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/random.h>
20 #include <asm/processor.h>
21 #include <asm/hypervisor.h>
22 #include <asm/hyperv-tlfs.h>
23 #include <asm/mshyperv.h>
25 #include <asm/idtentry.h>
26 #include <asm/irq_regs.h>
27 #include <asm/i8259.h>
29 #include <asm/timer.h>
30 #include <asm/reboot.h>
32 #include <clocksource/hyperv_timer.h>
36 /* Is Linux running as the root partition? */
37 bool hv_root_partition
;
38 /* Is Linux running on nested Microsoft Hypervisor */
40 struct ms_hyperv_info ms_hyperv
;
42 /* Used in modules via hv_do_hypercall(): see arch/x86/include/asm/mshyperv.h */
43 bool hyperv_paravisor_present __ro_after_init
;
44 EXPORT_SYMBOL_GPL(hyperv_paravisor_present
);
46 #if IS_ENABLED(CONFIG_HYPERV)
47 static inline unsigned int hv_get_nested_msr(unsigned int reg
)
49 if (hv_is_sint_msr(reg
))
50 return reg
- HV_X64_MSR_SINT0
+ HV_X64_MSR_NESTED_SINT0
;
54 return HV_X64_MSR_NESTED_SIMP
;
55 case HV_X64_MSR_SIEFP
:
56 return HV_X64_MSR_NESTED_SIEFP
;
57 case HV_X64_MSR_SVERSION
:
58 return HV_X64_MSR_NESTED_SVERSION
;
59 case HV_X64_MSR_SCONTROL
:
60 return HV_X64_MSR_NESTED_SCONTROL
;
62 return HV_X64_MSR_NESTED_EOM
;
68 u64
hv_get_non_nested_msr(unsigned int reg
)
72 if (hv_is_synic_msr(reg
) && ms_hyperv
.paravisor_present
)
73 hv_ivm_msr_read(reg
, &value
);
78 EXPORT_SYMBOL_GPL(hv_get_non_nested_msr
);
80 void hv_set_non_nested_msr(unsigned int reg
, u64 value
)
82 if (hv_is_synic_msr(reg
) && ms_hyperv
.paravisor_present
) {
83 hv_ivm_msr_write(reg
, value
);
85 /* Write proxy bit via wrmsl instruction */
86 if (hv_is_sint_msr(reg
))
87 wrmsrl(reg
, value
| 1 << 20);
92 EXPORT_SYMBOL_GPL(hv_set_non_nested_msr
);
94 u64
hv_get_msr(unsigned int reg
)
97 reg
= hv_get_nested_msr(reg
);
99 return hv_get_non_nested_msr(reg
);
101 EXPORT_SYMBOL_GPL(hv_get_msr
);
103 void hv_set_msr(unsigned int reg
, u64 value
)
106 reg
= hv_get_nested_msr(reg
);
108 hv_set_non_nested_msr(reg
, value
);
110 EXPORT_SYMBOL_GPL(hv_set_msr
);
112 static void (*vmbus_handler
)(void);
113 static void (*hv_stimer0_handler
)(void);
114 static void (*hv_kexec_handler
)(void);
115 static void (*hv_crash_handler
)(struct pt_regs
*regs
);
117 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback
)
119 struct pt_regs
*old_regs
= set_irq_regs(regs
);
121 inc_irq_stat(irq_hv_callback_count
);
125 if (ms_hyperv
.hints
& HV_DEPRECATING_AEOI_RECOMMENDED
)
128 set_irq_regs(old_regs
);
131 void hv_setup_vmbus_handler(void (*handler
)(void))
133 vmbus_handler
= handler
;
136 void hv_remove_vmbus_handler(void)
138 /* We have no way to deallocate the interrupt gate */
139 vmbus_handler
= NULL
;
143 * Routines to do per-architecture handling of stimer0
144 * interrupts when in Direct Mode
146 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0
)
148 struct pt_regs
*old_regs
= set_irq_regs(regs
);
150 inc_irq_stat(hyperv_stimer0_count
);
151 if (hv_stimer0_handler
)
152 hv_stimer0_handler();
153 add_interrupt_randomness(HYPERV_STIMER0_VECTOR
);
156 set_irq_regs(old_regs
);
159 /* For x86/x64, override weak placeholders in hyperv_timer.c */
160 void hv_setup_stimer0_handler(void (*handler
)(void))
162 hv_stimer0_handler
= handler
;
165 void hv_remove_stimer0_handler(void)
167 /* We have no way to deallocate the interrupt gate */
168 hv_stimer0_handler
= NULL
;
171 void hv_setup_kexec_handler(void (*handler
)(void))
173 hv_kexec_handler
= handler
;
176 void hv_remove_kexec_handler(void)
178 hv_kexec_handler
= NULL
;
181 void hv_setup_crash_handler(void (*handler
)(struct pt_regs
*regs
))
183 hv_crash_handler
= handler
;
186 void hv_remove_crash_handler(void)
188 hv_crash_handler
= NULL
;
191 #ifdef CONFIG_KEXEC_CORE
192 static void hv_machine_shutdown(void)
194 if (kexec_in_progress
&& hv_kexec_handler
)
198 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor
199 * corrupts the old VP Assist Pages and can crash the kexec kernel.
201 if (kexec_in_progress
)
202 cpuhp_remove_state(CPUHP_AP_HYPERV_ONLINE
);
204 /* The function calls stop_other_cpus(). */
205 native_machine_shutdown();
207 /* Disable the hypercall page when there is only 1 active CPU. */
208 if (kexec_in_progress
)
211 #endif /* CONFIG_KEXEC_CORE */
213 #ifdef CONFIG_CRASH_DUMP
214 static void hv_machine_crash_shutdown(struct pt_regs
*regs
)
216 if (hv_crash_handler
)
217 hv_crash_handler(regs
);
219 /* The function calls crash_smp_send_stop(). */
220 native_machine_crash_shutdown(regs
);
222 /* Disable the hypercall page when there is only 1 active CPU. */
225 #endif /* CONFIG_CRASH_DUMP */
226 #endif /* CONFIG_HYPERV */
228 static uint32_t __init
ms_hyperv_platform(void)
231 u32 hyp_signature
[3];
233 if (!boot_cpu_has(X86_FEATURE_HYPERVISOR
))
236 cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS
,
237 &eax
, &hyp_signature
[0], &hyp_signature
[1], &hyp_signature
[2]);
239 if (eax
< HYPERV_CPUID_MIN
|| eax
> HYPERV_CPUID_MAX
||
240 memcmp("Microsoft Hv", hyp_signature
, 12))
243 /* HYPERCALL and VP_INDEX MSRs are mandatory for all features. */
244 eax
= cpuid_eax(HYPERV_CPUID_FEATURES
);
245 if (!(eax
& HV_MSR_HYPERCALL_AVAILABLE
)) {
246 pr_warn("x86/hyperv: HYPERCALL MSR not available.\n");
249 if (!(eax
& HV_MSR_VP_INDEX_AVAILABLE
)) {
250 pr_warn("x86/hyperv: VP_INDEX MSR not available.\n");
254 return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS
;
257 #ifdef CONFIG_X86_LOCAL_APIC
259 * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes
260 * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle
261 * unknown NMI on the first CPU which gets it.
263 static int hv_nmi_unknown(unsigned int val
, struct pt_regs
*regs
)
265 static atomic_t nmi_cpu
= ATOMIC_INIT(-1);
266 unsigned int old_cpu
, this_cpu
;
268 if (!unknown_nmi_panic
)
272 this_cpu
= raw_smp_processor_id();
273 if (!atomic_try_cmpxchg(&nmi_cpu
, &old_cpu
, this_cpu
))
280 static unsigned long hv_get_tsc_khz(void)
284 rdmsrl(HV_X64_MSR_TSC_FREQUENCY
, freq
);
289 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
290 static void __init
hv_smp_prepare_boot_cpu(void)
292 native_smp_prepare_boot_cpu();
293 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
298 static void __init
hv_smp_prepare_cpus(unsigned int max_cpus
)
305 native_smp_prepare_cpus(max_cpus
);
308 * Override wakeup_secondary_cpu_64 callback for SEV-SNP
311 if (!ms_hyperv
.paravisor_present
&& hv_isolation_type_snp()) {
312 apic
->wakeup_secondary_cpu_64
= hv_snp_boot_ap
;
317 for_each_present_cpu(i
) {
320 ret
= hv_call_add_logical_proc(numa_cpu_node(i
), i
, cpu_physical_id(i
));
324 for_each_present_cpu(i
) {
327 ret
= hv_call_create_vp(numa_cpu_node(i
), hv_current_partition_id
, i
, i
);
335 * When a fully enlightened TDX VM runs on Hyper-V, the firmware sets the
336 * HW_REDUCED flag: refer to acpi_tb_create_local_fadt(). Consequently ttyS0
337 * interrupts can't work because request_irq() -> ... -> irq_to_desc() returns
338 * NULL for ttyS0. This happens because mp_config_acpi_legacy_irqs() sees a
339 * nr_legacy_irqs() of 0, so it doesn't initialize the array 'mp_irqs[]', and
340 * later setup_IO_APIC_irqs() -> find_irq_entry() fails to find the legacy irqs
341 * from the array and hence doesn't create the necessary irq description info.
343 * Clone arch/x86/kernel/acpi/boot.c: acpi_generic_reduced_hw_init() here,
344 * except don't change 'legacy_pic', which keeps its default value
345 * 'default_legacy_pic'. This way, mp_config_acpi_legacy_irqs() sees a non-zero
346 * nr_legacy_irqs() and eventually serial console interrupts works properly.
348 static void __init
reduced_hw_init(void)
350 x86_init
.timers
.timer_init
= x86_init_noop
;
351 x86_init
.irqs
.pre_vector_init
= x86_init_noop
;
354 int hv_get_hypervisor_version(union hv_hypervisor_version_info
*info
)
356 unsigned int hv_max_functions
;
358 hv_max_functions
= cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS
);
359 if (hv_max_functions
< HYPERV_CPUID_VERSION
) {
360 pr_err("%s: Could not detect Hyper-V version\n", __func__
);
364 cpuid(HYPERV_CPUID_VERSION
, &info
->eax
, &info
->ebx
, &info
->ecx
, &info
->edx
);
369 static void __init
ms_hyperv_init_platform(void)
371 int hv_max_functions_eax
;
373 #ifdef CONFIG_PARAVIRT
374 pv_info
.name
= "Hyper-V";
378 * Extract the features and hints
380 ms_hyperv
.features
= cpuid_eax(HYPERV_CPUID_FEATURES
);
381 ms_hyperv
.priv_high
= cpuid_ebx(HYPERV_CPUID_FEATURES
);
382 ms_hyperv
.misc_features
= cpuid_edx(HYPERV_CPUID_FEATURES
);
383 ms_hyperv
.hints
= cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO
);
385 hv_max_functions_eax
= cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS
);
387 pr_info("Hyper-V: privilege flags low 0x%x, high 0x%x, hints 0x%x, misc 0x%x\n",
388 ms_hyperv
.features
, ms_hyperv
.priv_high
, ms_hyperv
.hints
,
389 ms_hyperv
.misc_features
);
391 ms_hyperv
.max_vp_index
= cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS
);
392 ms_hyperv
.max_lp_index
= cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS
);
394 pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
395 ms_hyperv
.max_vp_index
, ms_hyperv
.max_lp_index
);
398 * Check CPU management privilege.
400 * To mirror what Windows does we should extract CPU management
401 * features and use the ReservedIdentityBit to detect if Linux is the
402 * root partition. But that requires negotiating CPU management
403 * interface (a process to be finalized). For now, use the privilege
404 * flag as the indicator for running as root.
406 * Hyper-V should never specify running as root and as a Confidential
407 * VM. But to protect against a compromised/malicious Hyper-V trying
408 * to exploit root behavior to expose Confidential VM memory, ignore
409 * the root partition setting if also a Confidential VM.
411 if ((ms_hyperv
.priv_high
& HV_CPU_MANAGEMENT
) &&
412 !(ms_hyperv
.priv_high
& HV_ISOLATION
)) {
413 hv_root_partition
= true;
414 pr_info("Hyper-V: running as root partition\n");
417 if (ms_hyperv
.hints
& HV_X64_HYPERV_NESTED
) {
419 pr_info("Hyper-V: running on a nested hypervisor\n");
422 if (ms_hyperv
.features
& HV_ACCESS_FREQUENCY_MSRS
&&
423 ms_hyperv
.misc_features
& HV_FEATURE_FREQUENCY_MSRS_AVAILABLE
) {
424 x86_platform
.calibrate_tsc
= hv_get_tsc_khz
;
425 x86_platform
.calibrate_cpu
= hv_get_tsc_khz
;
426 setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ
);
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
) {
453 * Mark the Hyper-V TSC page feature as disabled
454 * in a TDX VM without paravisor so that the
455 * Invariant TSC, which is a better clocksource
456 * anyway, is used instead.
458 ms_hyperv
.features
&= ~HV_MSR_REFERENCE_TSC_AVAILABLE
;
461 * The Invariant TSC is expected to be available
462 * in a TDX VM without paravisor, but if not,
463 * print a warning message. The slower Hyper-V MSR-based
464 * Ref Counter should end up being the clocksource.
466 if (!(ms_hyperv
.features
& HV_ACCESS_TSC_INVARIANT
))
467 pr_warn("Hyper-V: Invariant TSC is unavailable\n");
469 /* HV_MSR_CRASH_CTL is unsupported. */
470 ms_hyperv
.misc_features
&= ~HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE
;
472 /* Don't trust Hyper-V's TLB-flushing hypercalls. */
473 ms_hyperv
.hints
&= ~HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED
;
475 x86_init
.acpi
.reduced_hw_early_init
= reduced_hw_init
;
480 if (hv_max_functions_eax
>= HYPERV_CPUID_NESTED_FEATURES
) {
481 ms_hyperv
.nested_features
=
482 cpuid_eax(HYPERV_CPUID_NESTED_FEATURES
);
483 pr_info("Hyper-V: Nested features: 0x%x\n",
484 ms_hyperv
.nested_features
);
487 #ifdef CONFIG_X86_LOCAL_APIC
488 if (ms_hyperv
.features
& HV_ACCESS_FREQUENCY_MSRS
&&
489 ms_hyperv
.misc_features
& HV_FEATURE_FREQUENCY_MSRS_AVAILABLE
) {
491 * Get the APIC frequency.
493 u64 hv_lapic_frequency
;
495 rdmsrl(HV_X64_MSR_APIC_FREQUENCY
, hv_lapic_frequency
);
496 hv_lapic_frequency
= div_u64(hv_lapic_frequency
, HZ
);
497 lapic_timer_period
= hv_lapic_frequency
;
498 pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
502 register_nmi_handler(NMI_UNKNOWN
, hv_nmi_unknown
, NMI_FLAG_FIRST
,
506 #ifdef CONFIG_X86_IO_APIC
510 #if IS_ENABLED(CONFIG_HYPERV)
511 #if defined(CONFIG_KEXEC_CORE)
512 machine_ops
.shutdown
= hv_machine_shutdown
;
514 #if defined(CONFIG_CRASH_DUMP)
515 machine_ops
.crash_shutdown
= hv_machine_crash_shutdown
;
518 if (ms_hyperv
.features
& HV_ACCESS_TSC_INVARIANT
) {
520 * Writing to synthetic MSR 0x40000118 updates/changes the
521 * guest visible CPUIDs. Setting bit 0 of this MSR enables
522 * guests to report invariant TSC feature through CPUID
523 * instruction, CPUID 0x800000007/EDX, bit 8. See code in
524 * early_init_intel() where this bit is examined. The
525 * setting of this MSR bit should happen before init_intel()
528 wrmsrl(HV_X64_MSR_TSC_INVARIANT_CONTROL
, HV_EXPOSE_INVARIANT_TSC
);
529 setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE
);
533 * Generation 2 instances don't support reading the NMI status from
536 if (efi_enabled(EFI_BOOT
))
537 x86_platform
.get_nmi_reason
= hv_get_nmi_reason
;
539 #if IS_ENABLED(CONFIG_HYPERV)
540 if ((hv_get_isolation_type() == HV_ISOLATION_TYPE_VBS
) ||
541 ms_hyperv
.paravisor_present
)
544 * Setup the hook to get control post apic initialization.
546 x86_platform
.apic_post_init
= hyperv_init
;
547 hyperv_setup_mmu_ops();
549 /* Install system interrupt handler for hypervisor callback */
550 sysvec_install(HYPERVISOR_CALLBACK_VECTOR
, sysvec_hyperv_callback
);
552 /* Install system interrupt handler for reenlightenment notifications */
553 if (ms_hyperv
.features
& HV_ACCESS_REENLIGHTENMENT
) {
554 sysvec_install(HYPERV_REENLIGHTENMENT_VECTOR
, sysvec_hyperv_reenlightenment
);
557 /* Install system interrupt handler for stimer0 */
558 if (ms_hyperv
.misc_features
& HV_STIMER_DIRECT_MODE_AVAILABLE
) {
559 sysvec_install(HYPERV_STIMER0_VECTOR
, sysvec_hyperv_stimer0
);
563 smp_ops
.smp_prepare_boot_cpu
= hv_smp_prepare_boot_cpu
;
564 if (hv_root_partition
||
565 (!ms_hyperv
.paravisor_present
&& hv_isolation_type_snp()))
566 smp_ops
.smp_prepare_cpus
= hv_smp_prepare_cpus
;
570 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
571 * set x2apic destination mode to physical mode when x2apic is available
572 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
573 * have 8-bit APIC id.
575 # ifdef CONFIG_X86_X2APIC
576 if (x2apic_supported())
580 /* Register Hyper-V specific clocksource */
581 hv_init_clocksource();
582 hv_vtl_init_platform();
585 * TSC should be marked as unstable only after Hyper-V
586 * clocksource has been initialized. This ensures that the
587 * stability of the sched_clock is not altered.
589 if (!(ms_hyperv
.features
& HV_ACCESS_TSC_INVARIANT
))
590 mark_tsc_unstable("running on Hyper-V");
592 hardlockup_detector_disable();
595 static bool __init
ms_hyperv_x2apic_available(void)
597 return x2apic_supported();
601 * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping()
602 * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the
603 * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg().
605 * Note: for a VM on Hyper-V, the I/O-APIC is the only device which
606 * (logically) generates MSIs directly to the system APIC irq domain.
607 * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the
608 * pci-hyperv host bridge.
610 * Note: for a Hyper-V root partition, this will always return false.
611 * The hypervisor doesn't expose these HYPERV_CPUID_VIRT_STACK_* cpuids by
612 * default, they are implemented as intercepts by the Windows Hyper-V stack.
613 * Even a nested root partition (L2 root) will not get them because the
614 * nested (L1) hypervisor filters them out.
616 static bool __init
ms_hyperv_msi_ext_dest_id(void)
620 eax
= cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE
);
621 if (eax
!= HYPERV_VS_INTERFACE_EAX_SIGNATURE
)
624 eax
= cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES
);
625 return eax
& HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE
;
628 #ifdef CONFIG_AMD_MEM_ENCRYPT
629 static void hv_sev_es_hcall_prepare(struct ghcb
*ghcb
, struct pt_regs
*regs
)
631 /* RAX and CPL are already in the GHCB */
632 ghcb_set_rcx(ghcb
, regs
->cx
);
633 ghcb_set_rdx(ghcb
, regs
->dx
);
634 ghcb_set_r8(ghcb
, regs
->r8
);
637 static bool hv_sev_es_hcall_finish(struct ghcb
*ghcb
, struct pt_regs
*regs
)
639 /* No checking of the return state needed */
644 const __initconst
struct hypervisor_x86 x86_hyper_ms_hyperv
= {
645 .name
= "Microsoft Hyper-V",
646 .detect
= ms_hyperv_platform
,
647 .type
= X86_HYPER_MS_HYPERV
,
648 .init
.x2apic_available
= ms_hyperv_x2apic_available
,
649 .init
.msi_ext_dest_id
= ms_hyperv_msi_ext_dest_id
,
650 .init
.init_platform
= ms_hyperv_init_platform
,
651 .init
.guest_late_init
= ms_hyperv_late_init
,
652 #ifdef CONFIG_AMD_MEM_ENCRYPT
653 .runtime
.sev_es_hcall_prepare
= hv_sev_es_hcall_prepare
,
654 .runtime
.sev_es_hcall_finish
= hv_sev_es_hcall_finish
,