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