arch/x86/kernel/cpu/mshyperv.c

   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_max_functions_eax;
 257         int hv_host_info_eax;
 258         int hv_host_info_ebx;
 259         int hv_host_info_ecx;
 260         int hv_host_info_edx;
 261
 262 #ifdef CONFIG_PARAVIRT
 263         pv_info.name = "Hyper-V";
 264 #endif
 265
 266         /*
 267          * Extract the features and hints
 268          */
 269         ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
 270         ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES);
 271         ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES);
 272         ms_hyperv.hints    = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO);
 273
 274         hv_max_functions_eax = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
 275
 276         pr_info("Hyper-V: privilege flags low 0x%x, high 0x%x, hints 0x%x, misc 0x%x\n",
 277                 ms_hyperv.features, ms_hyperv.priv_high, ms_hyperv.hints,
 278                 ms_hyperv.misc_features);
 279
 280         ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
 281         ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
 282
 283         pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
 284                  ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
 285
 286         /*
 287          * Check CPU management privilege.
 288          *
 289          * To mirror what Windows does we should extract CPU management
 290          * features and use the ReservedIdentityBit to detect if Linux is the
 291          * root partition. But that requires negotiating CPU management
 292          * interface (a process to be finalized).
 293          *
 294          * For now, use the privilege flag as the indicator for running as
 295          * root.
 296          */
 297         if (cpuid_ebx(HYPERV_CPUID_FEATURES) & HV_CPU_MANAGEMENT) {
 298                 hv_root_partition = true;
 299                 pr_info("Hyper-V: running as root partition\n");
 300         }
 301
 302         /*
 303          * Extract host information.
 304          */
 305         if (hv_max_functions_eax >= HYPERV_CPUID_VERSION) {
 306                 hv_host_info_eax = cpuid_eax(HYPERV_CPUID_VERSION);
 307                 hv_host_info_ebx = cpuid_ebx(HYPERV_CPUID_VERSION);
 308                 hv_host_info_ecx = cpuid_ecx(HYPERV_CPUID_VERSION);
 309                 hv_host_info_edx = cpuid_edx(HYPERV_CPUID_VERSION);
 310
 311                 pr_info("Hyper-V Host Build:%d-%d.%d-%d-%d.%d\n",
 312                         hv_host_info_eax, hv_host_info_ebx >> 16,
 313                         hv_host_info_ebx & 0xFFFF, hv_host_info_ecx,
 314                         hv_host_info_edx >> 24, hv_host_info_edx & 0xFFFFFF);
 315         }
 316
 317         if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
 318             ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
 319                 x86_platform.calibrate_tsc = hv_get_tsc_khz;
 320                 x86_platform.calibrate_cpu = hv_get_tsc_khz;
 321         }
 322
 323         if (ms_hyperv.priv_high & HV_ISOLATION) {
 324                 ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG);
 325                 ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG);
 326
 327                 pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n",
 328                         ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b);
 329         }
 330
 331         if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) {
 332                 ms_hyperv.nested_features =
 333                         cpuid_eax(HYPERV_CPUID_NESTED_FEATURES);
 334                 pr_info("Hyper-V: Nested features: 0x%x\n",
 335                         ms_hyperv.nested_features);
 336         }
 337
 338         /*
 339          * Hyper-V expects to get crash register data or kmsg when
 340          * crash enlightment is available and system crashes. Set
 341          * crash_kexec_post_notifiers to be true to make sure that
 342          * calling crash enlightment interface before running kdump
 343          * kernel.
 344          */
 345         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE)
 346                 crash_kexec_post_notifiers = true;
 347
 348 #ifdef CONFIG_X86_LOCAL_APIC
 349         if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
 350             ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
 351                 /*
 352                  * Get the APIC frequency.
 353                  */
 354                 u64     hv_lapic_frequency;
 355
 356                 rdmsrl(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency);
 357                 hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ);
 358                 lapic_timer_period = hv_lapic_frequency;
 359                 pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
 360                         lapic_timer_period);
 361         }
 362
 363         register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST,
 364                              "hv_nmi_unknown");
 365 #endif
 366
 367 #ifdef CONFIG_X86_IO_APIC
 368         no_timer_check = 1;
 369 #endif
 370
 371 #if IS_ENABLED(CONFIG_HYPERV) && defined(CONFIG_KEXEC_CORE)
 372         machine_ops.shutdown = hv_machine_shutdown;
 373         machine_ops.crash_shutdown = hv_machine_crash_shutdown;
 374 #endif
 375         if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) {
 376                 wrmsrl(HV_X64_MSR_TSC_INVARIANT_CONTROL, 0x1);
 377                 setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
 378         } else {
 379                 mark_tsc_unstable("running on Hyper-V");
 380         }
 381
 382         /*
 383          * Generation 2 instances don't support reading the NMI status from
 384          * 0x61 port.
 385          */
 386         if (efi_enabled(EFI_BOOT))
 387                 x86_platform.get_nmi_reason = hv_get_nmi_reason;
 388
 389         /*
 390          * Hyper-V VMs have a PIT emulation quirk such that zeroing the
 391          * counter register during PIT shutdown restarts the PIT. So it
 392          * continues to interrupt @18.2 HZ. Setting i8253_clear_counter
 393          * to false tells pit_shutdown() not to zero the counter so that
 394          * the PIT really is shutdown. Generation 2 VMs don't have a PIT,
 395          * and setting this value has no effect.
 396          */
 397         i8253_clear_counter_on_shutdown = false;
 398
 399 #if IS_ENABLED(CONFIG_HYPERV)
 400         /*
 401          * Setup the hook to get control post apic initialization.
 402          */
 403         x86_platform.apic_post_init = hyperv_init;
 404         hyperv_setup_mmu_ops();
 405         /* Setup the IDT for hypervisor callback */
 406         alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, asm_sysvec_hyperv_callback);
 407
 408         /* Setup the IDT for reenlightenment notifications */
 409         if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) {
 410                 alloc_intr_gate(HYPERV_REENLIGHTENMENT_VECTOR,
 411                                 asm_sysvec_hyperv_reenlightenment);
 412         }
 413
 414         /* Setup the IDT for stimer0 */
 415         if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) {
 416                 alloc_intr_gate(HYPERV_STIMER0_VECTOR,
 417                                 asm_sysvec_hyperv_stimer0);
 418         }
 419
 420 # ifdef CONFIG_SMP
 421         smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
 422         if (hv_root_partition)
 423                 smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus;
 424 # endif
 425
 426         /*
 427          * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
 428          * set x2apic destination mode to physical mode when x2apic is available
 429          * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
 430          * have 8-bit APIC id.
 431          */
 432 # ifdef CONFIG_X86_X2APIC
 433         if (x2apic_supported())
 434                 x2apic_phys = 1;
 435 # endif
 436
 437         /* Register Hyper-V specific clocksource */
 438         hv_init_clocksource();
 439 #endif
 440 }
 441
 442 static bool __init ms_hyperv_x2apic_available(void)
 443 {
 444         return x2apic_supported();
 445 }
 446
 447 /*
 448  * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping()
 449  * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the
 450  * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg().
 451  *
 452  * Note: for a VM on Hyper-V, the I/O-APIC is the only device which
 453  * (logically) generates MSIs directly to the system APIC irq domain.
 454  * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the
 455  * pci-hyperv host bridge.
 456  */
 457 static bool __init ms_hyperv_msi_ext_dest_id(void)
 458 {
 459         u32 eax;
 460
 461         eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE);
 462         if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE)
 463                 return false;
 464
 465         eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES);
 466         return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE;
 467 }
 468
 469 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = {
 470         .name                   = "Microsoft Hyper-V",
 471         .detect                 = ms_hyperv_platform,
 472         .type                   = X86_HYPER_MS_HYPERV,
 473         .init.x2apic_available  = ms_hyperv_x2apic_available,
 474         .init.msi_ext_dest_id   = ms_hyperv_msi_ext_dest_id,
 475         .init.init_platform     = ms_hyperv_init_platform,
 476 };