xref: /linux/arch/x86/kernel/cpu/mshyperv.c (revision ff5ccdb8d5bd242f1064c6f7996603e47e28d095)
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 <asm/cpuid/api.h>
23 #include <hyperv/hvhdk.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/msr.h>
33 #include <asm/nmi.h>
34 #include <clocksource/hyperv_timer.h>
35 #include <asm/numa.h>
36 #include <asm/svm.h>
37 
38 /* Is Linux running on nested Microsoft Hypervisor */
39 bool hv_nested;
40 struct ms_hyperv_info ms_hyperv;
41 
42 #if IS_ENABLED(CONFIG_HYPERV)
43 /*
44  * When running with the paravisor, controls proxying the synthetic interrupts
45  * from the host
46  */
47 static bool hv_para_sint_proxy;
48 
49 static inline unsigned int hv_get_nested_msr(unsigned int reg)
50 {
51 	if (hv_is_sint_msr(reg))
52 		return reg - HV_X64_MSR_SINT0 + HV_X64_MSR_NESTED_SINT0;
53 
54 	switch (reg) {
55 	case HV_X64_MSR_SIMP:
56 		return HV_X64_MSR_NESTED_SIMP;
57 	case HV_X64_MSR_SIEFP:
58 		return HV_X64_MSR_NESTED_SIEFP;
59 	case HV_X64_MSR_SVERSION:
60 		return HV_X64_MSR_NESTED_SVERSION;
61 	case HV_X64_MSR_SCONTROL:
62 		return HV_X64_MSR_NESTED_SCONTROL;
63 	case HV_X64_MSR_EOM:
64 		return HV_X64_MSR_NESTED_EOM;
65 	default:
66 		return reg;
67 	}
68 }
69 
70 u64 hv_get_non_nested_msr(unsigned int reg)
71 {
72 	u64 value;
73 
74 	if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present)
75 		hv_ivm_msr_read(reg, &value);
76 	else
77 		rdmsrq(reg, value);
78 	return value;
79 }
80 EXPORT_SYMBOL_GPL(hv_get_non_nested_msr);
81 
82 void hv_set_non_nested_msr(unsigned int reg, u64 value)
83 {
84 	if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present) {
85 		/* The hypervisor will get the intercept. */
86 		hv_ivm_msr_write(reg, value);
87 
88 		/* Using wrmsrq so the following goes to the paravisor. */
89 		if (hv_is_sint_msr(reg)) {
90 			union hv_synic_sint sint = { .as_uint64 = value };
91 
92 			sint.proxy = hv_para_sint_proxy;
93 			native_wrmsrq(reg, sint.as_uint64);
94 		}
95 	} else {
96 		native_wrmsrq(reg, value);
97 	}
98 }
99 EXPORT_SYMBOL_GPL(hv_set_non_nested_msr);
100 
101 /*
102  * Enable or disable proxying synthetic interrupts
103  * to the paravisor.
104  */
105 void hv_para_set_sint_proxy(bool enable)
106 {
107 	hv_para_sint_proxy = enable;
108 }
109 
110 /*
111  * Get the SynIC register value from the paravisor.
112  */
113 u64 hv_para_get_synic_register(unsigned int reg)
114 {
115 	if (WARN_ON(!ms_hyperv.paravisor_present || !hv_is_synic_msr(reg)))
116 		return ~0ULL;
117 	return native_read_msr(reg);
118 }
119 
120 /*
121  * Set the SynIC register value with the paravisor.
122  */
123 void hv_para_set_synic_register(unsigned int reg, u64 val)
124 {
125 	if (WARN_ON(!ms_hyperv.paravisor_present || !hv_is_synic_msr(reg)))
126 		return;
127 	native_write_msr(reg, val);
128 }
129 
130 u64 hv_get_msr(unsigned int reg)
131 {
132 	if (hv_nested)
133 		reg = hv_get_nested_msr(reg);
134 
135 	return hv_get_non_nested_msr(reg);
136 }
137 EXPORT_SYMBOL_GPL(hv_get_msr);
138 
139 void hv_set_msr(unsigned int reg, u64 value)
140 {
141 	if (hv_nested)
142 		reg = hv_get_nested_msr(reg);
143 
144 	hv_set_non_nested_msr(reg, value);
145 }
146 EXPORT_SYMBOL_GPL(hv_set_msr);
147 
148 static void (*mshv_handler)(void);
149 static void (*vmbus_handler)(void);
150 static void (*hv_stimer0_handler)(void);
151 static void (*hv_kexec_handler)(void);
152 static void (*hv_crash_handler)(struct pt_regs *regs);
153 
154 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)
155 {
156 	struct pt_regs *old_regs = set_irq_regs(regs);
157 
158 	inc_irq_stat(HYPERVISOR_CALLBACK);
159 	if (mshv_handler)
160 		mshv_handler();
161 
162 	if (vmbus_handler)
163 		vmbus_handler();
164 
165 	add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR);
166 
167 	if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
168 		apic_eoi();
169 
170 	set_irq_regs(old_regs);
171 }
172 
173 void hv_setup_mshv_handler(void (*handler)(void))
174 {
175 	mshv_handler = handler;
176 }
177 
178 void hv_setup_vmbus_handler(void (*handler)(void))
179 {
180 	vmbus_handler = handler;
181 }
182 
183 void hv_remove_vmbus_handler(void)
184 {
185 	/* We have no way to deallocate the interrupt gate */
186 	vmbus_handler = NULL;
187 }
188 
189 /*
190  * Routines to do per-architecture handling of stimer0
191  * interrupts when in Direct Mode
192  */
193 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)
194 {
195 	struct pt_regs *old_regs = set_irq_regs(regs);
196 
197 	inc_irq_stat(HYPERV_STIMER0);
198 	if (hv_stimer0_handler)
199 		hv_stimer0_handler();
200 	add_interrupt_randomness(HYPERV_STIMER0_VECTOR);
201 	apic_eoi();
202 
203 	set_irq_regs(old_regs);
204 }
205 
206 /* For x86/x64, override weak placeholders in hyperv_timer.c */
207 void hv_setup_stimer0_handler(void (*handler)(void))
208 {
209 	hv_stimer0_handler = handler;
210 }
211 
212 void hv_remove_stimer0_handler(void)
213 {
214 	/* We have no way to deallocate the interrupt gate */
215 	hv_stimer0_handler = NULL;
216 }
217 
218 void hv_setup_kexec_handler(void (*handler)(void))
219 {
220 	hv_kexec_handler = handler;
221 }
222 
223 void hv_remove_kexec_handler(void)
224 {
225 	hv_kexec_handler = NULL;
226 }
227 
228 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs))
229 {
230 	hv_crash_handler = handler;
231 }
232 
233 void hv_remove_crash_handler(void)
234 {
235 	hv_crash_handler = NULL;
236 }
237 
238 #ifdef CONFIG_KEXEC_CORE
239 static void hv_machine_shutdown(void)
240 {
241 	if (kexec_in_progress) {
242 		hv_stimer_global_cleanup();
243 
244 		if (hv_kexec_handler)
245 			hv_kexec_handler();
246 	}
247 
248 	/*
249 	 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor
250 	 * corrupts the old VP Assist Pages and can crash the kexec kernel.
251 	 */
252 	if (kexec_in_progress)
253 		cpuhp_remove_state(CPUHP_AP_HYPERV_ONLINE);
254 
255 	/* The function calls stop_other_cpus(). */
256 	native_machine_shutdown();
257 
258 	/* Disable the hypercall page when there is only 1 active CPU. */
259 	if (kexec_in_progress)
260 		hyperv_cleanup();
261 }
262 #endif /* CONFIG_KEXEC_CORE */
263 
264 #ifdef CONFIG_CRASH_DUMP
265 static void hv_guest_crash_shutdown(struct pt_regs *regs)
266 {
267 	if (hv_crash_handler)
268 		hv_crash_handler(regs);
269 
270 	/* The function calls crash_smp_send_stop(). */
271 	native_machine_crash_shutdown(regs);
272 
273 	/* Disable the hypercall page when there is only 1 active CPU. */
274 	hyperv_cleanup();
275 }
276 #endif /* CONFIG_CRASH_DUMP */
277 
278 static u64 hv_ref_counter_at_suspend;
279 static void (*old_save_sched_clock_state)(void);
280 static void (*old_restore_sched_clock_state)(void);
281 
282 /*
283  * Hyper-V clock counter resets during hibernation. Save and restore clock
284  * offset during suspend/resume, while also considering the time passed
285  * before suspend. This is to make sure that sched_clock using hv tsc page
286  * based clocksource, proceeds from where it left off during suspend and
287  * it shows correct time for the timestamps of kernel messages after resume.
288  */
289 static void save_hv_clock_tsc_state(void)
290 {
291 	hv_ref_counter_at_suspend = hv_read_reference_counter();
292 }
293 
294 static void restore_hv_clock_tsc_state(void)
295 {
296 	/*
297 	 * Adjust the offsets used by hv tsc clocksource to
298 	 * account for the time spent before hibernation.
299 	 * adjusted value = reference counter (time) at suspend
300 	 *                - reference counter (time) now.
301 	 */
302 	hv_adj_sched_clock_offset(hv_ref_counter_at_suspend - hv_read_reference_counter());
303 }
304 
305 /*
306  * Functions to override save_sched_clock_state and restore_sched_clock_state
307  * functions of x86_platform. The Hyper-V clock counter is reset during
308  * suspend-resume and the offset used to measure time needs to be
309  * corrected, post resume.
310  */
311 static void hv_save_sched_clock_state(void)
312 {
313 	old_save_sched_clock_state();
314 	save_hv_clock_tsc_state();
315 }
316 
317 static void hv_restore_sched_clock_state(void)
318 {
319 	restore_hv_clock_tsc_state();
320 	old_restore_sched_clock_state();
321 }
322 
323 static void __init x86_setup_ops_for_tsc_pg_clock(void)
324 {
325 	if (!(ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE))
326 		return;
327 
328 	old_save_sched_clock_state = x86_platform.save_sched_clock_state;
329 	x86_platform.save_sched_clock_state = hv_save_sched_clock_state;
330 
331 	old_restore_sched_clock_state = x86_platform.restore_sched_clock_state;
332 	x86_platform.restore_sched_clock_state = hv_restore_sched_clock_state;
333 }
334 
335 #ifdef CONFIG_X86_64
336 DEFINE_STATIC_CALL(hv_hypercall, hv_std_hypercall);
337 EXPORT_STATIC_CALL_TRAMP_GPL(hv_hypercall);
338 #define hypercall_update(hc) static_call_update(hv_hypercall, hc)
339 #endif
340 #endif /* CONFIG_HYPERV */
341 
342 #ifndef hypercall_update
343 #define hypercall_update(hc) (void)hc
344 #endif
345 
346 static uint32_t  __init ms_hyperv_platform(void)
347 {
348 	u32 eax;
349 	u32 hyp_signature[3];
350 
351 	if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
352 		return 0;
353 
354 	cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS,
355 	      &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]);
356 
357 	if (eax < HYPERV_CPUID_MIN || eax > HYPERV_CPUID_MAX ||
358 	    memcmp("Microsoft Hv", hyp_signature, 12))
359 		return 0;
360 
361 	/* HYPERCALL and VP_INDEX MSRs are mandatory for all features. */
362 	eax = cpuid_eax(HYPERV_CPUID_FEATURES);
363 	if (!(eax & HV_MSR_HYPERCALL_AVAILABLE)) {
364 		pr_warn("x86/hyperv: HYPERCALL MSR not available.\n");
365 		return 0;
366 	}
367 	if (!(eax & HV_MSR_VP_INDEX_AVAILABLE)) {
368 		pr_warn("x86/hyperv: VP_INDEX MSR not available.\n");
369 		return 0;
370 	}
371 
372 	return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
373 }
374 
375 #ifdef CONFIG_X86_LOCAL_APIC
376 /*
377  * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes
378  * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle
379  * unknown NMI on the first CPU which gets it.
380  */
381 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)
382 {
383 	static atomic_t nmi_cpu = ATOMIC_INIT(-1);
384 	unsigned int old_cpu, this_cpu;
385 
386 	if (!unknown_nmi_panic)
387 		return NMI_DONE;
388 
389 	old_cpu = -1;
390 	this_cpu = raw_smp_processor_id();
391 	if (!atomic_try_cmpxchg(&nmi_cpu, &old_cpu, this_cpu))
392 		return NMI_HANDLED;
393 
394 	return NMI_DONE;
395 }
396 #endif
397 
398 static unsigned long hv_get_tsc_khz(void)
399 {
400 	unsigned long freq;
401 
402 	rdmsrq(HV_X64_MSR_TSC_FREQUENCY, freq);
403 
404 	return freq / 1000;
405 }
406 
407 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
408 static void __init hv_smp_prepare_boot_cpu(void)
409 {
410 	native_smp_prepare_boot_cpu();
411 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
412 	hv_init_spinlocks();
413 #endif
414 }
415 
416 static void __init hv_smp_prepare_cpus(unsigned int max_cpus)
417 {
418 #ifdef CONFIG_X86_64
419 	int i;
420 	int ret;
421 #endif
422 
423 	native_smp_prepare_cpus(max_cpus);
424 
425 	/*
426 	 *  Override wakeup_secondary_cpu_64 callback for SEV-SNP
427 	 *  enlightened guest.
428 	 */
429 	if (!ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
430 		apic->wakeup_secondary_cpu_64 = hv_snp_boot_ap;
431 		return;
432 	}
433 
434 #ifdef CONFIG_X86_64
435 	/* If AP LPs exist, we are in a kexec'd kernel and VPs already exist */
436 	if (num_present_cpus() == 1 || hv_lp_exists(1))
437 		return;
438 
439 	for_each_present_cpu(i) {
440 		if (i == 0)
441 			continue;
442 		ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i));
443 		BUG_ON(ret);
444 	}
445 
446 	ret = hv_call_notify_all_processors_started();
447 	WARN_ON(ret);
448 
449 	for_each_present_cpu(i) {
450 		if (i == 0)
451 			continue;
452 		ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i);
453 		BUG_ON(ret);
454 	}
455 #endif
456 }
457 #endif
458 
459 /*
460  * When a fully enlightened TDX VM runs on Hyper-V, the firmware sets the
461  * HW_REDUCED flag: refer to acpi_tb_create_local_fadt(). Consequently ttyS0
462  * interrupts can't work because request_irq() -> ... -> irq_to_desc() returns
463  * NULL for ttyS0. This happens because mp_config_acpi_legacy_irqs() sees a
464  * nr_legacy_irqs() of 0, so it doesn't initialize the array 'mp_irqs[]', and
465  * later setup_IO_APIC_irqs() -> find_irq_entry() fails to find the legacy irqs
466  * from the array and hence doesn't create the necessary irq description info.
467  *
468  * Clone arch/x86/kernel/acpi/boot.c: acpi_generic_reduced_hw_init() here,
469  * except don't change 'legacy_pic', which keeps its default value
470  * 'default_legacy_pic'. This way, mp_config_acpi_legacy_irqs() sees a non-zero
471  * nr_legacy_irqs() and eventually serial console interrupts works properly.
472  */
473 static void __init reduced_hw_init(void)
474 {
475 	x86_init.timers.timer_init	= x86_init_noop;
476 	x86_init.irqs.pre_vector_init	= x86_init_noop;
477 }
478 
479 int hv_get_hypervisor_version(union hv_hypervisor_version_info *info)
480 {
481 	unsigned int hv_max_functions;
482 
483 	hv_max_functions = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
484 	if (hv_max_functions < HYPERV_CPUID_VERSION) {
485 		pr_err("%s: Could not detect Hyper-V version\n", __func__);
486 		return -ENODEV;
487 	}
488 
489 	cpuid(HYPERV_CPUID_VERSION, &info->eax, &info->ebx, &info->ecx, &info->edx);
490 
491 	return 0;
492 }
493 EXPORT_SYMBOL_GPL(hv_get_hypervisor_version);
494 
495 /*
496  * Reserved vectors hard coded in the hypervisor. If used outside, the hypervisor
497  * will either crash or hang or attempt to break into debugger.
498  */
499 static void hv_reserve_irq_vectors(void)
500 {
501 	#define HYPERV_DBG_FASTFAIL_VECTOR	0x29
502 	#define HYPERV_DBG_ASSERT_VECTOR	0x2C
503 	#define HYPERV_DBG_SERVICE_VECTOR	0x2D
504 
505 	if (cpu_feature_enabled(X86_FEATURE_FRED))
506 		return;
507 
508 	if (test_and_set_bit(HYPERV_DBG_ASSERT_VECTOR, system_vectors) ||
509 	    test_and_set_bit(HYPERV_DBG_SERVICE_VECTOR, system_vectors) ||
510 	    test_and_set_bit(HYPERV_DBG_FASTFAIL_VECTOR, system_vectors))
511 		BUG();
512 
513 	pr_info("Hyper-V: reserve vectors: 0x%x 0x%x 0x%x\n",
514 		HYPERV_DBG_ASSERT_VECTOR, HYPERV_DBG_SERVICE_VECTOR,
515 		HYPERV_DBG_FASTFAIL_VECTOR);
516 }
517 
518 static void __init ms_hyperv_init_platform(void)
519 {
520 	int hv_max_functions_eax, eax;
521 
522 #ifdef CONFIG_PARAVIRT
523 	pv_info.name = "Hyper-V";
524 #endif
525 
526 	/*
527 	 * Extract the features and hints
528 	 */
529 	ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
530 	ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES);
531 	ms_hyperv.ext_features = cpuid_ecx(HYPERV_CPUID_FEATURES);
532 	ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES);
533 	ms_hyperv.hints    = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO);
534 
535 	hv_max_functions_eax = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
536 
537 	pr_info("Hyper-V: privilege flags low %#x, high %#x, ext %#x, hints %#x, misc %#x\n",
538 		ms_hyperv.features, ms_hyperv.priv_high,
539 		ms_hyperv.ext_features, ms_hyperv.hints,
540 		ms_hyperv.misc_features);
541 
542 	ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
543 	ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
544 
545 	pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
546 		 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
547 
548 	hv_identify_partition_type();
549 
550 	if (hv_root_partition())
551 		hv_reserve_irq_vectors();
552 
553 	if (cc_platform_has(CC_ATTR_SNP_SECURE_AVIC))
554 		ms_hyperv.hints |= HV_DEPRECATING_AEOI_RECOMMENDED;
555 
556 	if (ms_hyperv.hints & HV_X64_HYPERV_NESTED) {
557 		hv_nested = true;
558 		pr_info("Hyper-V: running on a nested hypervisor\n");
559 	}
560 
561 	/*
562 	 * There is no check against the max function for HYPERV_CPUID_VIRT_STACK_* CPUID
563 	 * leaves as the hypervisor doesn't handle them. Even a nested root partition (L2
564 	 * root) will not get them because the nested (L1) hypervisor filters them out.
565 	 * These are handled through intercept processing by the Windows Hyper-V stack
566 	 * or the paravisor.
567 	 */
568 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES);
569 	ms_hyperv.confidential_vmbus_available =
570 		eax & HYPERV_VS_PROPERTIES_EAX_CONFIDENTIAL_VMBUS_AVAILABLE;
571 	ms_hyperv.msi_ext_dest_id =
572 		eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE;
573 
574 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
575 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
576 		x86_platform.calibrate_tsc = hv_get_tsc_khz;
577 		x86_platform.calibrate_cpu = hv_get_tsc_khz;
578 		setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
579 	}
580 
581 	if (ms_hyperv.priv_high & HV_ISOLATION) {
582 		ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG);
583 		ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG);
584 
585 		if (ms_hyperv.shared_gpa_boundary_active)
586 			ms_hyperv.shared_gpa_boundary =
587 				BIT_ULL(ms_hyperv.shared_gpa_boundary_bits);
588 
589 		pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n",
590 			ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b);
591 
592 
593 		if (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP) {
594 			static_branch_enable(&isolation_type_snp);
595 			if (!ms_hyperv.paravisor_present)
596 				hypercall_update(hv_snp_hypercall);
597 		} else if (hv_get_isolation_type() == HV_ISOLATION_TYPE_TDX) {
598 			static_branch_enable(&isolation_type_tdx);
599 
600 			/* A TDX VM must use x2APIC and doesn't use lazy EOI. */
601 			ms_hyperv.hints &= ~HV_X64_APIC_ACCESS_RECOMMENDED;
602 
603 			if (!ms_hyperv.paravisor_present) {
604 				hypercall_update(hv_tdx_hypercall);
605 				/*
606 				 * Mark the Hyper-V TSC page feature as disabled
607 				 * in a TDX VM without paravisor so that the
608 				 * Invariant TSC, which is a better clocksource
609 				 * anyway, is used instead.
610 				 */
611 				ms_hyperv.features &= ~HV_MSR_REFERENCE_TSC_AVAILABLE;
612 
613 				/*
614 				 * The Invariant TSC is expected to be available
615 				 * in a TDX VM without paravisor, but if not,
616 				 * print a warning message. The slower Hyper-V MSR-based
617 				 * Ref Counter should end up being the clocksource.
618 				 */
619 				if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
620 					pr_warn("Hyper-V: Invariant TSC is unavailable\n");
621 
622 				/* HV_MSR_CRASH_CTL is unsupported. */
623 				ms_hyperv.misc_features &= ~HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
624 
625 				/* Don't trust Hyper-V's TLB-flushing hypercalls. */
626 				ms_hyperv.hints &= ~HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
627 
628 				x86_init.acpi.reduced_hw_early_init = reduced_hw_init;
629 			}
630 		}
631 	}
632 
633 	if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) {
634 		ms_hyperv.nested_features =
635 			cpuid_eax(HYPERV_CPUID_NESTED_FEATURES);
636 		pr_info("Hyper-V: Nested features: 0x%x\n",
637 			ms_hyperv.nested_features);
638 	}
639 
640 #ifdef CONFIG_X86_LOCAL_APIC
641 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
642 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
643 		/*
644 		 * Get the APIC frequency.
645 		 */
646 		u64	hv_lapic_frequency;
647 
648 		rdmsrq(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency);
649 		hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ);
650 		lapic_timer_period = hv_lapic_frequency;
651 		pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
652 			lapic_timer_period);
653 	}
654 
655 	register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST,
656 			     "hv_nmi_unknown");
657 #endif
658 
659 #ifdef CONFIG_X86_IO_APIC
660 	no_timer_check = 1;
661 #endif
662 
663 #if IS_ENABLED(CONFIG_HYPERV)
664 	if (hv_root_partition())
665 		machine_ops.power_off = hv_machine_power_off;
666 #if defined(CONFIG_KEXEC_CORE)
667 	machine_ops.shutdown = hv_machine_shutdown;
668 #endif
669 #if defined(CONFIG_CRASH_DUMP)
670 	if (!hv_root_partition())
671 		machine_ops.crash_shutdown = hv_guest_crash_shutdown;
672 #endif
673 #endif
674 	/*
675 	 * HV_ACCESS_TSC_INVARIANT is always zero for the root partition. Root
676 	 * partition doesn't need to write to synthetic MSR to enable invariant
677 	 * TSC feature. It sees what the hardware provides.
678 	 */
679 	if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) {
680 		/*
681 		 * Writing to synthetic MSR 0x40000118 updates/changes the
682 		 * guest visible CPUIDs. Setting bit 0 of this MSR  enables
683 		 * guests to report invariant TSC feature through CPUID
684 		 * instruction, CPUID 0x800000007/EDX, bit 8. See code in
685 		 * early_init_intel() where this bit is examined. The
686 		 * setting of this MSR bit should happen before init_intel()
687 		 * is called.
688 		 */
689 		wrmsrq(HV_X64_MSR_TSC_INVARIANT_CONTROL, HV_EXPOSE_INVARIANT_TSC);
690 		setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
691 	}
692 
693 	/*
694 	 * Generation 2 instances don't support reading the NMI status from
695 	 * 0x61 port.
696 	 */
697 	if (efi_enabled(EFI_BOOT))
698 		x86_platform.get_nmi_reason = hv_get_nmi_reason;
699 
700 #if IS_ENABLED(CONFIG_HYPERV)
701 	if ((hv_get_isolation_type() == HV_ISOLATION_TYPE_VBS) ||
702 	    ms_hyperv.paravisor_present)
703 		hv_vtom_init();
704 	/*
705 	 * Setup the hook to get control post apic initialization.
706 	 */
707 	x86_platform.apic_post_init = hyperv_init;
708 	hyperv_setup_mmu_ops();
709 
710 	/* Install system interrupt handler for hypervisor callback */
711 	sysvec_install(HYPERVISOR_CALLBACK_VECTOR, sysvec_hyperv_callback);
712 
713 	/* Install system interrupt handler for reenlightenment notifications */
714 	if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) {
715 		sysvec_install(HYPERV_REENLIGHTENMENT_VECTOR, sysvec_hyperv_reenlightenment);
716 	}
717 
718 	/* Install system interrupt handler for stimer0 */
719 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) {
720 		sysvec_install(HYPERV_STIMER0_VECTOR, sysvec_hyperv_stimer0);
721 	}
722 
723 # ifdef CONFIG_SMP
724 	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
725 	if (hv_root_partition() ||
726 	    (!ms_hyperv.paravisor_present && hv_isolation_type_snp()))
727 		smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus;
728 # endif
729 
730 	/*
731 	 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
732 	 * set x2apic destination mode to physical mode when x2apic is available
733 	 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
734 	 * have 8-bit APIC id.
735 	 */
736 # ifdef CONFIG_X86_X2APIC
737 	if (x2apic_supported())
738 		x2apic_phys = 1;
739 # endif
740 
741 	/* Register Hyper-V specific clocksource */
742 	hv_init_clocksource();
743 	x86_setup_ops_for_tsc_pg_clock();
744 	hv_vtl_init_platform();
745 #endif
746 	/*
747 	 * TSC should be marked as unstable only after Hyper-V
748 	 * clocksource has been initialized. This ensures that the
749 	 * stability of the sched_clock is not altered.
750 	 *
751 	 * HV_ACCESS_TSC_INVARIANT is always zero for the root partition. No
752 	 * need to check for it.
753 	 */
754 	if (!hv_root_partition() &&
755 	    !(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
756 		mark_tsc_unstable("running on Hyper-V");
757 
758 	hardlockup_detector_disable();
759 }
760 
761 static bool __init ms_hyperv_x2apic_available(void)
762 {
763 	return x2apic_supported();
764 }
765 
766 /*
767  * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping()
768  * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the
769  * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg().
770  *
771  * Note: for a VM on Hyper-V, the I/O-APIC is the only device which
772  * (logically) generates MSIs directly to the system APIC irq domain.
773  * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the
774  * pci-hyperv host bridge.
775  *
776  * Note: for a Hyper-V root partition, this will always return false.
777  */
778 static bool __init ms_hyperv_msi_ext_dest_id(void)
779 {
780 	return ms_hyperv.msi_ext_dest_id;
781 }
782 
783 #ifdef CONFIG_AMD_MEM_ENCRYPT
784 static void hv_sev_es_hcall_prepare(struct ghcb *ghcb, struct pt_regs *regs)
785 {
786 	/* RAX and CPL are already in the GHCB */
787 	ghcb_set_rcx(ghcb, regs->cx);
788 	ghcb_set_rdx(ghcb, regs->dx);
789 	ghcb_set_r8(ghcb, regs->r8);
790 }
791 
792 static bool hv_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs)
793 {
794 	/* No checking of the return state needed */
795 	return true;
796 }
797 #endif
798 
799 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = {
800 	.name			= "Microsoft Hyper-V",
801 	.detect			= ms_hyperv_platform,
802 	.type			= X86_HYPER_MS_HYPERV,
803 	.init.x2apic_available	= ms_hyperv_x2apic_available,
804 	.init.msi_ext_dest_id	= ms_hyperv_msi_ext_dest_id,
805 	.init.init_platform	= ms_hyperv_init_platform,
806 	.init.guest_late_init	= ms_hyperv_late_init,
807 #ifdef CONFIG_AMD_MEM_ENCRYPT
808 	.runtime.sev_es_hcall_prepare = hv_sev_es_hcall_prepare,
809 	.runtime.sev_es_hcall_finish = hv_sev_es_hcall_finish,
810 #endif
811 };
812