xref: /openbmc/linux/arch/x86/kernel/cpu/mshyperv.c (revision 844f5ed5)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * HyperV  Detection code.
4  *
5  * Copyright (C) 2010, Novell, Inc.
6  * Author : K. Y. Srinivasan <ksrinivasan@novell.com>
7  */
8 
9 #include <linux/types.h>
10 #include <linux/time.h>
11 #include <linux/clocksource.h>
12 #include <linux/init.h>
13 #include <linux/export.h>
14 #include <linux/hardirq.h>
15 #include <linux/efi.h>
16 #include <linux/interrupt.h>
17 #include <linux/irq.h>
18 #include <linux/kexec.h>
19 #include <linux/i8253.h>
20 #include <linux/random.h>
21 #include <linux/swiotlb.h>
22 #include <asm/processor.h>
23 #include <asm/hypervisor.h>
24 #include <asm/hyperv-tlfs.h>
25 #include <asm/mshyperv.h>
26 #include <asm/desc.h>
27 #include <asm/idtentry.h>
28 #include <asm/irq_regs.h>
29 #include <asm/i8259.h>
30 #include <asm/apic.h>
31 #include <asm/timer.h>
32 #include <asm/reboot.h>
33 #include <asm/nmi.h>
34 #include <clocksource/hyperv_timer.h>
35 #include <asm/numa.h>
36 #include <asm/coco.h>
37 
38 /* Is Linux running as the root partition? */
39 bool hv_root_partition;
40 /* Is Linux running on nested Microsoft Hypervisor */
41 bool hv_nested;
42 struct ms_hyperv_info ms_hyperv;
43 
44 #if IS_ENABLED(CONFIG_HYPERV)
45 static inline unsigned int hv_get_nested_reg(unsigned int reg)
46 {
47 	if (hv_is_sint_reg(reg))
48 		return reg - HV_REGISTER_SINT0 + HV_REGISTER_NESTED_SINT0;
49 
50 	switch (reg) {
51 	case HV_REGISTER_SIMP:
52 		return HV_REGISTER_NESTED_SIMP;
53 	case HV_REGISTER_SIEFP:
54 		return HV_REGISTER_NESTED_SIEFP;
55 	case HV_REGISTER_SVERSION:
56 		return HV_REGISTER_NESTED_SVERSION;
57 	case HV_REGISTER_SCONTROL:
58 		return HV_REGISTER_NESTED_SCONTROL;
59 	case HV_REGISTER_EOM:
60 		return HV_REGISTER_NESTED_EOM;
61 	default:
62 		return reg;
63 	}
64 }
65 
66 u64 hv_get_non_nested_register(unsigned int reg)
67 {
68 	u64 value;
69 
70 	if (hv_is_synic_reg(reg) && hv_isolation_type_snp())
71 		hv_ghcb_msr_read(reg, &value);
72 	else
73 		rdmsrl(reg, value);
74 	return value;
75 }
76 EXPORT_SYMBOL_GPL(hv_get_non_nested_register);
77 
78 void hv_set_non_nested_register(unsigned int reg, u64 value)
79 {
80 	if (hv_is_synic_reg(reg) && hv_isolation_type_snp()) {
81 		hv_ghcb_msr_write(reg, value);
82 
83 		/* Write proxy bit via wrmsl instruction */
84 		if (hv_is_sint_reg(reg))
85 			wrmsrl(reg, value | 1 << 20);
86 	} else {
87 		wrmsrl(reg, value);
88 	}
89 }
90 EXPORT_SYMBOL_GPL(hv_set_non_nested_register);
91 
92 u64 hv_get_register(unsigned int reg)
93 {
94 	if (hv_nested)
95 		reg = hv_get_nested_reg(reg);
96 
97 	return hv_get_non_nested_register(reg);
98 }
99 EXPORT_SYMBOL_GPL(hv_get_register);
100 
101 void hv_set_register(unsigned int reg, u64 value)
102 {
103 	if (hv_nested)
104 		reg = hv_get_nested_reg(reg);
105 
106 	hv_set_non_nested_register(reg, value);
107 }
108 EXPORT_SYMBOL_GPL(hv_set_register);
109 
110 static void (*vmbus_handler)(void);
111 static void (*hv_stimer0_handler)(void);
112 static void (*hv_kexec_handler)(void);
113 static void (*hv_crash_handler)(struct pt_regs *regs);
114 
115 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)
116 {
117 	struct pt_regs *old_regs = set_irq_regs(regs);
118 
119 	inc_irq_stat(irq_hv_callback_count);
120 	if (vmbus_handler)
121 		vmbus_handler();
122 
123 	if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
124 		ack_APIC_irq();
125 
126 	set_irq_regs(old_regs);
127 }
128 
129 void hv_setup_vmbus_handler(void (*handler)(void))
130 {
131 	vmbus_handler = handler;
132 }
133 
134 void hv_remove_vmbus_handler(void)
135 {
136 	/* We have no way to deallocate the interrupt gate */
137 	vmbus_handler = NULL;
138 }
139 
140 /*
141  * Routines to do per-architecture handling of stimer0
142  * interrupts when in Direct Mode
143  */
144 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)
145 {
146 	struct pt_regs *old_regs = set_irq_regs(regs);
147 
148 	inc_irq_stat(hyperv_stimer0_count);
149 	if (hv_stimer0_handler)
150 		hv_stimer0_handler();
151 	add_interrupt_randomness(HYPERV_STIMER0_VECTOR);
152 	ack_APIC_irq();
153 
154 	set_irq_regs(old_regs);
155 }
156 
157 /* For x86/x64, override weak placeholders in hyperv_timer.c */
158 void hv_setup_stimer0_handler(void (*handler)(void))
159 {
160 	hv_stimer0_handler = handler;
161 }
162 
163 void hv_remove_stimer0_handler(void)
164 {
165 	/* We have no way to deallocate the interrupt gate */
166 	hv_stimer0_handler = NULL;
167 }
168 
169 void hv_setup_kexec_handler(void (*handler)(void))
170 {
171 	hv_kexec_handler = handler;
172 }
173 
174 void hv_remove_kexec_handler(void)
175 {
176 	hv_kexec_handler = NULL;
177 }
178 
179 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs))
180 {
181 	hv_crash_handler = handler;
182 }
183 
184 void hv_remove_crash_handler(void)
185 {
186 	hv_crash_handler = NULL;
187 }
188 
189 #ifdef CONFIG_KEXEC_CORE
190 static void hv_machine_shutdown(void)
191 {
192 	if (kexec_in_progress && hv_kexec_handler)
193 		hv_kexec_handler();
194 
195 	/*
196 	 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor
197 	 * corrupts the old VP Assist Pages and can crash the kexec kernel.
198 	 */
199 	if (kexec_in_progress && hyperv_init_cpuhp > 0)
200 		cpuhp_remove_state(hyperv_init_cpuhp);
201 
202 	/* The function calls stop_other_cpus(). */
203 	native_machine_shutdown();
204 
205 	/* Disable the hypercall page when there is only 1 active CPU. */
206 	if (kexec_in_progress)
207 		hyperv_cleanup();
208 }
209 
210 static void hv_machine_crash_shutdown(struct pt_regs *regs)
211 {
212 	if (hv_crash_handler)
213 		hv_crash_handler(regs);
214 
215 	/* The function calls crash_smp_send_stop(). */
216 	native_machine_crash_shutdown(regs);
217 
218 	/* Disable the hypercall page when there is only 1 active CPU. */
219 	hyperv_cleanup();
220 }
221 #endif /* CONFIG_KEXEC_CORE */
222 #endif /* CONFIG_HYPERV */
223 
224 static uint32_t  __init ms_hyperv_platform(void)
225 {
226 	u32 eax;
227 	u32 hyp_signature[3];
228 
229 	if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
230 		return 0;
231 
232 	cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS,
233 	      &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]);
234 
235 	if (eax < HYPERV_CPUID_MIN || eax > HYPERV_CPUID_MAX ||
236 	    memcmp("Microsoft Hv", hyp_signature, 12))
237 		return 0;
238 
239 	/* HYPERCALL and VP_INDEX MSRs are mandatory for all features. */
240 	eax = cpuid_eax(HYPERV_CPUID_FEATURES);
241 	if (!(eax & HV_MSR_HYPERCALL_AVAILABLE)) {
242 		pr_warn("x86/hyperv: HYPERCALL MSR not available.\n");
243 		return 0;
244 	}
245 	if (!(eax & HV_MSR_VP_INDEX_AVAILABLE)) {
246 		pr_warn("x86/hyperv: VP_INDEX MSR not available.\n");
247 		return 0;
248 	}
249 
250 	return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
251 }
252 
253 static unsigned char hv_get_nmi_reason(void)
254 {
255 	return 0;
256 }
257 
258 #ifdef CONFIG_X86_LOCAL_APIC
259 /*
260  * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes
261  * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle
262  * unknown NMI on the first CPU which gets it.
263  */
264 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)
265 {
266 	static atomic_t nmi_cpu = ATOMIC_INIT(-1);
267 
268 	if (!unknown_nmi_panic)
269 		return NMI_DONE;
270 
271 	if (atomic_cmpxchg(&nmi_cpu, -1, raw_smp_processor_id()) != -1)
272 		return NMI_HANDLED;
273 
274 	return NMI_DONE;
275 }
276 #endif
277 
278 static unsigned long hv_get_tsc_khz(void)
279 {
280 	unsigned long freq;
281 
282 	rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
283 
284 	return freq / 1000;
285 }
286 
287 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
288 static void __init hv_smp_prepare_boot_cpu(void)
289 {
290 	native_smp_prepare_boot_cpu();
291 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
292 	hv_init_spinlocks();
293 #endif
294 }
295 
296 static void __init hv_smp_prepare_cpus(unsigned int max_cpus)
297 {
298 #ifdef CONFIG_X86_64
299 	int i;
300 	int ret;
301 #endif
302 
303 	native_smp_prepare_cpus(max_cpus);
304 
305 #ifdef CONFIG_X86_64
306 	for_each_present_cpu(i) {
307 		if (i == 0)
308 			continue;
309 		ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i));
310 		BUG_ON(ret);
311 	}
312 
313 	for_each_present_cpu(i) {
314 		if (i == 0)
315 			continue;
316 		ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i);
317 		BUG_ON(ret);
318 	}
319 #endif
320 }
321 #endif
322 
323 static void __init ms_hyperv_init_platform(void)
324 {
325 	int hv_max_functions_eax;
326 	int hv_host_info_eax;
327 	int hv_host_info_ebx;
328 	int hv_host_info_ecx;
329 	int hv_host_info_edx;
330 
331 #ifdef CONFIG_PARAVIRT
332 	pv_info.name = "Hyper-V";
333 #endif
334 
335 	/*
336 	 * Extract the features and hints
337 	 */
338 	ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
339 	ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES);
340 	ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES);
341 	ms_hyperv.hints    = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO);
342 
343 	hv_max_functions_eax = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
344 
345 	pr_info("Hyper-V: privilege flags low 0x%x, high 0x%x, hints 0x%x, misc 0x%x\n",
346 		ms_hyperv.features, ms_hyperv.priv_high, ms_hyperv.hints,
347 		ms_hyperv.misc_features);
348 
349 	ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
350 	ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
351 
352 	pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
353 		 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
354 
355 	/*
356 	 * Check CPU management privilege.
357 	 *
358 	 * To mirror what Windows does we should extract CPU management
359 	 * features and use the ReservedIdentityBit to detect if Linux is the
360 	 * root partition. But that requires negotiating CPU management
361 	 * interface (a process to be finalized).
362 	 *
363 	 * For now, use the privilege flag as the indicator for running as
364 	 * root.
365 	 */
366 	if (cpuid_ebx(HYPERV_CPUID_FEATURES) & HV_CPU_MANAGEMENT) {
367 		hv_root_partition = true;
368 		pr_info("Hyper-V: running as root partition\n");
369 	}
370 
371 	if (ms_hyperv.hints & HV_X64_HYPERV_NESTED) {
372 		hv_nested = true;
373 		pr_info("Hyper-V: running on a nested hypervisor\n");
374 	}
375 
376 	/*
377 	 * Extract host information.
378 	 */
379 	if (hv_max_functions_eax >= HYPERV_CPUID_VERSION) {
380 		hv_host_info_eax = cpuid_eax(HYPERV_CPUID_VERSION);
381 		hv_host_info_ebx = cpuid_ebx(HYPERV_CPUID_VERSION);
382 		hv_host_info_ecx = cpuid_ecx(HYPERV_CPUID_VERSION);
383 		hv_host_info_edx = cpuid_edx(HYPERV_CPUID_VERSION);
384 
385 		pr_info("Hyper-V: Host Build %d.%d.%d.%d-%d-%d\n",
386 			hv_host_info_ebx >> 16, hv_host_info_ebx & 0xFFFF,
387 			hv_host_info_eax, hv_host_info_edx & 0xFFFFFF,
388 			hv_host_info_ecx, hv_host_info_edx >> 24);
389 	}
390 
391 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
392 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
393 		x86_platform.calibrate_tsc = hv_get_tsc_khz;
394 		x86_platform.calibrate_cpu = hv_get_tsc_khz;
395 	}
396 
397 	if (ms_hyperv.priv_high & HV_ISOLATION) {
398 		ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG);
399 		ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG);
400 		ms_hyperv.shared_gpa_boundary =
401 			BIT_ULL(ms_hyperv.shared_gpa_boundary_bits);
402 
403 		pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n",
404 			ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b);
405 
406 		if (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP) {
407 			static_branch_enable(&isolation_type_snp);
408 #ifdef CONFIG_SWIOTLB
409 			swiotlb_unencrypted_base = ms_hyperv.shared_gpa_boundary;
410 #endif
411 		}
412 		/* Isolation VMs are unenlightened SEV-based VMs, thus this check: */
413 		if (IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT)) {
414 			if (hv_get_isolation_type() != HV_ISOLATION_TYPE_NONE)
415 				cc_set_vendor(CC_VENDOR_HYPERV);
416 		}
417 	}
418 
419 	if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) {
420 		ms_hyperv.nested_features =
421 			cpuid_eax(HYPERV_CPUID_NESTED_FEATURES);
422 		pr_info("Hyper-V: Nested features: 0x%x\n",
423 			ms_hyperv.nested_features);
424 	}
425 
426 #ifdef CONFIG_X86_LOCAL_APIC
427 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
428 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
429 		/*
430 		 * Get the APIC frequency.
431 		 */
432 		u64	hv_lapic_frequency;
433 
434 		rdmsrl(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency);
435 		hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ);
436 		lapic_timer_period = hv_lapic_frequency;
437 		pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
438 			lapic_timer_period);
439 	}
440 
441 	register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST,
442 			     "hv_nmi_unknown");
443 #endif
444 
445 #ifdef CONFIG_X86_IO_APIC
446 	no_timer_check = 1;
447 #endif
448 
449 #if IS_ENABLED(CONFIG_HYPERV) && defined(CONFIG_KEXEC_CORE)
450 	machine_ops.shutdown = hv_machine_shutdown;
451 	machine_ops.crash_shutdown = hv_machine_crash_shutdown;
452 #endif
453 	if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) {
454 		/*
455 		 * Writing to synthetic MSR 0x40000118 updates/changes the
456 		 * guest visible CPUIDs. Setting bit 0 of this MSR  enables
457 		 * guests to report invariant TSC feature through CPUID
458 		 * instruction, CPUID 0x800000007/EDX, bit 8. See code in
459 		 * early_init_intel() where this bit is examined. The
460 		 * setting of this MSR bit should happen before init_intel()
461 		 * is called.
462 		 */
463 		wrmsrl(HV_X64_MSR_TSC_INVARIANT_CONTROL, HV_EXPOSE_INVARIANT_TSC);
464 		setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
465 	}
466 
467 	/*
468 	 * Generation 2 instances don't support reading the NMI status from
469 	 * 0x61 port.
470 	 */
471 	if (efi_enabled(EFI_BOOT))
472 		x86_platform.get_nmi_reason = hv_get_nmi_reason;
473 
474 	/*
475 	 * Hyper-V VMs have a PIT emulation quirk such that zeroing the
476 	 * counter register during PIT shutdown restarts the PIT. So it
477 	 * continues to interrupt @18.2 HZ. Setting i8253_clear_counter
478 	 * to false tells pit_shutdown() not to zero the counter so that
479 	 * the PIT really is shutdown. Generation 2 VMs don't have a PIT,
480 	 * and setting this value has no effect.
481 	 */
482 	i8253_clear_counter_on_shutdown = false;
483 
484 #if IS_ENABLED(CONFIG_HYPERV)
485 	/*
486 	 * Setup the hook to get control post apic initialization.
487 	 */
488 	x86_platform.apic_post_init = hyperv_init;
489 	hyperv_setup_mmu_ops();
490 	/* Setup the IDT for hypervisor callback */
491 	alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, asm_sysvec_hyperv_callback);
492 
493 	/* Setup the IDT for reenlightenment notifications */
494 	if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) {
495 		alloc_intr_gate(HYPERV_REENLIGHTENMENT_VECTOR,
496 				asm_sysvec_hyperv_reenlightenment);
497 	}
498 
499 	/* Setup the IDT for stimer0 */
500 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) {
501 		alloc_intr_gate(HYPERV_STIMER0_VECTOR,
502 				asm_sysvec_hyperv_stimer0);
503 	}
504 
505 # ifdef CONFIG_SMP
506 	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
507 	if (hv_root_partition)
508 		smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus;
509 # endif
510 
511 	/*
512 	 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
513 	 * set x2apic destination mode to physical mode when x2apic is available
514 	 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
515 	 * have 8-bit APIC id.
516 	 */
517 # ifdef CONFIG_X86_X2APIC
518 	if (x2apic_supported())
519 		x2apic_phys = 1;
520 # endif
521 
522 	/* Register Hyper-V specific clocksource */
523 	hv_init_clocksource();
524 #endif
525 	/*
526 	 * TSC should be marked as unstable only after Hyper-V
527 	 * clocksource has been initialized. This ensures that the
528 	 * stability of the sched_clock is not altered.
529 	 */
530 	if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
531 		mark_tsc_unstable("running on Hyper-V");
532 
533 	hardlockup_detector_disable();
534 }
535 
536 static bool __init ms_hyperv_x2apic_available(void)
537 {
538 	return x2apic_supported();
539 }
540 
541 /*
542  * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping()
543  * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the
544  * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg().
545  *
546  * Note: for a VM on Hyper-V, the I/O-APIC is the only device which
547  * (logically) generates MSIs directly to the system APIC irq domain.
548  * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the
549  * pci-hyperv host bridge.
550  *
551  * Note: for a Hyper-V root partition, this will always return false.
552  * The hypervisor doesn't expose these HYPERV_CPUID_VIRT_STACK_* cpuids by
553  * default, they are implemented as intercepts by the Windows Hyper-V stack.
554  * Even a nested root partition (L2 root) will not get them because the
555  * nested (L1) hypervisor filters them out.
556  */
557 static bool __init ms_hyperv_msi_ext_dest_id(void)
558 {
559 	u32 eax;
560 
561 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE);
562 	if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE)
563 		return false;
564 
565 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES);
566 	return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE;
567 }
568 
569 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = {
570 	.name			= "Microsoft Hyper-V",
571 	.detect			= ms_hyperv_platform,
572 	.type			= X86_HYPER_MS_HYPERV,
573 	.init.x2apic_available	= ms_hyperv_x2apic_available,
574 	.init.msi_ext_dest_id	= ms_hyperv_msi_ext_dest_id,
575 	.init.init_platform	= ms_hyperv_init_platform,
576 };
577