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