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 36 /* Is Linux running as the root partition? */ 37 bool hv_root_partition; 38 EXPORT_SYMBOL_GPL(hv_root_partition); 39 40 struct ms_hyperv_info ms_hyperv; 41 EXPORT_SYMBOL_GPL(ms_hyperv); 42 43 #if IS_ENABLED(CONFIG_HYPERV) 44 static void (*vmbus_handler)(void); 45 static void (*hv_stimer0_handler)(void); 46 static void (*hv_kexec_handler)(void); 47 static void (*hv_crash_handler)(struct pt_regs *regs); 48 49 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback) 50 { 51 struct pt_regs *old_regs = set_irq_regs(regs); 52 53 inc_irq_stat(irq_hv_callback_count); 54 if (vmbus_handler) 55 vmbus_handler(); 56 57 if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED) 58 ack_APIC_irq(); 59 60 set_irq_regs(old_regs); 61 } 62 63 int hv_setup_vmbus_irq(int irq, void (*handler)(void)) 64 { 65 /* 66 * The 'irq' argument is ignored on x86/x64 because a hard-coded 67 * interrupt vector is used for Hyper-V interrupts. 68 */ 69 vmbus_handler = handler; 70 return 0; 71 } 72 73 void hv_remove_vmbus_irq(void) 74 { 75 /* We have no way to deallocate the interrupt gate */ 76 vmbus_handler = NULL; 77 } 78 EXPORT_SYMBOL_GPL(hv_setup_vmbus_irq); 79 EXPORT_SYMBOL_GPL(hv_remove_vmbus_irq); 80 81 /* 82 * Routines to do per-architecture handling of stimer0 83 * interrupts when in Direct Mode 84 */ 85 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0) 86 { 87 struct pt_regs *old_regs = set_irq_regs(regs); 88 89 inc_irq_stat(hyperv_stimer0_count); 90 if (hv_stimer0_handler) 91 hv_stimer0_handler(); 92 add_interrupt_randomness(HYPERV_STIMER0_VECTOR, 0); 93 ack_APIC_irq(); 94 95 set_irq_regs(old_regs); 96 } 97 98 int hv_setup_stimer0_irq(int *irq, int *vector, void (*handler)(void)) 99 { 100 *vector = HYPERV_STIMER0_VECTOR; 101 *irq = -1; /* Unused on x86/x64 */ 102 hv_stimer0_handler = handler; 103 return 0; 104 } 105 EXPORT_SYMBOL_GPL(hv_setup_stimer0_irq); 106 107 void hv_remove_stimer0_irq(int irq) 108 { 109 /* We have no way to deallocate the interrupt gate */ 110 hv_stimer0_handler = NULL; 111 } 112 EXPORT_SYMBOL_GPL(hv_remove_stimer0_irq); 113 114 void hv_setup_kexec_handler(void (*handler)(void)) 115 { 116 hv_kexec_handler = handler; 117 } 118 EXPORT_SYMBOL_GPL(hv_setup_kexec_handler); 119 120 void hv_remove_kexec_handler(void) 121 { 122 hv_kexec_handler = NULL; 123 } 124 EXPORT_SYMBOL_GPL(hv_remove_kexec_handler); 125 126 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs)) 127 { 128 hv_crash_handler = handler; 129 } 130 EXPORT_SYMBOL_GPL(hv_setup_crash_handler); 131 132 void hv_remove_crash_handler(void) 133 { 134 hv_crash_handler = NULL; 135 } 136 EXPORT_SYMBOL_GPL(hv_remove_crash_handler); 137 138 #ifdef CONFIG_KEXEC_CORE 139 static void hv_machine_shutdown(void) 140 { 141 if (kexec_in_progress && hv_kexec_handler) 142 hv_kexec_handler(); 143 144 /* 145 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor 146 * corrupts the old VP Assist Pages and can crash the kexec kernel. 147 */ 148 if (kexec_in_progress && hyperv_init_cpuhp > 0) 149 cpuhp_remove_state(hyperv_init_cpuhp); 150 151 /* The function calls stop_other_cpus(). */ 152 native_machine_shutdown(); 153 154 /* Disable the hypercall page when there is only 1 active CPU. */ 155 if (kexec_in_progress) 156 hyperv_cleanup(); 157 } 158 159 static void hv_machine_crash_shutdown(struct pt_regs *regs) 160 { 161 if (hv_crash_handler) 162 hv_crash_handler(regs); 163 164 /* The function calls crash_smp_send_stop(). */ 165 native_machine_crash_shutdown(regs); 166 167 /* Disable the hypercall page when there is only 1 active CPU. */ 168 hyperv_cleanup(); 169 } 170 #endif /* CONFIG_KEXEC_CORE */ 171 #endif /* CONFIG_HYPERV */ 172 173 static uint32_t __init ms_hyperv_platform(void) 174 { 175 u32 eax; 176 u32 hyp_signature[3]; 177 178 if (!boot_cpu_has(X86_FEATURE_HYPERVISOR)) 179 return 0; 180 181 cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS, 182 &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]); 183 184 if (eax >= HYPERV_CPUID_MIN && 185 eax <= HYPERV_CPUID_MAX && 186 !memcmp("Microsoft Hv", hyp_signature, 12)) 187 return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS; 188 189 return 0; 190 } 191 192 static unsigned char hv_get_nmi_reason(void) 193 { 194 return 0; 195 } 196 197 #ifdef CONFIG_X86_LOCAL_APIC 198 /* 199 * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes 200 * it dificult to process CHANNELMSG_UNLOAD in case of crash. Handle 201 * unknown NMI on the first CPU which gets it. 202 */ 203 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs) 204 { 205 static atomic_t nmi_cpu = ATOMIC_INIT(-1); 206 207 if (!unknown_nmi_panic) 208 return NMI_DONE; 209 210 if (atomic_cmpxchg(&nmi_cpu, -1, raw_smp_processor_id()) != -1) 211 return NMI_HANDLED; 212 213 return NMI_DONE; 214 } 215 #endif 216 217 static unsigned long hv_get_tsc_khz(void) 218 { 219 unsigned long freq; 220 221 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq); 222 223 return freq / 1000; 224 } 225 226 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV) 227 static void __init hv_smp_prepare_boot_cpu(void) 228 { 229 native_smp_prepare_boot_cpu(); 230 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS) 231 hv_init_spinlocks(); 232 #endif 233 } 234 235 static void __init hv_smp_prepare_cpus(unsigned int max_cpus) 236 { 237 #ifdef CONFIG_X86_64 238 int i; 239 int ret; 240 #endif 241 242 native_smp_prepare_cpus(max_cpus); 243 244 #ifdef CONFIG_X86_64 245 for_each_present_cpu(i) { 246 if (i == 0) 247 continue; 248 ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i)); 249 BUG_ON(ret); 250 } 251 252 for_each_present_cpu(i) { 253 if (i == 0) 254 continue; 255 ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i); 256 BUG_ON(ret); 257 } 258 #endif 259 } 260 #endif 261 262 static void __init ms_hyperv_init_platform(void) 263 { 264 int hv_host_info_eax; 265 int hv_host_info_ebx; 266 int hv_host_info_ecx; 267 int hv_host_info_edx; 268 269 #ifdef CONFIG_PARAVIRT 270 pv_info.name = "Hyper-V"; 271 #endif 272 273 /* 274 * Extract the features and hints 275 */ 276 ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES); 277 ms_hyperv.features_b = cpuid_ebx(HYPERV_CPUID_FEATURES); 278 ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES); 279 ms_hyperv.hints = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO); 280 281 pr_info("Hyper-V: features 0x%x, hints 0x%x, misc 0x%x\n", 282 ms_hyperv.features, ms_hyperv.hints, ms_hyperv.misc_features); 283 284 ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS); 285 ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS); 286 287 pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n", 288 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index); 289 290 /* 291 * Check CPU management privilege. 292 * 293 * To mirror what Windows does we should extract CPU management 294 * features and use the ReservedIdentityBit to detect if Linux is the 295 * root partition. But that requires negotiating CPU management 296 * interface (a process to be finalized). 297 * 298 * For now, use the privilege flag as the indicator for running as 299 * root. 300 */ 301 if (cpuid_ebx(HYPERV_CPUID_FEATURES) & HV_CPU_MANAGEMENT) { 302 hv_root_partition = true; 303 pr_info("Hyper-V: running as root partition\n"); 304 } 305 306 /* 307 * Extract host information. 308 */ 309 if (cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS) >= 310 HYPERV_CPUID_VERSION) { 311 hv_host_info_eax = cpuid_eax(HYPERV_CPUID_VERSION); 312 hv_host_info_ebx = cpuid_ebx(HYPERV_CPUID_VERSION); 313 hv_host_info_ecx = cpuid_ecx(HYPERV_CPUID_VERSION); 314 hv_host_info_edx = cpuid_edx(HYPERV_CPUID_VERSION); 315 316 pr_info("Hyper-V Host Build:%d-%d.%d-%d-%d.%d\n", 317 hv_host_info_eax, hv_host_info_ebx >> 16, 318 hv_host_info_ebx & 0xFFFF, hv_host_info_ecx, 319 hv_host_info_edx >> 24, hv_host_info_edx & 0xFFFFFF); 320 } 321 322 if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS && 323 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) { 324 x86_platform.calibrate_tsc = hv_get_tsc_khz; 325 x86_platform.calibrate_cpu = hv_get_tsc_khz; 326 } 327 328 if (ms_hyperv.features_b & HV_ISOLATION) { 329 ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG); 330 ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG); 331 332 pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n", 333 ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b); 334 } 335 336 if (ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED) { 337 ms_hyperv.nested_features = 338 cpuid_eax(HYPERV_CPUID_NESTED_FEATURES); 339 } 340 341 /* 342 * Hyper-V expects to get crash register data or kmsg when 343 * crash enlightment is available and system crashes. Set 344 * crash_kexec_post_notifiers to be true to make sure that 345 * calling crash enlightment interface before running kdump 346 * kernel. 347 */ 348 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) 349 crash_kexec_post_notifiers = true; 350 351 #ifdef CONFIG_X86_LOCAL_APIC 352 if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS && 353 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) { 354 /* 355 * Get the APIC frequency. 356 */ 357 u64 hv_lapic_frequency; 358 359 rdmsrl(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency); 360 hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ); 361 lapic_timer_period = hv_lapic_frequency; 362 pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n", 363 lapic_timer_period); 364 } 365 366 register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST, 367 "hv_nmi_unknown"); 368 #endif 369 370 #ifdef CONFIG_X86_IO_APIC 371 no_timer_check = 1; 372 #endif 373 374 #if IS_ENABLED(CONFIG_HYPERV) && defined(CONFIG_KEXEC_CORE) 375 machine_ops.shutdown = hv_machine_shutdown; 376 machine_ops.crash_shutdown = hv_machine_crash_shutdown; 377 #endif 378 if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) { 379 wrmsrl(HV_X64_MSR_TSC_INVARIANT_CONTROL, 0x1); 380 setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE); 381 } else { 382 mark_tsc_unstable("running on Hyper-V"); 383 } 384 385 /* 386 * Generation 2 instances don't support reading the NMI status from 387 * 0x61 port. 388 */ 389 if (efi_enabled(EFI_BOOT)) 390 x86_platform.get_nmi_reason = hv_get_nmi_reason; 391 392 /* 393 * Hyper-V VMs have a PIT emulation quirk such that zeroing the 394 * counter register during PIT shutdown restarts the PIT. So it 395 * continues to interrupt @18.2 HZ. Setting i8253_clear_counter 396 * to false tells pit_shutdown() not to zero the counter so that 397 * the PIT really is shutdown. Generation 2 VMs don't have a PIT, 398 * and setting this value has no effect. 399 */ 400 i8253_clear_counter_on_shutdown = false; 401 402 #if IS_ENABLED(CONFIG_HYPERV) 403 /* 404 * Setup the hook to get control post apic initialization. 405 */ 406 x86_platform.apic_post_init = hyperv_init; 407 hyperv_setup_mmu_ops(); 408 /* Setup the IDT for hypervisor callback */ 409 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, asm_sysvec_hyperv_callback); 410 411 /* Setup the IDT for reenlightenment notifications */ 412 if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) { 413 alloc_intr_gate(HYPERV_REENLIGHTENMENT_VECTOR, 414 asm_sysvec_hyperv_reenlightenment); 415 } 416 417 /* Setup the IDT for stimer0 */ 418 if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) { 419 alloc_intr_gate(HYPERV_STIMER0_VECTOR, 420 asm_sysvec_hyperv_stimer0); 421 } 422 423 # ifdef CONFIG_SMP 424 smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu; 425 if (hv_root_partition) 426 smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus; 427 # endif 428 429 /* 430 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic, 431 * set x2apic destination mode to physcial mode when x2apic is available 432 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs 433 * have 8-bit APIC id. 434 */ 435 # ifdef CONFIG_X86_X2APIC 436 if (x2apic_supported()) 437 x2apic_phys = 1; 438 # endif 439 440 /* Register Hyper-V specific clocksource */ 441 hv_init_clocksource(); 442 #endif 443 } 444 445 static bool __init ms_hyperv_x2apic_available(void) 446 { 447 return x2apic_supported(); 448 } 449 450 /* 451 * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping() 452 * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the 453 * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg(). 454 * 455 * Note: for a VM on Hyper-V, the I/O-APIC is the only device which 456 * (logically) generates MSIs directly to the system APIC irq domain. 457 * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the 458 * pci-hyperv host bridge. 459 */ 460 static bool __init ms_hyperv_msi_ext_dest_id(void) 461 { 462 u32 eax; 463 464 eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE); 465 if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE) 466 return false; 467 468 eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES); 469 return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE; 470 } 471 472 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = { 473 .name = "Microsoft Hyper-V", 474 .detect = ms_hyperv_platform, 475 .type = X86_HYPER_MS_HYPERV, 476 .init.x2apic_available = ms_hyperv_x2apic_available, 477 .init.msi_ext_dest_id = ms_hyperv_msi_ext_dest_id, 478 .init.init_platform = ms_hyperv_init_platform, 479 }; 480