xref: /linux/arch/x86/kernel/cpu/mshyperv.c (revision f5c31bcf604db54470868f3118a60dc4a9ba8813)
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/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>
25 #include <asm/desc.h>
26 #include <asm/idtentry.h>
27 #include <asm/irq_regs.h>
28 #include <asm/i8259.h>
29 #include <asm/apic.h>
30 #include <asm/timer.h>
31 #include <asm/reboot.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 as the root partition? */
38 bool hv_root_partition;
39 /* Is Linux running on nested Microsoft Hypervisor */
40 bool hv_nested;
41 struct ms_hyperv_info ms_hyperv;
42 
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);
46 
47 #if IS_ENABLED(CONFIG_HYPERV)
48 static inline unsigned int hv_get_nested_reg(unsigned int reg)
49 {
50 	if (hv_is_sint_reg(reg))
51 		return reg - HV_REGISTER_SINT0 + HV_REGISTER_NESTED_SINT0;
52 
53 	switch (reg) {
54 	case HV_REGISTER_SIMP:
55 		return HV_REGISTER_NESTED_SIMP;
56 	case HV_REGISTER_SIEFP:
57 		return HV_REGISTER_NESTED_SIEFP;
58 	case HV_REGISTER_SVERSION:
59 		return HV_REGISTER_NESTED_SVERSION;
60 	case HV_REGISTER_SCONTROL:
61 		return HV_REGISTER_NESTED_SCONTROL;
62 	case HV_REGISTER_EOM:
63 		return HV_REGISTER_NESTED_EOM;
64 	default:
65 		return reg;
66 	}
67 }
68 
69 u64 hv_get_non_nested_register(unsigned int reg)
70 {
71 	u64 value;
72 
73 	if (hv_is_synic_reg(reg) && ms_hyperv.paravisor_present)
74 		hv_ivm_msr_read(reg, &value);
75 	else
76 		rdmsrl(reg, value);
77 	return value;
78 }
79 EXPORT_SYMBOL_GPL(hv_get_non_nested_register);
80 
81 void hv_set_non_nested_register(unsigned int reg, u64 value)
82 {
83 	if (hv_is_synic_reg(reg) && ms_hyperv.paravisor_present) {
84 		hv_ivm_msr_write(reg, value);
85 
86 		/* Write proxy bit via wrmsl instruction */
87 		if (hv_is_sint_reg(reg))
88 			wrmsrl(reg, value | 1 << 20);
89 	} else {
90 		wrmsrl(reg, value);
91 	}
92 }
93 EXPORT_SYMBOL_GPL(hv_set_non_nested_register);
94 
95 u64 hv_get_register(unsigned int reg)
96 {
97 	if (hv_nested)
98 		reg = hv_get_nested_reg(reg);
99 
100 	return hv_get_non_nested_register(reg);
101 }
102 EXPORT_SYMBOL_GPL(hv_get_register);
103 
104 void hv_set_register(unsigned int reg, u64 value)
105 {
106 	if (hv_nested)
107 		reg = hv_get_nested_reg(reg);
108 
109 	hv_set_non_nested_register(reg, value);
110 }
111 EXPORT_SYMBOL_GPL(hv_set_register);
112 
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);
117 
118 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)
119 {
120 	struct pt_regs *old_regs = set_irq_regs(regs);
121 
122 	inc_irq_stat(irq_hv_callback_count);
123 	if (vmbus_handler)
124 		vmbus_handler();
125 
126 	if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
127 		apic_eoi();
128 
129 	set_irq_regs(old_regs);
130 }
131 
132 void hv_setup_vmbus_handler(void (*handler)(void))
133 {
134 	vmbus_handler = handler;
135 }
136 
137 void hv_remove_vmbus_handler(void)
138 {
139 	/* We have no way to deallocate the interrupt gate */
140 	vmbus_handler = NULL;
141 }
142 
143 /*
144  * Routines to do per-architecture handling of stimer0
145  * interrupts when in Direct Mode
146  */
147 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)
148 {
149 	struct pt_regs *old_regs = set_irq_regs(regs);
150 
151 	inc_irq_stat(hyperv_stimer0_count);
152 	if (hv_stimer0_handler)
153 		hv_stimer0_handler();
154 	add_interrupt_randomness(HYPERV_STIMER0_VECTOR);
155 	apic_eoi();
156 
157 	set_irq_regs(old_regs);
158 }
159 
160 /* For x86/x64, override weak placeholders in hyperv_timer.c */
161 void hv_setup_stimer0_handler(void (*handler)(void))
162 {
163 	hv_stimer0_handler = handler;
164 }
165 
166 void hv_remove_stimer0_handler(void)
167 {
168 	/* We have no way to deallocate the interrupt gate */
169 	hv_stimer0_handler = NULL;
170 }
171 
172 void hv_setup_kexec_handler(void (*handler)(void))
173 {
174 	hv_kexec_handler = handler;
175 }
176 
177 void hv_remove_kexec_handler(void)
178 {
179 	hv_kexec_handler = NULL;
180 }
181 
182 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs))
183 {
184 	hv_crash_handler = handler;
185 }
186 
187 void hv_remove_crash_handler(void)
188 {
189 	hv_crash_handler = NULL;
190 }
191 
192 #ifdef CONFIG_KEXEC_CORE
193 static void hv_machine_shutdown(void)
194 {
195 	if (kexec_in_progress && hv_kexec_handler)
196 		hv_kexec_handler();
197 
198 	/*
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.
201 	 */
202 	if (kexec_in_progress && hyperv_init_cpuhp > 0)
203 		cpuhp_remove_state(hyperv_init_cpuhp);
204 
205 	/* The function calls stop_other_cpus(). */
206 	native_machine_shutdown();
207 
208 	/* Disable the hypercall page when there is only 1 active CPU. */
209 	if (kexec_in_progress)
210 		hyperv_cleanup();
211 }
212 
213 static void hv_machine_crash_shutdown(struct pt_regs *regs)
214 {
215 	if (hv_crash_handler)
216 		hv_crash_handler(regs);
217 
218 	/* The function calls crash_smp_send_stop(). */
219 	native_machine_crash_shutdown(regs);
220 
221 	/* Disable the hypercall page when there is only 1 active CPU. */
222 	hyperv_cleanup();
223 }
224 #endif /* CONFIG_KEXEC_CORE */
225 #endif /* CONFIG_HYPERV */
226 
227 static uint32_t  __init ms_hyperv_platform(void)
228 {
229 	u32 eax;
230 	u32 hyp_signature[3];
231 
232 	if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
233 		return 0;
234 
235 	cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS,
236 	      &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]);
237 
238 	if (eax < HYPERV_CPUID_MIN || eax > HYPERV_CPUID_MAX ||
239 	    memcmp("Microsoft Hv", hyp_signature, 12))
240 		return 0;
241 
242 	/* HYPERCALL and VP_INDEX MSRs are mandatory for all features. */
243 	eax = cpuid_eax(HYPERV_CPUID_FEATURES);
244 	if (!(eax & HV_MSR_HYPERCALL_AVAILABLE)) {
245 		pr_warn("x86/hyperv: HYPERCALL MSR not available.\n");
246 		return 0;
247 	}
248 	if (!(eax & HV_MSR_VP_INDEX_AVAILABLE)) {
249 		pr_warn("x86/hyperv: VP_INDEX MSR not available.\n");
250 		return 0;
251 	}
252 
253 	return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
254 }
255 
256 #ifdef CONFIG_X86_LOCAL_APIC
257 /*
258  * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes
259  * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle
260  * unknown NMI on the first CPU which gets it.
261  */
262 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)
263 {
264 	static atomic_t nmi_cpu = ATOMIC_INIT(-1);
265 	unsigned int old_cpu, this_cpu;
266 
267 	if (!unknown_nmi_panic)
268 		return NMI_DONE;
269 
270 	old_cpu = -1;
271 	this_cpu = raw_smp_processor_id();
272 	if (!atomic_try_cmpxchg(&nmi_cpu, &old_cpu, this_cpu))
273 		return NMI_HANDLED;
274 
275 	return NMI_DONE;
276 }
277 #endif
278 
279 static unsigned long hv_get_tsc_khz(void)
280 {
281 	unsigned long freq;
282 
283 	rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
284 
285 	return freq / 1000;
286 }
287 
288 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
289 static void __init hv_smp_prepare_boot_cpu(void)
290 {
291 	native_smp_prepare_boot_cpu();
292 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
293 	hv_init_spinlocks();
294 #endif
295 }
296 
297 static void __init hv_smp_prepare_cpus(unsigned int max_cpus)
298 {
299 #ifdef CONFIG_X86_64
300 	int i;
301 	int ret;
302 #endif
303 
304 	native_smp_prepare_cpus(max_cpus);
305 
306 	/*
307 	 *  Override wakeup_secondary_cpu_64 callback for SEV-SNP
308 	 *  enlightened guest.
309 	 */
310 	if (!ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
311 		apic->wakeup_secondary_cpu_64 = hv_snp_boot_ap;
312 		return;
313 	}
314 
315 #ifdef CONFIG_X86_64
316 	for_each_present_cpu(i) {
317 		if (i == 0)
318 			continue;
319 		ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i));
320 		BUG_ON(ret);
321 	}
322 
323 	for_each_present_cpu(i) {
324 		if (i == 0)
325 			continue;
326 		ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i);
327 		BUG_ON(ret);
328 	}
329 #endif
330 }
331 #endif
332 
333 /*
334  * When a fully enlightened TDX VM runs on Hyper-V, the firmware sets the
335  * HW_REDUCED flag: refer to acpi_tb_create_local_fadt(). Consequently ttyS0
336  * interrupts can't work because request_irq() -> ... -> irq_to_desc() returns
337  * NULL for ttyS0. This happens because mp_config_acpi_legacy_irqs() sees a
338  * nr_legacy_irqs() of 0, so it doesn't initialize the array 'mp_irqs[]', and
339  * later setup_IO_APIC_irqs() -> find_irq_entry() fails to find the legacy irqs
340  * from the array and hence doesn't create the necessary irq description info.
341  *
342  * Clone arch/x86/kernel/acpi/boot.c: acpi_generic_reduced_hw_init() here,
343  * except don't change 'legacy_pic', which keeps its default value
344  * 'default_legacy_pic'. This way, mp_config_acpi_legacy_irqs() sees a non-zero
345  * nr_legacy_irqs() and eventually serial console interrupts works properly.
346  */
347 static void __init reduced_hw_init(void)
348 {
349 	x86_init.timers.timer_init	= x86_init_noop;
350 	x86_init.irqs.pre_vector_init	= x86_init_noop;
351 }
352 
353 static void __init ms_hyperv_init_platform(void)
354 {
355 	int hv_max_functions_eax;
356 	int hv_host_info_eax;
357 	int hv_host_info_ebx;
358 	int hv_host_info_ecx;
359 	int hv_host_info_edx;
360 
361 #ifdef CONFIG_PARAVIRT
362 	pv_info.name = "Hyper-V";
363 #endif
364 
365 	/*
366 	 * Extract the features and hints
367 	 */
368 	ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
369 	ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES);
370 	ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES);
371 	ms_hyperv.hints    = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO);
372 
373 	hv_max_functions_eax = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
374 
375 	pr_info("Hyper-V: privilege flags low 0x%x, high 0x%x, hints 0x%x, misc 0x%x\n",
376 		ms_hyperv.features, ms_hyperv.priv_high, ms_hyperv.hints,
377 		ms_hyperv.misc_features);
378 
379 	ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
380 	ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
381 
382 	pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
383 		 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
384 
385 	/*
386 	 * Check CPU management privilege.
387 	 *
388 	 * To mirror what Windows does we should extract CPU management
389 	 * features and use the ReservedIdentityBit to detect if Linux is the
390 	 * root partition. But that requires negotiating CPU management
391 	 * interface (a process to be finalized). For now, use the privilege
392 	 * flag as the indicator for running as root.
393 	 *
394 	 * Hyper-V should never specify running as root and as a Confidential
395 	 * VM. But to protect against a compromised/malicious Hyper-V trying
396 	 * to exploit root behavior to expose Confidential VM memory, ignore
397 	 * the root partition setting if also a Confidential VM.
398 	 */
399 	if ((ms_hyperv.priv_high & HV_CPU_MANAGEMENT) &&
400 	    !(ms_hyperv.priv_high & HV_ISOLATION)) {
401 		hv_root_partition = true;
402 		pr_info("Hyper-V: running as root partition\n");
403 	}
404 
405 	if (ms_hyperv.hints & HV_X64_HYPERV_NESTED) {
406 		hv_nested = true;
407 		pr_info("Hyper-V: running on a nested hypervisor\n");
408 	}
409 
410 	/*
411 	 * Extract host information.
412 	 */
413 	if (hv_max_functions_eax >= HYPERV_CPUID_VERSION) {
414 		hv_host_info_eax = cpuid_eax(HYPERV_CPUID_VERSION);
415 		hv_host_info_ebx = cpuid_ebx(HYPERV_CPUID_VERSION);
416 		hv_host_info_ecx = cpuid_ecx(HYPERV_CPUID_VERSION);
417 		hv_host_info_edx = cpuid_edx(HYPERV_CPUID_VERSION);
418 
419 		pr_info("Hyper-V: Host Build %d.%d.%d.%d-%d-%d\n",
420 			hv_host_info_ebx >> 16, hv_host_info_ebx & 0xFFFF,
421 			hv_host_info_eax, hv_host_info_edx & 0xFFFFFF,
422 			hv_host_info_ecx, hv_host_info_edx >> 24);
423 	}
424 
425 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
426 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
427 		x86_platform.calibrate_tsc = hv_get_tsc_khz;
428 		x86_platform.calibrate_cpu = hv_get_tsc_khz;
429 	}
430 
431 	if (ms_hyperv.priv_high & HV_ISOLATION) {
432 		ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG);
433 		ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG);
434 
435 		if (ms_hyperv.shared_gpa_boundary_active)
436 			ms_hyperv.shared_gpa_boundary =
437 				BIT_ULL(ms_hyperv.shared_gpa_boundary_bits);
438 
439 		hyperv_paravisor_present = !!ms_hyperv.paravisor_present;
440 
441 		pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n",
442 			ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b);
443 
444 
445 		if (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP) {
446 			static_branch_enable(&isolation_type_snp);
447 		} else if (hv_get_isolation_type() == HV_ISOLATION_TYPE_TDX) {
448 			static_branch_enable(&isolation_type_tdx);
449 
450 			/* A TDX VM must use x2APIC and doesn't use lazy EOI. */
451 			ms_hyperv.hints &= ~HV_X64_APIC_ACCESS_RECOMMENDED;
452 
453 			if (!ms_hyperv.paravisor_present) {
454 				/* To be supported: more work is required.  */
455 				ms_hyperv.features &= ~HV_MSR_REFERENCE_TSC_AVAILABLE;
456 
457 				/* HV_REGISTER_CRASH_CTL is unsupported. */
458 				ms_hyperv.misc_features &= ~HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
459 
460 				/* Don't trust Hyper-V's TLB-flushing hypercalls. */
461 				ms_hyperv.hints &= ~HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
462 
463 				x86_init.acpi.reduced_hw_early_init = reduced_hw_init;
464 			}
465 		}
466 	}
467 
468 	if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) {
469 		ms_hyperv.nested_features =
470 			cpuid_eax(HYPERV_CPUID_NESTED_FEATURES);
471 		pr_info("Hyper-V: Nested features: 0x%x\n",
472 			ms_hyperv.nested_features);
473 	}
474 
475 #ifdef CONFIG_X86_LOCAL_APIC
476 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
477 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
478 		/*
479 		 * Get the APIC frequency.
480 		 */
481 		u64	hv_lapic_frequency;
482 
483 		rdmsrl(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency);
484 		hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ);
485 		lapic_timer_period = hv_lapic_frequency;
486 		pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
487 			lapic_timer_period);
488 	}
489 
490 	register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST,
491 			     "hv_nmi_unknown");
492 #endif
493 
494 #ifdef CONFIG_X86_IO_APIC
495 	no_timer_check = 1;
496 #endif
497 
498 #if IS_ENABLED(CONFIG_HYPERV) && defined(CONFIG_KEXEC_CORE)
499 	machine_ops.shutdown = hv_machine_shutdown;
500 	machine_ops.crash_shutdown = hv_machine_crash_shutdown;
501 #endif
502 	if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) {
503 		/*
504 		 * Writing to synthetic MSR 0x40000118 updates/changes the
505 		 * guest visible CPUIDs. Setting bit 0 of this MSR  enables
506 		 * guests to report invariant TSC feature through CPUID
507 		 * instruction, CPUID 0x800000007/EDX, bit 8. See code in
508 		 * early_init_intel() where this bit is examined. The
509 		 * setting of this MSR bit should happen before init_intel()
510 		 * is called.
511 		 */
512 		wrmsrl(HV_X64_MSR_TSC_INVARIANT_CONTROL, HV_EXPOSE_INVARIANT_TSC);
513 		setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
514 	}
515 
516 	/*
517 	 * Generation 2 instances don't support reading the NMI status from
518 	 * 0x61 port.
519 	 */
520 	if (efi_enabled(EFI_BOOT))
521 		x86_platform.get_nmi_reason = hv_get_nmi_reason;
522 
523 	/*
524 	 * Hyper-V VMs have a PIT emulation quirk such that zeroing the
525 	 * counter register during PIT shutdown restarts the PIT. So it
526 	 * continues to interrupt @18.2 HZ. Setting i8253_clear_counter
527 	 * to false tells pit_shutdown() not to zero the counter so that
528 	 * the PIT really is shutdown. Generation 2 VMs don't have a PIT,
529 	 * and setting this value has no effect.
530 	 */
531 	i8253_clear_counter_on_shutdown = false;
532 
533 #if IS_ENABLED(CONFIG_HYPERV)
534 	if ((hv_get_isolation_type() == HV_ISOLATION_TYPE_VBS) ||
535 	    ms_hyperv.paravisor_present)
536 		hv_vtom_init();
537 	/*
538 	 * Setup the hook to get control post apic initialization.
539 	 */
540 	x86_platform.apic_post_init = hyperv_init;
541 	hyperv_setup_mmu_ops();
542 
543 	/* Install system interrupt handler for hypervisor callback */
544 	sysvec_install(HYPERVISOR_CALLBACK_VECTOR, sysvec_hyperv_callback);
545 
546 	/* Install system interrupt handler for reenlightenment notifications */
547 	if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) {
548 		sysvec_install(HYPERV_REENLIGHTENMENT_VECTOR, sysvec_hyperv_reenlightenment);
549 	}
550 
551 	/* Install system interrupt handler for stimer0 */
552 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) {
553 		sysvec_install(HYPERV_STIMER0_VECTOR, sysvec_hyperv_stimer0);
554 	}
555 
556 # ifdef CONFIG_SMP
557 	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
558 	if (hv_root_partition ||
559 	    (!ms_hyperv.paravisor_present && hv_isolation_type_snp()))
560 		smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus;
561 # endif
562 
563 	/*
564 	 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
565 	 * set x2apic destination mode to physical mode when x2apic is available
566 	 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
567 	 * have 8-bit APIC id.
568 	 */
569 # ifdef CONFIG_X86_X2APIC
570 	if (x2apic_supported())
571 		x2apic_phys = 1;
572 # endif
573 
574 	/* Register Hyper-V specific clocksource */
575 	hv_init_clocksource();
576 	hv_vtl_init_platform();
577 #endif
578 	/*
579 	 * TSC should be marked as unstable only after Hyper-V
580 	 * clocksource has been initialized. This ensures that the
581 	 * stability of the sched_clock is not altered.
582 	 */
583 	if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
584 		mark_tsc_unstable("running on Hyper-V");
585 
586 	hardlockup_detector_disable();
587 }
588 
589 static bool __init ms_hyperv_x2apic_available(void)
590 {
591 	return x2apic_supported();
592 }
593 
594 /*
595  * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping()
596  * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the
597  * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg().
598  *
599  * Note: for a VM on Hyper-V, the I/O-APIC is the only device which
600  * (logically) generates MSIs directly to the system APIC irq domain.
601  * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the
602  * pci-hyperv host bridge.
603  *
604  * Note: for a Hyper-V root partition, this will always return false.
605  * The hypervisor doesn't expose these HYPERV_CPUID_VIRT_STACK_* cpuids by
606  * default, they are implemented as intercepts by the Windows Hyper-V stack.
607  * Even a nested root partition (L2 root) will not get them because the
608  * nested (L1) hypervisor filters them out.
609  */
610 static bool __init ms_hyperv_msi_ext_dest_id(void)
611 {
612 	u32 eax;
613 
614 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE);
615 	if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE)
616 		return false;
617 
618 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES);
619 	return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE;
620 }
621 
622 #ifdef CONFIG_AMD_MEM_ENCRYPT
623 static void hv_sev_es_hcall_prepare(struct ghcb *ghcb, struct pt_regs *regs)
624 {
625 	/* RAX and CPL are already in the GHCB */
626 	ghcb_set_rcx(ghcb, regs->cx);
627 	ghcb_set_rdx(ghcb, regs->dx);
628 	ghcb_set_r8(ghcb, regs->r8);
629 }
630 
631 static bool hv_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs)
632 {
633 	/* No checking of the return state needed */
634 	return true;
635 }
636 #endif
637 
638 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = {
639 	.name			= "Microsoft Hyper-V",
640 	.detect			= ms_hyperv_platform,
641 	.type			= X86_HYPER_MS_HYPERV,
642 	.init.x2apic_available	= ms_hyperv_x2apic_available,
643 	.init.msi_ext_dest_id	= ms_hyperv_msi_ext_dest_id,
644 	.init.init_platform	= ms_hyperv_init_platform,
645 #ifdef CONFIG_AMD_MEM_ENCRYPT
646 	.runtime.sev_es_hcall_prepare = hv_sev_es_hcall_prepare,
647 	.runtime.sev_es_hcall_finish = hv_sev_es_hcall_finish,
648 #endif
649 };
650