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