arch/x86/hyperv/hv_init.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * X86 specific Hyper-V initialization code.
   4  *
   5  * Copyright (C) 2016, Microsoft, Inc.
   6  *
   7  * Author : K. Y. Srinivasan <kys@microsoft.com>
   8  */
   9
  10 #include <linux/acpi.h>
  11 #include <linux/efi.h>
  12 #include <linux/types.h>
  13 #include <asm/apic.h>
  14 #include <asm/desc.h>
  15 #include <asm/hypervisor.h>
  16 #include <asm/hyperv-tlfs.h>
  17 #include <asm/mshyperv.h>
  18 #include <linux/version.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/mm.h>
  21 #include <linux/hyperv.h>
  22 #include <linux/slab.h>
  23 #include <linux/kernel.h>
  24 #include <linux/cpuhotplug.h>
  25 #include <linux/syscore_ops.h>
  26 #include <clocksource/hyperv_timer.h>
  27
  28 void *hv_hypercall_pg;
  29 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
  30
  31 /* Storage to save the hypercall page temporarily for hibernation */
  32 static void *hv_hypercall_pg_saved;
  33
  34 u32 *hv_vp_index;
  35 EXPORT_SYMBOL_GPL(hv_vp_index);
  36
  37 struct hv_vp_assist_page **hv_vp_assist_page;
  38 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
  39
  40 void  __percpu **hyperv_pcpu_input_arg;
  41 EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg);
  42
  43 u32 hv_max_vp_index;
  44 EXPORT_SYMBOL_GPL(hv_max_vp_index);
  45
  46 void *hv_alloc_hyperv_page(void)
  47 {
  48         BUILD_BUG_ON(PAGE_SIZE != HV_HYP_PAGE_SIZE);
  49
  50         return (void *)__get_free_page(GFP_KERNEL);
  51 }
  52 EXPORT_SYMBOL_GPL(hv_alloc_hyperv_page);
  53
  54 void *hv_alloc_hyperv_zeroed_page(void)
  55 {
  56         BUILD_BUG_ON(PAGE_SIZE != HV_HYP_PAGE_SIZE);
  57
  58         return (void *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
  59 }
  60 EXPORT_SYMBOL_GPL(hv_alloc_hyperv_zeroed_page);
  61
  62 void hv_free_hyperv_page(unsigned long addr)
  63 {
  64         free_page(addr);
  65 }
  66 EXPORT_SYMBOL_GPL(hv_free_hyperv_page);
  67
  68 static int hv_cpu_init(unsigned int cpu)
  69 {
  70         u64 msr_vp_index;
  71         struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()];
  72         void **input_arg;
  73         struct page *pg;
  74
  75         input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
  76         /* hv_cpu_init() can be called with IRQs disabled from hv_resume() */
  77         pg = alloc_page(irqs_disabled() ? GFP_ATOMIC : GFP_KERNEL);
  78         if (unlikely(!pg))
  79                 return -ENOMEM;
  80         *input_arg = page_address(pg);
  81
  82         hv_get_vp_index(msr_vp_index);
  83
  84         hv_vp_index[smp_processor_id()] = msr_vp_index;
  85
  86         if (msr_vp_index > hv_max_vp_index)
  87                 hv_max_vp_index = msr_vp_index;
  88
  89         if (!hv_vp_assist_page)
  90                 return 0;
  91
  92         /*
  93          * The VP ASSIST PAGE is an "overlay" page (see Hyper-V TLFS's Section
  94          * 5.2.1 "GPA Overlay Pages"). Here it must be zeroed out to make sure
  95          * we always write the EOI MSR in hv_apic_eoi_write() *after* the
  96          * EOI optimization is disabled in hv_cpu_die(), otherwise a CPU may
  97          * not be stopped in the case of CPU offlining and the VM will hang.
  98          */
  99         if (!*hvp) {
 100                 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL | __GFP_ZERO,
 101                                  PAGE_KERNEL);
 102         }
 103
 104         if (*hvp) {
 105                 u64 val;
 106
 107                 val = vmalloc_to_pfn(*hvp);
 108                 val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) |
 109                         HV_X64_MSR_VP_ASSIST_PAGE_ENABLE;
 110
 111                 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val);
 112         }
 113
 114         return 0;
 115 }
 116
 117 static void (*hv_reenlightenment_cb)(void);
 118
 119 static void hv_reenlightenment_notify(struct work_struct *dummy)
 120 {
 121         struct hv_tsc_emulation_status emu_status;
 122
 123         rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 124
 125         /* Don't issue the callback if TSC accesses are not emulated */
 126         if (hv_reenlightenment_cb && emu_status.inprogress)
 127                 hv_reenlightenment_cb();
 128 }
 129 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
 130
 131 void hyperv_stop_tsc_emulation(void)
 132 {
 133         u64 freq;
 134         struct hv_tsc_emulation_status emu_status;
 135
 136         rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 137         emu_status.inprogress = 0;
 138         wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 139
 140         rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
 141         tsc_khz = div64_u64(freq, 1000);
 142 }
 143 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
 144
 145 static inline bool hv_reenlightenment_available(void)
 146 {
 147         /*
 148          * Check for required features and priviliges to make TSC frequency
 149          * change notifications work.
 150          */
 151         return ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
 152                 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
 153                 ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT;
 154 }
 155
 156 __visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs)
 157 {
 158         entering_ack_irq();
 159
 160         inc_irq_stat(irq_hv_reenlightenment_count);
 161
 162         schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
 163
 164         exiting_irq();
 165 }
 166
 167 void set_hv_tscchange_cb(void (*cb)(void))
 168 {
 169         struct hv_reenlightenment_control re_ctrl = {
 170                 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
 171                 .enabled = 1,
 172                 .target_vp = hv_vp_index[smp_processor_id()]
 173         };
 174         struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
 175
 176         if (!hv_reenlightenment_available()) {
 177                 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
 178                 return;
 179         }
 180
 181         hv_reenlightenment_cb = cb;
 182
 183         /* Make sure callback is registered before we write to MSRs */
 184         wmb();
 185
 186         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 187         wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
 188 }
 189 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
 190
 191 void clear_hv_tscchange_cb(void)
 192 {
 193         struct hv_reenlightenment_control re_ctrl;
 194
 195         if (!hv_reenlightenment_available())
 196                 return;
 197
 198         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 199         re_ctrl.enabled = 0;
 200         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 201
 202         hv_reenlightenment_cb = NULL;
 203 }
 204 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
 205
 206 static int hv_cpu_die(unsigned int cpu)
 207 {
 208         struct hv_reenlightenment_control re_ctrl;
 209         unsigned int new_cpu;
 210         unsigned long flags;
 211         void **input_arg;
 212         void *input_pg = NULL;
 213
 214         local_irq_save(flags);
 215         input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
 216         input_pg = *input_arg;
 217         *input_arg = NULL;
 218         local_irq_restore(flags);
 219         free_page((unsigned long)input_pg);
 220
 221         if (hv_vp_assist_page && hv_vp_assist_page[cpu])
 222                 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
 223
 224         if (hv_reenlightenment_cb == NULL)
 225                 return 0;
 226
 227         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 228         if (re_ctrl.target_vp == hv_vp_index[cpu]) {
 229                 /* Reassign to some other online CPU */
 230                 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
 231
 232                 re_ctrl.target_vp = hv_vp_index[new_cpu];
 233                 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 234         }
 235
 236         return 0;
 237 }
 238
 239 static int __init hv_pci_init(void)
 240 {
 241         int gen2vm = efi_enabled(EFI_BOOT);
 242
 243         /*
 244          * For Generation-2 VM, we exit from pci_arch_init() by returning 0.
 245          * The purpose is to suppress the harmless warning:
 246          * "PCI: Fatal: No config space access function found"
 247          */
 248         if (gen2vm)
 249                 return 0;
 250
 251         /* For Generation-1 VM, we'll proceed in pci_arch_init().  */
 252         return 1;
 253 }
 254
 255 static int hv_suspend(void)
 256 {
 257         union hv_x64_msr_hypercall_contents hypercall_msr;
 258         int ret;
 259
 260         /*
 261          * Reset the hypercall page as it is going to be invalidated
 262          * accross hibernation. Setting hv_hypercall_pg to NULL ensures
 263          * that any subsequent hypercall operation fails safely instead of
 264          * crashing due to an access of an invalid page. The hypercall page
 265          * pointer is restored on resume.
 266          */
 267         hv_hypercall_pg_saved = hv_hypercall_pg;
 268         hv_hypercall_pg = NULL;
 269
 270         /* Disable the hypercall page in the hypervisor */
 271         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 272         hypercall_msr.enable = 0;
 273         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 274
 275         ret = hv_cpu_die(0);
 276         return ret;
 277 }
 278
 279 static void hv_resume(void)
 280 {
 281         union hv_x64_msr_hypercall_contents hypercall_msr;
 282         int ret;
 283
 284         ret = hv_cpu_init(0);
 285         WARN_ON(ret);
 286
 287         /* Re-enable the hypercall page */
 288         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 289         hypercall_msr.enable = 1;
 290         hypercall_msr.guest_physical_address =
 291                 vmalloc_to_pfn(hv_hypercall_pg_saved);
 292         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 293
 294         hv_hypercall_pg = hv_hypercall_pg_saved;
 295         hv_hypercall_pg_saved = NULL;
 296 }
 297
 298 /* Note: when the ops are called, only CPU0 is online and IRQs are disabled. */
 299 static struct syscore_ops hv_syscore_ops = {
 300         .suspend        = hv_suspend,
 301         .resume         = hv_resume,
 302 };
 303
 304 /*
 305  * This function is to be invoked early in the boot sequence after the
 306  * hypervisor has been detected.
 307  *
 308  * 1. Setup the hypercall page.
 309  * 2. Register Hyper-V specific clocksource.
 310  * 3. Setup Hyper-V specific APIC entry points.
 311  */
 312 void __init hyperv_init(void)
 313 {
 314         u64 guest_id, required_msrs;
 315         union hv_x64_msr_hypercall_contents hypercall_msr;
 316         int cpuhp, i;
 317
 318         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 319                 return;
 320
 321         /* Absolutely required MSRs */
 322         required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
 323                 HV_X64_MSR_VP_INDEX_AVAILABLE;
 324
 325         if ((ms_hyperv.features & required_msrs) != required_msrs)
 326                 return;
 327
 328         /*
 329          * Allocate the per-CPU state for the hypercall input arg.
 330          * If this allocation fails, we will not be able to setup
 331          * (per-CPU) hypercall input page and thus this failure is
 332          * fatal on Hyper-V.
 333          */
 334         hyperv_pcpu_input_arg = alloc_percpu(void  *);
 335
 336         BUG_ON(hyperv_pcpu_input_arg == NULL);
 337
 338         /* Allocate percpu VP index */
 339         hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
 340                                     GFP_KERNEL);
 341         if (!hv_vp_index)
 342                 return;
 343
 344         for (i = 0; i < num_possible_cpus(); i++)
 345                 hv_vp_index[i] = VP_INVAL;
 346
 347         hv_vp_assist_page = kcalloc(num_possible_cpus(),
 348                                     sizeof(*hv_vp_assist_page), GFP_KERNEL);
 349         if (!hv_vp_assist_page) {
 350                 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
 351                 goto free_vp_index;
 352         }
 353
 354         cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
 355                                   hv_cpu_init, hv_cpu_die);
 356         if (cpuhp < 0)
 357                 goto free_vp_assist_page;
 358
 359         /*
 360          * Setup the hypercall page and enable hypercalls.
 361          * 1. Register the guest ID
 362          * 2. Enable the hypercall and register the hypercall page
 363          */
 364         guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
 365         wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 366
 367         hv_hypercall_pg  = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
 368         if (hv_hypercall_pg == NULL) {
 369                 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 370                 goto remove_cpuhp_state;
 371         }
 372
 373         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 374         hypercall_msr.enable = 1;
 375         hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
 376         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 377
 378         /*
 379          * Ignore any errors in setting up stimer clockevents
 380          * as we can run with the LAPIC timer as a fallback.
 381          */
 382         (void)hv_stimer_alloc();
 383
 384         hv_apic_init();
 385
 386         x86_init.pci.arch_init = hv_pci_init;
 387
 388         register_syscore_ops(&hv_syscore_ops);
 389
 390         return;
 391
 392 remove_cpuhp_state:
 393         cpuhp_remove_state(cpuhp);
 394 free_vp_assist_page:
 395         kfree(hv_vp_assist_page);
 396         hv_vp_assist_page = NULL;
 397 free_vp_index:
 398         kfree(hv_vp_index);
 399         hv_vp_index = NULL;
 400 }
 401
 402 /*
 403  * This routine is called before kexec/kdump, it does the required cleanup.
 404  */
 405 void hyperv_cleanup(void)
 406 {
 407         union hv_x64_msr_hypercall_contents hypercall_msr;
 408
 409         unregister_syscore_ops(&hv_syscore_ops);
 410
 411         /* Reset our OS id */
 412         wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 413
 414         /*
 415          * Reset hypercall page reference before reset the page,
 416          * let hypercall operations fail safely rather than
 417          * panic the kernel for using invalid hypercall page
 418          */
 419         hv_hypercall_pg = NULL;
 420
 421         /* Reset the hypercall page */
 422         hypercall_msr.as_uint64 = 0;
 423         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 424
 425         /* Reset the TSC page */
 426         hypercall_msr.as_uint64 = 0;
 427         wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
 428 }
 429 EXPORT_SYMBOL_GPL(hyperv_cleanup);
 430
 431 void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die)
 432 {
 433         static bool panic_reported;
 434         u64 guest_id;
 435
 436         if (in_die && !panic_on_oops)
 437                 return;
 438
 439         /*
 440          * We prefer to report panic on 'die' chain as we have proper
 441          * registers to report, but if we miss it (e.g. on BUG()) we need
 442          * to report it on 'panic'.
 443          */
 444         if (panic_reported)
 445                 return;
 446         panic_reported = true;
 447
 448         rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 449
 450         wrmsrl(HV_X64_MSR_CRASH_P0, err);
 451         wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
 452         wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
 453         wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
 454         wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
 455
 456         /*
 457          * Let Hyper-V know there is crash data available
 458          */
 459         wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
 460 }
 461 EXPORT_SYMBOL_GPL(hyperv_report_panic);
 462
 463 /**
 464  * hyperv_report_panic_msg - report panic message to Hyper-V
 465  * @pa: physical address of the panic page containing the message
 466  * @size: size of the message in the page
 467  */
 468 void hyperv_report_panic_msg(phys_addr_t pa, size_t size)
 469 {
 470         /*
 471          * P3 to contain the physical address of the panic page & P4 to
 472          * contain the size of the panic data in that page. Rest of the
 473          * registers are no-op when the NOTIFY_MSG flag is set.
 474          */
 475         wrmsrl(HV_X64_MSR_CRASH_P0, 0);
 476         wrmsrl(HV_X64_MSR_CRASH_P1, 0);
 477         wrmsrl(HV_X64_MSR_CRASH_P2, 0);
 478         wrmsrl(HV_X64_MSR_CRASH_P3, pa);
 479         wrmsrl(HV_X64_MSR_CRASH_P4, size);
 480
 481         /*
 482          * Let Hyper-V know there is crash data available along with
 483          * the panic message.
 484          */
 485         wrmsrl(HV_X64_MSR_CRASH_CTL,
 486                (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG));
 487 }
 488 EXPORT_SYMBOL_GPL(hyperv_report_panic_msg);
 489
 490 bool hv_is_hyperv_initialized(void)
 491 {
 492         union hv_x64_msr_hypercall_contents hypercall_msr;
 493
 494         /*
 495          * Ensure that we're really on Hyper-V, and not a KVM or Xen
 496          * emulation of Hyper-V
 497          */
 498         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 499                 return false;
 500
 501         /*
 502          * Verify that earlier initialization succeeded by checking
 503          * that the hypercall page is setup
 504          */
 505         hypercall_msr.as_uint64 = 0;
 506         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 507
 508         return hypercall_msr.enable;
 509 }
 510 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);
 511
 512 bool hv_is_hibernation_supported(void)
 513 {
 514         return acpi_sleep_state_supported(ACPI_STATE_S4);
 515 }
 516 EXPORT_SYMBOL_GPL(hv_is_hibernation_supported);