1 // SPDX-License-Identifier: GPL-2.0-only
3 * HyperV Detection code.
5 * Copyright (C) 2010, Novell, Inc.
6 * Author : K. Y. Srinivasan <ksrinivasan@novell.com>
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/i8253.h>
20 #include <linux/random.h>
21 #include <asm/processor.h>
22 #include <asm/hypervisor.h>
23 #include <asm/hyperv-tlfs.h>
24 #include <asm/mshyperv.h>
26 #include <asm/idtentry.h>
27 #include <asm/irq_regs.h>
28 #include <asm/i8259.h>
30 #include <asm/timer.h>
31 #include <asm/reboot.h>
33 #include <clocksource/hyperv_timer.h>
37 /* Is Linux running as the root partition? */
38 bool hv_root_partition;
39 /* Is Linux running on nested Microsoft Hypervisor */
41 struct ms_hyperv_info ms_hyperv;
43 /* Used in modules via hv_do_hypercall(): see arch/x86/include/asm/mshyperv.h */
44 bool hyperv_paravisor_present __ro_after_init;
45 EXPORT_SYMBOL_GPL(hyperv_paravisor_present);
47 #if IS_ENABLED(CONFIG_HYPERV)
48 static inline unsigned int hv_get_nested_msr(unsigned int reg)
50 if (hv_is_sint_msr(reg))
51 return reg - HV_X64_MSR_SINT0 + HV_X64_MSR_NESTED_SINT0;
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;
63 return HV_X64_MSR_NESTED_EOM;
69 u64 hv_get_non_nested_msr(unsigned int reg)
73 if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present)
74 hv_ivm_msr_read(reg, &value);
79 EXPORT_SYMBOL_GPL(hv_get_non_nested_msr);
81 void hv_set_non_nested_msr(unsigned int reg, u64 value)
83 if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present) {
84 hv_ivm_msr_write(reg, value);
86 /* Write proxy bit via wrmsl instruction */
87 if (hv_is_sint_msr(reg))
88 wrmsrl(reg, value | 1 << 20);
93 EXPORT_SYMBOL_GPL(hv_set_non_nested_msr);
95 u64 hv_get_msr(unsigned int reg)
98 reg = hv_get_nested_msr(reg);
100 return hv_get_non_nested_msr(reg);
102 EXPORT_SYMBOL_GPL(hv_get_msr);
104 void hv_set_msr(unsigned int reg, u64 value)
107 reg = hv_get_nested_msr(reg);
109 hv_set_non_nested_msr(reg, value);
111 EXPORT_SYMBOL_GPL(hv_set_msr);
113 static void (*vmbus_handler)(void);
114 static void (*hv_stimer0_handler)(void);
115 static void (*hv_kexec_handler)(void);
116 static void (*hv_crash_handler)(struct pt_regs *regs);
118 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)
120 struct pt_regs *old_regs = set_irq_regs(regs);
122 inc_irq_stat(irq_hv_callback_count);
126 if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
129 set_irq_regs(old_regs);
132 void hv_setup_vmbus_handler(void (*handler)(void))
134 vmbus_handler = handler;
137 void hv_remove_vmbus_handler(void)
139 /* We have no way to deallocate the interrupt gate */
140 vmbus_handler = NULL;
144 * Routines to do per-architecture handling of stimer0
145 * interrupts when in Direct Mode
147 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)
149 struct pt_regs *old_regs = set_irq_regs(regs);
151 inc_irq_stat(hyperv_stimer0_count);
152 if (hv_stimer0_handler)
153 hv_stimer0_handler();
154 add_interrupt_randomness(HYPERV_STIMER0_VECTOR);
157 set_irq_regs(old_regs);
160 /* For x86/x64, override weak placeholders in hyperv_timer.c */
161 void hv_setup_stimer0_handler(void (*handler)(void))
163 hv_stimer0_handler = handler;
166 void hv_remove_stimer0_handler(void)
168 /* We have no way to deallocate the interrupt gate */
169 hv_stimer0_handler = NULL;
172 void hv_setup_kexec_handler(void (*handler)(void))
174 hv_kexec_handler = handler;
177 void hv_remove_kexec_handler(void)
179 hv_kexec_handler = NULL;
182 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs))
184 hv_crash_handler = handler;
187 void hv_remove_crash_handler(void)
189 hv_crash_handler = NULL;
192 #ifdef CONFIG_KEXEC_CORE
193 static void hv_machine_shutdown(void)
195 if (kexec_in_progress && hv_kexec_handler)
199 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor
200 * corrupts the old VP Assist Pages and can crash the kexec kernel.
202 if (kexec_in_progress && hyperv_init_cpuhp > 0)
203 cpuhp_remove_state(hyperv_init_cpuhp);
205 /* The function calls stop_other_cpus(). */
206 native_machine_shutdown();
208 /* Disable the hypercall page when there is only 1 active CPU. */
209 if (kexec_in_progress)
212 #endif /* CONFIG_KEXEC_CORE */
214 #ifdef CONFIG_CRASH_DUMP
215 static void hv_machine_crash_shutdown(struct pt_regs *regs)
217 if (hv_crash_handler)
218 hv_crash_handler(regs);
220 /* The function calls crash_smp_send_stop(). */
221 native_machine_crash_shutdown(regs);
223 /* Disable the hypercall page when there is only 1 active CPU. */
226 #endif /* CONFIG_CRASH_DUMP */
227 #endif /* CONFIG_HYPERV */
229 static uint32_t __init ms_hyperv_platform(void)
232 u32 hyp_signature[3];
234 if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
237 cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS,
238 &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]);
240 if (eax < HYPERV_CPUID_MIN || eax > HYPERV_CPUID_MAX ||
241 memcmp("Microsoft Hv", hyp_signature, 12))
244 /* HYPERCALL and VP_INDEX MSRs are mandatory for all features. */
245 eax = cpuid_eax(HYPERV_CPUID_FEATURES);
246 if (!(eax & HV_MSR_HYPERCALL_AVAILABLE)) {
247 pr_warn("x86/hyperv: HYPERCALL MSR not available.\n");
250 if (!(eax & HV_MSR_VP_INDEX_AVAILABLE)) {
251 pr_warn("x86/hyperv: VP_INDEX MSR not available.\n");
255 return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
258 #ifdef CONFIG_X86_LOCAL_APIC
260 * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes
261 * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle
262 * unknown NMI on the first CPU which gets it.
264 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)
266 static atomic_t nmi_cpu = ATOMIC_INIT(-1);
267 unsigned int old_cpu, this_cpu;
269 if (!unknown_nmi_panic)
273 this_cpu = raw_smp_processor_id();
274 if (!atomic_try_cmpxchg(&nmi_cpu, &old_cpu, this_cpu))
281 static unsigned long hv_get_tsc_khz(void)
285 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
290 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
291 static void __init hv_smp_prepare_boot_cpu(void)
293 native_smp_prepare_boot_cpu();
294 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
299 static void __init hv_smp_prepare_cpus(unsigned int max_cpus)
306 native_smp_prepare_cpus(max_cpus);
309 * Override wakeup_secondary_cpu_64 callback for SEV-SNP
312 if (!ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
313 apic->wakeup_secondary_cpu_64 = hv_snp_boot_ap;
318 for_each_present_cpu(i) {
321 ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i));
325 for_each_present_cpu(i) {
328 ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i);
336 * When a fully enlightened TDX VM runs on Hyper-V, the firmware sets the
337 * HW_REDUCED flag: refer to acpi_tb_create_local_fadt(). Consequently ttyS0
338 * interrupts can't work because request_irq() -> ... -> irq_to_desc() returns
339 * NULL for ttyS0. This happens because mp_config_acpi_legacy_irqs() sees a
340 * nr_legacy_irqs() of 0, so it doesn't initialize the array 'mp_irqs[]', and
341 * later setup_IO_APIC_irqs() -> find_irq_entry() fails to find the legacy irqs
342 * from the array and hence doesn't create the necessary irq description info.
344 * Clone arch/x86/kernel/acpi/boot.c: acpi_generic_reduced_hw_init() here,
345 * except don't change 'legacy_pic', which keeps its default value
346 * 'default_legacy_pic'. This way, mp_config_acpi_legacy_irqs() sees a non-zero
347 * nr_legacy_irqs() and eventually serial console interrupts works properly.
349 static void __init reduced_hw_init(void)
351 x86_init.timers.timer_init = x86_init_noop;
352 x86_init.irqs.pre_vector_init = x86_init_noop;
355 int hv_get_hypervisor_version(union hv_hypervisor_version_info *info)
357 unsigned int hv_max_functions;
359 hv_max_functions = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
360 if (hv_max_functions < HYPERV_CPUID_VERSION) {
361 pr_err("%s: Could not detect Hyper-V version\n", __func__);
365 cpuid(HYPERV_CPUID_VERSION, &info->eax, &info->ebx, &info->ecx, &info->edx);
370 static void __init ms_hyperv_init_platform(void)
372 int hv_max_functions_eax;
374 #ifdef CONFIG_PARAVIRT
375 pv_info.name = "Hyper-V";
379 * Extract the features and hints
381 ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
382 ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES);
383 ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES);
384 ms_hyperv.hints = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO);
386 hv_max_functions_eax = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
388 pr_info("Hyper-V: privilege flags low 0x%x, high 0x%x, hints 0x%x, misc 0x%x\n",
389 ms_hyperv.features, ms_hyperv.priv_high, ms_hyperv.hints,
390 ms_hyperv.misc_features);
392 ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
393 ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
395 pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
396 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
399 * Check CPU management privilege.
401 * To mirror what Windows does we should extract CPU management
402 * features and use the ReservedIdentityBit to detect if Linux is the
403 * root partition. But that requires negotiating CPU management
404 * interface (a process to be finalized). For now, use the privilege
405 * flag as the indicator for running as root.
407 * Hyper-V should never specify running as root and as a Confidential
408 * VM. But to protect against a compromised/malicious Hyper-V trying
409 * to exploit root behavior to expose Confidential VM memory, ignore
410 * the root partition setting if also a Confidential VM.
412 if ((ms_hyperv.priv_high & HV_CPU_MANAGEMENT) &&
413 !(ms_hyperv.priv_high & HV_ISOLATION)) {
414 hv_root_partition = true;
415 pr_info("Hyper-V: running as root partition\n");
418 if (ms_hyperv.hints & HV_X64_HYPERV_NESTED) {
420 pr_info("Hyper-V: running on a nested hypervisor\n");
423 if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
424 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
425 x86_platform.calibrate_tsc = hv_get_tsc_khz;
426 x86_platform.calibrate_cpu = hv_get_tsc_khz;
429 if (ms_hyperv.priv_high & HV_ISOLATION) {
430 ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG);
431 ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG);
433 if (ms_hyperv.shared_gpa_boundary_active)
434 ms_hyperv.shared_gpa_boundary =
435 BIT_ULL(ms_hyperv.shared_gpa_boundary_bits);
437 hyperv_paravisor_present = !!ms_hyperv.paravisor_present;
439 pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n",
440 ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b);
443 if (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP) {
444 static_branch_enable(&isolation_type_snp);
445 } else if (hv_get_isolation_type() == HV_ISOLATION_TYPE_TDX) {
446 static_branch_enable(&isolation_type_tdx);
448 /* A TDX VM must use x2APIC and doesn't use lazy EOI. */
449 ms_hyperv.hints &= ~HV_X64_APIC_ACCESS_RECOMMENDED;
451 if (!ms_hyperv.paravisor_present) {
452 /* To be supported: more work is required. */
453 ms_hyperv.features &= ~HV_MSR_REFERENCE_TSC_AVAILABLE;
455 /* HV_MSR_CRASH_CTL is unsupported. */
456 ms_hyperv.misc_features &= ~HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
458 /* Don't trust Hyper-V's TLB-flushing hypercalls. */
459 ms_hyperv.hints &= ~HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
461 x86_init.acpi.reduced_hw_early_init = reduced_hw_init;
466 if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) {
467 ms_hyperv.nested_features =
468 cpuid_eax(HYPERV_CPUID_NESTED_FEATURES);
469 pr_info("Hyper-V: Nested features: 0x%x\n",
470 ms_hyperv.nested_features);
473 #ifdef CONFIG_X86_LOCAL_APIC
474 if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
475 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
477 * Get the APIC frequency.
479 u64 hv_lapic_frequency;
481 rdmsrl(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency);
482 hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ);
483 lapic_timer_period = hv_lapic_frequency;
484 pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
488 register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST,
492 #ifdef CONFIG_X86_IO_APIC
496 #if IS_ENABLED(CONFIG_HYPERV)
497 #if defined(CONFIG_KEXEC_CORE)
498 machine_ops.shutdown = hv_machine_shutdown;
500 #if defined(CONFIG_CRASH_DUMP)
501 machine_ops.crash_shutdown = hv_machine_crash_shutdown;
504 if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) {
506 * Writing to synthetic MSR 0x40000118 updates/changes the
507 * guest visible CPUIDs. Setting bit 0 of this MSR enables
508 * guests to report invariant TSC feature through CPUID
509 * instruction, CPUID 0x800000007/EDX, bit 8. See code in
510 * early_init_intel() where this bit is examined. The
511 * setting of this MSR bit should happen before init_intel()
514 wrmsrl(HV_X64_MSR_TSC_INVARIANT_CONTROL, HV_EXPOSE_INVARIANT_TSC);
515 setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
519 * Generation 2 instances don't support reading the NMI status from
522 if (efi_enabled(EFI_BOOT))
523 x86_platform.get_nmi_reason = hv_get_nmi_reason;
526 * Hyper-V VMs have a PIT emulation quirk such that zeroing the
527 * counter register during PIT shutdown restarts the PIT. So it
528 * continues to interrupt @18.2 HZ. Setting i8253_clear_counter
529 * to false tells pit_shutdown() not to zero the counter so that
530 * the PIT really is shutdown. Generation 2 VMs don't have a PIT,
531 * and setting this value has no effect.
533 i8253_clear_counter_on_shutdown = false;
535 #if IS_ENABLED(CONFIG_HYPERV)
536 if ((hv_get_isolation_type() == HV_ISOLATION_TYPE_VBS) ||
537 ms_hyperv.paravisor_present)
540 * Setup the hook to get control post apic initialization.
542 x86_platform.apic_post_init = hyperv_init;
543 hyperv_setup_mmu_ops();
545 /* Install system interrupt handler for hypervisor callback */
546 sysvec_install(HYPERVISOR_CALLBACK_VECTOR, sysvec_hyperv_callback);
548 /* Install system interrupt handler for reenlightenment notifications */
549 if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) {
550 sysvec_install(HYPERV_REENLIGHTENMENT_VECTOR, sysvec_hyperv_reenlightenment);
553 /* Install system interrupt handler for stimer0 */
554 if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) {
555 sysvec_install(HYPERV_STIMER0_VECTOR, sysvec_hyperv_stimer0);
559 smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
560 if (hv_root_partition ||
561 (!ms_hyperv.paravisor_present && hv_isolation_type_snp()))
562 smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus;
566 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
567 * set x2apic destination mode to physical mode when x2apic is available
568 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
569 * have 8-bit APIC id.
571 # ifdef CONFIG_X86_X2APIC
572 if (x2apic_supported())
576 /* Register Hyper-V specific clocksource */
577 hv_init_clocksource();
578 hv_vtl_init_platform();
581 * TSC should be marked as unstable only after Hyper-V
582 * clocksource has been initialized. This ensures that the
583 * stability of the sched_clock is not altered.
585 if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
586 mark_tsc_unstable("running on Hyper-V");
588 hardlockup_detector_disable();
591 static bool __init ms_hyperv_x2apic_available(void)
593 return x2apic_supported();
597 * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping()
598 * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the
599 * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg().
601 * Note: for a VM on Hyper-V, the I/O-APIC is the only device which
602 * (logically) generates MSIs directly to the system APIC irq domain.
603 * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the
604 * pci-hyperv host bridge.
606 * Note: for a Hyper-V root partition, this will always return false.
607 * The hypervisor doesn't expose these HYPERV_CPUID_VIRT_STACK_* cpuids by
608 * default, they are implemented as intercepts by the Windows Hyper-V stack.
609 * Even a nested root partition (L2 root) will not get them because the
610 * nested (L1) hypervisor filters them out.
612 static bool __init ms_hyperv_msi_ext_dest_id(void)
616 eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE);
617 if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE)
620 eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES);
621 return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE;
624 #ifdef CONFIG_AMD_MEM_ENCRYPT
625 static void hv_sev_es_hcall_prepare(struct ghcb *ghcb, struct pt_regs *regs)
627 /* RAX and CPL are already in the GHCB */
628 ghcb_set_rcx(ghcb, regs->cx);
629 ghcb_set_rdx(ghcb, regs->dx);
630 ghcb_set_r8(ghcb, regs->r8);
633 static bool hv_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs)
635 /* No checking of the return state needed */
640 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = {
641 .name = "Microsoft Hyper-V",
642 .detect = ms_hyperv_platform,
643 .type = X86_HYPER_MS_HYPERV,
644 .init.x2apic_available = ms_hyperv_x2apic_available,
645 .init.msi_ext_dest_id = ms_hyperv_msi_ext_dest_id,
646 .init.init_platform = ms_hyperv_init_platform,
647 .init.guest_late_init = ms_hyperv_late_init,
648 #ifdef CONFIG_AMD_MEM_ENCRYPT
649 .runtime.sev_es_hcall_prepare = hv_sev_es_hcall_prepare,
650 .runtime.sev_es_hcall_finish = hv_sev_es_hcall_finish,