arch/x86/hyperv/hv

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
hv_alloc_hyperv_page
hv_free_hyperv_page
hv_cpu_init
hv_reenlightenment_notify
hyperv_stop_tsc_emulation
hv_reenlightenment_available
hyperv_reenlightenment_intr
set_hv_tscchange_cb
clear_hv_tscchange_cb
hv_cpu_die
hv_pci_init
hyperv_init
hyperv_cleanup
hyperv_report_panic
hyperv_report_panic_msg
hv_is_hyperv_initialized
   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                 .target_vp = hv_vp_index[smp_processor_id()]
 159         };
 160         struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
 161 
 162         if (!hv_reenlightenment_available()) {
 163                 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
 164                 return;
 165         }
 166 
 167         hv_reenlightenment_cb = cb;
 168 
 169         /* Make sure callback is registered before we write to MSRs */
 170         wmb();
 171 
 172         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 173         wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
 174 }
 175 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
 176 
 177 void clear_hv_tscchange_cb(void)
 178 {
 179         struct hv_reenlightenment_control re_ctrl;
 180 
 181         if (!hv_reenlightenment_available())
 182                 return;
 183 
 184         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 185         re_ctrl.enabled = 0;
 186         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 187 
 188         hv_reenlightenment_cb = NULL;
 189 }
 190 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
 191 
 192 static int hv_cpu_die(unsigned int cpu)
 193 {
 194         struct hv_reenlightenment_control re_ctrl;
 195         unsigned int new_cpu;
 196         unsigned long flags;
 197         void **input_arg;
 198         void *input_pg = NULL;
 199 
 200         local_irq_save(flags);
 201         input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
 202         input_pg = *input_arg;
 203         *input_arg = NULL;
 204         local_irq_restore(flags);
 205         free_page((unsigned long)input_pg);
 206 
 207         if (hv_vp_assist_page && hv_vp_assist_page[cpu])
 208                 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
 209 
 210         if (hv_reenlightenment_cb == NULL)
 211                 return 0;
 212 
 213         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 214         if (re_ctrl.target_vp == hv_vp_index[cpu]) {
 215                 /* Reassign to some other online CPU */
 216                 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
 217 
 218                 re_ctrl.target_vp = hv_vp_index[new_cpu];
 219                 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 220         }
 221 
 222         return 0;
 223 }
 224 
 225 static int __init hv_pci_init(void)
 226 {
 227         int gen2vm = efi_enabled(EFI_BOOT);
 228 
 229         /*
 230          * For Generation-2 VM, we exit from pci_arch_init() by returning 0.
 231          * The purpose is to suppress the harmless warning:
 232          * "PCI: Fatal: No config space access function found"
 233          */
 234         if (gen2vm)
 235                 return 0;
 236 
 237         /* For Generation-1 VM, we'll proceed in pci_arch_init().  */
 238         return 1;
 239 }
 240 
 241 /*
 242  * This function is to be invoked early in the boot sequence after the
 243  * hypervisor has been detected.
 244  *
 245  * 1. Setup the hypercall page.
 246  * 2. Register Hyper-V specific clocksource.
 247  * 3. Setup Hyper-V specific APIC entry points.
 248  */
 249 void __init hyperv_init(void)
 250 {
 251         u64 guest_id, required_msrs;
 252         union hv_x64_msr_hypercall_contents hypercall_msr;
 253         int cpuhp, i;
 254 
 255         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 256                 return;
 257 
 258         /* Absolutely required MSRs */
 259         required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
 260                 HV_X64_MSR_VP_INDEX_AVAILABLE;
 261 
 262         if ((ms_hyperv.features & required_msrs) != required_msrs)
 263                 return;
 264 
 265         /*
 266          * Allocate the per-CPU state for the hypercall input arg.
 267          * If this allocation fails, we will not be able to setup
 268          * (per-CPU) hypercall input page and thus this failure is
 269          * fatal on Hyper-V.
 270          */
 271         hyperv_pcpu_input_arg = alloc_percpu(void  *);
 272 
 273         BUG_ON(hyperv_pcpu_input_arg == NULL);
 274 
 275         /* Allocate percpu VP index */
 276         hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
 277                                     GFP_KERNEL);
 278         if (!hv_vp_index)
 279                 return;
 280 
 281         for (i = 0; i < num_possible_cpus(); i++)
 282                 hv_vp_index[i] = VP_INVAL;
 283 
 284         hv_vp_assist_page = kcalloc(num_possible_cpus(),
 285                                     sizeof(*hv_vp_assist_page), GFP_KERNEL);
 286         if (!hv_vp_assist_page) {
 287                 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
 288                 goto free_vp_index;
 289         }
 290 
 291         cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
 292                                   hv_cpu_init, hv_cpu_die);
 293         if (cpuhp < 0)
 294                 goto free_vp_assist_page;
 295 
 296         /*
 297          * Setup the hypercall page and enable hypercalls.
 298          * 1. Register the guest ID
 299          * 2. Enable the hypercall and register the hypercall page
 300          */
 301         guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
 302         wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 303 
 304         hv_hypercall_pg  = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
 305         if (hv_hypercall_pg == NULL) {
 306                 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 307                 goto remove_cpuhp_state;
 308         }
 309 
 310         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 311         hypercall_msr.enable = 1;
 312         hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
 313         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 314 
 315         hv_apic_init();
 316 
 317         x86_init.pci.arch_init = hv_pci_init;
 318 
 319         return;
 320 
 321 remove_cpuhp_state:
 322         cpuhp_remove_state(cpuhp);
 323 free_vp_assist_page:
 324         kfree(hv_vp_assist_page);
 325         hv_vp_assist_page = NULL;
 326 free_vp_index:
 327         kfree(hv_vp_index);
 328         hv_vp_index = NULL;
 329 }
 330 
 331 /*
 332  * This routine is called before kexec/kdump, it does the required cleanup.
 333  */
 334 void hyperv_cleanup(void)
 335 {
 336         union hv_x64_msr_hypercall_contents hypercall_msr;
 337 
 338         /* Reset our OS id */
 339         wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 340 
 341         /*
 342          * Reset hypercall page reference before reset the page,
 343          * let hypercall operations fail safely rather than
 344          * panic the kernel for using invalid hypercall page
 345          */
 346         hv_hypercall_pg = NULL;
 347 
 348         /* Reset the hypercall page */
 349         hypercall_msr.as_uint64 = 0;
 350         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 351 
 352         /* Reset the TSC page */
 353         hypercall_msr.as_uint64 = 0;
 354         wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
 355 }
 356 EXPORT_SYMBOL_GPL(hyperv_cleanup);
 357 
 358 void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die)
 359 {
 360         static bool panic_reported;
 361         u64 guest_id;
 362 
 363         if (in_die && !panic_on_oops)
 364                 return;
 365 
 366         /*
 367          * We prefer to report panic on 'die' chain as we have proper
 368          * registers to report, but if we miss it (e.g. on BUG()) we need
 369          * to report it on 'panic'.
 370          */
 371         if (panic_reported)
 372                 return;
 373         panic_reported = true;
 374 
 375         rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 376 
 377         wrmsrl(HV_X64_MSR_CRASH_P0, err);
 378         wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
 379         wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
 380         wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
 381         wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
 382 
 383         /*
 384          * Let Hyper-V know there is crash data available
 385          */
 386         wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
 387 }
 388 EXPORT_SYMBOL_GPL(hyperv_report_panic);
 389 
 390 /**
 391  * hyperv_report_panic_msg - report panic message to Hyper-V
 392  * @pa: physical address of the panic page containing the message
 393  * @size: size of the message in the page
 394  */
 395 void hyperv_report_panic_msg(phys_addr_t pa, size_t size)
 396 {
 397         /*
 398          * P3 to contain the physical address of the panic page & P4 to
 399          * contain the size of the panic data in that page. Rest of the
 400          * registers are no-op when the NOTIFY_MSG flag is set.
 401          */
 402         wrmsrl(HV_X64_MSR_CRASH_P0, 0);
 403         wrmsrl(HV_X64_MSR_CRASH_P1, 0);
 404         wrmsrl(HV_X64_MSR_CRASH_P2, 0);
 405         wrmsrl(HV_X64_MSR_CRASH_P3, pa);
 406         wrmsrl(HV_X64_MSR_CRASH_P4, size);
 407 
 408         /*
 409          * Let Hyper-V know there is crash data available along with
 410          * the panic message.
 411          */
 412         wrmsrl(HV_X64_MSR_CRASH_CTL,
 413                (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG));
 414 }
 415 EXPORT_SYMBOL_GPL(hyperv_report_panic_msg);
 416 
 417 bool hv_is_hyperv_initialized(void)
 418 {
 419         union hv_x64_msr_hypercall_contents hypercall_msr;
 420 
 421         /*
 422          * Ensure that we're really on Hyper-V, and not a KVM or Xen
 423          * emulation of Hyper-V
 424          */
 425         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 426                 return false;
 427 
 428         /*
 429          * Verify that earlier initialization succeeded by checking
 430          * that the hypercall page is setup
 431          */
 432         hypercall_msr.as_uint64 = 0;
 433         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 434 
 435         return hypercall_msr.enable;
 436 }
 437 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);
/* [<][>][^][v][top][bottom][index][help] */
root/arch/x86/hyperv/hv_init.c

DEFINITIONS
