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/panic_notifier.h> 21 #include <linux/random.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 37 /* Is Linux running as the root partition? */ 38 bool hv_root_partition; 39 EXPORT_SYMBOL_GPL(hv_root_partition); 40 41 struct ms_hyperv_info ms_hyperv; 42 EXPORT_SYMBOL_GPL(ms_hyperv); 43 44 #if IS_ENABLED(CONFIG_HYPERV) 45 static void (*vmbus_handler)(void); 46 static void (*hv_stimer0_handler)(void); 47 static void (*hv_kexec_handler)(void); 48 static void (*hv_crash_handler)(struct pt_regs *regs); 49 50 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback) 51 { 52 struct pt_regs *old_regs = set_irq_regs(regs); 53 54 inc_irq_stat(irq_hv_callback_count); 55 if (vmbus_handler) 56 vmbus_handler(); 57 58 if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED) 59 ack_APIC_irq(); 60 61 set_irq_regs(old_regs); 62 } 63 64 void hv_setup_vmbus_handler(void (*handler)(void)) 65 { 66 vmbus_handler = handler; 67 } 68 EXPORT_SYMBOL_GPL(hv_setup_vmbus_handler); 69 70 void hv_remove_vmbus_handler(void) 71 { 72 /* We have no way to deallocate the interrupt gate */ 73 vmbus_handler = NULL; 74 } 75 EXPORT_SYMBOL_GPL(hv_remove_vmbus_handler); 76 77 /* 78 * Routines to do per-architecture handling of stimer0 79 * interrupts when in Direct Mode 80 */ 81 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0) 82 { 83 struct pt_regs *old_regs = set_irq_regs(regs); 84 85 inc_irq_stat(hyperv_stimer0_count); 86 if (hv_stimer0_handler) 87 hv_stimer0_handler(); 88 add_interrupt_randomness(HYPERV_STIMER0_VECTOR, 0); 89 ack_APIC_irq(); 90 91 set_irq_regs(old_regs); 92 } 93 94 /* For x86/x64, override weak placeholders in hyperv_timer.c */ 95 void hv_setup_stimer0_handler(void (*handler)(void)) 96 { 97 hv_stimer0_handler = handler; 98 } 99 100 void hv_remove_stimer0_handler(void) 101 { 102 /* We have no way to deallocate the interrupt gate */ 103 hv_stimer0_handler = NULL; 104 } 105 106 void hv_setup_kexec_handler(void (*handler)(void)) 107 { 108 hv_kexec_handler = handler; 109 } 110 EXPORT_SYMBOL_GPL(hv_setup_kexec_handler); 111 112 void hv_remove_kexec_handler(void) 113 { 114 hv_kexec_handler = NULL; 115 } 116 EXPORT_SYMBOL_GPL(hv_remove_kexec_handler); 117 118 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs)) 119 { 120 hv_crash_handler = handler; 121 } 122 EXPORT_SYMBOL_GPL(hv_setup_crash_handler); 123 124 void hv_remove_crash_handler(void) 125 { 126 hv_crash_handler = NULL; 127 } 128 EXPORT_SYMBOL_GPL(hv_remove_crash_handler); 129 130 #ifdef CONFIG_KEXEC_CORE 131 static void hv_machine_shutdown(void) 132 { 133 if (kexec_in_progress && hv_kexec_handler) 134 hv_kexec_handler(); 135 136 /* 137 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor 138 * corrupts the old VP Assist Pages and can crash the kexec kernel. 139 */ 140 if (kexec_in_progress && hyperv_init_cpuhp > 0) 141 cpuhp_remove_state(hyperv_init_cpuhp); 142 143 /* The function calls stop_other_cpus(). */ 144 native_machine_shutdown(); 145 146 /* Disable the hypercall page when there is only 1 active CPU. */ 147 if (kexec_in_progress) 148 hyperv_cleanup(); 149 } 150 151 static void hv_machine_crash_shutdown(struct pt_regs *regs) 152 { 153 if (hv_crash_handler) 154 hv_crash_handler(regs); 155 156 /* The function calls crash_smp_send_stop(). */ 157 native_machine_crash_shutdown(regs); 158 159 /* Disable the hypercall page when there is only 1 active CPU. */ 160 hyperv_cleanup(); 161 } 162 #endif /* CONFIG_KEXEC_CORE */ 163 #endif /* CONFIG_HYPERV */ 164 165 static uint32_t __init ms_hyperv_platform(void) 166 { 167 u32 eax; 168 u32 hyp_signature[3]; 169 170 if (!boot_cpu_has(X86_FEATURE_HYPERVISOR)) 171 return 0; 172 173 cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS, 174 &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]); 175 176 if (eax >= HYPERV_CPUID_MIN && 177 eax <= HYPERV_CPUID_MAX && 178 !memcmp("Microsoft Hv", hyp_signature, 12)) 179 return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS; 180 181 return 0; 182 } 183 184 static unsigned char hv_get_nmi_reason(void) 185 { 186 return 0; 187 } 188 189 #ifdef CONFIG_X86_LOCAL_APIC 190 /* 191 * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes 192 * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle 193 * unknown NMI on the first CPU which gets it. 194 */ 195 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs) 196 { 197 static atomic_t nmi_cpu = ATOMIC_INIT(-1); 198 199 if (!unknown_nmi_panic) 200 return NMI_DONE; 201 202 if (atomic_cmpxchg(&nmi_cpu, -1, raw_smp_processor_id()) != -1) 203 return NMI_HANDLED; 204 205 return NMI_DONE; 206 } 207 #endif 208 209 static unsigned long hv_get_tsc_khz(void) 210 { 211 unsigned long freq; 212 213 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq); 214 215 return freq / 1000; 216 } 217 218 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV) 219 static void __init hv_smp_prepare_boot_cpu(void) 220 { 221 native_smp_prepare_boot_cpu(); 222 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS) 223 hv_init_spinlocks(); 224 #endif 225 } 226 227 static void __init hv_smp_prepare_cpus(unsigned int max_cpus) 228 { 229 #ifdef CONFIG_X86_64 230 int i; 231 int ret; 232 #endif 233 234 native_smp_prepare_cpus(max_cpus); 235 236 #ifdef CONFIG_X86_64 237 for_each_present_cpu(i) { 238 if (i == 0) 239 continue; 240 ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i)); 241 BUG_ON(ret); 242 } 243 244 for_each_present_cpu(i) { 245 if (i == 0) 246 continue; 247 ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i); 248 BUG_ON(ret); 249 } 250 #endif 251 } 252 #endif 253 254 static void __init ms_hyperv_init_platform(void) 255 { 256 int hv_host_info_eax; 257 int hv_host_info_ebx; 258 int hv_host_info_ecx; 259 int hv_host_info_edx; 260 261 #ifdef CONFIG_PARAVIRT 262 pv_info.name = "Hyper-V"; 263 #endif 264 265 /* 266 * Extract the features and hints 267 */ 268 ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES); 269 ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES); 270 ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES); 271 ms_hyperv.hints = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO); 272 273 pr_info("Hyper-V: privilege flags low 0x%x, high 0x%x, hints 0x%x, misc 0x%x\n", 274 ms_hyperv.features, ms_hyperv.priv_high, ms_hyperv.hints, 275 ms_hyperv.misc_features); 276 277 ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS); 278 ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS); 279 280 pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n", 281 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index); 282 283 /* 284 * Check CPU management privilege. 285 * 286 * To mirror what Windows does we should extract CPU management 287 * features and use the ReservedIdentityBit to detect if Linux is the 288 * root partition. But that requires negotiating CPU management 289 * interface (a process to be finalized). 290 * 291 * For now, use the privilege flag as the indicator for running as 292 * root. 293 */ 294 if (cpuid_ebx(HYPERV_CPUID_FEATURES) & HV_CPU_MANAGEMENT) { 295 hv_root_partition = true; 296 pr_info("Hyper-V: running as root partition\n"); 297 } 298 299 /* 300 * Extract host information. 301 */ 302 if (cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS) >= 303 HYPERV_CPUID_VERSION) { 304 hv_host_info_eax = cpuid_eax(HYPERV_CPUID_VERSION); 305 hv_host_info_ebx = cpuid_ebx(HYPERV_CPUID_VERSION); 306 hv_host_info_ecx = cpuid_ecx(HYPERV_CPUID_VERSION); 307 hv_host_info_edx = cpuid_edx(HYPERV_CPUID_VERSION); 308 309 pr_info("Hyper-V Host Build:%d-%d.%d-%d-%d.%d\n", 310 hv_host_info_eax, hv_host_info_ebx >> 16, 311 hv_host_info_ebx & 0xFFFF, hv_host_info_ecx, 312 hv_host_info_edx >> 24, hv_host_info_edx & 0xFFFFFF); 313 } 314 315 if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS && 316 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) { 317 x86_platform.calibrate_tsc = hv_get_tsc_khz; 318 x86_platform.calibrate_cpu = hv_get_tsc_khz; 319 } 320 321 if (ms_hyperv.priv_high & HV_ISOLATION) { 322 ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG); 323 ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG); 324 325 pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n", 326 ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b); 327 } 328 329 if (ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED) { 330 ms_hyperv.nested_features = 331 cpuid_eax(HYPERV_CPUID_NESTED_FEATURES); 332 } 333 334 /* 335 * Hyper-V expects to get crash register data or kmsg when 336 * crash enlightment is available and system crashes. Set 337 * crash_kexec_post_notifiers to be true to make sure that 338 * calling crash enlightment interface before running kdump 339 * kernel. 340 */ 341 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) 342 crash_kexec_post_notifiers = true; 343 344 #ifdef CONFIG_X86_LOCAL_APIC 345 if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS && 346 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) { 347 /* 348 * Get the APIC frequency. 349 */ 350 u64 hv_lapic_frequency; 351 352 rdmsrl(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency); 353 hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ); 354 lapic_timer_period = hv_lapic_frequency; 355 pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n", 356 lapic_timer_period); 357 } 358 359 register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST, 360 "hv_nmi_unknown"); 361 #endif 362 363 #ifdef CONFIG_X86_IO_APIC 364 no_timer_check = 1; 365 #endif 366 367 #if IS_ENABLED(CONFIG_HYPERV) && defined(CONFIG_KEXEC_CORE) 368 machine_ops.shutdown = hv_machine_shutdown; 369 machine_ops.crash_shutdown = hv_machine_crash_shutdown; 370 #endif 371 if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) { 372 wrmsrl(HV_X64_MSR_TSC_INVARIANT_CONTROL, 0x1); 373 setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE); 374 } else { 375 mark_tsc_unstable("running on Hyper-V"); 376 } 377 378 /* 379 * Generation 2 instances don't support reading the NMI status from 380 * 0x61 port. 381 */ 382 if (efi_enabled(EFI_BOOT)) 383 x86_platform.get_nmi_reason = hv_get_nmi_reason; 384 385 /* 386 * Hyper-V VMs have a PIT emulation quirk such that zeroing the 387 * counter register during PIT shutdown restarts the PIT. So it 388 * continues to interrupt @18.2 HZ. Setting i8253_clear_counter 389 * to false tells pit_shutdown() not to zero the counter so that 390 * the PIT really is shutdown. Generation 2 VMs don't have a PIT, 391 * and setting this value has no effect. 392 */ 393 i8253_clear_counter_on_shutdown = false; 394 395 #if IS_ENABLED(CONFIG_HYPERV) 396 /* 397 * Setup the hook to get control post apic initialization. 398 */ 399 x86_platform.apic_post_init = hyperv_init; 400 hyperv_setup_mmu_ops(); 401 /* Setup the IDT for hypervisor callback */ 402 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, asm_sysvec_hyperv_callback); 403 404 /* Setup the IDT for reenlightenment notifications */ 405 if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) { 406 alloc_intr_gate(HYPERV_REENLIGHTENMENT_VECTOR, 407 asm_sysvec_hyperv_reenlightenment); 408 } 409 410 /* Setup the IDT for stimer0 */ 411 if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) { 412 alloc_intr_gate(HYPERV_STIMER0_VECTOR, 413 asm_sysvec_hyperv_stimer0); 414 } 415 416 # ifdef CONFIG_SMP 417 smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu; 418 if (hv_root_partition) 419 smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus; 420 # endif 421 422 /* 423 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic, 424 * set x2apic destination mode to physical mode when x2apic is available 425 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs 426 * have 8-bit APIC id. 427 */ 428 # ifdef CONFIG_X86_X2APIC 429 if (x2apic_supported()) 430 x2apic_phys = 1; 431 # endif 432 433 /* Register Hyper-V specific clocksource */ 434 hv_init_clocksource(); 435 #endif 436 } 437 438 static bool __init ms_hyperv_x2apic_available(void) 439 { 440 return x2apic_supported(); 441 } 442 443 /* 444 * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping() 445 * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the 446 * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg(). 447 * 448 * Note: for a VM on Hyper-V, the I/O-APIC is the only device which 449 * (logically) generates MSIs directly to the system APIC irq domain. 450 * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the 451 * pci-hyperv host bridge. 452 */ 453 static bool __init ms_hyperv_msi_ext_dest_id(void) 454 { 455 u32 eax; 456 457 eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE); 458 if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE) 459 return false; 460 461 eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES); 462 return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE; 463 } 464 465 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = { 466 .name = "Microsoft Hyper-V", 467 .detect = ms_hyperv_platform, 468 .type = X86_HYPER_MS_HYPERV, 469 .init.x2apic_available = ms_hyperv_x2apic_available, 470 .init.msi_ext_dest_id = ms_hyperv_msi_ext_dest_id, 471 .init.init_platform = ms_hyperv_init_platform, 472 }; 473