1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _ASM_X86_MSHYPER_H 3 #define _ASM_X86_MSHYPER_H 4 5 #include <linux/types.h> 6 #include <linux/atomic.h> 7 #include <linux/nmi.h> 8 #include <asm/io.h> 9 #include <asm/hyperv.h> 10 #include <asm/nospec-branch.h> 11 12 /* 13 * The below CPUID leaves are present if VersionAndFeatures.HypervisorPresent 14 * is set by CPUID(HVCPUID_VERSION_FEATURES). 15 */ 16 enum hv_cpuid_function { 17 HVCPUID_VERSION_FEATURES = 0x00000001, 18 HVCPUID_VENDOR_MAXFUNCTION = 0x40000000, 19 HVCPUID_INTERFACE = 0x40000001, 20 21 /* 22 * The remaining functions depend on the value of 23 * HVCPUID_INTERFACE 24 */ 25 HVCPUID_VERSION = 0x40000002, 26 HVCPUID_FEATURES = 0x40000003, 27 HVCPUID_ENLIGHTENMENT_INFO = 0x40000004, 28 HVCPUID_IMPLEMENTATION_LIMITS = 0x40000005, 29 }; 30 31 struct ms_hyperv_info { 32 u32 features; 33 u32 misc_features; 34 u32 hints; 35 u32 max_vp_index; 36 u32 max_lp_index; 37 }; 38 39 extern struct ms_hyperv_info ms_hyperv; 40 41 /* 42 * Declare the MSR used to setup pages used to communicate with the hypervisor. 43 */ 44 union hv_x64_msr_hypercall_contents { 45 u64 as_uint64; 46 struct { 47 u64 enable:1; 48 u64 reserved:11; 49 u64 guest_physical_address:52; 50 }; 51 }; 52 53 /* 54 * TSC page layout. 55 */ 56 57 struct ms_hyperv_tsc_page { 58 volatile u32 tsc_sequence; 59 u32 reserved1; 60 volatile u64 tsc_scale; 61 volatile s64 tsc_offset; 62 u64 reserved2[509]; 63 }; 64 65 /* 66 * The guest OS needs to register the guest ID with the hypervisor. 67 * The guest ID is a 64 bit entity and the structure of this ID is 68 * specified in the Hyper-V specification: 69 * 70 * msdn.microsoft.com/en-us/library/windows/hardware/ff542653%28v=vs.85%29.aspx 71 * 72 * While the current guideline does not specify how Linux guest ID(s) 73 * need to be generated, our plan is to publish the guidelines for 74 * Linux and other guest operating systems that currently are hosted 75 * on Hyper-V. The implementation here conforms to this yet 76 * unpublished guidelines. 77 * 78 * 79 * Bit(s) 80 * 63 - Indicates if the OS is Open Source or not; 1 is Open Source 81 * 62:56 - Os Type; Linux is 0x100 82 * 55:48 - Distro specific identification 83 * 47:16 - Linux kernel version number 84 * 15:0 - Distro specific identification 85 * 86 * 87 */ 88 89 #define HV_LINUX_VENDOR_ID 0x8100 90 91 /* 92 * Generate the guest ID based on the guideline described above. 93 */ 94 95 static inline __u64 generate_guest_id(__u64 d_info1, __u64 kernel_version, 96 __u64 d_info2) 97 { 98 __u64 guest_id = 0; 99 100 guest_id = (((__u64)HV_LINUX_VENDOR_ID) << 48); 101 guest_id |= (d_info1 << 48); 102 guest_id |= (kernel_version << 16); 103 guest_id |= d_info2; 104 105 return guest_id; 106 } 107 108 109 /* Free the message slot and signal end-of-message if required */ 110 static inline void vmbus_signal_eom(struct hv_message *msg, u32 old_msg_type) 111 { 112 /* 113 * On crash we're reading some other CPU's message page and we need 114 * to be careful: this other CPU may already had cleared the header 115 * and the host may already had delivered some other message there. 116 * In case we blindly write msg->header.message_type we're going 117 * to lose it. We can still lose a message of the same type but 118 * we count on the fact that there can only be one 119 * CHANNELMSG_UNLOAD_RESPONSE and we don't care about other messages 120 * on crash. 121 */ 122 if (cmpxchg(&msg->header.message_type, old_msg_type, 123 HVMSG_NONE) != old_msg_type) 124 return; 125 126 /* 127 * Make sure the write to MessageType (ie set to 128 * HVMSG_NONE) happens before we read the 129 * MessagePending and EOMing. Otherwise, the EOMing 130 * will not deliver any more messages since there is 131 * no empty slot 132 */ 133 mb(); 134 135 if (msg->header.message_flags.msg_pending) { 136 /* 137 * This will cause message queue rescan to 138 * possibly deliver another msg from the 139 * hypervisor 140 */ 141 wrmsrl(HV_X64_MSR_EOM, 0); 142 } 143 } 144 145 #define hv_init_timer(timer, tick) wrmsrl(timer, tick) 146 #define hv_init_timer_config(config, val) wrmsrl(config, val) 147 148 #define hv_get_simp(val) rdmsrl(HV_X64_MSR_SIMP, val) 149 #define hv_set_simp(val) wrmsrl(HV_X64_MSR_SIMP, val) 150 151 #define hv_get_siefp(val) rdmsrl(HV_X64_MSR_SIEFP, val) 152 #define hv_set_siefp(val) wrmsrl(HV_X64_MSR_SIEFP, val) 153 154 #define hv_get_synic_state(val) rdmsrl(HV_X64_MSR_SCONTROL, val) 155 #define hv_set_synic_state(val) wrmsrl(HV_X64_MSR_SCONTROL, val) 156 157 #define hv_get_vp_index(index) rdmsrl(HV_X64_MSR_VP_INDEX, index) 158 159 #define hv_get_synint_state(int_num, val) rdmsrl(int_num, val) 160 #define hv_set_synint_state(int_num, val) wrmsrl(int_num, val) 161 162 void hyperv_callback_vector(void); 163 void hyperv_reenlightenment_vector(void); 164 #ifdef CONFIG_TRACING 165 #define trace_hyperv_callback_vector hyperv_callback_vector 166 #endif 167 void hyperv_vector_handler(struct pt_regs *regs); 168 void hv_setup_vmbus_irq(void (*handler)(void)); 169 void hv_remove_vmbus_irq(void); 170 171 void hv_setup_kexec_handler(void (*handler)(void)); 172 void hv_remove_kexec_handler(void); 173 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs)); 174 void hv_remove_crash_handler(void); 175 176 /* 177 * Routines for stimer0 Direct Mode handling. 178 * On x86/x64, there are no percpu actions to take. 179 */ 180 void hv_stimer0_vector_handler(struct pt_regs *regs); 181 void hv_stimer0_callback_vector(void); 182 int hv_setup_stimer0_irq(int *irq, int *vector, void (*handler)(void)); 183 void hv_remove_stimer0_irq(int irq); 184 185 static inline void hv_enable_stimer0_percpu_irq(int irq) {} 186 static inline void hv_disable_stimer0_percpu_irq(int irq) {} 187 188 189 #if IS_ENABLED(CONFIG_HYPERV) 190 extern struct clocksource *hyperv_cs; 191 extern void *hv_hypercall_pg; 192 193 static inline u64 hv_do_hypercall(u64 control, void *input, void *output) 194 { 195 u64 input_address = input ? virt_to_phys(input) : 0; 196 u64 output_address = output ? virt_to_phys(output) : 0; 197 u64 hv_status; 198 199 #ifdef CONFIG_X86_64 200 if (!hv_hypercall_pg) 201 return U64_MAX; 202 203 __asm__ __volatile__("mov %4, %%r8\n" 204 CALL_NOSPEC 205 : "=a" (hv_status), ASM_CALL_CONSTRAINT, 206 "+c" (control), "+d" (input_address) 207 : "r" (output_address), 208 THUNK_TARGET(hv_hypercall_pg) 209 : "cc", "memory", "r8", "r9", "r10", "r11"); 210 #else 211 u32 input_address_hi = upper_32_bits(input_address); 212 u32 input_address_lo = lower_32_bits(input_address); 213 u32 output_address_hi = upper_32_bits(output_address); 214 u32 output_address_lo = lower_32_bits(output_address); 215 216 if (!hv_hypercall_pg) 217 return U64_MAX; 218 219 __asm__ __volatile__(CALL_NOSPEC 220 : "=A" (hv_status), 221 "+c" (input_address_lo), ASM_CALL_CONSTRAINT 222 : "A" (control), 223 "b" (input_address_hi), 224 "D"(output_address_hi), "S"(output_address_lo), 225 THUNK_TARGET(hv_hypercall_pg) 226 : "cc", "memory"); 227 #endif /* !x86_64 */ 228 return hv_status; 229 } 230 231 #define HV_HYPERCALL_RESULT_MASK GENMASK_ULL(15, 0) 232 #define HV_HYPERCALL_FAST_BIT BIT(16) 233 #define HV_HYPERCALL_VARHEAD_OFFSET 17 234 #define HV_HYPERCALL_REP_COMP_OFFSET 32 235 #define HV_HYPERCALL_REP_COMP_MASK GENMASK_ULL(43, 32) 236 #define HV_HYPERCALL_REP_START_OFFSET 48 237 #define HV_HYPERCALL_REP_START_MASK GENMASK_ULL(59, 48) 238 239 /* Fast hypercall with 8 bytes of input and no output */ 240 static inline u64 hv_do_fast_hypercall8(u16 code, u64 input1) 241 { 242 u64 hv_status, control = (u64)code | HV_HYPERCALL_FAST_BIT; 243 244 #ifdef CONFIG_X86_64 245 { 246 __asm__ __volatile__(CALL_NOSPEC 247 : "=a" (hv_status), ASM_CALL_CONSTRAINT, 248 "+c" (control), "+d" (input1) 249 : THUNK_TARGET(hv_hypercall_pg) 250 : "cc", "r8", "r9", "r10", "r11"); 251 } 252 #else 253 { 254 u32 input1_hi = upper_32_bits(input1); 255 u32 input1_lo = lower_32_bits(input1); 256 257 __asm__ __volatile__ (CALL_NOSPEC 258 : "=A"(hv_status), 259 "+c"(input1_lo), 260 ASM_CALL_CONSTRAINT 261 : "A" (control), 262 "b" (input1_hi), 263 THUNK_TARGET(hv_hypercall_pg) 264 : "cc", "edi", "esi"); 265 } 266 #endif 267 return hv_status; 268 } 269 270 /* 271 * Rep hypercalls. Callers of this functions are supposed to ensure that 272 * rep_count and varhead_size comply with Hyper-V hypercall definition. 273 */ 274 static inline u64 hv_do_rep_hypercall(u16 code, u16 rep_count, u16 varhead_size, 275 void *input, void *output) 276 { 277 u64 control = code; 278 u64 status; 279 u16 rep_comp; 280 281 control |= (u64)varhead_size << HV_HYPERCALL_VARHEAD_OFFSET; 282 control |= (u64)rep_count << HV_HYPERCALL_REP_COMP_OFFSET; 283 284 do { 285 status = hv_do_hypercall(control, input, output); 286 if ((status & HV_HYPERCALL_RESULT_MASK) != HV_STATUS_SUCCESS) 287 return status; 288 289 /* Bits 32-43 of status have 'Reps completed' data. */ 290 rep_comp = (status & HV_HYPERCALL_REP_COMP_MASK) >> 291 HV_HYPERCALL_REP_COMP_OFFSET; 292 293 control &= ~HV_HYPERCALL_REP_START_MASK; 294 control |= (u64)rep_comp << HV_HYPERCALL_REP_START_OFFSET; 295 296 touch_nmi_watchdog(); 297 } while (rep_comp < rep_count); 298 299 return status; 300 } 301 302 /* 303 * Hypervisor's notion of virtual processor ID is different from 304 * Linux' notion of CPU ID. This information can only be retrieved 305 * in the context of the calling CPU. Setup a map for easy access 306 * to this information. 307 */ 308 extern u32 *hv_vp_index; 309 extern u32 hv_max_vp_index; 310 311 /** 312 * hv_cpu_number_to_vp_number() - Map CPU to VP. 313 * @cpu_number: CPU number in Linux terms 314 * 315 * This function returns the mapping between the Linux processor 316 * number and the hypervisor's virtual processor number, useful 317 * in making hypercalls and such that talk about specific 318 * processors. 319 * 320 * Return: Virtual processor number in Hyper-V terms 321 */ 322 static inline int hv_cpu_number_to_vp_number(int cpu_number) 323 { 324 return hv_vp_index[cpu_number]; 325 } 326 327 void hyperv_init(void); 328 void hyperv_setup_mmu_ops(void); 329 void hyper_alloc_mmu(void); 330 void hyperv_report_panic(struct pt_regs *regs, long err); 331 bool hv_is_hyperv_initialized(void); 332 void hyperv_cleanup(void); 333 334 void hyperv_reenlightenment_intr(struct pt_regs *regs); 335 void set_hv_tscchange_cb(void (*cb)(void)); 336 void clear_hv_tscchange_cb(void); 337 void hyperv_stop_tsc_emulation(void); 338 #else /* CONFIG_HYPERV */ 339 static inline void hyperv_init(void) {} 340 static inline bool hv_is_hyperv_initialized(void) { return false; } 341 static inline void hyperv_cleanup(void) {} 342 static inline void hyperv_setup_mmu_ops(void) {} 343 static inline void set_hv_tscchange_cb(void (*cb)(void)) {} 344 static inline void clear_hv_tscchange_cb(void) {} 345 static inline void hyperv_stop_tsc_emulation(void) {}; 346 #endif /* CONFIG_HYPERV */ 347 348 #ifdef CONFIG_HYPERV_TSCPAGE 349 struct ms_hyperv_tsc_page *hv_get_tsc_page(void); 350 static inline u64 hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg, 351 u64 *cur_tsc) 352 { 353 u64 scale, offset; 354 u32 sequence; 355 356 /* 357 * The protocol for reading Hyper-V TSC page is specified in Hypervisor 358 * Top-Level Functional Specification ver. 3.0 and above. To get the 359 * reference time we must do the following: 360 * - READ ReferenceTscSequence 361 * A special '0' value indicates the time source is unreliable and we 362 * need to use something else. The currently published specification 363 * versions (up to 4.0b) contain a mistake and wrongly claim '-1' 364 * instead of '0' as the special value, see commit c35b82ef0294. 365 * - ReferenceTime = 366 * ((RDTSC() * ReferenceTscScale) >> 64) + ReferenceTscOffset 367 * - READ ReferenceTscSequence again. In case its value has changed 368 * since our first reading we need to discard ReferenceTime and repeat 369 * the whole sequence as the hypervisor was updating the page in 370 * between. 371 */ 372 do { 373 sequence = READ_ONCE(tsc_pg->tsc_sequence); 374 if (!sequence) 375 return U64_MAX; 376 /* 377 * Make sure we read sequence before we read other values from 378 * TSC page. 379 */ 380 smp_rmb(); 381 382 scale = READ_ONCE(tsc_pg->tsc_scale); 383 offset = READ_ONCE(tsc_pg->tsc_offset); 384 *cur_tsc = rdtsc_ordered(); 385 386 /* 387 * Make sure we read sequence after we read all other values 388 * from TSC page. 389 */ 390 smp_rmb(); 391 392 } while (READ_ONCE(tsc_pg->tsc_sequence) != sequence); 393 394 return mul_u64_u64_shr(*cur_tsc, scale, 64) + offset; 395 } 396 397 static inline u64 hv_read_tsc_page(const struct ms_hyperv_tsc_page *tsc_pg) 398 { 399 u64 cur_tsc; 400 401 return hv_read_tsc_page_tsc(tsc_pg, &cur_tsc); 402 } 403 404 #else 405 static inline struct ms_hyperv_tsc_page *hv_get_tsc_page(void) 406 { 407 return NULL; 408 } 409 410 static inline u64 hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg, 411 u64 *cur_tsc) 412 { 413 BUG(); 414 return U64_MAX; 415 } 416 #endif 417 #endif 418