xref: /linux/arch/x86/kernel/cpu/mshyperv.c (revision ab93e0dd72c37d378dd936f031ffb83ff2bd87ce)
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/msr.h>
34 #include <asm/numa.h>
35 #include <asm/svm.h>
36 
37 /* Is Linux running on nested Microsoft Hypervisor */
38 bool hv_nested;
39 struct ms_hyperv_info ms_hyperv;
40 
41 /* Used in modules via hv_do_hypercall(): see arch/x86/include/asm/mshyperv.h */
42 bool hyperv_paravisor_present __ro_after_init;
43 EXPORT_SYMBOL_GPL(hyperv_paravisor_present);
44 
45 #if IS_ENABLED(CONFIG_HYPERV)
hv_get_nested_msr(unsigned int reg)46 static inline unsigned int hv_get_nested_msr(unsigned int reg)
47 {
48 	if (hv_is_sint_msr(reg))
49 		return reg - HV_X64_MSR_SINT0 + HV_X64_MSR_NESTED_SINT0;
50 
51 	switch (reg) {
52 	case HV_X64_MSR_SIMP:
53 		return HV_X64_MSR_NESTED_SIMP;
54 	case HV_X64_MSR_SIEFP:
55 		return HV_X64_MSR_NESTED_SIEFP;
56 	case HV_X64_MSR_SVERSION:
57 		return HV_X64_MSR_NESTED_SVERSION;
58 	case HV_X64_MSR_SCONTROL:
59 		return HV_X64_MSR_NESTED_SCONTROL;
60 	case HV_X64_MSR_EOM:
61 		return HV_X64_MSR_NESTED_EOM;
62 	default:
63 		return reg;
64 	}
65 }
66 
hv_get_non_nested_msr(unsigned int reg)67 u64 hv_get_non_nested_msr(unsigned int reg)
68 {
69 	u64 value;
70 
71 	if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present)
72 		hv_ivm_msr_read(reg, &value);
73 	else
74 		rdmsrq(reg, value);
75 	return value;
76 }
77 EXPORT_SYMBOL_GPL(hv_get_non_nested_msr);
78 
hv_set_non_nested_msr(unsigned int reg,u64 value)79 void hv_set_non_nested_msr(unsigned int reg, u64 value)
80 {
81 	if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present) {
82 		hv_ivm_msr_write(reg, value);
83 
84 		/* Write proxy bit via wrmsl instruction */
85 		if (hv_is_sint_msr(reg))
86 			wrmsrq(reg, value | 1 << 20);
87 	} else {
88 		wrmsrq(reg, value);
89 	}
90 }
91 EXPORT_SYMBOL_GPL(hv_set_non_nested_msr);
92 
hv_get_msr(unsigned int reg)93 u64 hv_get_msr(unsigned int reg)
94 {
95 	if (hv_nested)
96 		reg = hv_get_nested_msr(reg);
97 
98 	return hv_get_non_nested_msr(reg);
99 }
100 EXPORT_SYMBOL_GPL(hv_get_msr);
101 
hv_set_msr(unsigned int reg,u64 value)102 void hv_set_msr(unsigned int reg, u64 value)
103 {
104 	if (hv_nested)
105 		reg = hv_get_nested_msr(reg);
106 
107 	hv_set_non_nested_msr(reg, value);
108 }
109 EXPORT_SYMBOL_GPL(hv_set_msr);
110 
111 static void (*mshv_handler)(void);
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);
116 
DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)117 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)
118 {
119 	struct pt_regs *old_regs = set_irq_regs(regs);
120 
121 	inc_irq_stat(irq_hv_callback_count);
122 	if (mshv_handler)
123 		mshv_handler();
124 
125 	if (vmbus_handler)
126 		vmbus_handler();
127 
128 	if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
129 		apic_eoi();
130 
131 	set_irq_regs(old_regs);
132 }
133 
hv_setup_mshv_handler(void (* handler)(void))134 void hv_setup_mshv_handler(void (*handler)(void))
135 {
136 	mshv_handler = handler;
137 }
138 
hv_setup_vmbus_handler(void (* handler)(void))139 void hv_setup_vmbus_handler(void (*handler)(void))
140 {
141 	vmbus_handler = handler;
142 }
143 
hv_remove_vmbus_handler(void)144 void hv_remove_vmbus_handler(void)
145 {
146 	/* We have no way to deallocate the interrupt gate */
147 	vmbus_handler = NULL;
148 }
149 
150 /*
151  * Routines to do per-architecture handling of stimer0
152  * interrupts when in Direct Mode
153  */
DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)154 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)
155 {
156 	struct pt_regs *old_regs = set_irq_regs(regs);
157 
158 	inc_irq_stat(hyperv_stimer0_count);
159 	if (hv_stimer0_handler)
160 		hv_stimer0_handler();
161 	add_interrupt_randomness(HYPERV_STIMER0_VECTOR);
162 	apic_eoi();
163 
164 	set_irq_regs(old_regs);
165 }
166 
167 /* For x86/x64, override weak placeholders in hyperv_timer.c */
hv_setup_stimer0_handler(void (* handler)(void))168 void hv_setup_stimer0_handler(void (*handler)(void))
169 {
170 	hv_stimer0_handler = handler;
171 }
172 
hv_remove_stimer0_handler(void)173 void hv_remove_stimer0_handler(void)
174 {
175 	/* We have no way to deallocate the interrupt gate */
176 	hv_stimer0_handler = NULL;
177 }
178 
hv_setup_kexec_handler(void (* handler)(void))179 void hv_setup_kexec_handler(void (*handler)(void))
180 {
181 	hv_kexec_handler = handler;
182 }
183 
hv_remove_kexec_handler(void)184 void hv_remove_kexec_handler(void)
185 {
186 	hv_kexec_handler = NULL;
187 }
188 
hv_setup_crash_handler(void (* handler)(struct pt_regs * regs))189 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs))
190 {
191 	hv_crash_handler = handler;
192 }
193 
hv_remove_crash_handler(void)194 void hv_remove_crash_handler(void)
195 {
196 	hv_crash_handler = NULL;
197 }
198 
199 #ifdef CONFIG_KEXEC_CORE
hv_machine_shutdown(void)200 static void hv_machine_shutdown(void)
201 {
202 	if (kexec_in_progress && hv_kexec_handler)
203 		hv_kexec_handler();
204 
205 	/*
206 	 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor
207 	 * corrupts the old VP Assist Pages and can crash the kexec kernel.
208 	 */
209 	if (kexec_in_progress)
210 		cpuhp_remove_state(CPUHP_AP_HYPERV_ONLINE);
211 
212 	/* The function calls stop_other_cpus(). */
213 	native_machine_shutdown();
214 
215 	/* Disable the hypercall page when there is only 1 active CPU. */
216 	if (kexec_in_progress)
217 		hyperv_cleanup();
218 }
219 #endif /* CONFIG_KEXEC_CORE */
220 
221 #ifdef CONFIG_CRASH_DUMP
hv_machine_crash_shutdown(struct pt_regs * regs)222 static void hv_machine_crash_shutdown(struct pt_regs *regs)
223 {
224 	if (hv_crash_handler)
225 		hv_crash_handler(regs);
226 
227 	/* The function calls crash_smp_send_stop(). */
228 	native_machine_crash_shutdown(regs);
229 
230 	/* Disable the hypercall page when there is only 1 active CPU. */
231 	hyperv_cleanup();
232 }
233 #endif /* CONFIG_CRASH_DUMP */
234 
235 static u64 hv_ref_counter_at_suspend;
236 static void (*old_save_sched_clock_state)(void);
237 static void (*old_restore_sched_clock_state)(void);
238 
239 /*
240  * Hyper-V clock counter resets during hibernation. Save and restore clock
241  * offset during suspend/resume, while also considering the time passed
242  * before suspend. This is to make sure that sched_clock using hv tsc page
243  * based clocksource, proceeds from where it left off during suspend and
244  * it shows correct time for the timestamps of kernel messages after resume.
245  */
save_hv_clock_tsc_state(void)246 static void save_hv_clock_tsc_state(void)
247 {
248 	hv_ref_counter_at_suspend = hv_read_reference_counter();
249 }
250 
restore_hv_clock_tsc_state(void)251 static void restore_hv_clock_tsc_state(void)
252 {
253 	/*
254 	 * Adjust the offsets used by hv tsc clocksource to
255 	 * account for the time spent before hibernation.
256 	 * adjusted value = reference counter (time) at suspend
257 	 *                - reference counter (time) now.
258 	 */
259 	hv_adj_sched_clock_offset(hv_ref_counter_at_suspend - hv_read_reference_counter());
260 }
261 
262 /*
263  * Functions to override save_sched_clock_state and restore_sched_clock_state
264  * functions of x86_platform. The Hyper-V clock counter is reset during
265  * suspend-resume and the offset used to measure time needs to be
266  * corrected, post resume.
267  */
hv_save_sched_clock_state(void)268 static void hv_save_sched_clock_state(void)
269 {
270 	old_save_sched_clock_state();
271 	save_hv_clock_tsc_state();
272 }
273 
hv_restore_sched_clock_state(void)274 static void hv_restore_sched_clock_state(void)
275 {
276 	restore_hv_clock_tsc_state();
277 	old_restore_sched_clock_state();
278 }
279 
x86_setup_ops_for_tsc_pg_clock(void)280 static void __init x86_setup_ops_for_tsc_pg_clock(void)
281 {
282 	if (!(ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE))
283 		return;
284 
285 	old_save_sched_clock_state = x86_platform.save_sched_clock_state;
286 	x86_platform.save_sched_clock_state = hv_save_sched_clock_state;
287 
288 	old_restore_sched_clock_state = x86_platform.restore_sched_clock_state;
289 	x86_platform.restore_sched_clock_state = hv_restore_sched_clock_state;
290 }
291 #endif /* CONFIG_HYPERV */
292 
ms_hyperv_platform(void)293 static uint32_t  __init ms_hyperv_platform(void)
294 {
295 	u32 eax;
296 	u32 hyp_signature[3];
297 
298 	if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
299 		return 0;
300 
301 	cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS,
302 	      &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]);
303 
304 	if (eax < HYPERV_CPUID_MIN || eax > HYPERV_CPUID_MAX ||
305 	    memcmp("Microsoft Hv", hyp_signature, 12))
306 		return 0;
307 
308 	/* HYPERCALL and VP_INDEX MSRs are mandatory for all features. */
309 	eax = cpuid_eax(HYPERV_CPUID_FEATURES);
310 	if (!(eax & HV_MSR_HYPERCALL_AVAILABLE)) {
311 		pr_warn("x86/hyperv: HYPERCALL MSR not available.\n");
312 		return 0;
313 	}
314 	if (!(eax & HV_MSR_VP_INDEX_AVAILABLE)) {
315 		pr_warn("x86/hyperv: VP_INDEX MSR not available.\n");
316 		return 0;
317 	}
318 
319 	return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
320 }
321 
322 #ifdef CONFIG_X86_LOCAL_APIC
323 /*
324  * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes
325  * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle
326  * unknown NMI on the first CPU which gets it.
327  */
hv_nmi_unknown(unsigned int val,struct pt_regs * regs)328 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)
329 {
330 	static atomic_t nmi_cpu = ATOMIC_INIT(-1);
331 	unsigned int old_cpu, this_cpu;
332 
333 	if (!unknown_nmi_panic)
334 		return NMI_DONE;
335 
336 	old_cpu = -1;
337 	this_cpu = raw_smp_processor_id();
338 	if (!atomic_try_cmpxchg(&nmi_cpu, &old_cpu, this_cpu))
339 		return NMI_HANDLED;
340 
341 	return NMI_DONE;
342 }
343 #endif
344 
hv_get_tsc_khz(void)345 static unsigned long hv_get_tsc_khz(void)
346 {
347 	unsigned long freq;
348 
349 	rdmsrq(HV_X64_MSR_TSC_FREQUENCY, freq);
350 
351 	return freq / 1000;
352 }
353 
354 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
hv_smp_prepare_boot_cpu(void)355 static void __init hv_smp_prepare_boot_cpu(void)
356 {
357 	native_smp_prepare_boot_cpu();
358 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
359 	hv_init_spinlocks();
360 #endif
361 }
362 
hv_smp_prepare_cpus(unsigned int max_cpus)363 static void __init hv_smp_prepare_cpus(unsigned int max_cpus)
364 {
365 #ifdef CONFIG_X86_64
366 	int i;
367 	int ret;
368 #endif
369 
370 	native_smp_prepare_cpus(max_cpus);
371 
372 	/*
373 	 *  Override wakeup_secondary_cpu_64 callback for SEV-SNP
374 	 *  enlightened guest.
375 	 */
376 	if (!ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
377 		apic->wakeup_secondary_cpu_64 = hv_snp_boot_ap;
378 		return;
379 	}
380 
381 #ifdef CONFIG_X86_64
382 	for_each_present_cpu(i) {
383 		if (i == 0)
384 			continue;
385 		ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i));
386 		BUG_ON(ret);
387 	}
388 
389 	for_each_present_cpu(i) {
390 		if (i == 0)
391 			continue;
392 		ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i);
393 		BUG_ON(ret);
394 	}
395 #endif
396 }
397 #endif
398 
399 /*
400  * When a fully enlightened TDX VM runs on Hyper-V, the firmware sets the
401  * HW_REDUCED flag: refer to acpi_tb_create_local_fadt(). Consequently ttyS0
402  * interrupts can't work because request_irq() -> ... -> irq_to_desc() returns
403  * NULL for ttyS0. This happens because mp_config_acpi_legacy_irqs() sees a
404  * nr_legacy_irqs() of 0, so it doesn't initialize the array 'mp_irqs[]', and
405  * later setup_IO_APIC_irqs() -> find_irq_entry() fails to find the legacy irqs
406  * from the array and hence doesn't create the necessary irq description info.
407  *
408  * Clone arch/x86/kernel/acpi/boot.c: acpi_generic_reduced_hw_init() here,
409  * except don't change 'legacy_pic', which keeps its default value
410  * 'default_legacy_pic'. This way, mp_config_acpi_legacy_irqs() sees a non-zero
411  * nr_legacy_irqs() and eventually serial console interrupts works properly.
412  */
reduced_hw_init(void)413 static void __init reduced_hw_init(void)
414 {
415 	x86_init.timers.timer_init	= x86_init_noop;
416 	x86_init.irqs.pre_vector_init	= x86_init_noop;
417 }
418 
hv_get_hypervisor_version(union hv_hypervisor_version_info * info)419 int hv_get_hypervisor_version(union hv_hypervisor_version_info *info)
420 {
421 	unsigned int hv_max_functions;
422 
423 	hv_max_functions = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
424 	if (hv_max_functions < HYPERV_CPUID_VERSION) {
425 		pr_err("%s: Could not detect Hyper-V version\n", __func__);
426 		return -ENODEV;
427 	}
428 
429 	cpuid(HYPERV_CPUID_VERSION, &info->eax, &info->ebx, &info->ecx, &info->edx);
430 
431 	return 0;
432 }
433 EXPORT_SYMBOL_GPL(hv_get_hypervisor_version);
434 
ms_hyperv_init_platform(void)435 static void __init ms_hyperv_init_platform(void)
436 {
437 	int hv_max_functions_eax;
438 
439 #ifdef CONFIG_PARAVIRT
440 	pv_info.name = "Hyper-V";
441 #endif
442 
443 	/*
444 	 * Extract the features and hints
445 	 */
446 	ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
447 	ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES);
448 	ms_hyperv.ext_features = cpuid_ecx(HYPERV_CPUID_FEATURES);
449 	ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES);
450 	ms_hyperv.hints    = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO);
451 
452 	hv_max_functions_eax = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
453 
454 	pr_info("Hyper-V: privilege flags low %#x, high %#x, ext %#x, hints %#x, misc %#x\n",
455 		ms_hyperv.features, ms_hyperv.priv_high,
456 		ms_hyperv.ext_features, ms_hyperv.hints,
457 		ms_hyperv.misc_features);
458 
459 	ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
460 	ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
461 
462 	pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
463 		 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
464 
465 	hv_identify_partition_type();
466 
467 	if (ms_hyperv.hints & HV_X64_HYPERV_NESTED) {
468 		hv_nested = true;
469 		pr_info("Hyper-V: running on a nested hypervisor\n");
470 	}
471 
472 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
473 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
474 		x86_platform.calibrate_tsc = hv_get_tsc_khz;
475 		x86_platform.calibrate_cpu = hv_get_tsc_khz;
476 		setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
477 	}
478 
479 	if (ms_hyperv.priv_high & HV_ISOLATION) {
480 		ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG);
481 		ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG);
482 
483 		if (ms_hyperv.shared_gpa_boundary_active)
484 			ms_hyperv.shared_gpa_boundary =
485 				BIT_ULL(ms_hyperv.shared_gpa_boundary_bits);
486 
487 		hyperv_paravisor_present = !!ms_hyperv.paravisor_present;
488 
489 		pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n",
490 			ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b);
491 
492 
493 		if (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP) {
494 			static_branch_enable(&isolation_type_snp);
495 		} else if (hv_get_isolation_type() == HV_ISOLATION_TYPE_TDX) {
496 			static_branch_enable(&isolation_type_tdx);
497 
498 			/* A TDX VM must use x2APIC and doesn't use lazy EOI. */
499 			ms_hyperv.hints &= ~HV_X64_APIC_ACCESS_RECOMMENDED;
500 
501 			if (!ms_hyperv.paravisor_present) {
502 				/*
503 				 * Mark the Hyper-V TSC page feature as disabled
504 				 * in a TDX VM without paravisor so that the
505 				 * Invariant TSC, which is a better clocksource
506 				 * anyway, is used instead.
507 				 */
508 				ms_hyperv.features &= ~HV_MSR_REFERENCE_TSC_AVAILABLE;
509 
510 				/*
511 				 * The Invariant TSC is expected to be available
512 				 * in a TDX VM without paravisor, but if not,
513 				 * print a warning message. The slower Hyper-V MSR-based
514 				 * Ref Counter should end up being the clocksource.
515 				 */
516 				if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
517 					pr_warn("Hyper-V: Invariant TSC is unavailable\n");
518 
519 				/* HV_MSR_CRASH_CTL is unsupported. */
520 				ms_hyperv.misc_features &= ~HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
521 
522 				/* Don't trust Hyper-V's TLB-flushing hypercalls. */
523 				ms_hyperv.hints &= ~HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
524 
525 				x86_init.acpi.reduced_hw_early_init = reduced_hw_init;
526 			}
527 		}
528 	}
529 
530 	if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) {
531 		ms_hyperv.nested_features =
532 			cpuid_eax(HYPERV_CPUID_NESTED_FEATURES);
533 		pr_info("Hyper-V: Nested features: 0x%x\n",
534 			ms_hyperv.nested_features);
535 	}
536 
537 #ifdef CONFIG_X86_LOCAL_APIC
538 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
539 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
540 		/*
541 		 * Get the APIC frequency.
542 		 */
543 		u64	hv_lapic_frequency;
544 
545 		rdmsrq(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency);
546 		hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ);
547 		lapic_timer_period = hv_lapic_frequency;
548 		pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
549 			lapic_timer_period);
550 	}
551 
552 	register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST,
553 			     "hv_nmi_unknown");
554 #endif
555 
556 #ifdef CONFIG_X86_IO_APIC
557 	no_timer_check = 1;
558 #endif
559 
560 #if IS_ENABLED(CONFIG_HYPERV)
561 #if defined(CONFIG_KEXEC_CORE)
562 	machine_ops.shutdown = hv_machine_shutdown;
563 #endif
564 #if defined(CONFIG_CRASH_DUMP)
565 	machine_ops.crash_shutdown = hv_machine_crash_shutdown;
566 #endif
567 #endif
568 	if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) {
569 		/*
570 		 * Writing to synthetic MSR 0x40000118 updates/changes the
571 		 * guest visible CPUIDs. Setting bit 0 of this MSR  enables
572 		 * guests to report invariant TSC feature through CPUID
573 		 * instruction, CPUID 0x800000007/EDX, bit 8. See code in
574 		 * early_init_intel() where this bit is examined. The
575 		 * setting of this MSR bit should happen before init_intel()
576 		 * is called.
577 		 */
578 		wrmsrq(HV_X64_MSR_TSC_INVARIANT_CONTROL, HV_EXPOSE_INVARIANT_TSC);
579 		setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
580 	}
581 
582 	/*
583 	 * Generation 2 instances don't support reading the NMI status from
584 	 * 0x61 port.
585 	 */
586 	if (efi_enabled(EFI_BOOT))
587 		x86_platform.get_nmi_reason = hv_get_nmi_reason;
588 
589 #if IS_ENABLED(CONFIG_HYPERV)
590 	if ((hv_get_isolation_type() == HV_ISOLATION_TYPE_VBS) ||
591 	    ms_hyperv.paravisor_present)
592 		hv_vtom_init();
593 	/*
594 	 * Setup the hook to get control post apic initialization.
595 	 */
596 	x86_platform.apic_post_init = hyperv_init;
597 	hyperv_setup_mmu_ops();
598 
599 	/* Install system interrupt handler for hypervisor callback */
600 	sysvec_install(HYPERVISOR_CALLBACK_VECTOR, sysvec_hyperv_callback);
601 
602 	/* Install system interrupt handler for reenlightenment notifications */
603 	if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) {
604 		sysvec_install(HYPERV_REENLIGHTENMENT_VECTOR, sysvec_hyperv_reenlightenment);
605 	}
606 
607 	/* Install system interrupt handler for stimer0 */
608 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) {
609 		sysvec_install(HYPERV_STIMER0_VECTOR, sysvec_hyperv_stimer0);
610 	}
611 
612 # ifdef CONFIG_SMP
613 	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
614 	if (hv_root_partition() ||
615 	    (!ms_hyperv.paravisor_present && hv_isolation_type_snp()))
616 		smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus;
617 # endif
618 
619 	/*
620 	 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
621 	 * set x2apic destination mode to physical mode when x2apic is available
622 	 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
623 	 * have 8-bit APIC id.
624 	 */
625 # ifdef CONFIG_X86_X2APIC
626 	if (x2apic_supported())
627 		x2apic_phys = 1;
628 # endif
629 
630 	/* Register Hyper-V specific clocksource */
631 	hv_init_clocksource();
632 	x86_setup_ops_for_tsc_pg_clock();
633 	hv_vtl_init_platform();
634 #endif
635 	/*
636 	 * TSC should be marked as unstable only after Hyper-V
637 	 * clocksource has been initialized. This ensures that the
638 	 * stability of the sched_clock is not altered.
639 	 */
640 	if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
641 		mark_tsc_unstable("running on Hyper-V");
642 
643 	hardlockup_detector_disable();
644 }
645 
ms_hyperv_x2apic_available(void)646 static bool __init ms_hyperv_x2apic_available(void)
647 {
648 	return x2apic_supported();
649 }
650 
651 /*
652  * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping()
653  * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the
654  * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg().
655  *
656  * Note: for a VM on Hyper-V, the I/O-APIC is the only device which
657  * (logically) generates MSIs directly to the system APIC irq domain.
658  * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the
659  * pci-hyperv host bridge.
660  *
661  * Note: for a Hyper-V root partition, this will always return false.
662  * The hypervisor doesn't expose these HYPERV_CPUID_VIRT_STACK_* cpuids by
663  * default, they are implemented as intercepts by the Windows Hyper-V stack.
664  * Even a nested root partition (L2 root) will not get them because the
665  * nested (L1) hypervisor filters them out.
666  */
ms_hyperv_msi_ext_dest_id(void)667 static bool __init ms_hyperv_msi_ext_dest_id(void)
668 {
669 	u32 eax;
670 
671 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE);
672 	if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE)
673 		return false;
674 
675 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES);
676 	return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE;
677 }
678 
679 #ifdef CONFIG_AMD_MEM_ENCRYPT
hv_sev_es_hcall_prepare(struct ghcb * ghcb,struct pt_regs * regs)680 static void hv_sev_es_hcall_prepare(struct ghcb *ghcb, struct pt_regs *regs)
681 {
682 	/* RAX and CPL are already in the GHCB */
683 	ghcb_set_rcx(ghcb, regs->cx);
684 	ghcb_set_rdx(ghcb, regs->dx);
685 	ghcb_set_r8(ghcb, regs->r8);
686 }
687 
hv_sev_es_hcall_finish(struct ghcb * ghcb,struct pt_regs * regs)688 static bool hv_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs)
689 {
690 	/* No checking of the return state needed */
691 	return true;
692 }
693 #endif
694 
695 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = {
696 	.name			= "Microsoft Hyper-V",
697 	.detect			= ms_hyperv_platform,
698 	.type			= X86_HYPER_MS_HYPERV,
699 	.init.x2apic_available	= ms_hyperv_x2apic_available,
700 	.init.msi_ext_dest_id	= ms_hyperv_msi_ext_dest_id,
701 	.init.init_platform	= ms_hyperv_init_platform,
702 	.init.guest_late_init	= ms_hyperv_late_init,
703 #ifdef CONFIG_AMD_MEM_ENCRYPT
704 	.runtime.sev_es_hcall_prepare = hv_sev_es_hcall_prepare,
705 	.runtime.sev_es_hcall_finish = hv_sev_es_hcall_finish,
706 #endif
707 };
708