1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * HyperV Detection code. 4 * 5 * Copyright (C) 2010, Novell, Inc. 6 * Author : K. Y. Srinivasan <ksrinivasan@novell.com> 7 */ 8 9 #include <linux/types.h> 10 #include <linux/time.h> 11 #include <linux/clocksource.h> 12 #include <linux/init.h> 13 #include <linux/export.h> 14 #include <linux/hardirq.h> 15 #include <linux/efi.h> 16 #include <linux/interrupt.h> 17 #include <linux/irq.h> 18 #include <linux/kexec.h> 19 #include <linux/random.h> 20 #include <asm/processor.h> 21 #include <asm/hypervisor.h> 22 #include <hyperv/hvhdk.h> 23 #include <asm/mshyperv.h> 24 #include <asm/desc.h> 25 #include <asm/idtentry.h> 26 #include <asm/irq_regs.h> 27 #include <asm/i8259.h> 28 #include <asm/apic.h> 29 #include <asm/timer.h> 30 #include <asm/reboot.h> 31 #include <asm/msr.h> 32 #include <asm/nmi.h> 33 #include <clocksource/hyperv_timer.h> 34 #include <asm/numa.h> 35 #include <asm/svm.h> 36 37 /* Is Linux running on nested Microsoft Hypervisor */ 38 bool hv_nested; 39 struct ms_hyperv_info ms_hyperv; 40 41 #if IS_ENABLED(CONFIG_HYPERV) 42 /* 43 * When running with the paravisor, controls proxying the synthetic interrupts 44 * from the host 45 */ 46 static bool hv_para_sint_proxy; 47 48 static inline unsigned int hv_get_nested_msr(unsigned int reg) 49 { 50 if (hv_is_sint_msr(reg)) 51 return reg - HV_X64_MSR_SINT0 + HV_X64_MSR_NESTED_SINT0; 52 53 switch (reg) { 54 case HV_X64_MSR_SIMP: 55 return HV_X64_MSR_NESTED_SIMP; 56 case HV_X64_MSR_SIEFP: 57 return HV_X64_MSR_NESTED_SIEFP; 58 case HV_X64_MSR_SVERSION: 59 return HV_X64_MSR_NESTED_SVERSION; 60 case HV_X64_MSR_SCONTROL: 61 return HV_X64_MSR_NESTED_SCONTROL; 62 case HV_X64_MSR_EOM: 63 return HV_X64_MSR_NESTED_EOM; 64 default: 65 return reg; 66 } 67 } 68 69 u64 hv_get_non_nested_msr(unsigned int reg) 70 { 71 u64 value; 72 73 if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present) 74 hv_ivm_msr_read(reg, &value); 75 else 76 rdmsrq(reg, value); 77 return value; 78 } 79 EXPORT_SYMBOL_GPL(hv_get_non_nested_msr); 80 81 void hv_set_non_nested_msr(unsigned int reg, u64 value) 82 { 83 if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present) { 84 /* The hypervisor will get the intercept. */ 85 hv_ivm_msr_write(reg, value); 86 87 /* Using wrmsrq so the following goes to the paravisor. */ 88 if (hv_is_sint_msr(reg)) { 89 union hv_synic_sint sint = { .as_uint64 = value }; 90 91 sint.proxy = hv_para_sint_proxy; 92 native_wrmsrq(reg, sint.as_uint64); 93 } 94 } else { 95 native_wrmsrq(reg, value); 96 } 97 } 98 EXPORT_SYMBOL_GPL(hv_set_non_nested_msr); 99 100 /* 101 * Enable or disable proxying synthetic interrupts 102 * to the paravisor. 103 */ 104 void hv_para_set_sint_proxy(bool enable) 105 { 106 hv_para_sint_proxy = enable; 107 } 108 109 /* 110 * Get the SynIC register value from the paravisor. 111 */ 112 u64 hv_para_get_synic_register(unsigned int reg) 113 { 114 if (WARN_ON(!ms_hyperv.paravisor_present || !hv_is_synic_msr(reg))) 115 return ~0ULL; 116 return native_read_msr(reg); 117 } 118 119 /* 120 * Set the SynIC register value with the paravisor. 121 */ 122 void hv_para_set_synic_register(unsigned int reg, u64 val) 123 { 124 if (WARN_ON(!ms_hyperv.paravisor_present || !hv_is_synic_msr(reg))) 125 return; 126 native_write_msr(reg, val); 127 } 128 129 u64 hv_get_msr(unsigned int reg) 130 { 131 if (hv_nested) 132 reg = hv_get_nested_msr(reg); 133 134 return hv_get_non_nested_msr(reg); 135 } 136 EXPORT_SYMBOL_GPL(hv_get_msr); 137 138 void hv_set_msr(unsigned int reg, u64 value) 139 { 140 if (hv_nested) 141 reg = hv_get_nested_msr(reg); 142 143 hv_set_non_nested_msr(reg, value); 144 } 145 EXPORT_SYMBOL_GPL(hv_set_msr); 146 147 static void (*mshv_handler)(void); 148 static void (*vmbus_handler)(void); 149 static void (*hv_stimer0_handler)(void); 150 static void (*hv_kexec_handler)(void); 151 static void (*hv_crash_handler)(struct pt_regs *regs); 152 153 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback) 154 { 155 struct pt_regs *old_regs = set_irq_regs(regs); 156 157 inc_irq_stat(irq_hv_callback_count); 158 if (mshv_handler) 159 mshv_handler(); 160 161 if (vmbus_handler) 162 vmbus_handler(); 163 164 add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR); 165 166 if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED) 167 apic_eoi(); 168 169 set_irq_regs(old_regs); 170 } 171 172 void hv_setup_mshv_handler(void (*handler)(void)) 173 { 174 mshv_handler = handler; 175 } 176 177 void hv_setup_vmbus_handler(void (*handler)(void)) 178 { 179 vmbus_handler = handler; 180 } 181 182 void hv_remove_vmbus_handler(void) 183 { 184 /* We have no way to deallocate the interrupt gate */ 185 vmbus_handler = NULL; 186 } 187 188 /* 189 * Routines to do per-architecture handling of stimer0 190 * interrupts when in Direct Mode 191 */ 192 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0) 193 { 194 struct pt_regs *old_regs = set_irq_regs(regs); 195 196 inc_irq_stat(hyperv_stimer0_count); 197 if (hv_stimer0_handler) 198 hv_stimer0_handler(); 199 add_interrupt_randomness(HYPERV_STIMER0_VECTOR); 200 apic_eoi(); 201 202 set_irq_regs(old_regs); 203 } 204 205 /* For x86/x64, override weak placeholders in hyperv_timer.c */ 206 void hv_setup_stimer0_handler(void (*handler)(void)) 207 { 208 hv_stimer0_handler = handler; 209 } 210 211 void hv_remove_stimer0_handler(void) 212 { 213 /* We have no way to deallocate the interrupt gate */ 214 hv_stimer0_handler = NULL; 215 } 216 217 void hv_setup_kexec_handler(void (*handler)(void)) 218 { 219 hv_kexec_handler = handler; 220 } 221 222 void hv_remove_kexec_handler(void) 223 { 224 hv_kexec_handler = NULL; 225 } 226 227 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs)) 228 { 229 hv_crash_handler = handler; 230 } 231 232 void hv_remove_crash_handler(void) 233 { 234 hv_crash_handler = NULL; 235 } 236 237 #ifdef CONFIG_KEXEC_CORE 238 static void hv_machine_shutdown(void) 239 { 240 if (kexec_in_progress && hv_kexec_handler) 241 hv_kexec_handler(); 242 243 /* 244 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor 245 * corrupts the old VP Assist Pages and can crash the kexec kernel. 246 */ 247 if (kexec_in_progress) 248 cpuhp_remove_state(CPUHP_AP_HYPERV_ONLINE); 249 250 /* The function calls stop_other_cpus(). */ 251 native_machine_shutdown(); 252 253 /* Disable the hypercall page when there is only 1 active CPU. */ 254 if (kexec_in_progress) 255 hyperv_cleanup(); 256 } 257 #endif /* CONFIG_KEXEC_CORE */ 258 259 #ifdef CONFIG_CRASH_DUMP 260 static void hv_guest_crash_shutdown(struct pt_regs *regs) 261 { 262 if (hv_crash_handler) 263 hv_crash_handler(regs); 264 265 /* The function calls crash_smp_send_stop(). */ 266 native_machine_crash_shutdown(regs); 267 268 /* Disable the hypercall page when there is only 1 active CPU. */ 269 hyperv_cleanup(); 270 } 271 #endif /* CONFIG_CRASH_DUMP */ 272 273 static u64 hv_ref_counter_at_suspend; 274 static void (*old_save_sched_clock_state)(void); 275 static void (*old_restore_sched_clock_state)(void); 276 277 /* 278 * Hyper-V clock counter resets during hibernation. Save and restore clock 279 * offset during suspend/resume, while also considering the time passed 280 * before suspend. This is to make sure that sched_clock using hv tsc page 281 * based clocksource, proceeds from where it left off during suspend and 282 * it shows correct time for the timestamps of kernel messages after resume. 283 */ 284 static void save_hv_clock_tsc_state(void) 285 { 286 hv_ref_counter_at_suspend = hv_read_reference_counter(); 287 } 288 289 static void restore_hv_clock_tsc_state(void) 290 { 291 /* 292 * Adjust the offsets used by hv tsc clocksource to 293 * account for the time spent before hibernation. 294 * adjusted value = reference counter (time) at suspend 295 * - reference counter (time) now. 296 */ 297 hv_adj_sched_clock_offset(hv_ref_counter_at_suspend - hv_read_reference_counter()); 298 } 299 300 /* 301 * Functions to override save_sched_clock_state and restore_sched_clock_state 302 * functions of x86_platform. The Hyper-V clock counter is reset during 303 * suspend-resume and the offset used to measure time needs to be 304 * corrected, post resume. 305 */ 306 static void hv_save_sched_clock_state(void) 307 { 308 old_save_sched_clock_state(); 309 save_hv_clock_tsc_state(); 310 } 311 312 static void hv_restore_sched_clock_state(void) 313 { 314 restore_hv_clock_tsc_state(); 315 old_restore_sched_clock_state(); 316 } 317 318 static void __init x86_setup_ops_for_tsc_pg_clock(void) 319 { 320 if (!(ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE)) 321 return; 322 323 old_save_sched_clock_state = x86_platform.save_sched_clock_state; 324 x86_platform.save_sched_clock_state = hv_save_sched_clock_state; 325 326 old_restore_sched_clock_state = x86_platform.restore_sched_clock_state; 327 x86_platform.restore_sched_clock_state = hv_restore_sched_clock_state; 328 } 329 330 #ifdef CONFIG_X86_64 331 DEFINE_STATIC_CALL(hv_hypercall, hv_std_hypercall); 332 EXPORT_STATIC_CALL_TRAMP_GPL(hv_hypercall); 333 #define hypercall_update(hc) static_call_update(hv_hypercall, hc) 334 #endif 335 #endif /* CONFIG_HYPERV */ 336 337 #ifndef hypercall_update 338 #define hypercall_update(hc) (void)hc 339 #endif 340 341 static uint32_t __init ms_hyperv_platform(void) 342 { 343 u32 eax; 344 u32 hyp_signature[3]; 345 346 if (!boot_cpu_has(X86_FEATURE_HYPERVISOR)) 347 return 0; 348 349 cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS, 350 &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]); 351 352 if (eax < HYPERV_CPUID_MIN || eax > HYPERV_CPUID_MAX || 353 memcmp("Microsoft Hv", hyp_signature, 12)) 354 return 0; 355 356 /* HYPERCALL and VP_INDEX MSRs are mandatory for all features. */ 357 eax = cpuid_eax(HYPERV_CPUID_FEATURES); 358 if (!(eax & HV_MSR_HYPERCALL_AVAILABLE)) { 359 pr_warn("x86/hyperv: HYPERCALL MSR not available.\n"); 360 return 0; 361 } 362 if (!(eax & HV_MSR_VP_INDEX_AVAILABLE)) { 363 pr_warn("x86/hyperv: VP_INDEX MSR not available.\n"); 364 return 0; 365 } 366 367 return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS; 368 } 369 370 #ifdef CONFIG_X86_LOCAL_APIC 371 /* 372 * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes 373 * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle 374 * unknown NMI on the first CPU which gets it. 375 */ 376 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs) 377 { 378 static atomic_t nmi_cpu = ATOMIC_INIT(-1); 379 unsigned int old_cpu, this_cpu; 380 381 if (!unknown_nmi_panic) 382 return NMI_DONE; 383 384 old_cpu = -1; 385 this_cpu = raw_smp_processor_id(); 386 if (!atomic_try_cmpxchg(&nmi_cpu, &old_cpu, this_cpu)) 387 return NMI_HANDLED; 388 389 return NMI_DONE; 390 } 391 #endif 392 393 static unsigned long hv_get_tsc_khz(void) 394 { 395 unsigned long freq; 396 397 rdmsrq(HV_X64_MSR_TSC_FREQUENCY, freq); 398 399 return freq / 1000; 400 } 401 402 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV) 403 static void __init hv_smp_prepare_boot_cpu(void) 404 { 405 native_smp_prepare_boot_cpu(); 406 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS) 407 hv_init_spinlocks(); 408 #endif 409 } 410 411 static void __init hv_smp_prepare_cpus(unsigned int max_cpus) 412 { 413 #ifdef CONFIG_X86_64 414 int i; 415 int ret; 416 #endif 417 418 native_smp_prepare_cpus(max_cpus); 419 420 /* 421 * Override wakeup_secondary_cpu_64 callback for SEV-SNP 422 * enlightened guest. 423 */ 424 if (!ms_hyperv.paravisor_present && hv_isolation_type_snp()) { 425 apic->wakeup_secondary_cpu_64 = hv_snp_boot_ap; 426 return; 427 } 428 429 #ifdef CONFIG_X86_64 430 for_each_present_cpu(i) { 431 if (i == 0) 432 continue; 433 ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i)); 434 BUG_ON(ret); 435 } 436 437 for_each_present_cpu(i) { 438 if (i == 0) 439 continue; 440 ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i); 441 BUG_ON(ret); 442 } 443 #endif 444 } 445 #endif 446 447 /* 448 * When a fully enlightened TDX VM runs on Hyper-V, the firmware sets the 449 * HW_REDUCED flag: refer to acpi_tb_create_local_fadt(). Consequently ttyS0 450 * interrupts can't work because request_irq() -> ... -> irq_to_desc() returns 451 * NULL for ttyS0. This happens because mp_config_acpi_legacy_irqs() sees a 452 * nr_legacy_irqs() of 0, so it doesn't initialize the array 'mp_irqs[]', and 453 * later setup_IO_APIC_irqs() -> find_irq_entry() fails to find the legacy irqs 454 * from the array and hence doesn't create the necessary irq description info. 455 * 456 * Clone arch/x86/kernel/acpi/boot.c: acpi_generic_reduced_hw_init() here, 457 * except don't change 'legacy_pic', which keeps its default value 458 * 'default_legacy_pic'. This way, mp_config_acpi_legacy_irqs() sees a non-zero 459 * nr_legacy_irqs() and eventually serial console interrupts works properly. 460 */ 461 static void __init reduced_hw_init(void) 462 { 463 x86_init.timers.timer_init = x86_init_noop; 464 x86_init.irqs.pre_vector_init = x86_init_noop; 465 } 466 467 int hv_get_hypervisor_version(union hv_hypervisor_version_info *info) 468 { 469 unsigned int hv_max_functions; 470 471 hv_max_functions = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS); 472 if (hv_max_functions < HYPERV_CPUID_VERSION) { 473 pr_err("%s: Could not detect Hyper-V version\n", __func__); 474 return -ENODEV; 475 } 476 477 cpuid(HYPERV_CPUID_VERSION, &info->eax, &info->ebx, &info->ecx, &info->edx); 478 479 return 0; 480 } 481 EXPORT_SYMBOL_GPL(hv_get_hypervisor_version); 482 483 /* 484 * Reserved vectors hard coded in the hypervisor. If used outside, the hypervisor 485 * will either crash or hang or attempt to break into debugger. 486 */ 487 static void hv_reserve_irq_vectors(void) 488 { 489 #define HYPERV_DBG_FASTFAIL_VECTOR 0x29 490 #define HYPERV_DBG_ASSERT_VECTOR 0x2C 491 #define HYPERV_DBG_SERVICE_VECTOR 0x2D 492 493 if (cpu_feature_enabled(X86_FEATURE_FRED)) 494 return; 495 496 if (test_and_set_bit(HYPERV_DBG_ASSERT_VECTOR, system_vectors) || 497 test_and_set_bit(HYPERV_DBG_SERVICE_VECTOR, system_vectors) || 498 test_and_set_bit(HYPERV_DBG_FASTFAIL_VECTOR, system_vectors)) 499 BUG(); 500 501 pr_info("Hyper-V: reserve vectors: 0x%x 0x%x 0x%x\n", 502 HYPERV_DBG_ASSERT_VECTOR, HYPERV_DBG_SERVICE_VECTOR, 503 HYPERV_DBG_FASTFAIL_VECTOR); 504 } 505 506 static void __init ms_hyperv_init_platform(void) 507 { 508 int hv_max_functions_eax, eax; 509 510 #ifdef CONFIG_PARAVIRT 511 pv_info.name = "Hyper-V"; 512 #endif 513 514 /* 515 * Extract the features and hints 516 */ 517 ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES); 518 ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES); 519 ms_hyperv.ext_features = cpuid_ecx(HYPERV_CPUID_FEATURES); 520 ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES); 521 ms_hyperv.hints = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO); 522 523 hv_max_functions_eax = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS); 524 525 pr_info("Hyper-V: privilege flags low %#x, high %#x, ext %#x, hints %#x, misc %#x\n", 526 ms_hyperv.features, ms_hyperv.priv_high, 527 ms_hyperv.ext_features, ms_hyperv.hints, 528 ms_hyperv.misc_features); 529 530 ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS); 531 ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS); 532 533 pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n", 534 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index); 535 536 hv_identify_partition_type(); 537 538 if (hv_root_partition()) 539 hv_reserve_irq_vectors(); 540 541 if (cc_platform_has(CC_ATTR_SNP_SECURE_AVIC)) 542 ms_hyperv.hints |= HV_DEPRECATING_AEOI_RECOMMENDED; 543 544 if (ms_hyperv.hints & HV_X64_HYPERV_NESTED) { 545 hv_nested = true; 546 pr_info("Hyper-V: running on a nested hypervisor\n"); 547 } 548 549 /* 550 * There is no check against the max function for HYPERV_CPUID_VIRT_STACK_* CPUID 551 * leaves as the hypervisor doesn't handle them. Even a nested root partition (L2 552 * root) will not get them because the nested (L1) hypervisor filters them out. 553 * These are handled through intercept processing by the Windows Hyper-V stack 554 * or the paravisor. 555 */ 556 eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES); 557 ms_hyperv.confidential_vmbus_available = 558 eax & HYPERV_VS_PROPERTIES_EAX_CONFIDENTIAL_VMBUS_AVAILABLE; 559 ms_hyperv.msi_ext_dest_id = 560 eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE; 561 562 if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS && 563 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) { 564 x86_platform.calibrate_tsc = hv_get_tsc_khz; 565 x86_platform.calibrate_cpu = hv_get_tsc_khz; 566 setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ); 567 } 568 569 if (ms_hyperv.priv_high & HV_ISOLATION) { 570 ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG); 571 ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG); 572 573 if (ms_hyperv.shared_gpa_boundary_active) 574 ms_hyperv.shared_gpa_boundary = 575 BIT_ULL(ms_hyperv.shared_gpa_boundary_bits); 576 577 pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n", 578 ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b); 579 580 581 if (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP) { 582 static_branch_enable(&isolation_type_snp); 583 if (!ms_hyperv.paravisor_present) 584 hypercall_update(hv_snp_hypercall); 585 } else if (hv_get_isolation_type() == HV_ISOLATION_TYPE_TDX) { 586 static_branch_enable(&isolation_type_tdx); 587 588 /* A TDX VM must use x2APIC and doesn't use lazy EOI. */ 589 ms_hyperv.hints &= ~HV_X64_APIC_ACCESS_RECOMMENDED; 590 591 if (!ms_hyperv.paravisor_present) { 592 hypercall_update(hv_tdx_hypercall); 593 /* 594 * Mark the Hyper-V TSC page feature as disabled 595 * in a TDX VM without paravisor so that the 596 * Invariant TSC, which is a better clocksource 597 * anyway, is used instead. 598 */ 599 ms_hyperv.features &= ~HV_MSR_REFERENCE_TSC_AVAILABLE; 600 601 /* 602 * The Invariant TSC is expected to be available 603 * in a TDX VM without paravisor, but if not, 604 * print a warning message. The slower Hyper-V MSR-based 605 * Ref Counter should end up being the clocksource. 606 */ 607 if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT)) 608 pr_warn("Hyper-V: Invariant TSC is unavailable\n"); 609 610 /* HV_MSR_CRASH_CTL is unsupported. */ 611 ms_hyperv.misc_features &= ~HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE; 612 613 /* Don't trust Hyper-V's TLB-flushing hypercalls. */ 614 ms_hyperv.hints &= ~HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED; 615 616 x86_init.acpi.reduced_hw_early_init = reduced_hw_init; 617 } 618 } 619 } 620 621 if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) { 622 ms_hyperv.nested_features = 623 cpuid_eax(HYPERV_CPUID_NESTED_FEATURES); 624 pr_info("Hyper-V: Nested features: 0x%x\n", 625 ms_hyperv.nested_features); 626 } 627 628 #ifdef CONFIG_X86_LOCAL_APIC 629 if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS && 630 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) { 631 /* 632 * Get the APIC frequency. 633 */ 634 u64 hv_lapic_frequency; 635 636 rdmsrq(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency); 637 hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ); 638 lapic_timer_period = hv_lapic_frequency; 639 pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n", 640 lapic_timer_period); 641 } 642 643 register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST, 644 "hv_nmi_unknown"); 645 #endif 646 647 #ifdef CONFIG_X86_IO_APIC 648 no_timer_check = 1; 649 #endif 650 651 #if IS_ENABLED(CONFIG_HYPERV) 652 if (hv_root_partition()) 653 machine_ops.power_off = hv_machine_power_off; 654 #if defined(CONFIG_KEXEC_CORE) 655 machine_ops.shutdown = hv_machine_shutdown; 656 #endif 657 #if defined(CONFIG_CRASH_DUMP) 658 if (!hv_root_partition()) 659 machine_ops.crash_shutdown = hv_guest_crash_shutdown; 660 #endif 661 #endif 662 /* 663 * HV_ACCESS_TSC_INVARIANT is always zero for the root partition. Root 664 * partition doesn't need to write to synthetic MSR to enable invariant 665 * TSC feature. It sees what the hardware provides. 666 */ 667 if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) { 668 /* 669 * Writing to synthetic MSR 0x40000118 updates/changes the 670 * guest visible CPUIDs. Setting bit 0 of this MSR enables 671 * guests to report invariant TSC feature through CPUID 672 * instruction, CPUID 0x800000007/EDX, bit 8. See code in 673 * early_init_intel() where this bit is examined. The 674 * setting of this MSR bit should happen before init_intel() 675 * is called. 676 */ 677 wrmsrq(HV_X64_MSR_TSC_INVARIANT_CONTROL, HV_EXPOSE_INVARIANT_TSC); 678 setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE); 679 } 680 681 /* 682 * Generation 2 instances don't support reading the NMI status from 683 * 0x61 port. 684 */ 685 if (efi_enabled(EFI_BOOT)) 686 x86_platform.get_nmi_reason = hv_get_nmi_reason; 687 688 #if IS_ENABLED(CONFIG_HYPERV) 689 if ((hv_get_isolation_type() == HV_ISOLATION_TYPE_VBS) || 690 ms_hyperv.paravisor_present) 691 hv_vtom_init(); 692 /* 693 * Setup the hook to get control post apic initialization. 694 */ 695 x86_platform.apic_post_init = hyperv_init; 696 hyperv_setup_mmu_ops(); 697 698 /* Install system interrupt handler for hypervisor callback */ 699 sysvec_install(HYPERVISOR_CALLBACK_VECTOR, sysvec_hyperv_callback); 700 701 /* Install system interrupt handler for reenlightenment notifications */ 702 if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) { 703 sysvec_install(HYPERV_REENLIGHTENMENT_VECTOR, sysvec_hyperv_reenlightenment); 704 } 705 706 /* Install system interrupt handler for stimer0 */ 707 if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) { 708 sysvec_install(HYPERV_STIMER0_VECTOR, sysvec_hyperv_stimer0); 709 } 710 711 # ifdef CONFIG_SMP 712 smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu; 713 if (hv_root_partition() || 714 (!ms_hyperv.paravisor_present && hv_isolation_type_snp())) 715 smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus; 716 # endif 717 718 /* 719 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic, 720 * set x2apic destination mode to physical mode when x2apic is available 721 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs 722 * have 8-bit APIC id. 723 */ 724 # ifdef CONFIG_X86_X2APIC 725 if (x2apic_supported()) 726 x2apic_phys = 1; 727 # endif 728 729 /* Register Hyper-V specific clocksource */ 730 hv_init_clocksource(); 731 x86_setup_ops_for_tsc_pg_clock(); 732 hv_vtl_init_platform(); 733 #endif 734 /* 735 * TSC should be marked as unstable only after Hyper-V 736 * clocksource has been initialized. This ensures that the 737 * stability of the sched_clock is not altered. 738 * 739 * HV_ACCESS_TSC_INVARIANT is always zero for the root partition. No 740 * need to check for it. 741 */ 742 if (!hv_root_partition() && 743 !(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT)) 744 mark_tsc_unstable("running on Hyper-V"); 745 746 hardlockup_detector_disable(); 747 } 748 749 static bool __init ms_hyperv_x2apic_available(void) 750 { 751 return x2apic_supported(); 752 } 753 754 /* 755 * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping() 756 * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the 757 * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg(). 758 * 759 * Note: for a VM on Hyper-V, the I/O-APIC is the only device which 760 * (logically) generates MSIs directly to the system APIC irq domain. 761 * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the 762 * pci-hyperv host bridge. 763 * 764 * Note: for a Hyper-V root partition, this will always return false. 765 */ 766 static bool __init ms_hyperv_msi_ext_dest_id(void) 767 { 768 return ms_hyperv.msi_ext_dest_id; 769 } 770 771 #ifdef CONFIG_AMD_MEM_ENCRYPT 772 static void hv_sev_es_hcall_prepare(struct ghcb *ghcb, struct pt_regs *regs) 773 { 774 /* RAX and CPL are already in the GHCB */ 775 ghcb_set_rcx(ghcb, regs->cx); 776 ghcb_set_rdx(ghcb, regs->dx); 777 ghcb_set_r8(ghcb, regs->r8); 778 } 779 780 static bool hv_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs) 781 { 782 /* No checking of the return state needed */ 783 return true; 784 } 785 #endif 786 787 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = { 788 .name = "Microsoft Hyper-V", 789 .detect = ms_hyperv_platform, 790 .type = X86_HYPER_MS_HYPERV, 791 .init.x2apic_available = ms_hyperv_x2apic_available, 792 .init.msi_ext_dest_id = ms_hyperv_msi_ext_dest_id, 793 .init.init_platform = ms_hyperv_init_platform, 794 .init.guest_late_init = ms_hyperv_late_init, 795 #ifdef CONFIG_AMD_MEM_ENCRYPT 796 .runtime.sev_es_hcall_prepare = hv_sev_es_hcall_prepare, 797 .runtime.sev_es_hcall_finish = hv_sev_es_hcall_finish, 798 #endif 799 }; 800