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/hyperv-tlfs.h> 19#include <linux/atomic.h> 20#include <linux/hyperv.h> 21#include <linux/interrupt.h> 22 23#include "hv_trace.h" 24 25/* 26 * Timeout for services such as KVP and fcopy. 27 */ 28#define HV_UTIL_TIMEOUT 30 29 30/* 31 * Timeout for guest-host handshake for services. 32 */ 33#define HV_UTIL_NEGO_TIMEOUT 55 34 35 36/* Definitions for the monitored notification facility */ 37union hv_monitor_trigger_group { 38 u64 as_uint64; 39 struct { 40 u32 pending; 41 u32 armed; 42 }; 43}; 44 45struct hv_monitor_parameter { 46 union hv_connection_id connectionid; 47 u16 flagnumber; 48 u16 rsvdz; 49}; 50 51union hv_monitor_trigger_state { 52 u32 asu32; 53 54 struct { 55 u32 group_enable:4; 56 u32 rsvdz:28; 57 }; 58}; 59 60/* struct hv_monitor_page Layout */ 61/* ------------------------------------------------------ */ 62/* | 0 | TriggerState (4 bytes) | Rsvd1 (4 bytes) | */ 63/* | 8 | TriggerGroup[0] | */ 64/* | 10 | TriggerGroup[1] | */ 65/* | 18 | TriggerGroup[2] | */ 66/* | 20 | TriggerGroup[3] | */ 67/* | 28 | Rsvd2[0] | */ 68/* | 30 | Rsvd2[1] | */ 69/* | 38 | Rsvd2[2] | */ 70/* | 40 | NextCheckTime[0][0] | NextCheckTime[0][1] | */ 71/* | ... | */ 72/* | 240 | Latency[0][0..3] | */ 73/* | 340 | Rsvz3[0] | */ 74/* | 440 | Parameter[0][0] | */ 75/* | 448 | Parameter[0][1] | */ 76/* | ... | */ 77/* | 840 | Rsvd4[0] | */ 78/* ------------------------------------------------------ */ 79struct hv_monitor_page { 80 union hv_monitor_trigger_state trigger_state; 81 u32 rsvdz1; 82 83 union hv_monitor_trigger_group trigger_group[4]; 84 u64 rsvdz2[3]; 85 86 s32 next_checktime[4][32]; 87 88 u16 latency[4][32]; 89 u64 rsvdz3[32]; 90 91 struct hv_monitor_parameter parameter[4][32]; 92 93 u8 rsvdz4[1984]; 94}; 95 96#define HV_HYPERCALL_PARAM_ALIGN sizeof(u64) 97 98/* Definition of the hv_post_message hypercall input structure. */ 99struct hv_input_post_message { 100 union hv_connection_id connectionid; 101 u32 reserved; 102 u32 message_type; 103 u32 payload_size; 104 u64 payload[HV_MESSAGE_PAYLOAD_QWORD_COUNT]; 105}; 106 107 108enum { 109 VMBUS_MESSAGE_CONNECTION_ID = 1, 110 VMBUS_MESSAGE_CONNECTION_ID_4 = 4, 111 VMBUS_MESSAGE_PORT_ID = 1, 112 VMBUS_EVENT_CONNECTION_ID = 2, 113 VMBUS_EVENT_PORT_ID = 2, 114 VMBUS_MONITOR_CONNECTION_ID = 3, 115 VMBUS_MONITOR_PORT_ID = 3, 116 VMBUS_MESSAGE_SINT = 2, 117}; 118 119/* 120 * Per cpu state for channel handling 121 */ 122struct hv_per_cpu_context { 123 void *synic_message_page; 124 void *synic_event_page; 125 /* 126 * buffer to post messages to the host. 127 */ 128 void *post_msg_page; 129 130 /* 131 * Starting with win8, we can take channel interrupts on any CPU; 132 * we will manage the tasklet that handles events messages on a per CPU 133 * basis. 134 */ 135 struct tasklet_struct msg_dpc; 136}; 137 138struct hv_context { 139 /* We only support running on top of Hyper-V 140 * So at this point this really can only contain the Hyper-V ID 141 */ 142 u64 guestid; 143 144 struct hv_per_cpu_context __percpu *cpu_context; 145 146 /* 147 * To manage allocations in a NUMA node. 148 * Array indexed by numa node ID. 149 */ 150 struct cpumask *hv_numa_map; 151}; 152 153extern struct hv_context hv_context; 154 155/* Hv Interface */ 156 157extern int hv_init(void); 158 159extern int hv_post_message(union hv_connection_id connection_id, 160 enum hv_message_type message_type, 161 void *payload, size_t payload_size); 162 163extern int hv_synic_alloc(void); 164 165extern void hv_synic_free(void); 166 167extern void hv_synic_enable_regs(unsigned int cpu); 168extern int hv_synic_init(unsigned int cpu); 169 170extern void hv_synic_disable_regs(unsigned int cpu); 171extern int hv_synic_cleanup(unsigned int cpu); 172 173/* Interface */ 174 175void hv_ringbuffer_pre_init(struct vmbus_channel *channel); 176 177int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info, 178 struct page *pages, u32 pagecnt); 179 180void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info); 181 182int hv_ringbuffer_write(struct vmbus_channel *channel, 183 const struct kvec *kv_list, u32 kv_count); 184 185int hv_ringbuffer_read(struct vmbus_channel *channel, 186 void *buffer, u32 buflen, u32 *buffer_actual_len, 187 u64 *requestid, bool raw); 188 189/* 190 * The Maximum number of channels (16348) is determined by the size of the 191 * interrupt page, which is HV_HYP_PAGE_SIZE. 1/2 of HV_HYP_PAGE_SIZE is to 192 * send endpoint interrupts, and the other is to receive endpoint interrupts. 193 */ 194#define MAX_NUM_CHANNELS ((HV_HYP_PAGE_SIZE >> 1) << 3) 195 196/* The value here must be in multiple of 32 */ 197/* TODO: Need to make this configurable */ 198#define MAX_NUM_CHANNELS_SUPPORTED 256 199 200#define MAX_CHANNEL_RELIDS \ 201 max(MAX_NUM_CHANNELS_SUPPORTED, HV_EVENT_FLAGS_COUNT) 202 203enum vmbus_connect_state { 204 DISCONNECTED, 205 CONNECTING, 206 CONNECTED, 207 DISCONNECTING 208}; 209 210#define MAX_SIZE_CHANNEL_MESSAGE HV_MESSAGE_PAYLOAD_BYTE_COUNT 211 212/* 213 * The CPU that Hyper-V will interrupt for VMBUS messages, such as 214 * CHANNELMSG_OFFERCHANNEL and CHANNELMSG_RESCIND_CHANNELOFFER. 215 */ 216#define VMBUS_CONNECT_CPU 0 217 218struct vmbus_connection { 219 u32 msg_conn_id; 220 221 atomic_t offer_in_progress; 222 223 enum vmbus_connect_state conn_state; 224 225 atomic_t next_gpadl_handle; 226 227 struct completion unload_event; 228 /* 229 * Represents channel interrupts. Each bit position represents a 230 * channel. When a channel sends an interrupt via VMBUS, it finds its 231 * bit in the sendInterruptPage, set it and calls Hv to generate a port 232 * event. The other end receives the port event and parse the 233 * recvInterruptPage to see which bit is set 234 */ 235 void *int_page; 236 void *send_int_page; 237 void *recv_int_page; 238 239 /* 240 * 2 pages - 1st page for parent->child notification and 2nd 241 * is child->parent notification 242 */ 243 struct hv_monitor_page *monitor_pages[2]; 244 struct list_head chn_msg_list; 245 spinlock_t channelmsg_lock; 246 247 /* List of channels */ 248 struct list_head chn_list; 249 struct mutex channel_mutex; 250 251 /* Array of channels */ 252 struct vmbus_channel **channels; 253 254 /* 255 * An offer message is handled first on the work_queue, and then 256 * is further handled on handle_primary_chan_wq or 257 * handle_sub_chan_wq. 258 */ 259 struct workqueue_struct *work_queue; 260 struct workqueue_struct *handle_primary_chan_wq; 261 struct workqueue_struct *handle_sub_chan_wq; 262 263 /* 264 * The number of sub-channels and hv_sock channels that should be 265 * cleaned up upon suspend: sub-channels will be re-created upon 266 * resume, and hv_sock channels should not survive suspend. 267 */ 268 atomic_t nr_chan_close_on_suspend; 269 /* 270 * vmbus_bus_suspend() waits for "nr_chan_close_on_suspend" to 271 * drop to zero. 272 */ 273 struct completion ready_for_suspend_event; 274 275 /* 276 * The number of primary channels that should be "fixed up" 277 * upon resume: these channels are re-offered upon resume, and some 278 * fields of the channel offers (i.e. child_relid and connection_id) 279 * can change, so the old offermsg must be fixed up, before the resume 280 * callbacks of the VSC drivers start to further touch the channels. 281 */ 282 atomic_t nr_chan_fixup_on_resume; 283 /* 284 * vmbus_bus_resume() waits for "nr_chan_fixup_on_resume" to 285 * drop to zero. 286 */ 287 struct completion ready_for_resume_event; 288}; 289 290 291struct vmbus_msginfo { 292 /* Bookkeeping stuff */ 293 struct list_head msglist_entry; 294 295 /* The message itself */ 296 unsigned char msg[]; 297}; 298 299 300extern struct vmbus_connection vmbus_connection; 301 302int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo, u32 version); 303 304static inline void vmbus_send_interrupt(u32 relid) 305{ 306 sync_set_bit(relid, vmbus_connection.send_int_page); 307} 308 309enum vmbus_message_handler_type { 310 /* The related handler can sleep. */ 311 VMHT_BLOCKING = 0, 312 313 /* The related handler must NOT sleep. */ 314 VMHT_NON_BLOCKING = 1, 315}; 316 317struct vmbus_channel_message_table_entry { 318 enum vmbus_channel_message_type message_type; 319 enum vmbus_message_handler_type handler_type; 320 void (*message_handler)(struct vmbus_channel_message_header *msg); 321 u32 min_payload_len; 322}; 323 324extern const struct vmbus_channel_message_table_entry 325 channel_message_table[CHANNELMSG_COUNT]; 326 327 328/* General vmbus interface */ 329 330struct hv_device *vmbus_device_create(const guid_t *type, 331 const guid_t *instance, 332 struct vmbus_channel *channel); 333 334int vmbus_device_register(struct hv_device *child_device_obj); 335void vmbus_device_unregister(struct hv_device *device_obj); 336int vmbus_add_channel_kobj(struct hv_device *device_obj, 337 struct vmbus_channel *channel); 338 339void vmbus_remove_channel_attr_group(struct vmbus_channel *channel); 340 341void vmbus_channel_map_relid(struct vmbus_channel *channel); 342void vmbus_channel_unmap_relid(struct vmbus_channel *channel); 343 344struct vmbus_channel *relid2channel(u32 relid); 345 346void vmbus_free_channels(void); 347 348/* Connection interface */ 349 350int vmbus_connect(void); 351void vmbus_disconnect(void); 352 353int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep); 354 355void vmbus_on_event(unsigned long data); 356void vmbus_on_msg_dpc(unsigned long data); 357 358int hv_kvp_init(struct hv_util_service *srv); 359void hv_kvp_deinit(void); 360int hv_kvp_pre_suspend(void); 361int hv_kvp_pre_resume(void); 362void hv_kvp_onchannelcallback(void *context); 363 364int hv_vss_init(struct hv_util_service *srv); 365void hv_vss_deinit(void); 366int hv_vss_pre_suspend(void); 367int hv_vss_pre_resume(void); 368void hv_vss_onchannelcallback(void *context); 369 370int hv_fcopy_init(struct hv_util_service *srv); 371void hv_fcopy_deinit(void); 372int hv_fcopy_pre_suspend(void); 373int hv_fcopy_pre_resume(void); 374void hv_fcopy_onchannelcallback(void *context); 375void vmbus_initiate_unload(bool crash); 376 377static inline void hv_poll_channel(struct vmbus_channel *channel, 378 void (*cb)(void *)) 379{ 380 if (!channel) 381 return; 382 cb(channel); 383} 384 385enum hvutil_device_state { 386 HVUTIL_DEVICE_INIT = 0, /* driver is loaded, waiting for userspace */ 387 HVUTIL_READY, /* userspace is registered */ 388 HVUTIL_HOSTMSG_RECEIVED, /* message from the host was received */ 389 HVUTIL_USERSPACE_REQ, /* request to userspace was sent */ 390 HVUTIL_USERSPACE_RECV, /* reply from userspace was received */ 391 HVUTIL_DEVICE_DYING, /* driver unload is in progress */ 392}; 393 394enum delay { 395 INTERRUPT_DELAY = 0, 396 MESSAGE_DELAY = 1, 397}; 398 399extern const struct vmbus_device vmbus_devs[]; 400 401static inline bool hv_is_perf_channel(struct vmbus_channel *channel) 402{ 403 return vmbus_devs[channel->device_id].perf_device; 404} 405 406static inline bool hv_is_alloced_cpu(unsigned int cpu) 407{ 408 struct vmbus_channel *channel, *sc; 409 410 lockdep_assert_held(&vmbus_connection.channel_mutex); 411 /* 412 * List additions/deletions as well as updates of the target CPUs are 413 * protected by channel_mutex. 414 */ 415 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) { 416 if (!hv_is_perf_channel(channel)) 417 continue; 418 if (channel->target_cpu == cpu) 419 return true; 420 list_for_each_entry(sc, &channel->sc_list, sc_list) { 421 if (sc->target_cpu == cpu) 422 return true; 423 } 424 } 425 return false; 426} 427 428static inline void hv_set_alloced_cpu(unsigned int cpu) 429{ 430 cpumask_set_cpu(cpu, &hv_context.hv_numa_map[cpu_to_node(cpu)]); 431} 432 433static inline void hv_clear_alloced_cpu(unsigned int cpu) 434{ 435 if (hv_is_alloced_cpu(cpu)) 436 return; 437 cpumask_clear_cpu(cpu, &hv_context.hv_numa_map[cpu_to_node(cpu)]); 438} 439 440static inline void hv_update_alloced_cpus(unsigned int old_cpu, 441 unsigned int new_cpu) 442{ 443 hv_set_alloced_cpu(new_cpu); 444 hv_clear_alloced_cpu(old_cpu); 445} 446 447#ifdef CONFIG_HYPERV_TESTING 448 449int hv_debug_add_dev_dir(struct hv_device *dev); 450void hv_debug_rm_dev_dir(struct hv_device *dev); 451void hv_debug_rm_all_dir(void); 452int hv_debug_init(void); 453void hv_debug_delay_test(struct vmbus_channel *channel, enum delay delay_type); 454 455#else /* CONFIG_HYPERV_TESTING */ 456 457static inline void hv_debug_rm_dev_dir(struct hv_device *dev) {}; 458static inline void hv_debug_rm_all_dir(void) {}; 459static inline void hv_debug_delay_test(struct vmbus_channel *channel, 460 enum delay delay_type) {}; 461static inline int hv_debug_init(void) 462{ 463 return -1; 464} 465 466static inline int hv_debug_add_dev_dir(struct hv_device *dev) 467{ 468 return -1; 469} 470 471#endif /* CONFIG_HYPERV_TESTING */ 472 473#endif /* _HYPERV_VMBUS_H */ 474