Loading...
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 <asm/processor.h>
22#include <asm/hypervisor.h>
23#include <asm/hyperv-tlfs.h>
24#include <asm/mshyperv.h>
25#include <asm/desc.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
34struct ms_hyperv_info ms_hyperv;
35EXPORT_SYMBOL_GPL(ms_hyperv);
36
37#if IS_ENABLED(CONFIG_HYPERV)
38static void (*vmbus_handler)(void);
39static void (*hv_stimer0_handler)(void);
40static void (*hv_kexec_handler)(void);
41static void (*hv_crash_handler)(struct pt_regs *regs);
42
43__visible void __irq_entry hyperv_vector_handler(struct pt_regs *regs)
44{
45 struct pt_regs *old_regs = set_irq_regs(regs);
46
47 entering_irq();
48 inc_irq_stat(irq_hv_callback_count);
49 if (vmbus_handler)
50 vmbus_handler();
51
52 if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
53 ack_APIC_irq();
54
55 exiting_irq();
56 set_irq_regs(old_regs);
57}
58
59void hv_setup_vmbus_irq(void (*handler)(void))
60{
61 vmbus_handler = handler;
62}
63
64void hv_remove_vmbus_irq(void)
65{
66 /* We have no way to deallocate the interrupt gate */
67 vmbus_handler = NULL;
68}
69EXPORT_SYMBOL_GPL(hv_setup_vmbus_irq);
70EXPORT_SYMBOL_GPL(hv_remove_vmbus_irq);
71
72/*
73 * Routines to do per-architecture handling of stimer0
74 * interrupts when in Direct Mode
75 */
76
77__visible void __irq_entry hv_stimer0_vector_handler(struct pt_regs *regs)
78{
79 struct pt_regs *old_regs = set_irq_regs(regs);
80
81 entering_irq();
82 inc_irq_stat(hyperv_stimer0_count);
83 if (hv_stimer0_handler)
84 hv_stimer0_handler();
85 add_interrupt_randomness(HYPERV_STIMER0_VECTOR, 0);
86 ack_APIC_irq();
87
88 exiting_irq();
89 set_irq_regs(old_regs);
90}
91
92int hv_setup_stimer0_irq(int *irq, int *vector, void (*handler)(void))
93{
94 *vector = HYPERV_STIMER0_VECTOR;
95 *irq = -1; /* Unused on x86/x64 */
96 hv_stimer0_handler = handler;
97 return 0;
98}
99EXPORT_SYMBOL_GPL(hv_setup_stimer0_irq);
100
101void hv_remove_stimer0_irq(int irq)
102{
103 /* We have no way to deallocate the interrupt gate */
104 hv_stimer0_handler = NULL;
105}
106EXPORT_SYMBOL_GPL(hv_remove_stimer0_irq);
107
108void hv_setup_kexec_handler(void (*handler)(void))
109{
110 hv_kexec_handler = handler;
111}
112EXPORT_SYMBOL_GPL(hv_setup_kexec_handler);
113
114void hv_remove_kexec_handler(void)
115{
116 hv_kexec_handler = NULL;
117}
118EXPORT_SYMBOL_GPL(hv_remove_kexec_handler);
119
120void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs))
121{
122 hv_crash_handler = handler;
123}
124EXPORT_SYMBOL_GPL(hv_setup_crash_handler);
125
126void hv_remove_crash_handler(void)
127{
128 hv_crash_handler = NULL;
129}
130EXPORT_SYMBOL_GPL(hv_remove_crash_handler);
131
132#ifdef CONFIG_KEXEC_CORE
133static void hv_machine_shutdown(void)
134{
135 if (kexec_in_progress && hv_kexec_handler)
136 hv_kexec_handler();
137 native_machine_shutdown();
138}
139
140static void hv_machine_crash_shutdown(struct pt_regs *regs)
141{
142 if (hv_crash_handler)
143 hv_crash_handler(regs);
144 native_machine_crash_shutdown(regs);
145}
146#endif /* CONFIG_KEXEC_CORE */
147#endif /* CONFIG_HYPERV */
148
149static uint32_t __init ms_hyperv_platform(void)
150{
151 u32 eax;
152 u32 hyp_signature[3];
153
154 if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
155 return 0;
156
157 cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS,
158 &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]);
159
160 if (eax >= HYPERV_CPUID_MIN &&
161 eax <= HYPERV_CPUID_MAX &&
162 !memcmp("Microsoft Hv", hyp_signature, 12))
163 return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
164
165 return 0;
166}
167
168static unsigned char hv_get_nmi_reason(void)
169{
170 return 0;
171}
172
173#ifdef CONFIG_X86_LOCAL_APIC
174/*
175 * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes
176 * it dificult to process CHANNELMSG_UNLOAD in case of crash. Handle
177 * unknown NMI on the first CPU which gets it.
178 */
179static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)
180{
181 static atomic_t nmi_cpu = ATOMIC_INIT(-1);
182
183 if (!unknown_nmi_panic)
184 return NMI_DONE;
185
186 if (atomic_cmpxchg(&nmi_cpu, -1, raw_smp_processor_id()) != -1)
187 return NMI_HANDLED;
188
189 return NMI_DONE;
190}
191#endif
192
193static unsigned long hv_get_tsc_khz(void)
194{
195 unsigned long freq;
196
197 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
198
199 return freq / 1000;
200}
201
202#if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
203static void __init hv_smp_prepare_boot_cpu(void)
204{
205 native_smp_prepare_boot_cpu();
206#if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
207 hv_init_spinlocks();
208#endif
209}
210#endif
211
212static void __init ms_hyperv_init_platform(void)
213{
214 int hv_host_info_eax;
215 int hv_host_info_ebx;
216 int hv_host_info_ecx;
217 int hv_host_info_edx;
218
219#ifdef CONFIG_PARAVIRT
220 pv_info.name = "Hyper-V";
221#endif
222
223 /*
224 * Extract the features and hints
225 */
226 ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
227 ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES);
228 ms_hyperv.hints = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO);
229
230 pr_info("Hyper-V: features 0x%x, hints 0x%x\n",
231 ms_hyperv.features, ms_hyperv.hints);
232
233 ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
234 ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
235
236 pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
237 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
238
239 /*
240 * Extract host information.
241 */
242 if (cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS) >=
243 HYPERV_CPUID_VERSION) {
244 hv_host_info_eax = cpuid_eax(HYPERV_CPUID_VERSION);
245 hv_host_info_ebx = cpuid_ebx(HYPERV_CPUID_VERSION);
246 hv_host_info_ecx = cpuid_ecx(HYPERV_CPUID_VERSION);
247 hv_host_info_edx = cpuid_edx(HYPERV_CPUID_VERSION);
248
249 pr_info("Hyper-V Host Build:%d-%d.%d-%d-%d.%d\n",
250 hv_host_info_eax, hv_host_info_ebx >> 16,
251 hv_host_info_ebx & 0xFFFF, hv_host_info_ecx,
252 hv_host_info_edx >> 24, hv_host_info_edx & 0xFFFFFF);
253 }
254
255 if (ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
256 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
257 x86_platform.calibrate_tsc = hv_get_tsc_khz;
258 x86_platform.calibrate_cpu = hv_get_tsc_khz;
259 }
260
261 if (ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED) {
262 ms_hyperv.nested_features =
263 cpuid_eax(HYPERV_CPUID_NESTED_FEATURES);
264 }
265
266#ifdef CONFIG_X86_LOCAL_APIC
267 if (ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
268 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
269 /*
270 * Get the APIC frequency.
271 */
272 u64 hv_lapic_frequency;
273
274 rdmsrl(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency);
275 hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ);
276 lapic_timer_period = hv_lapic_frequency;
277 pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
278 lapic_timer_period);
279 }
280
281 register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST,
282 "hv_nmi_unknown");
283#endif
284
285#ifdef CONFIG_X86_IO_APIC
286 no_timer_check = 1;
287#endif
288
289#if IS_ENABLED(CONFIG_HYPERV) && defined(CONFIG_KEXEC_CORE)
290 machine_ops.shutdown = hv_machine_shutdown;
291 machine_ops.crash_shutdown = hv_machine_crash_shutdown;
292#endif
293 mark_tsc_unstable("running on Hyper-V");
294
295 /*
296 * Generation 2 instances don't support reading the NMI status from
297 * 0x61 port.
298 */
299 if (efi_enabled(EFI_BOOT))
300 x86_platform.get_nmi_reason = hv_get_nmi_reason;
301
302 /*
303 * Hyper-V VMs have a PIT emulation quirk such that zeroing the
304 * counter register during PIT shutdown restarts the PIT. So it
305 * continues to interrupt @18.2 HZ. Setting i8253_clear_counter
306 * to false tells pit_shutdown() not to zero the counter so that
307 * the PIT really is shutdown. Generation 2 VMs don't have a PIT,
308 * and setting this value has no effect.
309 */
310 i8253_clear_counter_on_shutdown = false;
311
312#if IS_ENABLED(CONFIG_HYPERV)
313 /*
314 * Setup the hook to get control post apic initialization.
315 */
316 x86_platform.apic_post_init = hyperv_init;
317 hyperv_setup_mmu_ops();
318 /* Setup the IDT for hypervisor callback */
319 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, hyperv_callback_vector);
320
321 /* Setup the IDT for reenlightenment notifications */
322 if (ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT)
323 alloc_intr_gate(HYPERV_REENLIGHTENMENT_VECTOR,
324 hyperv_reenlightenment_vector);
325
326 /* Setup the IDT for stimer0 */
327 if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE)
328 alloc_intr_gate(HYPERV_STIMER0_VECTOR,
329 hv_stimer0_callback_vector);
330
331# ifdef CONFIG_SMP
332 smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
333# endif
334
335 /*
336 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
337 * set x2apic destination mode to physcial mode when x2apic is available
338 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
339 * have 8-bit APIC id.
340 */
341# ifdef CONFIG_X86_X2APIC
342 if (x2apic_supported())
343 x2apic_phys = 1;
344# endif
345
346 /* Register Hyper-V specific clocksource */
347 hv_init_clocksource();
348#endif
349}
350
351void hv_setup_sched_clock(void *sched_clock)
352{
353#ifdef CONFIG_PARAVIRT
354 pv_ops.time.sched_clock = sched_clock;
355#endif
356}
357
358const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = {
359 .name = "Microsoft Hyper-V",
360 .detect = ms_hyperv_platform,
361 .type = X86_HYPER_MS_HYPERV,
362 .init.init_platform = ms_hyperv_init_platform,
363};
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 <asm/processor.h>
22#include <asm/hypervisor.h>
23#include <asm/hyperv-tlfs.h>
24#include <asm/mshyperv.h>
25#include <asm/desc.h>
26#include <asm/idtentry.h>
27#include <asm/irq_regs.h>
28#include <asm/i8259.h>
29#include <asm/apic.h>
30#include <asm/timer.h>
31#include <asm/reboot.h>
32#include <asm/nmi.h>
33#include <clocksource/hyperv_timer.h>
34#include <asm/numa.h>
35#include <asm/svm.h>
36
37/* Is Linux running as the root partition? */
38bool hv_root_partition;
39/* Is Linux running on nested Microsoft Hypervisor */
40bool hv_nested;
41struct ms_hyperv_info ms_hyperv;
42
43/* Used in modules via hv_do_hypercall(): see arch/x86/include/asm/mshyperv.h */
44bool hyperv_paravisor_present __ro_after_init;
45EXPORT_SYMBOL_GPL(hyperv_paravisor_present);
46
47#if IS_ENABLED(CONFIG_HYPERV)
48static inline unsigned int hv_get_nested_reg(unsigned int reg)
49{
50 if (hv_is_sint_reg(reg))
51 return reg - HV_REGISTER_SINT0 + HV_REGISTER_NESTED_SINT0;
52
53 switch (reg) {
54 case HV_REGISTER_SIMP:
55 return HV_REGISTER_NESTED_SIMP;
56 case HV_REGISTER_SIEFP:
57 return HV_REGISTER_NESTED_SIEFP;
58 case HV_REGISTER_SVERSION:
59 return HV_REGISTER_NESTED_SVERSION;
60 case HV_REGISTER_SCONTROL:
61 return HV_REGISTER_NESTED_SCONTROL;
62 case HV_REGISTER_EOM:
63 return HV_REGISTER_NESTED_EOM;
64 default:
65 return reg;
66 }
67}
68
69u64 hv_get_non_nested_register(unsigned int reg)
70{
71 u64 value;
72
73 if (hv_is_synic_reg(reg) && ms_hyperv.paravisor_present)
74 hv_ivm_msr_read(reg, &value);
75 else
76 rdmsrl(reg, value);
77 return value;
78}
79EXPORT_SYMBOL_GPL(hv_get_non_nested_register);
80
81void hv_set_non_nested_register(unsigned int reg, u64 value)
82{
83 if (hv_is_synic_reg(reg) && ms_hyperv.paravisor_present) {
84 hv_ivm_msr_write(reg, value);
85
86 /* Write proxy bit via wrmsl instruction */
87 if (hv_is_sint_reg(reg))
88 wrmsrl(reg, value | 1 << 20);
89 } else {
90 wrmsrl(reg, value);
91 }
92}
93EXPORT_SYMBOL_GPL(hv_set_non_nested_register);
94
95u64 hv_get_register(unsigned int reg)
96{
97 if (hv_nested)
98 reg = hv_get_nested_reg(reg);
99
100 return hv_get_non_nested_register(reg);
101}
102EXPORT_SYMBOL_GPL(hv_get_register);
103
104void hv_set_register(unsigned int reg, u64 value)
105{
106 if (hv_nested)
107 reg = hv_get_nested_reg(reg);
108
109 hv_set_non_nested_register(reg, value);
110}
111EXPORT_SYMBOL_GPL(hv_set_register);
112
113static void (*vmbus_handler)(void);
114static void (*hv_stimer0_handler)(void);
115static void (*hv_kexec_handler)(void);
116static void (*hv_crash_handler)(struct pt_regs *regs);
117
118DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)
119{
120 struct pt_regs *old_regs = set_irq_regs(regs);
121
122 inc_irq_stat(irq_hv_callback_count);
123 if (vmbus_handler)
124 vmbus_handler();
125
126 if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
127 apic_eoi();
128
129 set_irq_regs(old_regs);
130}
131
132void hv_setup_vmbus_handler(void (*handler)(void))
133{
134 vmbus_handler = handler;
135}
136
137void hv_remove_vmbus_handler(void)
138{
139 /* We have no way to deallocate the interrupt gate */
140 vmbus_handler = NULL;
141}
142
143/*
144 * Routines to do per-architecture handling of stimer0
145 * interrupts when in Direct Mode
146 */
147DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)
148{
149 struct pt_regs *old_regs = set_irq_regs(regs);
150
151 inc_irq_stat(hyperv_stimer0_count);
152 if (hv_stimer0_handler)
153 hv_stimer0_handler();
154 add_interrupt_randomness(HYPERV_STIMER0_VECTOR);
155 apic_eoi();
156
157 set_irq_regs(old_regs);
158}
159
160/* For x86/x64, override weak placeholders in hyperv_timer.c */
161void hv_setup_stimer0_handler(void (*handler)(void))
162{
163 hv_stimer0_handler = handler;
164}
165
166void hv_remove_stimer0_handler(void)
167{
168 /* We have no way to deallocate the interrupt gate */
169 hv_stimer0_handler = NULL;
170}
171
172void hv_setup_kexec_handler(void (*handler)(void))
173{
174 hv_kexec_handler = handler;
175}
176
177void hv_remove_kexec_handler(void)
178{
179 hv_kexec_handler = NULL;
180}
181
182void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs))
183{
184 hv_crash_handler = handler;
185}
186
187void hv_remove_crash_handler(void)
188{
189 hv_crash_handler = NULL;
190}
191
192#ifdef CONFIG_KEXEC_CORE
193static void hv_machine_shutdown(void)
194{
195 if (kexec_in_progress && hv_kexec_handler)
196 hv_kexec_handler();
197
198 /*
199 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor
200 * corrupts the old VP Assist Pages and can crash the kexec kernel.
201 */
202 if (kexec_in_progress && hyperv_init_cpuhp > 0)
203 cpuhp_remove_state(hyperv_init_cpuhp);
204
205 /* The function calls stop_other_cpus(). */
206 native_machine_shutdown();
207
208 /* Disable the hypercall page when there is only 1 active CPU. */
209 if (kexec_in_progress)
210 hyperv_cleanup();
211}
212
213static void hv_machine_crash_shutdown(struct pt_regs *regs)
214{
215 if (hv_crash_handler)
216 hv_crash_handler(regs);
217
218 /* The function calls crash_smp_send_stop(). */
219 native_machine_crash_shutdown(regs);
220
221 /* Disable the hypercall page when there is only 1 active CPU. */
222 hyperv_cleanup();
223}
224#endif /* CONFIG_KEXEC_CORE */
225#endif /* CONFIG_HYPERV */
226
227static uint32_t __init ms_hyperv_platform(void)
228{
229 u32 eax;
230 u32 hyp_signature[3];
231
232 if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
233 return 0;
234
235 cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS,
236 &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]);
237
238 if (eax < HYPERV_CPUID_MIN || eax > HYPERV_CPUID_MAX ||
239 memcmp("Microsoft Hv", hyp_signature, 12))
240 return 0;
241
242 /* HYPERCALL and VP_INDEX MSRs are mandatory for all features. */
243 eax = cpuid_eax(HYPERV_CPUID_FEATURES);
244 if (!(eax & HV_MSR_HYPERCALL_AVAILABLE)) {
245 pr_warn("x86/hyperv: HYPERCALL MSR not available.\n");
246 return 0;
247 }
248 if (!(eax & HV_MSR_VP_INDEX_AVAILABLE)) {
249 pr_warn("x86/hyperv: VP_INDEX MSR not available.\n");
250 return 0;
251 }
252
253 return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
254}
255
256#ifdef CONFIG_X86_LOCAL_APIC
257/*
258 * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes
259 * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle
260 * unknown NMI on the first CPU which gets it.
261 */
262static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)
263{
264 static atomic_t nmi_cpu = ATOMIC_INIT(-1);
265 unsigned int old_cpu, this_cpu;
266
267 if (!unknown_nmi_panic)
268 return NMI_DONE;
269
270 old_cpu = -1;
271 this_cpu = raw_smp_processor_id();
272 if (!atomic_try_cmpxchg(&nmi_cpu, &old_cpu, this_cpu))
273 return NMI_HANDLED;
274
275 return NMI_DONE;
276}
277#endif
278
279static unsigned long hv_get_tsc_khz(void)
280{
281 unsigned long freq;
282
283 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
284
285 return freq / 1000;
286}
287
288#if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
289static void __init hv_smp_prepare_boot_cpu(void)
290{
291 native_smp_prepare_boot_cpu();
292#if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
293 hv_init_spinlocks();
294#endif
295}
296
297static void __init hv_smp_prepare_cpus(unsigned int max_cpus)
298{
299#ifdef CONFIG_X86_64
300 int i;
301 int ret;
302#endif
303
304 native_smp_prepare_cpus(max_cpus);
305
306 /*
307 * Override wakeup_secondary_cpu_64 callback for SEV-SNP
308 * enlightened guest.
309 */
310 if (!ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
311 apic->wakeup_secondary_cpu_64 = hv_snp_boot_ap;
312 return;
313 }
314
315#ifdef CONFIG_X86_64
316 for_each_present_cpu(i) {
317 if (i == 0)
318 continue;
319 ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i));
320 BUG_ON(ret);
321 }
322
323 for_each_present_cpu(i) {
324 if (i == 0)
325 continue;
326 ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i);
327 BUG_ON(ret);
328 }
329#endif
330}
331#endif
332
333/*
334 * When a fully enlightened TDX VM runs on Hyper-V, the firmware sets the
335 * HW_REDUCED flag: refer to acpi_tb_create_local_fadt(). Consequently ttyS0
336 * interrupts can't work because request_irq() -> ... -> irq_to_desc() returns
337 * NULL for ttyS0. This happens because mp_config_acpi_legacy_irqs() sees a
338 * nr_legacy_irqs() of 0, so it doesn't initialize the array 'mp_irqs[]', and
339 * later setup_IO_APIC_irqs() -> find_irq_entry() fails to find the legacy irqs
340 * from the array and hence doesn't create the necessary irq description info.
341 *
342 * Clone arch/x86/kernel/acpi/boot.c: acpi_generic_reduced_hw_init() here,
343 * except don't change 'legacy_pic', which keeps its default value
344 * 'default_legacy_pic'. This way, mp_config_acpi_legacy_irqs() sees a non-zero
345 * nr_legacy_irqs() and eventually serial console interrupts works properly.
346 */
347static void __init reduced_hw_init(void)
348{
349 x86_init.timers.timer_init = x86_init_noop;
350 x86_init.irqs.pre_vector_init = x86_init_noop;
351}
352
353static void __init ms_hyperv_init_platform(void)
354{
355 int hv_max_functions_eax;
356 int hv_host_info_eax;
357 int hv_host_info_ebx;
358 int hv_host_info_ecx;
359 int hv_host_info_edx;
360
361#ifdef CONFIG_PARAVIRT
362 pv_info.name = "Hyper-V";
363#endif
364
365 /*
366 * Extract the features and hints
367 */
368 ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
369 ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES);
370 ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES);
371 ms_hyperv.hints = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO);
372
373 hv_max_functions_eax = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
374
375 pr_info("Hyper-V: privilege flags low 0x%x, high 0x%x, hints 0x%x, misc 0x%x\n",
376 ms_hyperv.features, ms_hyperv.priv_high, ms_hyperv.hints,
377 ms_hyperv.misc_features);
378
379 ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
380 ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
381
382 pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
383 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
384
385 /*
386 * Check CPU management privilege.
387 *
388 * To mirror what Windows does we should extract CPU management
389 * features and use the ReservedIdentityBit to detect if Linux is the
390 * root partition. But that requires negotiating CPU management
391 * interface (a process to be finalized). For now, use the privilege
392 * flag as the indicator for running as root.
393 *
394 * Hyper-V should never specify running as root and as a Confidential
395 * VM. But to protect against a compromised/malicious Hyper-V trying
396 * to exploit root behavior to expose Confidential VM memory, ignore
397 * the root partition setting if also a Confidential VM.
398 */
399 if ((ms_hyperv.priv_high & HV_CPU_MANAGEMENT) &&
400 !(ms_hyperv.priv_high & HV_ISOLATION)) {
401 hv_root_partition = true;
402 pr_info("Hyper-V: running as root partition\n");
403 }
404
405 if (ms_hyperv.hints & HV_X64_HYPERV_NESTED) {
406 hv_nested = true;
407 pr_info("Hyper-V: running on a nested hypervisor\n");
408 }
409
410 /*
411 * Extract host information.
412 */
413 if (hv_max_functions_eax >= HYPERV_CPUID_VERSION) {
414 hv_host_info_eax = cpuid_eax(HYPERV_CPUID_VERSION);
415 hv_host_info_ebx = cpuid_ebx(HYPERV_CPUID_VERSION);
416 hv_host_info_ecx = cpuid_ecx(HYPERV_CPUID_VERSION);
417 hv_host_info_edx = cpuid_edx(HYPERV_CPUID_VERSION);
418
419 pr_info("Hyper-V: Host Build %d.%d.%d.%d-%d-%d\n",
420 hv_host_info_ebx >> 16, hv_host_info_ebx & 0xFFFF,
421 hv_host_info_eax, hv_host_info_edx & 0xFFFFFF,
422 hv_host_info_ecx, hv_host_info_edx >> 24);
423 }
424
425 if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
426 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
427 x86_platform.calibrate_tsc = hv_get_tsc_khz;
428 x86_platform.calibrate_cpu = hv_get_tsc_khz;
429 }
430
431 if (ms_hyperv.priv_high & HV_ISOLATION) {
432 ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG);
433 ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG);
434
435 if (ms_hyperv.shared_gpa_boundary_active)
436 ms_hyperv.shared_gpa_boundary =
437 BIT_ULL(ms_hyperv.shared_gpa_boundary_bits);
438
439 hyperv_paravisor_present = !!ms_hyperv.paravisor_present;
440
441 pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n",
442 ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b);
443
444
445 if (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP) {
446 static_branch_enable(&isolation_type_snp);
447 } else if (hv_get_isolation_type() == HV_ISOLATION_TYPE_TDX) {
448 static_branch_enable(&isolation_type_tdx);
449
450 /* A TDX VM must use x2APIC and doesn't use lazy EOI. */
451 ms_hyperv.hints &= ~HV_X64_APIC_ACCESS_RECOMMENDED;
452
453 if (!ms_hyperv.paravisor_present) {
454 /* To be supported: more work is required. */
455 ms_hyperv.features &= ~HV_MSR_REFERENCE_TSC_AVAILABLE;
456
457 /* HV_REGISTER_CRASH_CTL is unsupported. */
458 ms_hyperv.misc_features &= ~HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
459
460 /* Don't trust Hyper-V's TLB-flushing hypercalls. */
461 ms_hyperv.hints &= ~HV_X64_REMOTE_TLB_FLUSH_RECOMMENDED;
462
463 x86_init.acpi.reduced_hw_early_init = reduced_hw_init;
464 }
465 }
466 }
467
468 if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) {
469 ms_hyperv.nested_features =
470 cpuid_eax(HYPERV_CPUID_NESTED_FEATURES);
471 pr_info("Hyper-V: Nested features: 0x%x\n",
472 ms_hyperv.nested_features);
473 }
474
475#ifdef CONFIG_X86_LOCAL_APIC
476 if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
477 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
478 /*
479 * Get the APIC frequency.
480 */
481 u64 hv_lapic_frequency;
482
483 rdmsrl(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency);
484 hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ);
485 lapic_timer_period = hv_lapic_frequency;
486 pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
487 lapic_timer_period);
488 }
489
490 register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST,
491 "hv_nmi_unknown");
492#endif
493
494#ifdef CONFIG_X86_IO_APIC
495 no_timer_check = 1;
496#endif
497
498#if IS_ENABLED(CONFIG_HYPERV) && defined(CONFIG_KEXEC_CORE)
499 machine_ops.shutdown = hv_machine_shutdown;
500 machine_ops.crash_shutdown = hv_machine_crash_shutdown;
501#endif
502 if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) {
503 /*
504 * Writing to synthetic MSR 0x40000118 updates/changes the
505 * guest visible CPUIDs. Setting bit 0 of this MSR enables
506 * guests to report invariant TSC feature through CPUID
507 * instruction, CPUID 0x800000007/EDX, bit 8. See code in
508 * early_init_intel() where this bit is examined. The
509 * setting of this MSR bit should happen before init_intel()
510 * is called.
511 */
512 wrmsrl(HV_X64_MSR_TSC_INVARIANT_CONTROL, HV_EXPOSE_INVARIANT_TSC);
513 setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
514 }
515
516 /*
517 * Generation 2 instances don't support reading the NMI status from
518 * 0x61 port.
519 */
520 if (efi_enabled(EFI_BOOT))
521 x86_platform.get_nmi_reason = hv_get_nmi_reason;
522
523 /*
524 * Hyper-V VMs have a PIT emulation quirk such that zeroing the
525 * counter register during PIT shutdown restarts the PIT. So it
526 * continues to interrupt @18.2 HZ. Setting i8253_clear_counter
527 * to false tells pit_shutdown() not to zero the counter so that
528 * the PIT really is shutdown. Generation 2 VMs don't have a PIT,
529 * and setting this value has no effect.
530 */
531 i8253_clear_counter_on_shutdown = false;
532
533#if IS_ENABLED(CONFIG_HYPERV)
534 if ((hv_get_isolation_type() == HV_ISOLATION_TYPE_VBS) ||
535 ms_hyperv.paravisor_present)
536 hv_vtom_init();
537 /*
538 * Setup the hook to get control post apic initialization.
539 */
540 x86_platform.apic_post_init = hyperv_init;
541 hyperv_setup_mmu_ops();
542 /* Setup the IDT for hypervisor callback */
543 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, asm_sysvec_hyperv_callback);
544
545 /* Setup the IDT for reenlightenment notifications */
546 if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) {
547 alloc_intr_gate(HYPERV_REENLIGHTENMENT_VECTOR,
548 asm_sysvec_hyperv_reenlightenment);
549 }
550
551 /* Setup the IDT for stimer0 */
552 if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) {
553 alloc_intr_gate(HYPERV_STIMER0_VECTOR,
554 asm_sysvec_hyperv_stimer0);
555 }
556
557# ifdef CONFIG_SMP
558 smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
559 if (hv_root_partition ||
560 (!ms_hyperv.paravisor_present && hv_isolation_type_snp()))
561 smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus;
562# endif
563
564 /*
565 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
566 * set x2apic destination mode to physical mode when x2apic is available
567 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
568 * have 8-bit APIC id.
569 */
570# ifdef CONFIG_X86_X2APIC
571 if (x2apic_supported())
572 x2apic_phys = 1;
573# endif
574
575 /* Register Hyper-V specific clocksource */
576 hv_init_clocksource();
577 hv_vtl_init_platform();
578#endif
579 /*
580 * TSC should be marked as unstable only after Hyper-V
581 * clocksource has been initialized. This ensures that the
582 * stability of the sched_clock is not altered.
583 */
584 if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
585 mark_tsc_unstable("running on Hyper-V");
586
587 hardlockup_detector_disable();
588}
589
590static bool __init ms_hyperv_x2apic_available(void)
591{
592 return x2apic_supported();
593}
594
595/*
596 * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping()
597 * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the
598 * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg().
599 *
600 * Note: for a VM on Hyper-V, the I/O-APIC is the only device which
601 * (logically) generates MSIs directly to the system APIC irq domain.
602 * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the
603 * pci-hyperv host bridge.
604 *
605 * Note: for a Hyper-V root partition, this will always return false.
606 * The hypervisor doesn't expose these HYPERV_CPUID_VIRT_STACK_* cpuids by
607 * default, they are implemented as intercepts by the Windows Hyper-V stack.
608 * Even a nested root partition (L2 root) will not get them because the
609 * nested (L1) hypervisor filters them out.
610 */
611static bool __init ms_hyperv_msi_ext_dest_id(void)
612{
613 u32 eax;
614
615 eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE);
616 if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE)
617 return false;
618
619 eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES);
620 return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE;
621}
622
623#ifdef CONFIG_AMD_MEM_ENCRYPT
624static void hv_sev_es_hcall_prepare(struct ghcb *ghcb, struct pt_regs *regs)
625{
626 /* RAX and CPL are already in the GHCB */
627 ghcb_set_rcx(ghcb, regs->cx);
628 ghcb_set_rdx(ghcb, regs->dx);
629 ghcb_set_r8(ghcb, regs->r8);
630}
631
632static bool hv_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs)
633{
634 /* No checking of the return state needed */
635 return true;
636}
637#endif
638
639const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = {
640 .name = "Microsoft Hyper-V",
641 .detect = ms_hyperv_platform,
642 .type = X86_HYPER_MS_HYPERV,
643 .init.x2apic_available = ms_hyperv_x2apic_available,
644 .init.msi_ext_dest_id = ms_hyperv_msi_ext_dest_id,
645 .init.init_platform = ms_hyperv_init_platform,
646#ifdef CONFIG_AMD_MEM_ENCRYPT
647 .runtime.sev_es_hcall_prepare = hv_sev_es_hcall_prepare,
648 .runtime.sev_es_hcall_finish = hv_sev_es_hcall_finish,
649#endif
650};