arch/x86/hyperv/hv_init.c

   1 /*
   2  * X86 specific Hyper-V initialization code.
   3  *
   4  * Copyright (C) 2016, Microsoft, Inc.
   5  *
   6  * Author : K. Y. Srinivasan <kys@microsoft.com>
   7  *
   8  * This program is free software; you can redistribute it and/or modify it
   9  * under the terms of the GNU General Public License version 2 as published
  10  * by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  15  * NON INFRINGEMENT.  See the GNU General Public License for more
  16  * details.
  17  *
  18  */
  19
  20 #include <linux/efi.h>
  21 #include <linux/types.h>
  22 #include <asm/apic.h>
  23 #include <asm/desc.h>
  24 #include <asm/hypervisor.h>
  25 #include <asm/hyperv-tlfs.h>
  26 #include <asm/mshyperv.h>
  27 #include <linux/version.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/mm.h>
  30 #include <linux/clockchips.h>
  31 #include <linux/hyperv.h>
  32 #include <linux/slab.h>
  33 #include <linux/cpuhotplug.h>
  34
  35 #ifdef CONFIG_HYPERV_TSCPAGE
  36
  37 static struct ms_hyperv_tsc_page *tsc_pg;
  38
  39 struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
  40 {
  41         return tsc_pg;
  42 }
  43 EXPORT_SYMBOL_GPL(hv_get_tsc_page);
  44
  45 static u64 read_hv_clock_tsc(struct clocksource *arg)
  46 {
  47         u64 current_tick = hv_read_tsc_page(tsc_pg);
  48
  49         if (current_tick == U64_MAX)
  50                 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
  51
  52         return current_tick;
  53 }
  54
  55 static struct clocksource hyperv_cs_tsc = {
  56                 .name           = "hyperv_clocksource_tsc_page",
  57                 .rating         = 400,
  58                 .read           = read_hv_clock_tsc,
  59                 .mask           = CLOCKSOURCE_MASK(64),
  60                 .flags          = CLOCK_SOURCE_IS_CONTINUOUS,
  61 };
  62 #endif
  63
  64 static u64 read_hv_clock_msr(struct clocksource *arg)
  65 {
  66         u64 current_tick;
  67         /*
  68          * Read the partition counter to get the current tick count. This count
  69          * is set to 0 when the partition is created and is incremented in
  70          * 100 nanosecond units.
  71          */
  72         rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
  73         return current_tick;
  74 }
  75
  76 static struct clocksource hyperv_cs_msr = {
  77         .name           = "hyperv_clocksource_msr",
  78         .rating         = 400,
  79         .read           = read_hv_clock_msr,
  80         .mask           = CLOCKSOURCE_MASK(64),
  81         .flags          = CLOCK_SOURCE_IS_CONTINUOUS,
  82 };
  83
  84 void *hv_hypercall_pg;
  85 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
  86 struct clocksource *hyperv_cs;
  87 EXPORT_SYMBOL_GPL(hyperv_cs);
  88
  89 u32 *hv_vp_index;
  90 EXPORT_SYMBOL_GPL(hv_vp_index);
  91
  92 struct hv_vp_assist_page **hv_vp_assist_page;
  93 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
  94
  95 void  __percpu **hyperv_pcpu_input_arg;
  96 EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg);
  97
  98 u32 hv_max_vp_index;
  99 EXPORT_SYMBOL_GPL(hv_max_vp_index);
 100
 101 static int hv_cpu_init(unsigned int cpu)
 102 {
 103         u64 msr_vp_index;
 104         struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()];
 105         void **input_arg;
 106
 107         input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
 108         *input_arg = page_address(alloc_page(GFP_KERNEL));
 109
 110         hv_get_vp_index(msr_vp_index);
 111
 112         hv_vp_index[smp_processor_id()] = msr_vp_index;
 113
 114         if (msr_vp_index > hv_max_vp_index)
 115                 hv_max_vp_index = msr_vp_index;
 116
 117         if (!hv_vp_assist_page)
 118                 return 0;
 119
 120         if (!*hvp)
 121                 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
 122
 123         if (*hvp) {
 124                 u64 val;
 125
 126                 val = vmalloc_to_pfn(*hvp);
 127                 val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) |
 128                         HV_X64_MSR_VP_ASSIST_PAGE_ENABLE;
 129
 130                 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val);
 131         }
 132
 133         return 0;
 134 }
 135
 136 static void (*hv_reenlightenment_cb)(void);
 137
 138 static void hv_reenlightenment_notify(struct work_struct *dummy)
 139 {
 140         struct hv_tsc_emulation_status emu_status;
 141
 142         rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 143
 144         /* Don't issue the callback if TSC accesses are not emulated */
 145         if (hv_reenlightenment_cb && emu_status.inprogress)
 146                 hv_reenlightenment_cb();
 147 }
 148 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
 149
 150 void hyperv_stop_tsc_emulation(void)
 151 {
 152         u64 freq;
 153         struct hv_tsc_emulation_status emu_status;
 154
 155         rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 156         emu_status.inprogress = 0;
 157         wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 158
 159         rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
 160         tsc_khz = div64_u64(freq, 1000);
 161 }
 162 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
 163
 164 static inline bool hv_reenlightenment_available(void)
 165 {
 166         /*
 167          * Check for required features and priviliges to make TSC frequency
 168          * change notifications work.
 169          */
 170         return ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
 171                 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
 172                 ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT;
 173 }
 174
 175 __visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs)
 176 {
 177         entering_ack_irq();
 178
 179         inc_irq_stat(irq_hv_reenlightenment_count);
 180
 181         schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
 182
 183         exiting_irq();
 184 }
 185
 186 void set_hv_tscchange_cb(void (*cb)(void))
 187 {
 188         struct hv_reenlightenment_control re_ctrl = {
 189                 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
 190                 .enabled = 1,
 191                 .target_vp = hv_vp_index[smp_processor_id()]
 192         };
 193         struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
 194
 195         if (!hv_reenlightenment_available()) {
 196                 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
 197                 return;
 198         }
 199
 200         hv_reenlightenment_cb = cb;
 201
 202         /* Make sure callback is registered before we write to MSRs */
 203         wmb();
 204
 205         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 206         wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
 207 }
 208 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
 209
 210 void clear_hv_tscchange_cb(void)
 211 {
 212         struct hv_reenlightenment_control re_ctrl;
 213
 214         if (!hv_reenlightenment_available())
 215                 return;
 216
 217         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 218         re_ctrl.enabled = 0;
 219         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 220
 221         hv_reenlightenment_cb = NULL;
 222 }
 223 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
 224
 225 static int hv_cpu_die(unsigned int cpu)
 226 {
 227         struct hv_reenlightenment_control re_ctrl;
 228         unsigned int new_cpu;
 229         unsigned long flags;
 230         void **input_arg;
 231         void *input_pg = NULL;
 232
 233         local_irq_save(flags);
 234         input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
 235         input_pg = *input_arg;
 236         *input_arg = NULL;
 237         local_irq_restore(flags);
 238         free_page((unsigned long)input_pg);
 239
 240         if (hv_vp_assist_page && hv_vp_assist_page[cpu])
 241                 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
 242
 243         if (hv_reenlightenment_cb == NULL)
 244                 return 0;
 245
 246         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 247         if (re_ctrl.target_vp == hv_vp_index[cpu]) {
 248                 /* Reassign to some other online CPU */
 249                 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
 250
 251                 re_ctrl.target_vp = hv_vp_index[new_cpu];
 252                 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 253         }
 254
 255         return 0;
 256 }
 257
 258 static int __init hv_pci_init(void)
 259 {
 260         int gen2vm = efi_enabled(EFI_BOOT);
 261
 262         /*
 263          * For Generation-2 VM, we exit from pci_arch_init() by returning 0.
 264          * The purpose is to suppress the harmless warning:
 265          * "PCI: Fatal: No config space access function found"
 266          */
 267         if (gen2vm)
 268                 return 0;
 269
 270         /* For Generation-1 VM, we'll proceed in pci_arch_init().  */
 271         return 1;
 272 }
 273
 274 /*
 275  * This function is to be invoked early in the boot sequence after the
 276  * hypervisor has been detected.
 277  *
 278  * 1. Setup the hypercall page.
 279  * 2. Register Hyper-V specific clocksource.
 280  * 3. Setup Hyper-V specific APIC entry points.
 281  */
 282 void __init hyperv_init(void)
 283 {
 284         u64 guest_id, required_msrs;
 285         union hv_x64_msr_hypercall_contents hypercall_msr;
 286         int cpuhp, i;
 287
 288         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 289                 return;
 290
 291         /* Absolutely required MSRs */
 292         required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
 293                 HV_X64_MSR_VP_INDEX_AVAILABLE;
 294
 295         if ((ms_hyperv.features & required_msrs) != required_msrs)
 296                 return;
 297
 298         /*
 299          * Allocate the per-CPU state for the hypercall input arg.
 300          * If this allocation fails, we will not be able to setup
 301          * (per-CPU) hypercall input page and thus this failure is
 302          * fatal on Hyper-V.
 303          */
 304         hyperv_pcpu_input_arg = alloc_percpu(void  *);
 305
 306         BUG_ON(hyperv_pcpu_input_arg == NULL);
 307
 308         /* Allocate percpu VP index */
 309         hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
 310                                     GFP_KERNEL);
 311         if (!hv_vp_index)
 312                 return;
 313
 314         for (i = 0; i < num_possible_cpus(); i++)
 315                 hv_vp_index[i] = VP_INVAL;
 316
 317         hv_vp_assist_page = kcalloc(num_possible_cpus(),
 318                                     sizeof(*hv_vp_assist_page), GFP_KERNEL);
 319         if (!hv_vp_assist_page) {
 320                 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
 321                 goto free_vp_index;
 322         }
 323
 324         cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
 325                                   hv_cpu_init, hv_cpu_die);
 326         if (cpuhp < 0)
 327                 goto free_vp_assist_page;
 328
 329         /*
 330          * Setup the hypercall page and enable hypercalls.
 331          * 1. Register the guest ID
 332          * 2. Enable the hypercall and register the hypercall page
 333          */
 334         guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
 335         wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 336
 337         hv_hypercall_pg  = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
 338         if (hv_hypercall_pg == NULL) {
 339                 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 340                 goto remove_cpuhp_state;
 341         }
 342
 343         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 344         hypercall_msr.enable = 1;
 345         hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
 346         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 347
 348         hv_apic_init();
 349
 350         x86_init.pci.arch_init = hv_pci_init;
 351
 352         /*
 353          * Register Hyper-V specific clocksource.
 354          */
 355 #ifdef CONFIG_HYPERV_TSCPAGE
 356         if (ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE) {
 357                 union hv_x64_msr_hypercall_contents tsc_msr;
 358
 359                 tsc_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
 360                 if (!tsc_pg)
 361                         goto register_msr_cs;
 362
 363                 hyperv_cs = &hyperv_cs_tsc;
 364
 365                 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
 366
 367                 tsc_msr.enable = 1;
 368                 tsc_msr.guest_physical_address = vmalloc_to_pfn(tsc_pg);
 369
 370                 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
 371
 372                 hyperv_cs_tsc.archdata.vclock_mode = VCLOCK_HVCLOCK;
 373
 374                 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
 375                 return;
 376         }
 377 register_msr_cs:
 378 #endif
 379         /*
 380          * For 32 bit guests just use the MSR based mechanism for reading
 381          * the partition counter.
 382          */
 383
 384         hyperv_cs = &hyperv_cs_msr;
 385         if (ms_hyperv.features & HV_MSR_TIME_REF_COUNT_AVAILABLE)
 386                 clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
 387
 388         return;
 389
 390 remove_cpuhp_state:
 391         cpuhp_remove_state(cpuhp);
 392 free_vp_assist_page:
 393         kfree(hv_vp_assist_page);
 394         hv_vp_assist_page = NULL;
 395 free_vp_index:
 396         kfree(hv_vp_index);
 397         hv_vp_index = NULL;
 398 }
 399
 400 /*
 401  * This routine is called before kexec/kdump, it does the required cleanup.
 402  */
 403 void hyperv_cleanup(void)
 404 {
 405         union hv_x64_msr_hypercall_contents hypercall_msr;
 406
 407         /* Reset our OS id */
 408         wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 409
 410         /*
 411          * Reset hypercall page reference before reset the page,
 412          * let hypercall operations fail safely rather than
 413          * panic the kernel for using invalid hypercall page
 414          */
 415         hv_hypercall_pg = NULL;
 416
 417         /* Reset the hypercall page */
 418         hypercall_msr.as_uint64 = 0;
 419         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 420
 421         /* Reset the TSC page */
 422         hypercall_msr.as_uint64 = 0;
 423         wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
 424 }
 425 EXPORT_SYMBOL_GPL(hyperv_cleanup);
 426
 427 void hyperv_report_panic(struct pt_regs *regs, long err)
 428 {
 429         static bool panic_reported;
 430         u64 guest_id;
 431
 432         /*
 433          * We prefer to report panic on 'die' chain as we have proper
 434          * registers to report, but if we miss it (e.g. on BUG()) we need
 435          * to report it on 'panic'.
 436          */
 437         if (panic_reported)
 438                 return;
 439         panic_reported = true;
 440
 441         rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 442
 443         wrmsrl(HV_X64_MSR_CRASH_P0, err);
 444         wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
 445         wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
 446         wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
 447         wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
 448
 449         /*
 450          * Let Hyper-V know there is crash data available
 451          */
 452         wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
 453 }
 454 EXPORT_SYMBOL_GPL(hyperv_report_panic);
 455
 456 /**
 457  * hyperv_report_panic_msg - report panic message to Hyper-V
 458  * @pa: physical address of the panic page containing the message
 459  * @size: size of the message in the page
 460  */
 461 void hyperv_report_panic_msg(phys_addr_t pa, size_t size)
 462 {
 463         /*
 464          * P3 to contain the physical address of the panic page & P4 to
 465          * contain the size of the panic data in that page. Rest of the
 466          * registers are no-op when the NOTIFY_MSG flag is set.
 467          */
 468         wrmsrl(HV_X64_MSR_CRASH_P0, 0);
 469         wrmsrl(HV_X64_MSR_CRASH_P1, 0);
 470         wrmsrl(HV_X64_MSR_CRASH_P2, 0);
 471         wrmsrl(HV_X64_MSR_CRASH_P3, pa);
 472         wrmsrl(HV_X64_MSR_CRASH_P4, size);
 473
 474         /*
 475          * Let Hyper-V know there is crash data available along with
 476          * the panic message.
 477          */
 478         wrmsrl(HV_X64_MSR_CRASH_CTL,
 479                (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG));
 480 }
 481 EXPORT_SYMBOL_GPL(hyperv_report_panic_msg);
 482
 483 bool hv_is_hyperv_initialized(void)
 484 {
 485         union hv_x64_msr_hypercall_contents hypercall_msr;
 486
 487         /*
 488          * Ensure that we're really on Hyper-V, and not a KVM or Xen
 489          * emulation of Hyper-V
 490          */
 491         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 492                 return false;
 493
 494         /*
 495          * Verify that earlier initialization succeeded by checking
 496          * that the hypercall page is setup
 497          */
 498         hypercall_msr.as_uint64 = 0;
 499         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 500
 501         return hypercall_msr.enable;
 502 }
 503 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);