xref: /linux/arch/x86/kernel/cpu/mshyperv.c (revision 37cb0c76ac6c3bf3d82d25a7c95950f2dac3ca41)
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/hyperv-tlfs.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 as the root partition? */
37 bool hv_root_partition;
38 /* Is Linux running on nested Microsoft Hypervisor */
39 bool hv_nested;
40 struct ms_hyperv_info ms_hyperv;
41 
42 /* Used in modules via hv_do_hypercall(): see arch/x86/include/asm/mshyperv.h */
43 bool hyperv_paravisor_present __ro_after_init;
44 EXPORT_SYMBOL_GPL(hyperv_paravisor_present);
45 
46 #if IS_ENABLED(CONFIG_HYPERV)
hv_get_nested_msr(unsigned int reg)47 static inline unsigned int hv_get_nested_msr(unsigned int reg)
48 {
49 	if (hv_is_sint_msr(reg))
50 		return reg - HV_X64_MSR_SINT0 + HV_X64_MSR_NESTED_SINT0;
51 
52 	switch (reg) {
53 	case HV_X64_MSR_SIMP:
54 		return HV_X64_MSR_NESTED_SIMP;
55 	case HV_X64_MSR_SIEFP:
56 		return HV_X64_MSR_NESTED_SIEFP;
57 	case HV_X64_MSR_SVERSION:
58 		return HV_X64_MSR_NESTED_SVERSION;
59 	case HV_X64_MSR_SCONTROL:
60 		return HV_X64_MSR_NESTED_SCONTROL;
61 	case HV_X64_MSR_EOM:
62 		return HV_X64_MSR_NESTED_EOM;
63 	default:
64 		return reg;
65 	}
66 }
67 
hv_get_non_nested_msr(unsigned int reg)68 u64 hv_get_non_nested_msr(unsigned int reg)
69 {
70 	u64 value;
71 
72 	if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present)
73 		hv_ivm_msr_read(reg, &value);
74 	else
75 		rdmsrl(reg, value);
76 	return value;
77 }
78 EXPORT_SYMBOL_GPL(hv_get_non_nested_msr);
79 
hv_set_non_nested_msr(unsigned int reg,u64 value)80 void hv_set_non_nested_msr(unsigned int reg, u64 value)
81 {
82 	if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present) {
83 		hv_ivm_msr_write(reg, value);
84 
85 		/* Write proxy bit via wrmsl instruction */
86 		if (hv_is_sint_msr(reg))
87 			wrmsrl(reg, value | 1 << 20);
88 	} else {
89 		wrmsrl(reg, value);
90 	}
91 }
92 EXPORT_SYMBOL_GPL(hv_set_non_nested_msr);
93 
hv_get_msr(unsigned int reg)94 u64 hv_get_msr(unsigned int reg)
95 {
96 	if (hv_nested)
97 		reg = hv_get_nested_msr(reg);
98 
99 	return hv_get_non_nested_msr(reg);
100 }
101 EXPORT_SYMBOL_GPL(hv_get_msr);
102 
hv_set_msr(unsigned int reg,u64 value)103 void hv_set_msr(unsigned int reg, u64 value)
104 {
105 	if (hv_nested)
106 		reg = hv_get_nested_msr(reg);
107 
108 	hv_set_non_nested_msr(reg, value);
109 }
110 EXPORT_SYMBOL_GPL(hv_set_msr);
111 
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 (vmbus_handler)
123 		vmbus_handler();
124 
125 	if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
126 		apic_eoi();
127 
128 	set_irq_regs(old_regs);
129 }
130 
hv_setup_vmbus_handler(void (* handler)(void))131 void hv_setup_vmbus_handler(void (*handler)(void))
132 {
133 	vmbus_handler = handler;
134 }
135 
hv_remove_vmbus_handler(void)136 void hv_remove_vmbus_handler(void)
137 {
138 	/* We have no way to deallocate the interrupt gate */
139 	vmbus_handler = NULL;
140 }
141 
142 /*
143  * Routines to do per-architecture handling of stimer0
144  * interrupts when in Direct Mode
145  */
DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)146 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)
147 {
148 	struct pt_regs *old_regs = set_irq_regs(regs);
149 
150 	inc_irq_stat(hyperv_stimer0_count);
151 	if (hv_stimer0_handler)
152 		hv_stimer0_handler();
153 	add_interrupt_randomness(HYPERV_STIMER0_VECTOR);
154 	apic_eoi();
155 
156 	set_irq_regs(old_regs);
157 }
158 
159 /* For x86/x64, override weak placeholders in hyperv_timer.c */
hv_setup_stimer0_handler(void (* handler)(void))160 void hv_setup_stimer0_handler(void (*handler)(void))
161 {
162 	hv_stimer0_handler = handler;
163 }
164 
hv_remove_stimer0_handler(void)165 void hv_remove_stimer0_handler(void)
166 {
167 	/* We have no way to deallocate the interrupt gate */
168 	hv_stimer0_handler = NULL;
169 }
170 
hv_setup_kexec_handler(void (* handler)(void))171 void hv_setup_kexec_handler(void (*handler)(void))
172 {
173 	hv_kexec_handler = handler;
174 }
175 
hv_remove_kexec_handler(void)176 void hv_remove_kexec_handler(void)
177 {
178 	hv_kexec_handler = NULL;
179 }
180 
hv_setup_crash_handler(void (* handler)(struct pt_regs * regs))181 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs))
182 {
183 	hv_crash_handler = handler;
184 }
185 
hv_remove_crash_handler(void)186 void hv_remove_crash_handler(void)
187 {
188 	hv_crash_handler = NULL;
189 }
190 
191 #ifdef CONFIG_KEXEC_CORE
hv_machine_shutdown(void)192 static void hv_machine_shutdown(void)
193 {
194 	if (kexec_in_progress && hv_kexec_handler)
195 		hv_kexec_handler();
196 
197 	/*
198 	 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor
199 	 * corrupts the old VP Assist Pages and can crash the kexec kernel.
200 	 */
201 	if (kexec_in_progress)
202 		cpuhp_remove_state(CPUHP_AP_HYPERV_ONLINE);
203 
204 	/* The function calls stop_other_cpus(). */
205 	native_machine_shutdown();
206 
207 	/* Disable the hypercall page when there is only 1 active CPU. */
208 	if (kexec_in_progress)
209 		hyperv_cleanup();
210 }
211 #endif /* CONFIG_KEXEC_CORE */
212 
213 #ifdef CONFIG_CRASH_DUMP
hv_machine_crash_shutdown(struct pt_regs * regs)214 static void hv_machine_crash_shutdown(struct pt_regs *regs)
215 {
216 	if (hv_crash_handler)
217 		hv_crash_handler(regs);
218 
219 	/* The function calls crash_smp_send_stop(). */
220 	native_machine_crash_shutdown(regs);
221 
222 	/* Disable the hypercall page when there is only 1 active CPU. */
223 	hyperv_cleanup();
224 }
225 #endif /* CONFIG_CRASH_DUMP */
226 
227 static u64 hv_ref_counter_at_suspend;
228 static void (*old_save_sched_clock_state)(void);
229 static void (*old_restore_sched_clock_state)(void);
230 
231 /*
232  * Hyper-V clock counter resets during hibernation. Save and restore clock
233  * offset during suspend/resume, while also considering the time passed
234  * before suspend. This is to make sure that sched_clock using hv tsc page
235  * based clocksource, proceeds from where it left off during suspend and
236  * it shows correct time for the timestamps of kernel messages after resume.
237  */
save_hv_clock_tsc_state(void)238 static void save_hv_clock_tsc_state(void)
239 {
240 	hv_ref_counter_at_suspend = hv_read_reference_counter();
241 }
242 
restore_hv_clock_tsc_state(void)243 static void restore_hv_clock_tsc_state(void)
244 {
245 	/*
246 	 * Adjust the offsets used by hv tsc clocksource to
247 	 * account for the time spent before hibernation.
248 	 * adjusted value = reference counter (time) at suspend
249 	 *                - reference counter (time) now.
250 	 */
251 	hv_adj_sched_clock_offset(hv_ref_counter_at_suspend - hv_read_reference_counter());
252 }
253 
254 /*
255  * Functions to override save_sched_clock_state and restore_sched_clock_state
256  * functions of x86_platform. The Hyper-V clock counter is reset during
257  * suspend-resume and the offset used to measure time needs to be
258  * corrected, post resume.
259  */
hv_save_sched_clock_state(void)260 static void hv_save_sched_clock_state(void)
261 {
262 	old_save_sched_clock_state();
263 	save_hv_clock_tsc_state();
264 }
265 
hv_restore_sched_clock_state(void)266 static void hv_restore_sched_clock_state(void)
267 {
268 	restore_hv_clock_tsc_state();
269 	old_restore_sched_clock_state();
270 }
271 
x86_setup_ops_for_tsc_pg_clock(void)272 static void __init x86_setup_ops_for_tsc_pg_clock(void)
273 {
274 	if (!(ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE))
275 		return;
276 
277 	old_save_sched_clock_state = x86_platform.save_sched_clock_state;
278 	x86_platform.save_sched_clock_state = hv_save_sched_clock_state;
279 
280 	old_restore_sched_clock_state = x86_platform.restore_sched_clock_state;
281 	x86_platform.restore_sched_clock_state = hv_restore_sched_clock_state;
282 }
283 #endif /* CONFIG_HYPERV */
284 
ms_hyperv_platform(void)285 static uint32_t  __init ms_hyperv_platform(void)
286 {
287 	u32 eax;
288 	u32 hyp_signature[3];
289 
290 	if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
291 		return 0;
292 
293 	cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS,
294 	      &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]);
295 
296 	if (eax < HYPERV_CPUID_MIN || eax > HYPERV_CPUID_MAX ||
297 	    memcmp("Microsoft Hv", hyp_signature, 12))
298 		return 0;
299 
300 	/* HYPERCALL and VP_INDEX MSRs are mandatory for all features. */
301 	eax = cpuid_eax(HYPERV_CPUID_FEATURES);
302 	if (!(eax & HV_MSR_HYPERCALL_AVAILABLE)) {
303 		pr_warn("x86/hyperv: HYPERCALL MSR not available.\n");
304 		return 0;
305 	}
306 	if (!(eax & HV_MSR_VP_INDEX_AVAILABLE)) {
307 		pr_warn("x86/hyperv: VP_INDEX MSR not available.\n");
308 		return 0;
309 	}
310 
311 	return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
312 }
313 
314 #ifdef CONFIG_X86_LOCAL_APIC
315 /*
316  * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes
317  * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle
318  * unknown NMI on the first CPU which gets it.
319  */
hv_nmi_unknown(unsigned int val,struct pt_regs * regs)320 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)
321 {
322 	static atomic_t nmi_cpu = ATOMIC_INIT(-1);
323 	unsigned int old_cpu, this_cpu;
324 
325 	if (!unknown_nmi_panic)
326 		return NMI_DONE;
327 
328 	old_cpu = -1;
329 	this_cpu = raw_smp_processor_id();
330 	if (!atomic_try_cmpxchg(&nmi_cpu, &old_cpu, this_cpu))
331 		return NMI_HANDLED;
332 
333 	return NMI_DONE;
334 }
335 #endif
336 
hv_get_tsc_khz(void)337 static unsigned long hv_get_tsc_khz(void)
338 {
339 	unsigned long freq;
340 
341 	rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
342 
343 	return freq / 1000;
344 }
345 
346 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
hv_smp_prepare_boot_cpu(void)347 static void __init hv_smp_prepare_boot_cpu(void)
348 {
349 	native_smp_prepare_boot_cpu();
350 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
351 	hv_init_spinlocks();
352 #endif
353 }
354 
hv_smp_prepare_cpus(unsigned int max_cpus)355 static void __init hv_smp_prepare_cpus(unsigned int max_cpus)
356 {
357 #ifdef CONFIG_X86_64
358 	int i;
359 	int ret;
360 #endif
361 
362 	native_smp_prepare_cpus(max_cpus);
363 
364 	/*
365 	 *  Override wakeup_secondary_cpu_64 callback for SEV-SNP
366 	 *  enlightened guest.
367 	 */
368 	if (!ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
369 		apic->wakeup_secondary_cpu_64 = hv_snp_boot_ap;
370 		return;
371 	}
372 
373 #ifdef CONFIG_X86_64
374 	for_each_present_cpu(i) {
375 		if (i == 0)
376 			continue;
377 		ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i));
378 		BUG_ON(ret);
379 	}
380 
381 	for_each_present_cpu(i) {
382 		if (i == 0)
383 			continue;
384 		ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i);
385 		BUG_ON(ret);
386 	}
387 #endif
388 }
389 #endif
390 
391 /*
392  * When a fully enlightened TDX VM runs on Hyper-V, the firmware sets the
393  * HW_REDUCED flag: refer to acpi_tb_create_local_fadt(). Consequently ttyS0
394  * interrupts can't work because request_irq() -> ... -> irq_to_desc() returns
395  * NULL for ttyS0. This happens because mp_config_acpi_legacy_irqs() sees a
396  * nr_legacy_irqs() of 0, so it doesn't initialize the array 'mp_irqs[]', and
397  * later setup_IO_APIC_irqs() -> find_irq_entry() fails to find the legacy irqs
398  * from the array and hence doesn't create the necessary irq description info.
399  *
400  * Clone arch/x86/kernel/acpi/boot.c: acpi_generic_reduced_hw_init() here,
401  * except don't change 'legacy_pic', which keeps its default value
402  * 'default_legacy_pic'. This way, mp_config_acpi_legacy_irqs() sees a non-zero
403  * nr_legacy_irqs() and eventually serial console interrupts works properly.
404  */
reduced_hw_init(void)405 static void __init reduced_hw_init(void)
406 {
407 	x86_init.timers.timer_init	= x86_init_noop;
408 	x86_init.irqs.pre_vector_init	= x86_init_noop;
409 }
410 
hv_get_hypervisor_version(union hv_hypervisor_version_info * info)411 int hv_get_hypervisor_version(union hv_hypervisor_version_info *info)
412 {
413 	unsigned int hv_max_functions;
414 
415 	hv_max_functions = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
416 	if (hv_max_functions < HYPERV_CPUID_VERSION) {
417 		pr_err("%s: Could not detect Hyper-V version\n", __func__);
418 		return -ENODEV;
419 	}
420 
421 	cpuid(HYPERV_CPUID_VERSION, &info->eax, &info->ebx, &info->ecx, &info->edx);
422 
423 	return 0;
424 }
425 
ms_hyperv_init_platform(void)426 static void __init ms_hyperv_init_platform(void)
427 {
428 	int hv_max_functions_eax;
429 
430 #ifdef CONFIG_PARAVIRT
431 	pv_info.name = "Hyper-V";
432 #endif
433 
434 	/*
435 	 * Extract the features and hints
436 	 */
437 	ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
438 	ms_hyperv.priv_high = cpuid_ebx(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 0x%x, high 0x%x, hints 0x%x, misc 0x%x\n",
445 		ms_hyperv.features, ms_hyperv.priv_high, ms_hyperv.hints,
446 		ms_hyperv.misc_features);
447 
448 	ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
449 	ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
450 
451 	pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
452 		 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
453 
454 	/*
455 	 * Check CPU management privilege.
456 	 *
457 	 * To mirror what Windows does we should extract CPU management
458 	 * features and use the ReservedIdentityBit to detect if Linux is the
459 	 * root partition. But that requires negotiating CPU management
460 	 * interface (a process to be finalized). For now, use the privilege
461 	 * flag as the indicator for running as root.
462 	 *
463 	 * Hyper-V should never specify running as root and as a Confidential
464 	 * VM. But to protect against a compromised/malicious Hyper-V trying
465 	 * to exploit root behavior to expose Confidential VM memory, ignore
466 	 * the root partition setting if also a Confidential VM.
467 	 */
468 	if ((ms_hyperv.priv_high & HV_CPU_MANAGEMENT) &&
469 	    !(ms_hyperv.priv_high & HV_ISOLATION)) {
470 		hv_root_partition = true;
471 		pr_info("Hyper-V: running as root partition\n");
472 	}
473 
474 	if (ms_hyperv.hints & HV_X64_HYPERV_NESTED) {
475 		hv_nested = true;
476 		pr_info("Hyper-V: running on a nested hypervisor\n");
477 	}
478 
479 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
480 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
481 		x86_platform.calibrate_tsc = hv_get_tsc_khz;
482 		x86_platform.calibrate_cpu = hv_get_tsc_khz;
483 		setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
484 	}
485 
486 	if (ms_hyperv.priv_high & HV_ISOLATION) {
487 		ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG);
488 		ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG);
489 
490 		if (ms_hyperv.shared_gpa_boundary_active)
491 			ms_hyperv.shared_gpa_boundary =
492 				BIT_ULL(ms_hyperv.shared_gpa_boundary_bits);
493 
494 		hyperv_paravisor_present = !!ms_hyperv.paravisor_present;
495 
496 		pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n",
497 			ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b);
498 
499 
500 		if (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP) {
501 			static_branch_enable(&isolation_type_snp);
502 		} else if (hv_get_isolation_type() == HV_ISOLATION_TYPE_TDX) {
503 			static_branch_enable(&isolation_type_tdx);
504 
505 			/* A TDX VM must use x2APIC and doesn't use lazy EOI. */
506 			ms_hyperv.hints &= ~HV_X64_APIC_ACCESS_RECOMMENDED;
507 
508 			if (!ms_hyperv.paravisor_present) {
509 				/*
510 				 * Mark the Hyper-V TSC page feature as disabled
511 				 * in a TDX VM without paravisor so that the
512 				 * Invariant TSC, which is a better clocksource
513 				 * anyway, is used instead.
514 				 */
515 				ms_hyperv.features &= ~HV_MSR_REFERENCE_TSC_AVAILABLE;
516 
517 				/*
518 				 * The Invariant TSC is expected to be available
519 				 * in a TDX VM without paravisor, but if not,
520 				 * print a warning message. The slower Hyper-V MSR-based
521 				 * Ref Counter should end up being the clocksource.
522 				 */
523 				if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
524 					pr_warn("Hyper-V: Invariant TSC is unavailable\n");
525 
526 				/* HV_MSR_CRASH_CTL is unsupported. */
527 				ms_hyperv.misc_features &= ~HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
528 
529 				/* Don't trust Hyper-V's TLB-flushing hypercalls. */
530 				ms_hyperv.hints &= ~HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
531 
532 				x86_init.acpi.reduced_hw_early_init = reduced_hw_init;
533 			}
534 		}
535 	}
536 
537 	if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) {
538 		ms_hyperv.nested_features =
539 			cpuid_eax(HYPERV_CPUID_NESTED_FEATURES);
540 		pr_info("Hyper-V: Nested features: 0x%x\n",
541 			ms_hyperv.nested_features);
542 	}
543 
544 #ifdef CONFIG_X86_LOCAL_APIC
545 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
546 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
547 		/*
548 		 * Get the APIC frequency.
549 		 */
550 		u64	hv_lapic_frequency;
551 
552 		rdmsrl(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency);
553 		hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ);
554 		lapic_timer_period = hv_lapic_frequency;
555 		pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
556 			lapic_timer_period);
557 	}
558 
559 	register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST,
560 			     "hv_nmi_unknown");
561 #endif
562 
563 #ifdef CONFIG_X86_IO_APIC
564 	no_timer_check = 1;
565 #endif
566 
567 #if IS_ENABLED(CONFIG_HYPERV)
568 #if defined(CONFIG_KEXEC_CORE)
569 	machine_ops.shutdown = hv_machine_shutdown;
570 #endif
571 #if defined(CONFIG_CRASH_DUMP)
572 	machine_ops.crash_shutdown = hv_machine_crash_shutdown;
573 #endif
574 #endif
575 	if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) {
576 		/*
577 		 * Writing to synthetic MSR 0x40000118 updates/changes the
578 		 * guest visible CPUIDs. Setting bit 0 of this MSR  enables
579 		 * guests to report invariant TSC feature through CPUID
580 		 * instruction, CPUID 0x800000007/EDX, bit 8. See code in
581 		 * early_init_intel() where this bit is examined. The
582 		 * setting of this MSR bit should happen before init_intel()
583 		 * is called.
584 		 */
585 		wrmsrl(HV_X64_MSR_TSC_INVARIANT_CONTROL, HV_EXPOSE_INVARIANT_TSC);
586 		setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
587 	}
588 
589 	/*
590 	 * Generation 2 instances don't support reading the NMI status from
591 	 * 0x61 port.
592 	 */
593 	if (efi_enabled(EFI_BOOT))
594 		x86_platform.get_nmi_reason = hv_get_nmi_reason;
595 
596 #if IS_ENABLED(CONFIG_HYPERV)
597 	if ((hv_get_isolation_type() == HV_ISOLATION_TYPE_VBS) ||
598 	    ms_hyperv.paravisor_present)
599 		hv_vtom_init();
600 	/*
601 	 * Setup the hook to get control post apic initialization.
602 	 */
603 	x86_platform.apic_post_init = hyperv_init;
604 	hyperv_setup_mmu_ops();
605 
606 	/* Install system interrupt handler for hypervisor callback */
607 	sysvec_install(HYPERVISOR_CALLBACK_VECTOR, sysvec_hyperv_callback);
608 
609 	/* Install system interrupt handler for reenlightenment notifications */
610 	if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) {
611 		sysvec_install(HYPERV_REENLIGHTENMENT_VECTOR, sysvec_hyperv_reenlightenment);
612 	}
613 
614 	/* Install system interrupt handler for stimer0 */
615 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) {
616 		sysvec_install(HYPERV_STIMER0_VECTOR, sysvec_hyperv_stimer0);
617 	}
618 
619 # ifdef CONFIG_SMP
620 	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
621 	if (hv_root_partition ||
622 	    (!ms_hyperv.paravisor_present && hv_isolation_type_snp()))
623 		smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus;
624 # endif
625 
626 	/*
627 	 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
628 	 * set x2apic destination mode to physical mode when x2apic is available
629 	 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
630 	 * have 8-bit APIC id.
631 	 */
632 # ifdef CONFIG_X86_X2APIC
633 	if (x2apic_supported())
634 		x2apic_phys = 1;
635 # endif
636 
637 	/* Register Hyper-V specific clocksource */
638 	hv_init_clocksource();
639 	x86_setup_ops_for_tsc_pg_clock();
640 	hv_vtl_init_platform();
641 #endif
642 	/*
643 	 * TSC should be marked as unstable only after Hyper-V
644 	 * clocksource has been initialized. This ensures that the
645 	 * stability of the sched_clock is not altered.
646 	 */
647 	if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
648 		mark_tsc_unstable("running on Hyper-V");
649 
650 	hardlockup_detector_disable();
651 }
652 
ms_hyperv_x2apic_available(void)653 static bool __init ms_hyperv_x2apic_available(void)
654 {
655 	return x2apic_supported();
656 }
657 
658 /*
659  * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping()
660  * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the
661  * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg().
662  *
663  * Note: for a VM on Hyper-V, the I/O-APIC is the only device which
664  * (logically) generates MSIs directly to the system APIC irq domain.
665  * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the
666  * pci-hyperv host bridge.
667  *
668  * Note: for a Hyper-V root partition, this will always return false.
669  * The hypervisor doesn't expose these HYPERV_CPUID_VIRT_STACK_* cpuids by
670  * default, they are implemented as intercepts by the Windows Hyper-V stack.
671  * Even a nested root partition (L2 root) will not get them because the
672  * nested (L1) hypervisor filters them out.
673  */
ms_hyperv_msi_ext_dest_id(void)674 static bool __init ms_hyperv_msi_ext_dest_id(void)
675 {
676 	u32 eax;
677 
678 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE);
679 	if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE)
680 		return false;
681 
682 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES);
683 	return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE;
684 }
685 
686 #ifdef CONFIG_AMD_MEM_ENCRYPT
hv_sev_es_hcall_prepare(struct ghcb * ghcb,struct pt_regs * regs)687 static void hv_sev_es_hcall_prepare(struct ghcb *ghcb, struct pt_regs *regs)
688 {
689 	/* RAX and CPL are already in the GHCB */
690 	ghcb_set_rcx(ghcb, regs->cx);
691 	ghcb_set_rdx(ghcb, regs->dx);
692 	ghcb_set_r8(ghcb, regs->r8);
693 }
694 
hv_sev_es_hcall_finish(struct ghcb * ghcb,struct pt_regs * regs)695 static bool hv_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs)
696 {
697 	/* No checking of the return state needed */
698 	return true;
699 }
700 #endif
701 
702 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = {
703 	.name			= "Microsoft Hyper-V",
704 	.detect			= ms_hyperv_platform,
705 	.type			= X86_HYPER_MS_HYPERV,
706 	.init.x2apic_available	= ms_hyperv_x2apic_available,
707 	.init.msi_ext_dest_id	= ms_hyperv_msi_ext_dest_id,
708 	.init.init_platform	= ms_hyperv_init_platform,
709 	.init.guest_late_init	= ms_hyperv_late_init,
710 #ifdef CONFIG_AMD_MEM_ENCRYPT
711 	.runtime.sev_es_hcall_prepare = hv_sev_es_hcall_prepare,
712 	.runtime.sev_es_hcall_finish = hv_sev_es_hcall_finish,
713 #endif
714 };
715