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/clockchips.h>
  21 #include <linux/hyperv.h>
  22 #include <linux/slab.h>
  23 #include <linux/cpuhotplug.h>
  24
  25 #ifdef CONFIG_HYPERV_TSCPAGE
  26
  27 static struct ms_hyperv_tsc_page *tsc_pg;
  28
  29 struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
  30 {
  31         return tsc_pg;
  32 }
  33 EXPORT_SYMBOL_GPL(hv_get_tsc_page);
  34
  35 static u64 read_hv_clock_tsc(struct clocksource *arg)
  36 {
  37         u64 current_tick = hv_read_tsc_page(tsc_pg);
  38
  39         if (current_tick == U64_MAX)
  40                 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
  41
  42         return current_tick;
  43 }
  44
  45 static struct clocksource hyperv_cs_tsc = {
  46                 .name           = "hyperv_clocksource_tsc_page",
  47                 .rating         = 400,
  48                 .read           = read_hv_clock_tsc,
  49                 .mask           = CLOCKSOURCE_MASK(64),
  50                 .flags          = CLOCK_SOURCE_IS_CONTINUOUS,
  51 };
  52 #endif
  53
  54 static u64 read_hv_clock_msr(struct clocksource *arg)
  55 {
  56         u64 current_tick;
  57         /*
  58          * Read the partition counter to get the current tick count. This count
  59          * is set to 0 when the partition is created and is incremented in
  60          * 100 nanosecond units.
  61          */
  62         rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
  63         return current_tick;
  64 }
  65
  66 static struct clocksource hyperv_cs_msr = {
  67         .name           = "hyperv_clocksource_msr",
  68         .rating         = 400,
  69         .read           = read_hv_clock_msr,
  70         .mask           = CLOCKSOURCE_MASK(64),
  71         .flags          = CLOCK_SOURCE_IS_CONTINUOUS,
  72 };
  73
  74 void *hv_hypercall_pg;
  75 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
  76 struct clocksource *hyperv_cs;
  77 EXPORT_SYMBOL_GPL(hyperv_cs);
  78
  79 u32 *hv_vp_index;
  80 EXPORT_SYMBOL_GPL(hv_vp_index);
  81
  82 struct hv_vp_assist_page **hv_vp_assist_page;
  83 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
  84
  85 void  __percpu **hyperv_pcpu_input_arg;
  86 EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg);
  87
  88 u32 hv_max_vp_index;
  89 EXPORT_SYMBOL_GPL(hv_max_vp_index);
  90
  91 static int hv_cpu_init(unsigned int cpu)
  92 {
  93         u64 msr_vp_index;
  94         struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()];
  95         void **input_arg;
  96         struct page *pg;
  97
  98         input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
  99         pg = alloc_page(GFP_KERNEL);
 100         if (unlikely(!pg))
 101                 return -ENOMEM;
 102         *input_arg = page_address(pg);
 103
 104         hv_get_vp_index(msr_vp_index);
 105
 106         hv_vp_index[smp_processor_id()] = msr_vp_index;
 107
 108         if (msr_vp_index > hv_max_vp_index)
 109                 hv_max_vp_index = msr_vp_index;
 110
 111         if (!hv_vp_assist_page)
 112                 return 0;
 113
 114         if (!*hvp)
 115                 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
 116
 117         if (*hvp) {
 118                 u64 val;
 119
 120                 val = vmalloc_to_pfn(*hvp);
 121                 val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) |
 122                         HV_X64_MSR_VP_ASSIST_PAGE_ENABLE;
 123
 124                 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val);
 125         }
 126
 127         return 0;
 128 }
 129
 130 static void (*hv_reenlightenment_cb)(void);
 131
 132 static void hv_reenlightenment_notify(struct work_struct *dummy)
 133 {
 134         struct hv_tsc_emulation_status emu_status;
 135
 136         rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 137
 138         /* Don't issue the callback if TSC accesses are not emulated */
 139         if (hv_reenlightenment_cb && emu_status.inprogress)
 140                 hv_reenlightenment_cb();
 141 }
 142 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
 143
 144 void hyperv_stop_tsc_emulation(void)
 145 {
 146         u64 freq;
 147         struct hv_tsc_emulation_status emu_status;
 148
 149         rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 150         emu_status.inprogress = 0;
 151         wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 152
 153         rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
 154         tsc_khz = div64_u64(freq, 1000);
 155 }
 156 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
 157
 158 static inline bool hv_reenlightenment_available(void)
 159 {
 160         /*
 161          * Check for required features and priviliges to make TSC frequency
 162          * change notifications work.
 163          */
 164         return ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
 165                 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
 166                 ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT;
 167 }
 168
 169 __visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs)
 170 {
 171         entering_ack_irq();
 172
 173         inc_irq_stat(irq_hv_reenlightenment_count);
 174
 175         schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
 176
 177         exiting_irq();
 178 }
 179
 180 void set_hv_tscchange_cb(void (*cb)(void))
 181 {
 182         struct hv_reenlightenment_control re_ctrl = {
 183                 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
 184                 .enabled = 1,
 185                 .target_vp = hv_vp_index[smp_processor_id()]
 186         };
 187         struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
 188
 189         if (!hv_reenlightenment_available()) {
 190                 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
 191                 return;
 192         }
 193
 194         hv_reenlightenment_cb = cb;
 195
 196         /* Make sure callback is registered before we write to MSRs */
 197         wmb();
 198
 199         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 200         wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
 201 }
 202 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
 203
 204 void clear_hv_tscchange_cb(void)
 205 {
 206         struct hv_reenlightenment_control re_ctrl;
 207
 208         if (!hv_reenlightenment_available())
 209                 return;
 210
 211         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 212         re_ctrl.enabled = 0;
 213         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 214
 215         hv_reenlightenment_cb = NULL;
 216 }
 217 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
 218
 219 static int hv_cpu_die(unsigned int cpu)
 220 {
 221         struct hv_reenlightenment_control re_ctrl;
 222         unsigned int new_cpu;
 223         unsigned long flags;
 224         void **input_arg;
 225         void *input_pg = NULL;
 226
 227         local_irq_save(flags);
 228         input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
 229         input_pg = *input_arg;
 230         *input_arg = NULL;
 231         local_irq_restore(flags);
 232         free_page((unsigned long)input_pg);
 233
 234         if (hv_vp_assist_page && hv_vp_assist_page[cpu])
 235                 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
 236
 237         if (hv_reenlightenment_cb == NULL)
 238                 return 0;
 239
 240         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 241         if (re_ctrl.target_vp == hv_vp_index[cpu]) {
 242                 /* Reassign to some other online CPU */
 243                 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
 244
 245                 re_ctrl.target_vp = hv_vp_index[new_cpu];
 246                 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 247         }
 248
 249         return 0;
 250 }
 251
 252 static int __init hv_pci_init(void)
 253 {
 254         int gen2vm = efi_enabled(EFI_BOOT);
 255
 256         /*
 257          * For Generation-2 VM, we exit from pci_arch_init() by returning 0.
 258          * The purpose is to suppress the harmless warning:
 259          * "PCI: Fatal: No config space access function found"
 260          */
 261         if (gen2vm)
 262                 return 0;
 263
 264         /* For Generation-1 VM, we'll proceed in pci_arch_init().  */
 265         return 1;
 266 }
 267
 268 /*
 269  * This function is to be invoked early in the boot sequence after the
 270  * hypervisor has been detected.
 271  *
 272  * 1. Setup the hypercall page.
 273  * 2. Register Hyper-V specific clocksource.
 274  * 3. Setup Hyper-V specific APIC entry points.
 275  */
 276 void __init hyperv_init(void)
 277 {
 278         u64 guest_id, required_msrs;
 279         union hv_x64_msr_hypercall_contents hypercall_msr;
 280         int cpuhp, i;
 281
 282         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 283                 return;
 284
 285         /* Absolutely required MSRs */
 286         required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
 287                 HV_X64_MSR_VP_INDEX_AVAILABLE;
 288
 289         if ((ms_hyperv.features & required_msrs) != required_msrs)
 290                 return;
 291
 292         /*
 293          * Allocate the per-CPU state for the hypercall input arg.
 294          * If this allocation fails, we will not be able to setup
 295          * (per-CPU) hypercall input page and thus this failure is
 296          * fatal on Hyper-V.
 297          */
 298         hyperv_pcpu_input_arg = alloc_percpu(void  *);
 299
 300         BUG_ON(hyperv_pcpu_input_arg == NULL);
 301
 302         /* Allocate percpu VP index */
 303         hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
 304                                     GFP_KERNEL);
 305         if (!hv_vp_index)
 306                 return;
 307
 308         for (i = 0; i < num_possible_cpus(); i++)
 309                 hv_vp_index[i] = VP_INVAL;
 310
 311         hv_vp_assist_page = kcalloc(num_possible_cpus(),
 312                                     sizeof(*hv_vp_assist_page), GFP_KERNEL);
 313         if (!hv_vp_assist_page) {
 314                 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
 315                 goto free_vp_index;
 316         }
 317
 318         cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
 319                                   hv_cpu_init, hv_cpu_die);
 320         if (cpuhp < 0)
 321                 goto free_vp_assist_page;
 322
 323         /*
 324          * Setup the hypercall page and enable hypercalls.
 325          * 1. Register the guest ID
 326          * 2. Enable the hypercall and register the hypercall page
 327          */
 328         guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
 329         wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 330
 331         hv_hypercall_pg  = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
 332         if (hv_hypercall_pg == NULL) {
 333                 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 334                 goto remove_cpuhp_state;
 335         }
 336
 337         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 338         hypercall_msr.enable = 1;
 339         hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
 340         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 341
 342         hv_apic_init();
 343
 344         x86_init.pci.arch_init = hv_pci_init;
 345
 346         /*
 347          * Register Hyper-V specific clocksource.
 348          */
 349 #ifdef CONFIG_HYPERV_TSCPAGE
 350         if (ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE) {
 351                 union hv_x64_msr_hypercall_contents tsc_msr;
 352
 353                 tsc_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
 354                 if (!tsc_pg)
 355                         goto register_msr_cs;
 356
 357                 hyperv_cs = &hyperv_cs_tsc;
 358
 359                 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
 360
 361                 tsc_msr.enable = 1;
 362                 tsc_msr.guest_physical_address = vmalloc_to_pfn(tsc_pg);
 363
 364                 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
 365
 366                 hyperv_cs_tsc.archdata.vclock_mode = VCLOCK_HVCLOCK;
 367
 368                 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
 369                 return;
 370         }
 371 register_msr_cs:
 372 #endif
 373         /*
 374          * For 32 bit guests just use the MSR based mechanism for reading
 375          * the partition counter.
 376          */
 377
 378         hyperv_cs = &hyperv_cs_msr;
 379         if (ms_hyperv.features & HV_MSR_TIME_REF_COUNT_AVAILABLE)
 380                 clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
 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         /* Reset our OS id */
 402         wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 403
 404         /*
 405          * Reset hypercall page reference before reset the page,
 406          * let hypercall operations fail safely rather than
 407          * panic the kernel for using invalid hypercall page
 408          */
 409         hv_hypercall_pg = NULL;
 410
 411         /* Reset the hypercall page */
 412         hypercall_msr.as_uint64 = 0;
 413         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 414
 415         /* Reset the TSC page */
 416         hypercall_msr.as_uint64 = 0;
 417         wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
 418 }
 419 EXPORT_SYMBOL_GPL(hyperv_cleanup);
 420
 421 void hyperv_report_panic(struct pt_regs *regs, long err)
 422 {
 423         static bool panic_reported;
 424         u64 guest_id;
 425
 426         /*
 427          * We prefer to report panic on 'die' chain as we have proper
 428          * registers to report, but if we miss it (e.g. on BUG()) we need
 429          * to report it on 'panic'.
 430          */
 431         if (panic_reported)
 432                 return;
 433         panic_reported = true;
 434
 435         rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 436
 437         wrmsrl(HV_X64_MSR_CRASH_P0, err);
 438         wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
 439         wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
 440         wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
 441         wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
 442
 443         /*
 444          * Let Hyper-V know there is crash data available
 445          */
 446         wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
 447 }
 448 EXPORT_SYMBOL_GPL(hyperv_report_panic);
 449
 450 /**
 451  * hyperv_report_panic_msg - report panic message to Hyper-V
 452  * @pa: physical address of the panic page containing the message
 453  * @size: size of the message in the page
 454  */
 455 void hyperv_report_panic_msg(phys_addr_t pa, size_t size)
 456 {
 457         /*
 458          * P3 to contain the physical address of the panic page & P4 to
 459          * contain the size of the panic data in that page. Rest of the
 460          * registers are no-op when the NOTIFY_MSG flag is set.
 461          */
 462         wrmsrl(HV_X64_MSR_CRASH_P0, 0);
 463         wrmsrl(HV_X64_MSR_CRASH_P1, 0);
 464         wrmsrl(HV_X64_MSR_CRASH_P2, 0);
 465         wrmsrl(HV_X64_MSR_CRASH_P3, pa);
 466         wrmsrl(HV_X64_MSR_CRASH_P4, size);
 467
 468         /*
 469          * Let Hyper-V know there is crash data available along with
 470          * the panic message.
 471          */
 472         wrmsrl(HV_X64_MSR_CRASH_CTL,
 473                (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG));
 474 }
 475 EXPORT_SYMBOL_GPL(hyperv_report_panic_msg);
 476
 477 bool hv_is_hyperv_initialized(void)
 478 {
 479         union hv_x64_msr_hypercall_contents hypercall_msr;
 480
 481         /*
 482          * Ensure that we're really on Hyper-V, and not a KVM or Xen
 483          * emulation of Hyper-V
 484          */
 485         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 486                 return false;
 487
 488         /*
 489          * Verify that earlier initialization succeeded by checking
 490          * that the hypercall page is setup
 491          */
 492         hypercall_msr.as_uint64 = 0;
 493         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 494
 495         return hypercall_msr.enable;
 496 }
 497 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);