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