xref: /linux/arch/x86/kernel/cpu/mshyperv.c (revision 21050f619720029e89d365b0e1328eafe088dbd8)
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/nmi.h>
32 #include <clocksource/hyperv_timer.h>
33 #include <asm/numa.h>
34 #include <asm/svm.h>
35 
36 /* Is Linux running on nested Microsoft Hypervisor */
37 bool hv_nested;
38 struct ms_hyperv_info ms_hyperv;
39 
40 /* Used in modules via hv_do_hypercall(): see arch/x86/include/asm/mshyperv.h */
41 bool hyperv_paravisor_present __ro_after_init;
42 EXPORT_SYMBOL_GPL(hyperv_paravisor_present);
43 
44 #if IS_ENABLED(CONFIG_HYPERV)
45 static inline unsigned int hv_get_nested_msr(unsigned int reg)
46 {
47 	if (hv_is_sint_msr(reg))
48 		return reg - HV_X64_MSR_SINT0 + HV_X64_MSR_NESTED_SINT0;
49 
50 	switch (reg) {
51 	case HV_X64_MSR_SIMP:
52 		return HV_X64_MSR_NESTED_SIMP;
53 	case HV_X64_MSR_SIEFP:
54 		return HV_X64_MSR_NESTED_SIEFP;
55 	case HV_X64_MSR_SVERSION:
56 		return HV_X64_MSR_NESTED_SVERSION;
57 	case HV_X64_MSR_SCONTROL:
58 		return HV_X64_MSR_NESTED_SCONTROL;
59 	case HV_X64_MSR_EOM:
60 		return HV_X64_MSR_NESTED_EOM;
61 	default:
62 		return reg;
63 	}
64 }
65 
66 u64 hv_get_non_nested_msr(unsigned int reg)
67 {
68 	u64 value;
69 
70 	if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present)
71 		hv_ivm_msr_read(reg, &value);
72 	else
73 		rdmsrl(reg, value);
74 	return value;
75 }
76 EXPORT_SYMBOL_GPL(hv_get_non_nested_msr);
77 
78 void hv_set_non_nested_msr(unsigned int reg, u64 value)
79 {
80 	if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present) {
81 		hv_ivm_msr_write(reg, value);
82 
83 		/* Write proxy bit via wrmsl instruction */
84 		if (hv_is_sint_msr(reg))
85 			wrmsrl(reg, value | 1 << 20);
86 	} else {
87 		wrmsrl(reg, value);
88 	}
89 }
90 EXPORT_SYMBOL_GPL(hv_set_non_nested_msr);
91 
92 u64 hv_get_msr(unsigned int reg)
93 {
94 	if (hv_nested)
95 		reg = hv_get_nested_msr(reg);
96 
97 	return hv_get_non_nested_msr(reg);
98 }
99 EXPORT_SYMBOL_GPL(hv_get_msr);
100 
101 void hv_set_msr(unsigned int reg, u64 value)
102 {
103 	if (hv_nested)
104 		reg = hv_get_nested_msr(reg);
105 
106 	hv_set_non_nested_msr(reg, value);
107 }
108 EXPORT_SYMBOL_GPL(hv_set_msr);
109 
110 static void (*vmbus_handler)(void);
111 static void (*hv_stimer0_handler)(void);
112 static void (*hv_kexec_handler)(void);
113 static void (*hv_crash_handler)(struct pt_regs *regs);
114 
115 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)
116 {
117 	struct pt_regs *old_regs = set_irq_regs(regs);
118 
119 	inc_irq_stat(irq_hv_callback_count);
120 	if (vmbus_handler)
121 		vmbus_handler();
122 
123 	if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
124 		apic_eoi();
125 
126 	set_irq_regs(old_regs);
127 }
128 
129 void hv_setup_vmbus_handler(void (*handler)(void))
130 {
131 	vmbus_handler = handler;
132 }
133 
134 void hv_remove_vmbus_handler(void)
135 {
136 	/* We have no way to deallocate the interrupt gate */
137 	vmbus_handler = NULL;
138 }
139 
140 /*
141  * Routines to do per-architecture handling of stimer0
142  * interrupts when in Direct Mode
143  */
144 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)
145 {
146 	struct pt_regs *old_regs = set_irq_regs(regs);
147 
148 	inc_irq_stat(hyperv_stimer0_count);
149 	if (hv_stimer0_handler)
150 		hv_stimer0_handler();
151 	add_interrupt_randomness(HYPERV_STIMER0_VECTOR);
152 	apic_eoi();
153 
154 	set_irq_regs(old_regs);
155 }
156 
157 /* For x86/x64, override weak placeholders in hyperv_timer.c */
158 void hv_setup_stimer0_handler(void (*handler)(void))
159 {
160 	hv_stimer0_handler = handler;
161 }
162 
163 void hv_remove_stimer0_handler(void)
164 {
165 	/* We have no way to deallocate the interrupt gate */
166 	hv_stimer0_handler = NULL;
167 }
168 
169 void hv_setup_kexec_handler(void (*handler)(void))
170 {
171 	hv_kexec_handler = handler;
172 }
173 
174 void hv_remove_kexec_handler(void)
175 {
176 	hv_kexec_handler = NULL;
177 }
178 
179 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs))
180 {
181 	hv_crash_handler = handler;
182 }
183 
184 void hv_remove_crash_handler(void)
185 {
186 	hv_crash_handler = NULL;
187 }
188 
189 #ifdef CONFIG_KEXEC_CORE
190 static void hv_machine_shutdown(void)
191 {
192 	if (kexec_in_progress && hv_kexec_handler)
193 		hv_kexec_handler();
194 
195 	/*
196 	 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor
197 	 * corrupts the old VP Assist Pages and can crash the kexec kernel.
198 	 */
199 	if (kexec_in_progress)
200 		cpuhp_remove_state(CPUHP_AP_HYPERV_ONLINE);
201 
202 	/* The function calls stop_other_cpus(). */
203 	native_machine_shutdown();
204 
205 	/* Disable the hypercall page when there is only 1 active CPU. */
206 	if (kexec_in_progress)
207 		hyperv_cleanup();
208 }
209 #endif /* CONFIG_KEXEC_CORE */
210 
211 #ifdef CONFIG_CRASH_DUMP
212 static void hv_machine_crash_shutdown(struct pt_regs *regs)
213 {
214 	if (hv_crash_handler)
215 		hv_crash_handler(regs);
216 
217 	/* The function calls crash_smp_send_stop(). */
218 	native_machine_crash_shutdown(regs);
219 
220 	/* Disable the hypercall page when there is only 1 active CPU. */
221 	hyperv_cleanup();
222 }
223 #endif /* CONFIG_CRASH_DUMP */
224 
225 static u64 hv_ref_counter_at_suspend;
226 static void (*old_save_sched_clock_state)(void);
227 static void (*old_restore_sched_clock_state)(void);
228 
229 /*
230  * Hyper-V clock counter resets during hibernation. Save and restore clock
231  * offset during suspend/resume, while also considering the time passed
232  * before suspend. This is to make sure that sched_clock using hv tsc page
233  * based clocksource, proceeds from where it left off during suspend and
234  * it shows correct time for the timestamps of kernel messages after resume.
235  */
236 static void save_hv_clock_tsc_state(void)
237 {
238 	hv_ref_counter_at_suspend = hv_read_reference_counter();
239 }
240 
241 static void restore_hv_clock_tsc_state(void)
242 {
243 	/*
244 	 * Adjust the offsets used by hv tsc clocksource to
245 	 * account for the time spent before hibernation.
246 	 * adjusted value = reference counter (time) at suspend
247 	 *                - reference counter (time) now.
248 	 */
249 	hv_adj_sched_clock_offset(hv_ref_counter_at_suspend - hv_read_reference_counter());
250 }
251 
252 /*
253  * Functions to override save_sched_clock_state and restore_sched_clock_state
254  * functions of x86_platform. The Hyper-V clock counter is reset during
255  * suspend-resume and the offset used to measure time needs to be
256  * corrected, post resume.
257  */
258 static void hv_save_sched_clock_state(void)
259 {
260 	old_save_sched_clock_state();
261 	save_hv_clock_tsc_state();
262 }
263 
264 static void hv_restore_sched_clock_state(void)
265 {
266 	restore_hv_clock_tsc_state();
267 	old_restore_sched_clock_state();
268 }
269 
270 static void __init x86_setup_ops_for_tsc_pg_clock(void)
271 {
272 	if (!(ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE))
273 		return;
274 
275 	old_save_sched_clock_state = x86_platform.save_sched_clock_state;
276 	x86_platform.save_sched_clock_state = hv_save_sched_clock_state;
277 
278 	old_restore_sched_clock_state = x86_platform.restore_sched_clock_state;
279 	x86_platform.restore_sched_clock_state = hv_restore_sched_clock_state;
280 }
281 #endif /* CONFIG_HYPERV */
282 
283 static uint32_t  __init ms_hyperv_platform(void)
284 {
285 	u32 eax;
286 	u32 hyp_signature[3];
287 
288 	if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
289 		return 0;
290 
291 	cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS,
292 	      &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]);
293 
294 	if (eax < HYPERV_CPUID_MIN || eax > HYPERV_CPUID_MAX ||
295 	    memcmp("Microsoft Hv", hyp_signature, 12))
296 		return 0;
297 
298 	/* HYPERCALL and VP_INDEX MSRs are mandatory for all features. */
299 	eax = cpuid_eax(HYPERV_CPUID_FEATURES);
300 	if (!(eax & HV_MSR_HYPERCALL_AVAILABLE)) {
301 		pr_warn("x86/hyperv: HYPERCALL MSR not available.\n");
302 		return 0;
303 	}
304 	if (!(eax & HV_MSR_VP_INDEX_AVAILABLE)) {
305 		pr_warn("x86/hyperv: VP_INDEX MSR not available.\n");
306 		return 0;
307 	}
308 
309 	return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
310 }
311 
312 #ifdef CONFIG_X86_LOCAL_APIC
313 /*
314  * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes
315  * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle
316  * unknown NMI on the first CPU which gets it.
317  */
318 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)
319 {
320 	static atomic_t nmi_cpu = ATOMIC_INIT(-1);
321 	unsigned int old_cpu, this_cpu;
322 
323 	if (!unknown_nmi_panic)
324 		return NMI_DONE;
325 
326 	old_cpu = -1;
327 	this_cpu = raw_smp_processor_id();
328 	if (!atomic_try_cmpxchg(&nmi_cpu, &old_cpu, this_cpu))
329 		return NMI_HANDLED;
330 
331 	return NMI_DONE;
332 }
333 #endif
334 
335 static unsigned long hv_get_tsc_khz(void)
336 {
337 	unsigned long freq;
338 
339 	rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
340 
341 	return freq / 1000;
342 }
343 
344 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
345 static void __init hv_smp_prepare_boot_cpu(void)
346 {
347 	native_smp_prepare_boot_cpu();
348 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
349 	hv_init_spinlocks();
350 #endif
351 }
352 
353 static void __init hv_smp_prepare_cpus(unsigned int max_cpus)
354 {
355 #ifdef CONFIG_X86_64
356 	int i;
357 	int ret;
358 #endif
359 
360 	native_smp_prepare_cpus(max_cpus);
361 
362 	/*
363 	 *  Override wakeup_secondary_cpu_64 callback for SEV-SNP
364 	 *  enlightened guest.
365 	 */
366 	if (!ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
367 		apic->wakeup_secondary_cpu_64 = hv_snp_boot_ap;
368 		return;
369 	}
370 
371 #ifdef CONFIG_X86_64
372 	for_each_present_cpu(i) {
373 		if (i == 0)
374 			continue;
375 		ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i));
376 		BUG_ON(ret);
377 	}
378 
379 	for_each_present_cpu(i) {
380 		if (i == 0)
381 			continue;
382 		ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i);
383 		BUG_ON(ret);
384 	}
385 #endif
386 }
387 #endif
388 
389 /*
390  * When a fully enlightened TDX VM runs on Hyper-V, the firmware sets the
391  * HW_REDUCED flag: refer to acpi_tb_create_local_fadt(). Consequently ttyS0
392  * interrupts can't work because request_irq() -> ... -> irq_to_desc() returns
393  * NULL for ttyS0. This happens because mp_config_acpi_legacy_irqs() sees a
394  * nr_legacy_irqs() of 0, so it doesn't initialize the array 'mp_irqs[]', and
395  * later setup_IO_APIC_irqs() -> find_irq_entry() fails to find the legacy irqs
396  * from the array and hence doesn't create the necessary irq description info.
397  *
398  * Clone arch/x86/kernel/acpi/boot.c: acpi_generic_reduced_hw_init() here,
399  * except don't change 'legacy_pic', which keeps its default value
400  * 'default_legacy_pic'. This way, mp_config_acpi_legacy_irqs() sees a non-zero
401  * nr_legacy_irqs() and eventually serial console interrupts works properly.
402  */
403 static void __init reduced_hw_init(void)
404 {
405 	x86_init.timers.timer_init	= x86_init_noop;
406 	x86_init.irqs.pre_vector_init	= x86_init_noop;
407 }
408 
409 int hv_get_hypervisor_version(union hv_hypervisor_version_info *info)
410 {
411 	unsigned int hv_max_functions;
412 
413 	hv_max_functions = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
414 	if (hv_max_functions < HYPERV_CPUID_VERSION) {
415 		pr_err("%s: Could not detect Hyper-V version\n", __func__);
416 		return -ENODEV;
417 	}
418 
419 	cpuid(HYPERV_CPUID_VERSION, &info->eax, &info->ebx, &info->ecx, &info->edx);
420 
421 	return 0;
422 }
423 EXPORT_SYMBOL_GPL(hv_get_hypervisor_version);
424 
425 static void __init ms_hyperv_init_platform(void)
426 {
427 	int hv_max_functions_eax;
428 
429 #ifdef CONFIG_PARAVIRT
430 	pv_info.name = "Hyper-V";
431 #endif
432 
433 	/*
434 	 * Extract the features and hints
435 	 */
436 	ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
437 	ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES);
438 	ms_hyperv.ext_features = cpuid_ecx(HYPERV_CPUID_FEATURES);
439 	ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES);
440 	ms_hyperv.hints    = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO);
441 
442 	hv_max_functions_eax = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
443 
444 	pr_info("Hyper-V: privilege flags low %#x, high %#x, ext %#x, hints %#x, misc %#x\n",
445 		ms_hyperv.features, ms_hyperv.priv_high,
446 		ms_hyperv.ext_features, ms_hyperv.hints,
447 		ms_hyperv.misc_features);
448 
449 	ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
450 	ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
451 
452 	pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
453 		 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
454 
455 	hv_identify_partition_type();
456 
457 	if (ms_hyperv.hints & HV_X64_HYPERV_NESTED) {
458 		hv_nested = true;
459 		pr_info("Hyper-V: running on a nested hypervisor\n");
460 	}
461 
462 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
463 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
464 		x86_platform.calibrate_tsc = hv_get_tsc_khz;
465 		x86_platform.calibrate_cpu = hv_get_tsc_khz;
466 		setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
467 	}
468 
469 	if (ms_hyperv.priv_high & HV_ISOLATION) {
470 		ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG);
471 		ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG);
472 
473 		if (ms_hyperv.shared_gpa_boundary_active)
474 			ms_hyperv.shared_gpa_boundary =
475 				BIT_ULL(ms_hyperv.shared_gpa_boundary_bits);
476 
477 		hyperv_paravisor_present = !!ms_hyperv.paravisor_present;
478 
479 		pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n",
480 			ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b);
481 
482 
483 		if (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP) {
484 			static_branch_enable(&isolation_type_snp);
485 		} else if (hv_get_isolation_type() == HV_ISOLATION_TYPE_TDX) {
486 			static_branch_enable(&isolation_type_tdx);
487 
488 			/* A TDX VM must use x2APIC and doesn't use lazy EOI. */
489 			ms_hyperv.hints &= ~HV_X64_APIC_ACCESS_RECOMMENDED;
490 
491 			if (!ms_hyperv.paravisor_present) {
492 				/*
493 				 * Mark the Hyper-V TSC page feature as disabled
494 				 * in a TDX VM without paravisor so that the
495 				 * Invariant TSC, which is a better clocksource
496 				 * anyway, is used instead.
497 				 */
498 				ms_hyperv.features &= ~HV_MSR_REFERENCE_TSC_AVAILABLE;
499 
500 				/*
501 				 * The Invariant TSC is expected to be available
502 				 * in a TDX VM without paravisor, but if not,
503 				 * print a warning message. The slower Hyper-V MSR-based
504 				 * Ref Counter should end up being the clocksource.
505 				 */
506 				if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
507 					pr_warn("Hyper-V: Invariant TSC is unavailable\n");
508 
509 				/* HV_MSR_CRASH_CTL is unsupported. */
510 				ms_hyperv.misc_features &= ~HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
511 
512 				/* Don't trust Hyper-V's TLB-flushing hypercalls. */
513 				ms_hyperv.hints &= ~HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
514 
515 				x86_init.acpi.reduced_hw_early_init = reduced_hw_init;
516 			}
517 		}
518 	}
519 
520 	if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) {
521 		ms_hyperv.nested_features =
522 			cpuid_eax(HYPERV_CPUID_NESTED_FEATURES);
523 		pr_info("Hyper-V: Nested features: 0x%x\n",
524 			ms_hyperv.nested_features);
525 	}
526 
527 #ifdef CONFIG_X86_LOCAL_APIC
528 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
529 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
530 		/*
531 		 * Get the APIC frequency.
532 		 */
533 		u64	hv_lapic_frequency;
534 
535 		rdmsrl(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency);
536 		hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ);
537 		lapic_timer_period = hv_lapic_frequency;
538 		pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
539 			lapic_timer_period);
540 	}
541 
542 	register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST,
543 			     "hv_nmi_unknown");
544 #endif
545 
546 #ifdef CONFIG_X86_IO_APIC
547 	no_timer_check = 1;
548 #endif
549 
550 #if IS_ENABLED(CONFIG_HYPERV)
551 #if defined(CONFIG_KEXEC_CORE)
552 	machine_ops.shutdown = hv_machine_shutdown;
553 #endif
554 #if defined(CONFIG_CRASH_DUMP)
555 	machine_ops.crash_shutdown = hv_machine_crash_shutdown;
556 #endif
557 #endif
558 	if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) {
559 		/*
560 		 * Writing to synthetic MSR 0x40000118 updates/changes the
561 		 * guest visible CPUIDs. Setting bit 0 of this MSR  enables
562 		 * guests to report invariant TSC feature through CPUID
563 		 * instruction, CPUID 0x800000007/EDX, bit 8. See code in
564 		 * early_init_intel() where this bit is examined. The
565 		 * setting of this MSR bit should happen before init_intel()
566 		 * is called.
567 		 */
568 		wrmsrl(HV_X64_MSR_TSC_INVARIANT_CONTROL, HV_EXPOSE_INVARIANT_TSC);
569 		setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
570 	}
571 
572 	/*
573 	 * Generation 2 instances don't support reading the NMI status from
574 	 * 0x61 port.
575 	 */
576 	if (efi_enabled(EFI_BOOT))
577 		x86_platform.get_nmi_reason = hv_get_nmi_reason;
578 
579 #if IS_ENABLED(CONFIG_HYPERV)
580 	if ((hv_get_isolation_type() == HV_ISOLATION_TYPE_VBS) ||
581 	    ms_hyperv.paravisor_present)
582 		hv_vtom_init();
583 	/*
584 	 * Setup the hook to get control post apic initialization.
585 	 */
586 	x86_platform.apic_post_init = hyperv_init;
587 	hyperv_setup_mmu_ops();
588 
589 	/* Install system interrupt handler for hypervisor callback */
590 	sysvec_install(HYPERVISOR_CALLBACK_VECTOR, sysvec_hyperv_callback);
591 
592 	/* Install system interrupt handler for reenlightenment notifications */
593 	if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) {
594 		sysvec_install(HYPERV_REENLIGHTENMENT_VECTOR, sysvec_hyperv_reenlightenment);
595 	}
596 
597 	/* Install system interrupt handler for stimer0 */
598 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) {
599 		sysvec_install(HYPERV_STIMER0_VECTOR, sysvec_hyperv_stimer0);
600 	}
601 
602 # ifdef CONFIG_SMP
603 	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
604 	if (hv_root_partition() ||
605 	    (!ms_hyperv.paravisor_present && hv_isolation_type_snp()))
606 		smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus;
607 # endif
608 
609 	/*
610 	 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
611 	 * set x2apic destination mode to physical mode when x2apic is available
612 	 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
613 	 * have 8-bit APIC id.
614 	 */
615 # ifdef CONFIG_X86_X2APIC
616 	if (x2apic_supported())
617 		x2apic_phys = 1;
618 # endif
619 
620 	/* Register Hyper-V specific clocksource */
621 	hv_init_clocksource();
622 	x86_setup_ops_for_tsc_pg_clock();
623 	hv_vtl_init_platform();
624 #endif
625 	/*
626 	 * TSC should be marked as unstable only after Hyper-V
627 	 * clocksource has been initialized. This ensures that the
628 	 * stability of the sched_clock is not altered.
629 	 */
630 	if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
631 		mark_tsc_unstable("running on Hyper-V");
632 
633 	hardlockup_detector_disable();
634 }
635 
636 static bool __init ms_hyperv_x2apic_available(void)
637 {
638 	return x2apic_supported();
639 }
640 
641 /*
642  * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping()
643  * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the
644  * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg().
645  *
646  * Note: for a VM on Hyper-V, the I/O-APIC is the only device which
647  * (logically) generates MSIs directly to the system APIC irq domain.
648  * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the
649  * pci-hyperv host bridge.
650  *
651  * Note: for a Hyper-V root partition, this will always return false.
652  * The hypervisor doesn't expose these HYPERV_CPUID_VIRT_STACK_* cpuids by
653  * default, they are implemented as intercepts by the Windows Hyper-V stack.
654  * Even a nested root partition (L2 root) will not get them because the
655  * nested (L1) hypervisor filters them out.
656  */
657 static bool __init ms_hyperv_msi_ext_dest_id(void)
658 {
659 	u32 eax;
660 
661 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE);
662 	if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE)
663 		return false;
664 
665 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES);
666 	return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE;
667 }
668 
669 #ifdef CONFIG_AMD_MEM_ENCRYPT
670 static void hv_sev_es_hcall_prepare(struct ghcb *ghcb, struct pt_regs *regs)
671 {
672 	/* RAX and CPL are already in the GHCB */
673 	ghcb_set_rcx(ghcb, regs->cx);
674 	ghcb_set_rdx(ghcb, regs->dx);
675 	ghcb_set_r8(ghcb, regs->r8);
676 }
677 
678 static bool hv_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs)
679 {
680 	/* No checking of the return state needed */
681 	return true;
682 }
683 #endif
684 
685 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = {
686 	.name			= "Microsoft Hyper-V",
687 	.detect			= ms_hyperv_platform,
688 	.type			= X86_HYPER_MS_HYPERV,
689 	.init.x2apic_available	= ms_hyperv_x2apic_available,
690 	.init.msi_ext_dest_id	= ms_hyperv_msi_ext_dest_id,
691 	.init.init_platform	= ms_hyperv_init_platform,
692 	.init.guest_late_init	= ms_hyperv_late_init,
693 #ifdef CONFIG_AMD_MEM_ENCRYPT
694 	.runtime.sev_es_hcall_prepare = hv_sev_es_hcall_prepare,
695 	.runtime.sev_es_hcall_finish = hv_sev_es_hcall_finish,
696 #endif
697 };
698