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