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                 .target_vp = hv_vp_index[smp_processor_id()]
 196         };
 197         struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
 198
 199         if (!hv_reenlightenment_available()) {
 200                 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
 201                 return;
 202         }
 203
 204         hv_reenlightenment_cb = cb;
 205
 206         /* Make sure callback is registered before we write to MSRs */
 207         wmb();
 208
 209         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 210         wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
 211 }
 212 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
 213
 214 void clear_hv_tscchange_cb(void)
 215 {
 216         struct hv_reenlightenment_control re_ctrl;
 217
 218         if (!hv_reenlightenment_available())
 219                 return;
 220
 221         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 222         re_ctrl.enabled = 0;
 223         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 224
 225         hv_reenlightenment_cb = NULL;
 226 }
 227 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
 228
 229 static int hv_cpu_die(unsigned int cpu)
 230 {
 231         struct hv_reenlightenment_control re_ctrl;
 232         unsigned int new_cpu;
 233         unsigned long flags;
 234         void **input_arg;
 235         void *input_pg = NULL;
 236
 237         local_irq_save(flags);
 238         input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
 239         input_pg = *input_arg;
 240         *input_arg = NULL;
 241         local_irq_restore(flags);
 242         free_page((unsigned long)input_pg);
 243
 244         if (hv_vp_assist_page && hv_vp_assist_page[cpu])
 245                 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
 246
 247         if (hv_reenlightenment_cb == NULL)
 248                 return 0;
 249
 250         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 251         if (re_ctrl.target_vp == hv_vp_index[cpu]) {
 252                 /* Reassign to some other online CPU */
 253                 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
 254
 255                 re_ctrl.target_vp = hv_vp_index[new_cpu];
 256                 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 257         }
 258
 259         return 0;
 260 }
 261
 262 static int __init hv_pci_init(void)
 263 {
 264         int gen2vm = efi_enabled(EFI_BOOT);
 265
 266         /*
 267          * For Generation-2 VM, we exit from pci_arch_init() by returning 0.
 268          * The purpose is to suppress the harmless warning:
 269          * "PCI: Fatal: No config space access function found"
 270          */
 271         if (gen2vm)
 272                 return 0;
 273
 274         /* For Generation-1 VM, we'll proceed in pci_arch_init().  */
 275         return 1;
 276 }
 277
 278 /*
 279  * This function is to be invoked early in the boot sequence after the
 280  * hypervisor has been detected.
 281  *
 282  * 1. Setup the hypercall page.
 283  * 2. Register Hyper-V specific clocksource.
 284  * 3. Setup Hyper-V specific APIC entry points.
 285  */
 286 void __init hyperv_init(void)
 287 {
 288         u64 guest_id, required_msrs;
 289         union hv_x64_msr_hypercall_contents hypercall_msr;
 290         int cpuhp, i;
 291
 292         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 293                 return;
 294
 295         /* Absolutely required MSRs */
 296         required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
 297                 HV_X64_MSR_VP_INDEX_AVAILABLE;
 298
 299         if ((ms_hyperv.features & required_msrs) != required_msrs)
 300                 return;
 301
 302         /*
 303          * Allocate the per-CPU state for the hypercall input arg.
 304          * If this allocation fails, we will not be able to setup
 305          * (per-CPU) hypercall input page and thus this failure is
 306          * fatal on Hyper-V.
 307          */
 308         hyperv_pcpu_input_arg = alloc_percpu(void  *);
 309
 310         BUG_ON(hyperv_pcpu_input_arg == NULL);
 311
 312         /* Allocate percpu VP index */
 313         hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
 314                                     GFP_KERNEL);
 315         if (!hv_vp_index)
 316                 return;
 317
 318         for (i = 0; i < num_possible_cpus(); i++)
 319                 hv_vp_index[i] = VP_INVAL;
 320
 321         hv_vp_assist_page = kcalloc(num_possible_cpus(),
 322                                     sizeof(*hv_vp_assist_page), GFP_KERNEL);
 323         if (!hv_vp_assist_page) {
 324                 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
 325                 goto free_vp_index;
 326         }
 327
 328         cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
 329                                   hv_cpu_init, hv_cpu_die);
 330         if (cpuhp < 0)
 331                 goto free_vp_assist_page;
 332
 333         /*
 334          * Setup the hypercall page and enable hypercalls.
 335          * 1. Register the guest ID
 336          * 2. Enable the hypercall and register the hypercall page
 337          */
 338         guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
 339         wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 340
 341         hv_hypercall_pg  = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
 342         if (hv_hypercall_pg == NULL) {
 343                 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 344                 goto remove_cpuhp_state;
 345         }
 346
 347         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 348         hypercall_msr.enable = 1;
 349         hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
 350         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 351
 352         hv_apic_init();
 353
 354         x86_init.pci.arch_init = hv_pci_init;
 355
 356         /*
 357          * Register Hyper-V specific clocksource.
 358          */
 359 #ifdef CONFIG_HYPERV_TSCPAGE
 360         if (ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE) {
 361                 union hv_x64_msr_hypercall_contents tsc_msr;
 362
 363                 tsc_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
 364                 if (!tsc_pg)
 365                         goto register_msr_cs;
 366
 367                 hyperv_cs = &hyperv_cs_tsc;
 368
 369                 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
 370
 371                 tsc_msr.enable = 1;
 372                 tsc_msr.guest_physical_address = vmalloc_to_pfn(tsc_pg);
 373
 374                 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
 375
 376                 hyperv_cs_tsc.archdata.vclock_mode = VCLOCK_HVCLOCK;
 377
 378                 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
 379                 return;
 380         }
 381 register_msr_cs:
 382 #endif
 383         /*
 384          * For 32 bit guests just use the MSR based mechanism for reading
 385          * the partition counter.
 386          */
 387
 388         hyperv_cs = &hyperv_cs_msr;
 389         if (ms_hyperv.features & HV_MSR_TIME_REF_COUNT_AVAILABLE)
 390                 clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
 391
 392         return;
 393
 394 remove_cpuhp_state:
 395         cpuhp_remove_state(cpuhp);
 396 free_vp_assist_page:
 397         kfree(hv_vp_assist_page);
 398         hv_vp_assist_page = NULL;
 399 free_vp_index:
 400         kfree(hv_vp_index);
 401         hv_vp_index = NULL;
 402 }
 403
 404 /*
 405  * This routine is called before kexec/kdump, it does the required cleanup.
 406  */
 407 void hyperv_cleanup(void)
 408 {
 409         union hv_x64_msr_hypercall_contents hypercall_msr;
 410
 411         /* Reset our OS id */
 412         wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 413
 414         /*
 415          * Reset hypercall page reference before reset the page,
 416          * let hypercall operations fail safely rather than
 417          * panic the kernel for using invalid hypercall page
 418          */
 419         hv_hypercall_pg = NULL;
 420
 421         /* Reset the hypercall page */
 422         hypercall_msr.as_uint64 = 0;
 423         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 424
 425         /* Reset the TSC page */
 426         hypercall_msr.as_uint64 = 0;
 427         wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
 428 }
 429 EXPORT_SYMBOL_GPL(hyperv_cleanup);
 430
 431 void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die)
 432 {
 433         static bool panic_reported;
 434         u64 guest_id;
 435
 436         if (in_die && !panic_on_oops)
 437                 return;
 438
 439         /*
 440          * We prefer to report panic on 'die' chain as we have proper
 441          * registers to report, but if we miss it (e.g. on BUG()) we need
 442          * to report it on 'panic'.
 443          */
 444         if (panic_reported)
 445                 return;
 446         panic_reported = true;
 447
 448         rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 449
 450         wrmsrl(HV_X64_MSR_CRASH_P0, err);
 451         wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
 452         wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
 453         wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
 454         wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
 455
 456         /*
 457          * Let Hyper-V know there is crash data available
 458          */
 459         wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
 460 }
 461 EXPORT_SYMBOL_GPL(hyperv_report_panic);
 462
 463 /**
 464  * hyperv_report_panic_msg - report panic message to Hyper-V
 465  * @pa: physical address of the panic page containing the message
 466  * @size: size of the message in the page
 467  */
 468 void hyperv_report_panic_msg(phys_addr_t pa, size_t size)
 469 {
 470         /*
 471          * P3 to contain the physical address of the panic page & P4 to
 472          * contain the size of the panic data in that page. Rest of the
 473          * registers are no-op when the NOTIFY_MSG flag is set.
 474          */
 475         wrmsrl(HV_X64_MSR_CRASH_P0, 0);
 476         wrmsrl(HV_X64_MSR_CRASH_P1, 0);
 477         wrmsrl(HV_X64_MSR_CRASH_P2, 0);
 478         wrmsrl(HV_X64_MSR_CRASH_P3, pa);
 479         wrmsrl(HV_X64_MSR_CRASH_P4, size);
 480
 481         /*
 482          * Let Hyper-V know there is crash data available along with
 483          * the panic message.
 484          */
 485         wrmsrl(HV_X64_MSR_CRASH_CTL,
 486                (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG));
 487 }
 488 EXPORT_SYMBOL_GPL(hyperv_report_panic_msg);
 489
 490 bool hv_is_hyperv_initialized(void)
 491 {
 492         union hv_x64_msr_hypercall_contents hypercall_msr;
 493
 494         /*
 495          * Ensure that we're really on Hyper-V, and not a KVM or Xen
 496          * emulation of Hyper-V
 497          */
 498         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 499                 return false;
 500
 501         /*
 502          * Verify that earlier initialization succeeded by checking
 503          * that the hypercall page is setup
 504          */
 505         hypercall_msr.as_uint64 = 0;
 506         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 507
 508         return hypercall_msr.enable;
 509 }
 510 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);