xref: /linux/arch/x86/kernel/cpu/mshyperv.c (revision 9591fdb0611dccdeeeeacb99d89f0098737d209b)
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 #if IS_ENABLED(CONFIG_HYPERV)
hv_get_nested_msr(unsigned int reg)42 static inline unsigned int hv_get_nested_msr(unsigned int reg)
43 {
44 	if (hv_is_sint_msr(reg))
45 		return reg - HV_X64_MSR_SINT0 + HV_X64_MSR_NESTED_SINT0;
46 
47 	switch (reg) {
48 	case HV_X64_MSR_SIMP:
49 		return HV_X64_MSR_NESTED_SIMP;
50 	case HV_X64_MSR_SIEFP:
51 		return HV_X64_MSR_NESTED_SIEFP;
52 	case HV_X64_MSR_SVERSION:
53 		return HV_X64_MSR_NESTED_SVERSION;
54 	case HV_X64_MSR_SCONTROL:
55 		return HV_X64_MSR_NESTED_SCONTROL;
56 	case HV_X64_MSR_EOM:
57 		return HV_X64_MSR_NESTED_EOM;
58 	default:
59 		return reg;
60 	}
61 }
62 
hv_get_non_nested_msr(unsigned int reg)63 u64 hv_get_non_nested_msr(unsigned int reg)
64 {
65 	u64 value;
66 
67 	if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present)
68 		hv_ivm_msr_read(reg, &value);
69 	else
70 		rdmsrq(reg, value);
71 	return value;
72 }
73 EXPORT_SYMBOL_GPL(hv_get_non_nested_msr);
74 
hv_set_non_nested_msr(unsigned int reg,u64 value)75 void hv_set_non_nested_msr(unsigned int reg, u64 value)
76 {
77 	if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present) {
78 		hv_ivm_msr_write(reg, value);
79 
80 		/* Write proxy bit via wrmsl instruction */
81 		if (hv_is_sint_msr(reg))
82 			wrmsrq(reg, value | 1 << 20);
83 	} else {
84 		wrmsrq(reg, value);
85 	}
86 }
87 EXPORT_SYMBOL_GPL(hv_set_non_nested_msr);
88 
hv_get_msr(unsigned int reg)89 u64 hv_get_msr(unsigned int reg)
90 {
91 	if (hv_nested)
92 		reg = hv_get_nested_msr(reg);
93 
94 	return hv_get_non_nested_msr(reg);
95 }
96 EXPORT_SYMBOL_GPL(hv_get_msr);
97 
hv_set_msr(unsigned int reg,u64 value)98 void hv_set_msr(unsigned int reg, u64 value)
99 {
100 	if (hv_nested)
101 		reg = hv_get_nested_msr(reg);
102 
103 	hv_set_non_nested_msr(reg, value);
104 }
105 EXPORT_SYMBOL_GPL(hv_set_msr);
106 
107 static void (*mshv_handler)(void);
108 static void (*vmbus_handler)(void);
109 static void (*hv_stimer0_handler)(void);
110 static void (*hv_kexec_handler)(void);
111 static void (*hv_crash_handler)(struct pt_regs *regs);
112 
DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)113 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)
114 {
115 	struct pt_regs *old_regs = set_irq_regs(regs);
116 
117 	inc_irq_stat(irq_hv_callback_count);
118 	if (mshv_handler)
119 		mshv_handler();
120 
121 	if (vmbus_handler)
122 		vmbus_handler();
123 
124 	if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
125 		apic_eoi();
126 
127 	set_irq_regs(old_regs);
128 }
129 
hv_setup_mshv_handler(void (* handler)(void))130 void hv_setup_mshv_handler(void (*handler)(void))
131 {
132 	mshv_handler = handler;
133 }
134 
hv_setup_vmbus_handler(void (* handler)(void))135 void hv_setup_vmbus_handler(void (*handler)(void))
136 {
137 	vmbus_handler = handler;
138 }
139 
hv_remove_vmbus_handler(void)140 void hv_remove_vmbus_handler(void)
141 {
142 	/* We have no way to deallocate the interrupt gate */
143 	vmbus_handler = NULL;
144 }
145 
146 /*
147  * Routines to do per-architecture handling of stimer0
148  * interrupts when in Direct Mode
149  */
DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)150 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)
151 {
152 	struct pt_regs *old_regs = set_irq_regs(regs);
153 
154 	inc_irq_stat(hyperv_stimer0_count);
155 	if (hv_stimer0_handler)
156 		hv_stimer0_handler();
157 	add_interrupt_randomness(HYPERV_STIMER0_VECTOR);
158 	apic_eoi();
159 
160 	set_irq_regs(old_regs);
161 }
162 
163 /* For x86/x64, override weak placeholders in hyperv_timer.c */
hv_setup_stimer0_handler(void (* handler)(void))164 void hv_setup_stimer0_handler(void (*handler)(void))
165 {
166 	hv_stimer0_handler = handler;
167 }
168 
hv_remove_stimer0_handler(void)169 void hv_remove_stimer0_handler(void)
170 {
171 	/* We have no way to deallocate the interrupt gate */
172 	hv_stimer0_handler = NULL;
173 }
174 
hv_setup_kexec_handler(void (* handler)(void))175 void hv_setup_kexec_handler(void (*handler)(void))
176 {
177 	hv_kexec_handler = handler;
178 }
179 
hv_remove_kexec_handler(void)180 void hv_remove_kexec_handler(void)
181 {
182 	hv_kexec_handler = NULL;
183 }
184 
hv_setup_crash_handler(void (* handler)(struct pt_regs * regs))185 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs))
186 {
187 	hv_crash_handler = handler;
188 }
189 
hv_remove_crash_handler(void)190 void hv_remove_crash_handler(void)
191 {
192 	hv_crash_handler = NULL;
193 }
194 
195 #ifdef CONFIG_KEXEC_CORE
hv_machine_shutdown(void)196 static void hv_machine_shutdown(void)
197 {
198 	if (kexec_in_progress && hv_kexec_handler)
199 		hv_kexec_handler();
200 
201 	/*
202 	 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor
203 	 * corrupts the old VP Assist Pages and can crash the kexec kernel.
204 	 */
205 	if (kexec_in_progress)
206 		cpuhp_remove_state(CPUHP_AP_HYPERV_ONLINE);
207 
208 	/* The function calls stop_other_cpus(). */
209 	native_machine_shutdown();
210 
211 	/* Disable the hypercall page when there is only 1 active CPU. */
212 	if (kexec_in_progress)
213 		hyperv_cleanup();
214 }
215 #endif /* CONFIG_KEXEC_CORE */
216 
217 #ifdef CONFIG_CRASH_DUMP
hv_machine_crash_shutdown(struct pt_regs * regs)218 static void hv_machine_crash_shutdown(struct pt_regs *regs)
219 {
220 	if (hv_crash_handler)
221 		hv_crash_handler(regs);
222 
223 	/* The function calls crash_smp_send_stop(). */
224 	native_machine_crash_shutdown(regs);
225 
226 	/* Disable the hypercall page when there is only 1 active CPU. */
227 	hyperv_cleanup();
228 }
229 #endif /* CONFIG_CRASH_DUMP */
230 
231 static u64 hv_ref_counter_at_suspend;
232 static void (*old_save_sched_clock_state)(void);
233 static void (*old_restore_sched_clock_state)(void);
234 
235 /*
236  * Hyper-V clock counter resets during hibernation. Save and restore clock
237  * offset during suspend/resume, while also considering the time passed
238  * before suspend. This is to make sure that sched_clock using hv tsc page
239  * based clocksource, proceeds from where it left off during suspend and
240  * it shows correct time for the timestamps of kernel messages after resume.
241  */
save_hv_clock_tsc_state(void)242 static void save_hv_clock_tsc_state(void)
243 {
244 	hv_ref_counter_at_suspend = hv_read_reference_counter();
245 }
246 
restore_hv_clock_tsc_state(void)247 static void restore_hv_clock_tsc_state(void)
248 {
249 	/*
250 	 * Adjust the offsets used by hv tsc clocksource to
251 	 * account for the time spent before hibernation.
252 	 * adjusted value = reference counter (time) at suspend
253 	 *                - reference counter (time) now.
254 	 */
255 	hv_adj_sched_clock_offset(hv_ref_counter_at_suspend - hv_read_reference_counter());
256 }
257 
258 /*
259  * Functions to override save_sched_clock_state and restore_sched_clock_state
260  * functions of x86_platform. The Hyper-V clock counter is reset during
261  * suspend-resume and the offset used to measure time needs to be
262  * corrected, post resume.
263  */
hv_save_sched_clock_state(void)264 static void hv_save_sched_clock_state(void)
265 {
266 	old_save_sched_clock_state();
267 	save_hv_clock_tsc_state();
268 }
269 
hv_restore_sched_clock_state(void)270 static void hv_restore_sched_clock_state(void)
271 {
272 	restore_hv_clock_tsc_state();
273 	old_restore_sched_clock_state();
274 }
275 
x86_setup_ops_for_tsc_pg_clock(void)276 static void __init x86_setup_ops_for_tsc_pg_clock(void)
277 {
278 	if (!(ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE))
279 		return;
280 
281 	old_save_sched_clock_state = x86_platform.save_sched_clock_state;
282 	x86_platform.save_sched_clock_state = hv_save_sched_clock_state;
283 
284 	old_restore_sched_clock_state = x86_platform.restore_sched_clock_state;
285 	x86_platform.restore_sched_clock_state = hv_restore_sched_clock_state;
286 }
287 
288 #ifdef CONFIG_X86_64
289 DEFINE_STATIC_CALL(hv_hypercall, hv_std_hypercall);
290 EXPORT_STATIC_CALL_TRAMP_GPL(hv_hypercall);
291 #define hypercall_update(hc) static_call_update(hv_hypercall, hc)
292 #endif
293 #endif /* CONFIG_HYPERV */
294 
295 #ifndef hypercall_update
296 #define hypercall_update(hc) (void)hc
297 #endif
298 
ms_hyperv_platform(void)299 static uint32_t  __init ms_hyperv_platform(void)
300 {
301 	u32 eax;
302 	u32 hyp_signature[3];
303 
304 	if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
305 		return 0;
306 
307 	cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS,
308 	      &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]);
309 
310 	if (eax < HYPERV_CPUID_MIN || eax > HYPERV_CPUID_MAX ||
311 	    memcmp("Microsoft Hv", hyp_signature, 12))
312 		return 0;
313 
314 	/* HYPERCALL and VP_INDEX MSRs are mandatory for all features. */
315 	eax = cpuid_eax(HYPERV_CPUID_FEATURES);
316 	if (!(eax & HV_MSR_HYPERCALL_AVAILABLE)) {
317 		pr_warn("x86/hyperv: HYPERCALL MSR not available.\n");
318 		return 0;
319 	}
320 	if (!(eax & HV_MSR_VP_INDEX_AVAILABLE)) {
321 		pr_warn("x86/hyperv: VP_INDEX MSR not available.\n");
322 		return 0;
323 	}
324 
325 	return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
326 }
327 
328 #ifdef CONFIG_X86_LOCAL_APIC
329 /*
330  * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes
331  * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle
332  * unknown NMI on the first CPU which gets it.
333  */
hv_nmi_unknown(unsigned int val,struct pt_regs * regs)334 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)
335 {
336 	static atomic_t nmi_cpu = ATOMIC_INIT(-1);
337 	unsigned int old_cpu, this_cpu;
338 
339 	if (!unknown_nmi_panic)
340 		return NMI_DONE;
341 
342 	old_cpu = -1;
343 	this_cpu = raw_smp_processor_id();
344 	if (!atomic_try_cmpxchg(&nmi_cpu, &old_cpu, this_cpu))
345 		return NMI_HANDLED;
346 
347 	return NMI_DONE;
348 }
349 #endif
350 
hv_get_tsc_khz(void)351 static unsigned long hv_get_tsc_khz(void)
352 {
353 	unsigned long freq;
354 
355 	rdmsrq(HV_X64_MSR_TSC_FREQUENCY, freq);
356 
357 	return freq / 1000;
358 }
359 
360 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
hv_smp_prepare_boot_cpu(void)361 static void __init hv_smp_prepare_boot_cpu(void)
362 {
363 	native_smp_prepare_boot_cpu();
364 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
365 	hv_init_spinlocks();
366 #endif
367 }
368 
hv_smp_prepare_cpus(unsigned int max_cpus)369 static void __init hv_smp_prepare_cpus(unsigned int max_cpus)
370 {
371 #ifdef CONFIG_X86_64
372 	int i;
373 	int ret;
374 #endif
375 
376 	native_smp_prepare_cpus(max_cpus);
377 
378 	/*
379 	 *  Override wakeup_secondary_cpu_64 callback for SEV-SNP
380 	 *  enlightened guest.
381 	 */
382 	if (!ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
383 		apic->wakeup_secondary_cpu_64 = hv_snp_boot_ap;
384 		return;
385 	}
386 
387 #ifdef CONFIG_X86_64
388 	for_each_present_cpu(i) {
389 		if (i == 0)
390 			continue;
391 		ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i));
392 		BUG_ON(ret);
393 	}
394 
395 	for_each_present_cpu(i) {
396 		if (i == 0)
397 			continue;
398 		ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i);
399 		BUG_ON(ret);
400 	}
401 #endif
402 }
403 #endif
404 
405 /*
406  * When a fully enlightened TDX VM runs on Hyper-V, the firmware sets the
407  * HW_REDUCED flag: refer to acpi_tb_create_local_fadt(). Consequently ttyS0
408  * interrupts can't work because request_irq() -> ... -> irq_to_desc() returns
409  * NULL for ttyS0. This happens because mp_config_acpi_legacy_irqs() sees a
410  * nr_legacy_irqs() of 0, so it doesn't initialize the array 'mp_irqs[]', and
411  * later setup_IO_APIC_irqs() -> find_irq_entry() fails to find the legacy irqs
412  * from the array and hence doesn't create the necessary irq description info.
413  *
414  * Clone arch/x86/kernel/acpi/boot.c: acpi_generic_reduced_hw_init() here,
415  * except don't change 'legacy_pic', which keeps its default value
416  * 'default_legacy_pic'. This way, mp_config_acpi_legacy_irqs() sees a non-zero
417  * nr_legacy_irqs() and eventually serial console interrupts works properly.
418  */
reduced_hw_init(void)419 static void __init reduced_hw_init(void)
420 {
421 	x86_init.timers.timer_init	= x86_init_noop;
422 	x86_init.irqs.pre_vector_init	= x86_init_noop;
423 }
424 
hv_get_hypervisor_version(union hv_hypervisor_version_info * info)425 int hv_get_hypervisor_version(union hv_hypervisor_version_info *info)
426 {
427 	unsigned int hv_max_functions;
428 
429 	hv_max_functions = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
430 	if (hv_max_functions < HYPERV_CPUID_VERSION) {
431 		pr_err("%s: Could not detect Hyper-V version\n", __func__);
432 		return -ENODEV;
433 	}
434 
435 	cpuid(HYPERV_CPUID_VERSION, &info->eax, &info->ebx, &info->ecx, &info->edx);
436 
437 	return 0;
438 }
439 EXPORT_SYMBOL_GPL(hv_get_hypervisor_version);
440 
ms_hyperv_init_platform(void)441 static void __init ms_hyperv_init_platform(void)
442 {
443 	int hv_max_functions_eax;
444 
445 #ifdef CONFIG_PARAVIRT
446 	pv_info.name = "Hyper-V";
447 #endif
448 
449 	/*
450 	 * Extract the features and hints
451 	 */
452 	ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
453 	ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES);
454 	ms_hyperv.ext_features = cpuid_ecx(HYPERV_CPUID_FEATURES);
455 	ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES);
456 	ms_hyperv.hints    = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO);
457 
458 	hv_max_functions_eax = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
459 
460 	pr_info("Hyper-V: privilege flags low %#x, high %#x, ext %#x, hints %#x, misc %#x\n",
461 		ms_hyperv.features, ms_hyperv.priv_high,
462 		ms_hyperv.ext_features, ms_hyperv.hints,
463 		ms_hyperv.misc_features);
464 
465 	ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
466 	ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
467 
468 	pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
469 		 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
470 
471 	hv_identify_partition_type();
472 
473 	if (ms_hyperv.hints & HV_X64_HYPERV_NESTED) {
474 		hv_nested = true;
475 		pr_info("Hyper-V: running on a nested hypervisor\n");
476 	}
477 
478 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
479 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
480 		x86_platform.calibrate_tsc = hv_get_tsc_khz;
481 		x86_platform.calibrate_cpu = hv_get_tsc_khz;
482 		setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
483 	}
484 
485 	if (ms_hyperv.priv_high & HV_ISOLATION) {
486 		ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG);
487 		ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG);
488 
489 		if (ms_hyperv.shared_gpa_boundary_active)
490 			ms_hyperv.shared_gpa_boundary =
491 				BIT_ULL(ms_hyperv.shared_gpa_boundary_bits);
492 
493 		pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n",
494 			ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b);
495 
496 
497 		if (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP) {
498 			static_branch_enable(&isolation_type_snp);
499 			if (!ms_hyperv.paravisor_present)
500 				hypercall_update(hv_snp_hypercall);
501 		} else if (hv_get_isolation_type() == HV_ISOLATION_TYPE_TDX) {
502 			static_branch_enable(&isolation_type_tdx);
503 
504 			/* A TDX VM must use x2APIC and doesn't use lazy EOI. */
505 			ms_hyperv.hints &= ~HV_X64_APIC_ACCESS_RECOMMENDED;
506 
507 			if (!ms_hyperv.paravisor_present) {
508 				hypercall_update(hv_tdx_hypercall);
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 		rdmsrq(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 	/*
576 	 * HV_ACCESS_TSC_INVARIANT is always zero for the root partition. Root
577 	 * partition doesn't need to write to synthetic MSR to enable invariant
578 	 * TSC feature. It sees what the hardware provides.
579 	 */
580 	if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) {
581 		/*
582 		 * Writing to synthetic MSR 0x40000118 updates/changes the
583 		 * guest visible CPUIDs. Setting bit 0 of this MSR  enables
584 		 * guests to report invariant TSC feature through CPUID
585 		 * instruction, CPUID 0x800000007/EDX, bit 8. See code in
586 		 * early_init_intel() where this bit is examined. The
587 		 * setting of this MSR bit should happen before init_intel()
588 		 * is called.
589 		 */
590 		wrmsrq(HV_X64_MSR_TSC_INVARIANT_CONTROL, HV_EXPOSE_INVARIANT_TSC);
591 		setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
592 	}
593 
594 	/*
595 	 * Generation 2 instances don't support reading the NMI status from
596 	 * 0x61 port.
597 	 */
598 	if (efi_enabled(EFI_BOOT))
599 		x86_platform.get_nmi_reason = hv_get_nmi_reason;
600 
601 #if IS_ENABLED(CONFIG_HYPERV)
602 	if ((hv_get_isolation_type() == HV_ISOLATION_TYPE_VBS) ||
603 	    ms_hyperv.paravisor_present)
604 		hv_vtom_init();
605 	/*
606 	 * Setup the hook to get control post apic initialization.
607 	 */
608 	x86_platform.apic_post_init = hyperv_init;
609 	hyperv_setup_mmu_ops();
610 
611 	/* Install system interrupt handler for hypervisor callback */
612 	sysvec_install(HYPERVISOR_CALLBACK_VECTOR, sysvec_hyperv_callback);
613 
614 	/* Install system interrupt handler for reenlightenment notifications */
615 	if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) {
616 		sysvec_install(HYPERV_REENLIGHTENMENT_VECTOR, sysvec_hyperv_reenlightenment);
617 	}
618 
619 	/* Install system interrupt handler for stimer0 */
620 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) {
621 		sysvec_install(HYPERV_STIMER0_VECTOR, sysvec_hyperv_stimer0);
622 	}
623 
624 # ifdef CONFIG_SMP
625 	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
626 	if (hv_root_partition() ||
627 	    (!ms_hyperv.paravisor_present && hv_isolation_type_snp()))
628 		smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus;
629 # endif
630 
631 	/*
632 	 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
633 	 * set x2apic destination mode to physical mode when x2apic is available
634 	 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
635 	 * have 8-bit APIC id.
636 	 */
637 # ifdef CONFIG_X86_X2APIC
638 	if (x2apic_supported())
639 		x2apic_phys = 1;
640 # endif
641 
642 	/* Register Hyper-V specific clocksource */
643 	hv_init_clocksource();
644 	x86_setup_ops_for_tsc_pg_clock();
645 	hv_vtl_init_platform();
646 #endif
647 	/*
648 	 * TSC should be marked as unstable only after Hyper-V
649 	 * clocksource has been initialized. This ensures that the
650 	 * stability of the sched_clock is not altered.
651 	 *
652 	 * HV_ACCESS_TSC_INVARIANT is always zero for the root partition. No
653 	 * need to check for it.
654 	 */
655 	if (!hv_root_partition() &&
656 	    !(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
657 		mark_tsc_unstable("running on Hyper-V");
658 
659 	hardlockup_detector_disable();
660 }
661 
ms_hyperv_x2apic_available(void)662 static bool __init ms_hyperv_x2apic_available(void)
663 {
664 	return x2apic_supported();
665 }
666 
667 /*
668  * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping()
669  * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the
670  * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg().
671  *
672  * Note: for a VM on Hyper-V, the I/O-APIC is the only device which
673  * (logically) generates MSIs directly to the system APIC irq domain.
674  * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the
675  * pci-hyperv host bridge.
676  *
677  * Note: for a Hyper-V root partition, this will always return false.
678  * The hypervisor doesn't expose these HYPERV_CPUID_VIRT_STACK_* cpuids by
679  * default, they are implemented as intercepts by the Windows Hyper-V stack.
680  * Even a nested root partition (L2 root) will not get them because the
681  * nested (L1) hypervisor filters them out.
682  */
ms_hyperv_msi_ext_dest_id(void)683 static bool __init ms_hyperv_msi_ext_dest_id(void)
684 {
685 	u32 eax;
686 
687 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE);
688 	if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE)
689 		return false;
690 
691 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES);
692 	return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE;
693 }
694 
695 #ifdef CONFIG_AMD_MEM_ENCRYPT
hv_sev_es_hcall_prepare(struct ghcb * ghcb,struct pt_regs * regs)696 static void hv_sev_es_hcall_prepare(struct ghcb *ghcb, struct pt_regs *regs)
697 {
698 	/* RAX and CPL are already in the GHCB */
699 	ghcb_set_rcx(ghcb, regs->cx);
700 	ghcb_set_rdx(ghcb, regs->dx);
701 	ghcb_set_r8(ghcb, regs->r8);
702 }
703 
hv_sev_es_hcall_finish(struct ghcb * ghcb,struct pt_regs * regs)704 static bool hv_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs)
705 {
706 	/* No checking of the return state needed */
707 	return true;
708 }
709 #endif
710 
711 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = {
712 	.name			= "Microsoft Hyper-V",
713 	.detect			= ms_hyperv_platform,
714 	.type			= X86_HYPER_MS_HYPERV,
715 	.init.x2apic_available	= ms_hyperv_x2apic_available,
716 	.init.msi_ext_dest_id	= ms_hyperv_msi_ext_dest_id,
717 	.init.init_platform	= ms_hyperv_init_platform,
718 	.init.guest_late_init	= ms_hyperv_late_init,
719 #ifdef CONFIG_AMD_MEM_ENCRYPT
720 	.runtime.sev_es_hcall_prepare = hv_sev_es_hcall_prepare,
721 	.runtime.sev_es_hcall_finish = hv_sev_es_hcall_finish,
722 #endif
723 };
724