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