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/kernel.h>
  34 #include <linux/cpuhotplug.h>
  35
  36 #ifdef CONFIG_HYPERV_TSCPAGE
  37
  38 static struct ms_hyperv_tsc_page *tsc_pg;
  39
  40 struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
  41 {
  42         return tsc_pg;
  43 }
  44 EXPORT_SYMBOL_GPL(hv_get_tsc_page);
  45
  46 static u64 read_hv_clock_tsc(struct clocksource *arg)
  47 {
  48         u64 current_tick = hv_read_tsc_page(tsc_pg);
  49
  50         if (current_tick == U64_MAX)
  51                 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
  52
  53         return current_tick;
  54 }
  55
  56 static struct clocksource hyperv_cs_tsc = {
  57                 .name           = "hyperv_clocksource_tsc_page",
  58                 .rating         = 400,
  59                 .read           = read_hv_clock_tsc,
  60                 .mask           = CLOCKSOURCE_MASK(64),
  61                 .flags          = CLOCK_SOURCE_IS_CONTINUOUS,
  62 };
  63 #endif
  64
  65 static u64 read_hv_clock_msr(struct clocksource *arg)
  66 {
  67         u64 current_tick;
  68         /*
  69          * Read the partition counter to get the current tick count. This count
  70          * is set to 0 when the partition is created and is incremented in
  71          * 100 nanosecond units.
  72          */
  73         rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
  74         return current_tick;
  75 }
  76
  77 static struct clocksource hyperv_cs_msr = {
  78         .name           = "hyperv_clocksource_msr",
  79         .rating         = 400,
  80         .read           = read_hv_clock_msr,
  81         .mask           = CLOCKSOURCE_MASK(64),
  82         .flags          = CLOCK_SOURCE_IS_CONTINUOUS,
  83 };
  84
  85 void *hv_hypercall_pg;
  86 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
  87 struct clocksource *hyperv_cs;
  88 EXPORT_SYMBOL_GPL(hyperv_cs);
  89
  90 u32 *hv_vp_index;
  91 EXPORT_SYMBOL_GPL(hv_vp_index);
  92
  93 struct hv_vp_assist_page **hv_vp_assist_page;
  94 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
  95
  96 void  __percpu **hyperv_pcpu_input_arg;
  97 EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg);
  98
  99 u32 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         struct page *pg;
 107
 108         input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
 109         pg = alloc_page(GFP_KERNEL);
 110         if (unlikely(!pg))
 111                 return -ENOMEM;
 112         *input_arg = page_address(pg);
 113
 114         hv_get_vp_index(msr_vp_index);
 115
 116         hv_vp_index[smp_processor_id()] = msr_vp_index;
 117
 118         if (msr_vp_index > hv_max_vp_index)
 119                 hv_max_vp_index = msr_vp_index;
 120
 121         if (!hv_vp_assist_page)
 122                 return 0;
 123
 124         if (!*hvp)
 125                 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
 126
 127         if (*hvp) {
 128                 u64 val;
 129
 130                 val = vmalloc_to_pfn(*hvp);
 131                 val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) |
 132                         HV_X64_MSR_VP_ASSIST_PAGE_ENABLE;
 133
 134                 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val);
 135         }
 136
 137         return 0;
 138 }
 139
 140 static void (*hv_reenlightenment_cb)(void);
 141
 142 static void hv_reenlightenment_notify(struct work_struct *dummy)
 143 {
 144         struct hv_tsc_emulation_status emu_status;
 145
 146         rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 147
 148         /* Don't issue the callback if TSC accesses are not emulated */
 149         if (hv_reenlightenment_cb && emu_status.inprogress)
 150                 hv_reenlightenment_cb();
 151 }
 152 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
 153
 154 void hyperv_stop_tsc_emulation(void)
 155 {
 156         u64 freq;
 157         struct hv_tsc_emulation_status emu_status;
 158
 159         rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 160         emu_status.inprogress = 0;
 161         wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 162
 163         rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
 164         tsc_khz = div64_u64(freq, 1000);
 165 }
 166 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
 167
 168 static inline bool hv_reenlightenment_available(void)
 169 {
 170         /*
 171          * Check for required features and priviliges to make TSC frequency
 172          * change notifications work.
 173          */
 174         return ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
 175                 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
 176                 ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT;
 177 }
 178
 179 __visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs)
 180 {
 181         entering_ack_irq();
 182
 183         inc_irq_stat(irq_hv_reenlightenment_count);
 184
 185         schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
 186
 187         exiting_irq();
 188 }
 189
 190 void set_hv_tscchange_cb(void (*cb)(void))
 191 {
 192         struct hv_reenlightenment_control re_ctrl = {
 193                 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
 194                 .enabled = 1,
 195         };
 196         struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
 197
 198         if (!hv_reenlightenment_available()) {
 199                 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
 200                 return;
 201         }
 202
 203         if (!hv_vp_index)
 204                 return;
 205
 206         hv_reenlightenment_cb = cb;
 207
 208         /* Make sure callback is registered before we write to MSRs */
 209         wmb();
 210
 211         re_ctrl.target_vp = hv_vp_index[get_cpu()];
 212
 213         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 214         wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
 215
 216         put_cpu();
 217 }
 218 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
 219
 220 void clear_hv_tscchange_cb(void)
 221 {
 222         struct hv_reenlightenment_control re_ctrl;
 223
 224         if (!hv_reenlightenment_available())
 225                 return;
 226
 227         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 228         re_ctrl.enabled = 0;
 229         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 230
 231         hv_reenlightenment_cb = NULL;
 232 }
 233 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
 234
 235 static int hv_cpu_die(unsigned int cpu)
 236 {
 237         struct hv_reenlightenment_control re_ctrl;
 238         unsigned int new_cpu;
 239         unsigned long flags;
 240         void **input_arg;
 241         void *input_pg = NULL;
 242
 243         local_irq_save(flags);
 244         input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
 245         input_pg = *input_arg;
 246         *input_arg = NULL;
 247         local_irq_restore(flags);
 248         free_page((unsigned long)input_pg);
 249
 250         if (hv_vp_assist_page && hv_vp_assist_page[cpu])
 251                 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
 252
 253         if (hv_reenlightenment_cb == NULL)
 254                 return 0;
 255
 256         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 257         if (re_ctrl.target_vp == hv_vp_index[cpu]) {
 258                 /* Reassign to some other online CPU */
 259                 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
 260
 261                 re_ctrl.target_vp = hv_vp_index[new_cpu];
 262                 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 263         }
 264
 265         return 0;
 266 }
 267
 268 static int __init hv_pci_init(void)
 269 {
 270         int gen2vm = efi_enabled(EFI_BOOT);
 271
 272         /*
 273          * For Generation-2 VM, we exit from pci_arch_init() by returning 0.
 274          * The purpose is to suppress the harmless warning:
 275          * "PCI: Fatal: No config space access function found"
 276          */
 277         if (gen2vm)
 278                 return 0;
 279
 280         /* For Generation-1 VM, we'll proceed in pci_arch_init().  */
 281         return 1;
 282 }
 283
 284 /*
 285  * This function is to be invoked early in the boot sequence after the
 286  * hypervisor has been detected.
 287  *
 288  * 1. Setup the hypercall page.
 289  * 2. Register Hyper-V specific clocksource.
 290  * 3. Setup Hyper-V specific APIC entry points.
 291  */
 292 void __init hyperv_init(void)
 293 {
 294         u64 guest_id, required_msrs;
 295         union hv_x64_msr_hypercall_contents hypercall_msr;
 296         int cpuhp, i;
 297
 298         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 299                 return;
 300
 301         /* Absolutely required MSRs */
 302         required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
 303                 HV_X64_MSR_VP_INDEX_AVAILABLE;
 304
 305         if ((ms_hyperv.features & required_msrs) != required_msrs)
 306                 return;
 307
 308         /*
 309          * Allocate the per-CPU state for the hypercall input arg.
 310          * If this allocation fails, we will not be able to setup
 311          * (per-CPU) hypercall input page and thus this failure is
 312          * fatal on Hyper-V.
 313          */
 314         hyperv_pcpu_input_arg = alloc_percpu(void  *);
 315
 316         BUG_ON(hyperv_pcpu_input_arg == NULL);
 317
 318         /* Allocate percpu VP index */
 319         hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
 320                                     GFP_KERNEL);
 321         if (!hv_vp_index)
 322                 return;
 323
 324         for (i = 0; i < num_possible_cpus(); i++)
 325                 hv_vp_index[i] = VP_INVAL;
 326
 327         hv_vp_assist_page = kcalloc(num_possible_cpus(),
 328                                     sizeof(*hv_vp_assist_page), GFP_KERNEL);
 329         if (!hv_vp_assist_page) {
 330                 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
 331                 goto free_vp_index;
 332         }
 333
 334         cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
 335                                   hv_cpu_init, hv_cpu_die);
 336         if (cpuhp < 0)
 337                 goto free_vp_assist_page;
 338
 339         /*
 340          * Setup the hypercall page and enable hypercalls.
 341          * 1. Register the guest ID
 342          * 2. Enable the hypercall and register the hypercall page
 343          */
 344         guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
 345         wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 346
 347         hv_hypercall_pg  = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
 348         if (hv_hypercall_pg == NULL) {
 349                 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 350                 goto remove_cpuhp_state;
 351         }
 352
 353         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 354         hypercall_msr.enable = 1;
 355         hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
 356         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 357
 358         hv_apic_init();
 359
 360         x86_init.pci.arch_init = hv_pci_init;
 361
 362         /*
 363          * Register Hyper-V specific clocksource.
 364          */
 365 #ifdef CONFIG_HYPERV_TSCPAGE
 366         if (ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE) {
 367                 union hv_x64_msr_hypercall_contents tsc_msr;
 368
 369                 tsc_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
 370                 if (!tsc_pg)
 371                         goto register_msr_cs;
 372
 373                 hyperv_cs = &hyperv_cs_tsc;
 374
 375                 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
 376
 377                 tsc_msr.enable = 1;
 378                 tsc_msr.guest_physical_address = vmalloc_to_pfn(tsc_pg);
 379
 380                 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
 381
 382                 hyperv_cs_tsc.archdata.vclock_mode = VCLOCK_HVCLOCK;
 383
 384                 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
 385                 return;
 386         }
 387 register_msr_cs:
 388 #endif
 389         /*
 390          * For 32 bit guests just use the MSR based mechanism for reading
 391          * the partition counter.
 392          */
 393
 394         hyperv_cs = &hyperv_cs_msr;
 395         if (ms_hyperv.features & HV_MSR_TIME_REF_COUNT_AVAILABLE)
 396                 clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
 397
 398         return;
 399
 400 remove_cpuhp_state:
 401         cpuhp_remove_state(cpuhp);
 402 free_vp_assist_page:
 403         kfree(hv_vp_assist_page);
 404         hv_vp_assist_page = NULL;
 405 free_vp_index:
 406         kfree(hv_vp_index);
 407         hv_vp_index = NULL;
 408 }
 409
 410 /*
 411  * This routine is called before kexec/kdump, it does the required cleanup.
 412  */
 413 void hyperv_cleanup(void)
 414 {
 415         union hv_x64_msr_hypercall_contents hypercall_msr;
 416
 417         /* Reset our OS id */
 418         wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 419
 420         /*
 421          * Reset hypercall page reference before reset the page,
 422          * let hypercall operations fail safely rather than
 423          * panic the kernel for using invalid hypercall page
 424          */
 425         hv_hypercall_pg = NULL;
 426
 427         /* Reset the hypercall page */
 428         hypercall_msr.as_uint64 = 0;
 429         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 430
 431         /* Reset the TSC page */
 432         hypercall_msr.as_uint64 = 0;
 433         wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
 434 }
 435 EXPORT_SYMBOL_GPL(hyperv_cleanup);
 436
 437 void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die)
 438 {
 439         static bool panic_reported;
 440         u64 guest_id;
 441
 442         if (in_die && !panic_on_oops)
 443                 return;
 444
 445         /*
 446          * We prefer to report panic on 'die' chain as we have proper
 447          * registers to report, but if we miss it (e.g. on BUG()) we need
 448          * to report it on 'panic'.
 449          */
 450         if (panic_reported)
 451                 return;
 452         panic_reported = true;
 453
 454         rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 455
 456         wrmsrl(HV_X64_MSR_CRASH_P0, err);
 457         wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
 458         wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
 459         wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
 460         wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
 461
 462         /*
 463          * Let Hyper-V know there is crash data available
 464          */
 465         wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
 466 }
 467 EXPORT_SYMBOL_GPL(hyperv_report_panic);
 468
 469 /**
 470  * hyperv_report_panic_msg - report panic message to Hyper-V
 471  * @pa: physical address of the panic page containing the message
 472  * @size: size of the message in the page
 473  */
 474 void hyperv_report_panic_msg(phys_addr_t pa, size_t size)
 475 {
 476         /*
 477          * P3 to contain the physical address of the panic page & P4 to
 478          * contain the size of the panic data in that page. Rest of the
 479          * registers are no-op when the NOTIFY_MSG flag is set.
 480          */
 481         wrmsrl(HV_X64_MSR_CRASH_P0, 0);
 482         wrmsrl(HV_X64_MSR_CRASH_P1, 0);
 483         wrmsrl(HV_X64_MSR_CRASH_P2, 0);
 484         wrmsrl(HV_X64_MSR_CRASH_P3, pa);
 485         wrmsrl(HV_X64_MSR_CRASH_P4, size);
 486
 487         /*
 488          * Let Hyper-V know there is crash data available along with
 489          * the panic message.
 490          */
 491         wrmsrl(HV_X64_MSR_CRASH_CTL,
 492                (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG));
 493 }
 494 EXPORT_SYMBOL_GPL(hyperv_report_panic_msg);
 495
 496 bool hv_is_hyperv_initialized(void)
 497 {
 498         union hv_x64_msr_hypercall_contents hypercall_msr;
 499
 500         /*
 501          * Ensure that we're really on Hyper-V, and not a KVM or Xen
 502          * emulation of Hyper-V
 503          */
 504         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 505                 return false;
 506
 507         /*
 508          * Verify that earlier initialization succeeded by checking
 509          * that the hypercall page is setup
 510          */
 511         hypercall_msr.as_uint64 = 0;
 512         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 513
 514         return hypercall_msr.enable;
 515 }
 516 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);