBUS_H */ 555