xref: /linux/arch/x86/kernel/cpu/mshyperv.c (revision 36f353a1ebf88280f58d1ebfe2731251d9159456)
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 #endif /* CONFIG_KEXEC_CORE */
213 
214 #ifdef CONFIG_CRASH_DUMP
215 static void hv_machine_crash_shutdown(struct pt_regs *regs)
216 {
217 	if (hv_crash_handler)
218 		hv_crash_handler(regs);
219 
220 	/* The function calls crash_smp_send_stop(). */
221 	native_machine_crash_shutdown(regs);
222 
223 	/* Disable the hypercall page when there is only 1 active CPU. */
224 	hyperv_cleanup();
225 }
226 #endif /* CONFIG_CRASH_DUMP */
227 #endif /* CONFIG_HYPERV */
228 
229 static uint32_t  __init ms_hyperv_platform(void)
230 {
231 	u32 eax;
232 	u32 hyp_signature[3];
233 
234 	if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
235 		return 0;
236 
237 	cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS,
238 	      &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]);
239 
240 	if (eax < HYPERV_CPUID_MIN || eax > HYPERV_CPUID_MAX ||
241 	    memcmp("Microsoft Hv", hyp_signature, 12))
242 		return 0;
243 
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");
248 		return 0;
249 	}
250 	if (!(eax & HV_MSR_VP_INDEX_AVAILABLE)) {
251 		pr_warn("x86/hyperv: VP_INDEX MSR not available.\n");
252 		return 0;
253 	}
254 
255 	return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
256 }
257 
258 #ifdef CONFIG_X86_LOCAL_APIC
259 /*
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.
263  */
264 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)
265 {
266 	static atomic_t nmi_cpu = ATOMIC_INIT(-1);
267 	unsigned int old_cpu, this_cpu;
268 
269 	if (!unknown_nmi_panic)
270 		return NMI_DONE;
271 
272 	old_cpu = -1;
273 	this_cpu = raw_smp_processor_id();
274 	if (!atomic_try_cmpxchg(&nmi_cpu, &old_cpu, this_cpu))
275 		return NMI_HANDLED;
276 
277 	return NMI_DONE;
278 }
279 #endif
280 
281 static unsigned long hv_get_tsc_khz(void)
282 {
283 	unsigned long freq;
284 
285 	rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
286 
287 	return freq / 1000;
288 }
289 
290 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
291 static void __init hv_smp_prepare_boot_cpu(void)
292 {
293 	native_smp_prepare_boot_cpu();
294 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
295 	hv_init_spinlocks();
296 #endif
297 }
298 
299 static void __init hv_smp_prepare_cpus(unsigned int max_cpus)
300 {
301 #ifdef CONFIG_X86_64
302 	int i;
303 	int ret;
304 #endif
305 
306 	native_smp_prepare_cpus(max_cpus);
307 
308 	/*
309 	 *  Override wakeup_secondary_cpu_64 callback for SEV-SNP
310 	 *  enlightened guest.
311 	 */
312 	if (!ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
313 		apic->wakeup_secondary_cpu_64 = hv_snp_boot_ap;
314 		return;
315 	}
316 
317 #ifdef CONFIG_X86_64
318 	for_each_present_cpu(i) {
319 		if (i == 0)
320 			continue;
321 		ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i));
322 		BUG_ON(ret);
323 	}
324 
325 	for_each_present_cpu(i) {
326 		if (i == 0)
327 			continue;
328 		ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i);
329 		BUG_ON(ret);
330 	}
331 #endif
332 }
333 #endif
334 
335 /*
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.
343  *
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.
348  */
349 static void __init reduced_hw_init(void)
350 {
351 	x86_init.timers.timer_init	= x86_init_noop;
352 	x86_init.irqs.pre_vector_init	= x86_init_noop;
353 }
354 
355 static void __init ms_hyperv_init_platform(void)
356 {
357 	int hv_max_functions_eax;
358 	int hv_host_info_eax;
359 	int hv_host_info_ebx;
360 	int hv_host_info_ecx;
361 	int hv_host_info_edx;
362 
363 #ifdef CONFIG_PARAVIRT
364 	pv_info.name = "Hyper-V";
365 #endif
366 
367 	/*
368 	 * Extract the features and hints
369 	 */
370 	ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
371 	ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES);
372 	ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES);
373 	ms_hyperv.hints    = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO);
374 
375 	hv_max_functions_eax = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
376 
377 	pr_info("Hyper-V: privilege flags low 0x%x, high 0x%x, hints 0x%x, misc 0x%x\n",
378 		ms_hyperv.features, ms_hyperv.priv_high, ms_hyperv.hints,
379 		ms_hyperv.misc_features);
380 
381 	ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
382 	ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
383 
384 	pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
385 		 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
386 
387 	/*
388 	 * Check CPU management privilege.
389 	 *
390 	 * To mirror what Windows does we should extract CPU management
391 	 * features and use the ReservedIdentityBit to detect if Linux is the
392 	 * root partition. But that requires negotiating CPU management
393 	 * interface (a process to be finalized). For now, use the privilege
394 	 * flag as the indicator for running as root.
395 	 *
396 	 * Hyper-V should never specify running as root and as a Confidential
397 	 * VM. But to protect against a compromised/malicious Hyper-V trying
398 	 * to exploit root behavior to expose Confidential VM memory, ignore
399 	 * the root partition setting if also a Confidential VM.
400 	 */
401 	if ((ms_hyperv.priv_high & HV_CPU_MANAGEMENT) &&
402 	    !(ms_hyperv.priv_high & HV_ISOLATION)) {
403 		hv_root_partition = true;
404 		pr_info("Hyper-V: running as root partition\n");
405 	}
406 
407 	if (ms_hyperv.hints & HV_X64_HYPERV_NESTED) {
408 		hv_nested = true;
409 		pr_info("Hyper-V: running on a nested hypervisor\n");
410 	}
411 
412 	/*
413 	 * Extract host information.
414 	 */
415 	if (hv_max_functions_eax >= HYPERV_CPUID_VERSION) {
416 		hv_host_info_eax = cpuid_eax(HYPERV_CPUID_VERSION);
417 		hv_host_info_ebx = cpuid_ebx(HYPERV_CPUID_VERSION);
418 		hv_host_info_ecx = cpuid_ecx(HYPERV_CPUID_VERSION);
419 		hv_host_info_edx = cpuid_edx(HYPERV_CPUID_VERSION);
420 
421 		pr_info("Hyper-V: Host Build %d.%d.%d.%d-%d-%d\n",
422 			hv_host_info_ebx >> 16, hv_host_info_ebx & 0xFFFF,
423 			hv_host_info_eax, hv_host_info_edx & 0xFFFFFF,
424 			hv_host_info_ecx, hv_host_info_edx >> 24);
425 	}
426 
427 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
428 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
429 		x86_platform.calibrate_tsc = hv_get_tsc_khz;
430 		x86_platform.calibrate_cpu = hv_get_tsc_khz;
431 	}
432 
433 	if (ms_hyperv.priv_high & HV_ISOLATION) {
434 		ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG);
435 		ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG);
436 
437 		if (ms_hyperv.shared_gpa_boundary_active)
438 			ms_hyperv.shared_gpa_boundary =
439 				BIT_ULL(ms_hyperv.shared_gpa_boundary_bits);
440 
441 		hyperv_paravisor_present = !!ms_hyperv.paravisor_present;
442 
443 		pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n",
444 			ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b);
445 
446 
447 		if (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP) {
448 			static_branch_enable(&isolation_type_snp);
449 		} else if (hv_get_isolation_type() == HV_ISOLATION_TYPE_TDX) {
450 			static_branch_enable(&isolation_type_tdx);
451 
452 			/* A TDX VM must use x2APIC and doesn't use lazy EOI. */
453 			ms_hyperv.hints &= ~HV_X64_APIC_ACCESS_RECOMMENDED;
454 
455 			if (!ms_hyperv.paravisor_present) {
456 				/* To be supported: more work is required.  */
457 				ms_hyperv.features &= ~HV_MSR_REFERENCE_TSC_AVAILABLE;
458 
459 				/* HV_REGISTER_CRASH_CTL is unsupported. */
460 				ms_hyperv.misc_features &= ~HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
461 
462 				/* Don't trust Hyper-V's TLB-flushing hypercalls. */
463 				ms_hyperv.hints &= ~HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
464 
465 				x86_init.acpi.reduced_hw_early_init = reduced_hw_init;
466 			}
467 		}
468 	}
469 
470 	if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) {
471 		ms_hyperv.nested_features =
472 			cpuid_eax(HYPERV_CPUID_NESTED_FEATURES);
473 		pr_info("Hyper-V: Nested features: 0x%x\n",
474 			ms_hyperv.nested_features);
475 	}
476 
477 #ifdef CONFIG_X86_LOCAL_APIC
478 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
479 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
480 		/*
481 		 * Get the APIC frequency.
482 		 */
483 		u64	hv_lapic_frequency;
484 
485 		rdmsrl(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency);
486 		hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ);
487 		lapic_timer_period = hv_lapic_frequency;
488 		pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
489 			lapic_timer_period);
490 	}
491 
492 	register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST,
493 			     "hv_nmi_unknown");
494 #endif
495 
496 #ifdef CONFIG_X86_IO_APIC
497 	no_timer_check = 1;
498 #endif
499 
500 #if IS_ENABLED(CONFIG_HYPERV)
501 #if defined(CONFIG_KEXEC_CORE)
502 	machine_ops.shutdown = hv_machine_shutdown;
503 #endif
504 #if defined(CONFIG_CRASH_DUMP)
505 	machine_ops.crash_shutdown = hv_machine_crash_shutdown;
506 #endif
507 #endif
508 	if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) {
509 		/*
510 		 * Writing to synthetic MSR 0x40000118 updates/changes the
511 		 * guest visible CPUIDs. Setting bit 0 of this MSR  enables
512 		 * guests to report invariant TSC feature through CPUID
513 		 * instruction, CPUID 0x800000007/EDX, bit 8. See code in
514 		 * early_init_intel() where this bit is examined. The
515 		 * setting of this MSR bit should happen before init_intel()
516 		 * is called.
517 		 */
518 		wrmsrl(HV_X64_MSR_TSC_INVARIANT_CONTROL, HV_EXPOSE_INVARIANT_TSC);
519 		setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
520 	}
521 
522 	/*
523 	 * Generation 2 instances don't support reading the NMI status from
524 	 * 0x61 port.
525 	 */
526 	if (efi_enabled(EFI_BOOT))
527 		x86_platform.get_nmi_reason = hv_get_nmi_reason;
528 
529 	/*
530 	 * Hyper-V VMs have a PIT emulation quirk such that zeroing the
531 	 * counter register during PIT shutdown restarts the PIT. So it
532 	 * continues to interrupt @18.2 HZ. Setting i8253_clear_counter
533 	 * to false tells pit_shutdown() not to zero the counter so that
534 	 * the PIT really is shutdown. Generation 2 VMs don't have a PIT,
535 	 * and setting this value has no effect.
536 	 */
537 	i8253_clear_counter_on_shutdown = false;
538 
539 #if IS_ENABLED(CONFIG_HYPERV)
540 	if ((hv_get_isolation_type() == HV_ISOLATION_TYPE_VBS) ||
541 	    ms_hyperv.paravisor_present)
542 		hv_vtom_init();
543 	/*
544 	 * Setup the hook to get control post apic initialization.
545 	 */
546 	x86_platform.apic_post_init = hyperv_init;
547 	hyperv_setup_mmu_ops();
548 
549 	/* Install system interrupt handler for hypervisor callback */
550 	sysvec_install(HYPERVISOR_CALLBACK_VECTOR, sysvec_hyperv_callback);
551 
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);
555 	}
556 
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);
560 	}
561 
562 # ifdef CONFIG_SMP
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;
567 # endif
568 
569 	/*
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.
574 	 */
575 # ifdef CONFIG_X86_X2APIC
576 	if (x2apic_supported())
577 		x2apic_phys = 1;
578 # endif
579 
580 	/* Register Hyper-V specific clocksource */
581 	hv_init_clocksource();
582 	hv_vtl_init_platform();
583 #endif
584 	/*
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.
588 	 */
589 	if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
590 		mark_tsc_unstable("running on Hyper-V");
591 
592 	hardlockup_detector_disable();
593 }
594 
595 static bool __init ms_hyperv_x2apic_available(void)
596 {
597 	return x2apic_supported();
598 }
599 
600 /*
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().
604  *
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.
609  *
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.
615  */
616 static bool __init ms_hyperv_msi_ext_dest_id(void)
617 {
618 	u32 eax;
619 
620 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE);
621 	if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE)
622 		return false;
623 
624 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES);
625 	return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE;
626 }
627 
628 #ifdef CONFIG_AMD_MEM_ENCRYPT
629 static void hv_sev_es_hcall_prepare(struct ghcb *ghcb, struct pt_regs *regs)
630 {
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);
635 }
636 
637 static bool hv_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs)
638 {
639 	/* No checking of the return state needed */
640 	return true;
641 }
642 #endif
643 
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 #ifdef CONFIG_AMD_MEM_ENCRYPT
652 	.runtime.sev_es_hcall_prepare = hv_sev_es_hcall_prepare,
653 	.runtime.sev_es_hcall_finish = hv_sev_es_hcall_finish,
654 #endif
655 };
656