xref: /openbmc/linux/arch/x86/kernel/cpu/mshyperv.c (revision 65844828)
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 struct ms_hyperv_info ms_hyperv;
41 
42 #if IS_ENABLED(CONFIG_HYPERV)
43 static void (*vmbus_handler)(void);
44 static void (*hv_stimer0_handler)(void);
45 static void (*hv_kexec_handler)(void);
46 static void (*hv_crash_handler)(struct pt_regs *regs);
47 
48 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)
49 {
50 	struct pt_regs *old_regs = set_irq_regs(regs);
51 
52 	inc_irq_stat(irq_hv_callback_count);
53 	if (vmbus_handler)
54 		vmbus_handler();
55 
56 	if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
57 		ack_APIC_irq();
58 
59 	set_irq_regs(old_regs);
60 }
61 
62 void hv_setup_vmbus_handler(void (*handler)(void))
63 {
64 	vmbus_handler = handler;
65 }
66 
67 void hv_remove_vmbus_handler(void)
68 {
69 	/* We have no way to deallocate the interrupt gate */
70 	vmbus_handler = NULL;
71 }
72 
73 /*
74  * Routines to do per-architecture handling of stimer0
75  * interrupts when in Direct Mode
76  */
77 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)
78 {
79 	struct pt_regs *old_regs = set_irq_regs(regs);
80 
81 	inc_irq_stat(hyperv_stimer0_count);
82 	if (hv_stimer0_handler)
83 		hv_stimer0_handler();
84 	add_interrupt_randomness(HYPERV_STIMER0_VECTOR);
85 	ack_APIC_irq();
86 
87 	set_irq_regs(old_regs);
88 }
89 
90 /* For x86/x64, override weak placeholders in hyperv_timer.c */
91 void hv_setup_stimer0_handler(void (*handler)(void))
92 {
93 	hv_stimer0_handler = handler;
94 }
95 
96 void hv_remove_stimer0_handler(void)
97 {
98 	/* We have no way to deallocate the interrupt gate */
99 	hv_stimer0_handler = NULL;
100 }
101 
102 void hv_setup_kexec_handler(void (*handler)(void))
103 {
104 	hv_kexec_handler = handler;
105 }
106 
107 void hv_remove_kexec_handler(void)
108 {
109 	hv_kexec_handler = NULL;
110 }
111 
112 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs))
113 {
114 	hv_crash_handler = handler;
115 }
116 
117 void hv_remove_crash_handler(void)
118 {
119 	hv_crash_handler = NULL;
120 }
121 
122 #ifdef CONFIG_KEXEC_CORE
123 static void hv_machine_shutdown(void)
124 {
125 	if (kexec_in_progress && hv_kexec_handler)
126 		hv_kexec_handler();
127 
128 	/*
129 	 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor
130 	 * corrupts the old VP Assist Pages and can crash the kexec kernel.
131 	 */
132 	if (kexec_in_progress && hyperv_init_cpuhp > 0)
133 		cpuhp_remove_state(hyperv_init_cpuhp);
134 
135 	/* The function calls stop_other_cpus(). */
136 	native_machine_shutdown();
137 
138 	/* Disable the hypercall page when there is only 1 active CPU. */
139 	if (kexec_in_progress)
140 		hyperv_cleanup();
141 }
142 
143 static void hv_machine_crash_shutdown(struct pt_regs *regs)
144 {
145 	if (hv_crash_handler)
146 		hv_crash_handler(regs);
147 
148 	/* The function calls crash_smp_send_stop(). */
149 	native_machine_crash_shutdown(regs);
150 
151 	/* Disable the hypercall page when there is only 1 active CPU. */
152 	hyperv_cleanup();
153 }
154 #endif /* CONFIG_KEXEC_CORE */
155 #endif /* CONFIG_HYPERV */
156 
157 static uint32_t  __init ms_hyperv_platform(void)
158 {
159 	u32 eax;
160 	u32 hyp_signature[3];
161 
162 	if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
163 		return 0;
164 
165 	cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS,
166 	      &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]);
167 
168 	if (eax < HYPERV_CPUID_MIN || eax > HYPERV_CPUID_MAX ||
169 	    memcmp("Microsoft Hv", hyp_signature, 12))
170 		return 0;
171 
172 	/* HYPERCALL and VP_INDEX MSRs are mandatory for all features. */
173 	eax = cpuid_eax(HYPERV_CPUID_FEATURES);
174 	if (!(eax & HV_MSR_HYPERCALL_AVAILABLE)) {
175 		pr_warn("x86/hyperv: HYPERCALL MSR not available.\n");
176 		return 0;
177 	}
178 	if (!(eax & HV_MSR_VP_INDEX_AVAILABLE)) {
179 		pr_warn("x86/hyperv: VP_INDEX MSR not available.\n");
180 		return 0;
181 	}
182 
183 	return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
184 }
185 
186 static unsigned char hv_get_nmi_reason(void)
187 {
188 	return 0;
189 }
190 
191 #ifdef CONFIG_X86_LOCAL_APIC
192 /*
193  * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes
194  * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle
195  * unknown NMI on the first CPU which gets it.
196  */
197 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)
198 {
199 	static atomic_t nmi_cpu = ATOMIC_INIT(-1);
200 
201 	if (!unknown_nmi_panic)
202 		return NMI_DONE;
203 
204 	if (atomic_cmpxchg(&nmi_cpu, -1, raw_smp_processor_id()) != -1)
205 		return NMI_HANDLED;
206 
207 	return NMI_DONE;
208 }
209 #endif
210 
211 static unsigned long hv_get_tsc_khz(void)
212 {
213 	unsigned long freq;
214 
215 	rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
216 
217 	return freq / 1000;
218 }
219 
220 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
221 static void __init hv_smp_prepare_boot_cpu(void)
222 {
223 	native_smp_prepare_boot_cpu();
224 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
225 	hv_init_spinlocks();
226 #endif
227 }
228 
229 static void __init hv_smp_prepare_cpus(unsigned int max_cpus)
230 {
231 #ifdef CONFIG_X86_64
232 	int i;
233 	int ret;
234 #endif
235 
236 	native_smp_prepare_cpus(max_cpus);
237 
238 #ifdef CONFIG_X86_64
239 	for_each_present_cpu(i) {
240 		if (i == 0)
241 			continue;
242 		ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i));
243 		BUG_ON(ret);
244 	}
245 
246 	for_each_present_cpu(i) {
247 		if (i == 0)
248 			continue;
249 		ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i);
250 		BUG_ON(ret);
251 	}
252 #endif
253 }
254 #endif
255 
256 static void __init ms_hyperv_init_platform(void)
257 {
258 	int hv_max_functions_eax;
259 	int hv_host_info_eax;
260 	int hv_host_info_ebx;
261 	int hv_host_info_ecx;
262 	int hv_host_info_edx;
263 
264 #ifdef CONFIG_PARAVIRT
265 	pv_info.name = "Hyper-V";
266 #endif
267 
268 	/*
269 	 * Extract the features and hints
270 	 */
271 	ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
272 	ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES);
273 	ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES);
274 	ms_hyperv.hints    = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO);
275 
276 	hv_max_functions_eax = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
277 
278 	pr_info("Hyper-V: privilege flags low 0x%x, high 0x%x, hints 0x%x, misc 0x%x\n",
279 		ms_hyperv.features, ms_hyperv.priv_high, ms_hyperv.hints,
280 		ms_hyperv.misc_features);
281 
282 	ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
283 	ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
284 
285 	pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
286 		 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
287 
288 	/*
289 	 * Check CPU management privilege.
290 	 *
291 	 * To mirror what Windows does we should extract CPU management
292 	 * features and use the ReservedIdentityBit to detect if Linux is the
293 	 * root partition. But that requires negotiating CPU management
294 	 * interface (a process to be finalized).
295 	 *
296 	 * For now, use the privilege flag as the indicator for running as
297 	 * root.
298 	 */
299 	if (cpuid_ebx(HYPERV_CPUID_FEATURES) & HV_CPU_MANAGEMENT) {
300 		hv_root_partition = true;
301 		pr_info("Hyper-V: running as root partition\n");
302 	}
303 
304 	/*
305 	 * Extract host information.
306 	 */
307 	if (hv_max_functions_eax >= HYPERV_CPUID_VERSION) {
308 		hv_host_info_eax = cpuid_eax(HYPERV_CPUID_VERSION);
309 		hv_host_info_ebx = cpuid_ebx(HYPERV_CPUID_VERSION);
310 		hv_host_info_ecx = cpuid_ecx(HYPERV_CPUID_VERSION);
311 		hv_host_info_edx = cpuid_edx(HYPERV_CPUID_VERSION);
312 
313 		pr_info("Hyper-V: Host Build %d.%d.%d.%d-%d-%d\n",
314 			hv_host_info_ebx >> 16, hv_host_info_ebx & 0xFFFF,
315 			hv_host_info_eax, hv_host_info_edx & 0xFFFFFF,
316 			hv_host_info_ecx, hv_host_info_edx >> 24);
317 	}
318 
319 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
320 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
321 		x86_platform.calibrate_tsc = hv_get_tsc_khz;
322 		x86_platform.calibrate_cpu = hv_get_tsc_khz;
323 	}
324 
325 	if (ms_hyperv.priv_high & HV_ISOLATION) {
326 		ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG);
327 		ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG);
328 		ms_hyperv.shared_gpa_boundary =
329 			BIT_ULL(ms_hyperv.shared_gpa_boundary_bits);
330 
331 		pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n",
332 			ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b);
333 
334 		if (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP) {
335 			static_branch_enable(&isolation_type_snp);
336 #ifdef CONFIG_SWIOTLB
337 			swiotlb_unencrypted_base = ms_hyperv.shared_gpa_boundary;
338 #endif
339 		}
340 		/* Isolation VMs are unenlightened SEV-based VMs, thus this check: */
341 		if (IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT)) {
342 			if (hv_get_isolation_type() != HV_ISOLATION_TYPE_NONE)
343 				cc_set_vendor(CC_VENDOR_HYPERV);
344 		}
345 	}
346 
347 	if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) {
348 		ms_hyperv.nested_features =
349 			cpuid_eax(HYPERV_CPUID_NESTED_FEATURES);
350 		pr_info("Hyper-V: Nested features: 0x%x\n",
351 			ms_hyperv.nested_features);
352 	}
353 
354 #ifdef CONFIG_X86_LOCAL_APIC
355 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
356 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
357 		/*
358 		 * Get the APIC frequency.
359 		 */
360 		u64	hv_lapic_frequency;
361 
362 		rdmsrl(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency);
363 		hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ);
364 		lapic_timer_period = hv_lapic_frequency;
365 		pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
366 			lapic_timer_period);
367 	}
368 
369 	register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST,
370 			     "hv_nmi_unknown");
371 #endif
372 
373 #ifdef CONFIG_X86_IO_APIC
374 	no_timer_check = 1;
375 #endif
376 
377 #if IS_ENABLED(CONFIG_HYPERV) && defined(CONFIG_KEXEC_CORE)
378 	machine_ops.shutdown = hv_machine_shutdown;
379 	machine_ops.crash_shutdown = hv_machine_crash_shutdown;
380 #endif
381 	if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) {
382 		/*
383 		 * Writing to synthetic MSR 0x40000118 updates/changes the
384 		 * guest visible CPUIDs. Setting bit 0 of this MSR  enables
385 		 * guests to report invariant TSC feature through CPUID
386 		 * instruction, CPUID 0x800000007/EDX, bit 8. See code in
387 		 * early_init_intel() where this bit is examined. The
388 		 * setting of this MSR bit should happen before init_intel()
389 		 * is called.
390 		 */
391 		wrmsrl(HV_X64_MSR_TSC_INVARIANT_CONTROL, 0x1);
392 		setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
393 	}
394 
395 	/*
396 	 * Generation 2 instances don't support reading the NMI status from
397 	 * 0x61 port.
398 	 */
399 	if (efi_enabled(EFI_BOOT))
400 		x86_platform.get_nmi_reason = hv_get_nmi_reason;
401 
402 	/*
403 	 * Hyper-V VMs have a PIT emulation quirk such that zeroing the
404 	 * counter register during PIT shutdown restarts the PIT. So it
405 	 * continues to interrupt @18.2 HZ. Setting i8253_clear_counter
406 	 * to false tells pit_shutdown() not to zero the counter so that
407 	 * the PIT really is shutdown. Generation 2 VMs don't have a PIT,
408 	 * and setting this value has no effect.
409 	 */
410 	i8253_clear_counter_on_shutdown = false;
411 
412 #if IS_ENABLED(CONFIG_HYPERV)
413 	/*
414 	 * Setup the hook to get control post apic initialization.
415 	 */
416 	x86_platform.apic_post_init = hyperv_init;
417 	hyperv_setup_mmu_ops();
418 	/* Setup the IDT for hypervisor callback */
419 	alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, asm_sysvec_hyperv_callback);
420 
421 	/* Setup the IDT for reenlightenment notifications */
422 	if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) {
423 		alloc_intr_gate(HYPERV_REENLIGHTENMENT_VECTOR,
424 				asm_sysvec_hyperv_reenlightenment);
425 	}
426 
427 	/* Setup the IDT for stimer0 */
428 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) {
429 		alloc_intr_gate(HYPERV_STIMER0_VECTOR,
430 				asm_sysvec_hyperv_stimer0);
431 	}
432 
433 # ifdef CONFIG_SMP
434 	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
435 	if (hv_root_partition)
436 		smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus;
437 # endif
438 
439 	/*
440 	 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
441 	 * set x2apic destination mode to physical mode when x2apic is available
442 	 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
443 	 * have 8-bit APIC id.
444 	 */
445 # ifdef CONFIG_X86_X2APIC
446 	if (x2apic_supported())
447 		x2apic_phys = 1;
448 # endif
449 
450 	/* Register Hyper-V specific clocksource */
451 	hv_init_clocksource();
452 #endif
453 	/*
454 	 * TSC should be marked as unstable only after Hyper-V
455 	 * clocksource has been initialized. This ensures that the
456 	 * stability of the sched_clock is not altered.
457 	 */
458 	if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
459 		mark_tsc_unstable("running on Hyper-V");
460 
461 	hardlockup_detector_disable();
462 }
463 
464 static bool __init ms_hyperv_x2apic_available(void)
465 {
466 	return x2apic_supported();
467 }
468 
469 /*
470  * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping()
471  * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the
472  * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg().
473  *
474  * Note: for a VM on Hyper-V, the I/O-APIC is the only device which
475  * (logically) generates MSIs directly to the system APIC irq domain.
476  * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the
477  * pci-hyperv host bridge.
478  *
479  * Note: for a Hyper-V root partition, this will always return false.
480  * The hypervisor doesn't expose these HYPERV_CPUID_VIRT_STACK_* cpuids by
481  * default, they are implemented as intercepts by the Windows Hyper-V stack.
482  * Even a nested root partition (L2 root) will not get them because the
483  * nested (L1) hypervisor filters them out.
484  */
485 static bool __init ms_hyperv_msi_ext_dest_id(void)
486 {
487 	u32 eax;
488 
489 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE);
490 	if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE)
491 		return false;
492 
493 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES);
494 	return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE;
495 }
496 
497 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = {
498 	.name			= "Microsoft Hyper-V",
499 	.detect			= ms_hyperv_platform,
500 	.type			= X86_HYPER_MS_HYPERV,
501 	.init.x2apic_available	= ms_hyperv_x2apic_available,
502 	.init.msi_ext_dest_id	= ms_hyperv_msi_ext_dest_id,
503 	.init.init_platform	= ms_hyperv_init_platform,
504 };
505