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/types.h>
  21 #include <asm/apic.h>
  22 #include <asm/desc.h>
  23 #include <asm/hypervisor.h>
  24 #include <asm/hyperv-tlfs.h>
  25 #include <asm/mshyperv.h>
  26 #include <linux/version.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/mm.h>
  29 #include <linux/clockchips.h>
  30 #include <linux/hyperv.h>
  31 #include <linux/slab.h>
  32 #include <linux/cpuhotplug.h>
  33
  34 #ifdef CONFIG_HYPERV_TSCPAGE
  35
  36 static struct ms_hyperv_tsc_page *tsc_pg;
  37
  38 struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
  39 {
  40         return tsc_pg;
  41 }
  42 EXPORT_SYMBOL_GPL(hv_get_tsc_page);
  43
  44 static u64 read_hv_clock_tsc(struct clocksource *arg)
  45 {
  46         u64 current_tick = hv_read_tsc_page(tsc_pg);
  47
  48         if (current_tick == U64_MAX)
  49                 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
  50
  51         return current_tick;
  52 }
  53
  54 static struct clocksource hyperv_cs_tsc = {
  55                 .name           = "hyperv_clocksource_tsc_page",
  56                 .rating         = 400,
  57                 .read           = read_hv_clock_tsc,
  58                 .mask           = CLOCKSOURCE_MASK(64),
  59                 .flags          = CLOCK_SOURCE_IS_CONTINUOUS,
  60 };
  61 #endif
  62
  63 static u64 read_hv_clock_msr(struct clocksource *arg)
  64 {
  65         u64 current_tick;
  66         /*
  67          * Read the partition counter to get the current tick count. This count
  68          * is set to 0 when the partition is created and is incremented in
  69          * 100 nanosecond units.
  70          */
  71         rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
  72         return current_tick;
  73 }
  74
  75 static struct clocksource hyperv_cs_msr = {
  76         .name           = "hyperv_clocksource_msr",
  77         .rating         = 400,
  78         .read           = read_hv_clock_msr,
  79         .mask           = CLOCKSOURCE_MASK(64),
  80         .flags          = CLOCK_SOURCE_IS_CONTINUOUS,
  81 };
  82
  83 void *hv_hypercall_pg;
  84 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
  85 struct clocksource *hyperv_cs;
  86 EXPORT_SYMBOL_GPL(hyperv_cs);
  87
  88 u32 *hv_vp_index;
  89 EXPORT_SYMBOL_GPL(hv_vp_index);
  90
  91 struct hv_vp_assist_page **hv_vp_assist_page;
  92 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
  93
  94 u32 hv_max_vp_index;
  95
  96 static int hv_cpu_init(unsigned int cpu)
  97 {
  98         u64 msr_vp_index;
  99         struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()];
 100
 101         hv_get_vp_index(msr_vp_index);
 102
 103         hv_vp_index[smp_processor_id()] = msr_vp_index;
 104
 105         if (msr_vp_index > hv_max_vp_index)
 106                 hv_max_vp_index = msr_vp_index;
 107
 108         if (!hv_vp_assist_page)
 109                 return 0;
 110
 111         if (!*hvp)
 112                 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
 113
 114         if (*hvp) {
 115                 u64 val;
 116
 117                 val = vmalloc_to_pfn(*hvp);
 118                 val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) |
 119                         HV_X64_MSR_VP_ASSIST_PAGE_ENABLE;
 120
 121                 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val);
 122         }
 123
 124         return 0;
 125 }
 126
 127 static void (*hv_reenlightenment_cb)(void);
 128
 129 static void hv_reenlightenment_notify(struct work_struct *dummy)
 130 {
 131         struct hv_tsc_emulation_status emu_status;
 132
 133         rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 134
 135         /* Don't issue the callback if TSC accesses are not emulated */
 136         if (hv_reenlightenment_cb && emu_status.inprogress)
 137                 hv_reenlightenment_cb();
 138 }
 139 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
 140
 141 void hyperv_stop_tsc_emulation(void)
 142 {
 143         u64 freq;
 144         struct hv_tsc_emulation_status emu_status;
 145
 146         rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 147         emu_status.inprogress = 0;
 148         wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
 149
 150         rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
 151         tsc_khz = div64_u64(freq, 1000);
 152 }
 153 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
 154
 155 static inline bool hv_reenlightenment_available(void)
 156 {
 157         /*
 158          * Check for required features and priviliges to make TSC frequency
 159          * change notifications work.
 160          */
 161         return ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
 162                 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
 163                 ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT;
 164 }
 165
 166 __visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs)
 167 {
 168         entering_ack_irq();
 169
 170         inc_irq_stat(irq_hv_reenlightenment_count);
 171
 172         schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
 173
 174         exiting_irq();
 175 }
 176
 177 void set_hv_tscchange_cb(void (*cb)(void))
 178 {
 179         struct hv_reenlightenment_control re_ctrl = {
 180                 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
 181                 .enabled = 1,
 182                 .target_vp = hv_vp_index[smp_processor_id()]
 183         };
 184         struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
 185
 186         if (!hv_reenlightenment_available()) {
 187                 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
 188                 return;
 189         }
 190
 191         hv_reenlightenment_cb = cb;
 192
 193         /* Make sure callback is registered before we write to MSRs */
 194         wmb();
 195
 196         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 197         wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
 198 }
 199 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
 200
 201 void clear_hv_tscchange_cb(void)
 202 {
 203         struct hv_reenlightenment_control re_ctrl;
 204
 205         if (!hv_reenlightenment_available())
 206                 return;
 207
 208         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 209         re_ctrl.enabled = 0;
 210         wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
 211
 212         hv_reenlightenment_cb = NULL;
 213 }
 214 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
 215
 216 static int hv_cpu_die(unsigned int cpu)
 217 {
 218         struct hv_reenlightenment_control re_ctrl;
 219         unsigned int new_cpu;
 220
 221         if (hv_vp_assist_page && hv_vp_assist_page[cpu])
 222                 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
 223
 224         if (hv_reenlightenment_cb == NULL)
 225                 return 0;
 226
 227         rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 228         if (re_ctrl.target_vp == hv_vp_index[cpu]) {
 229                 /* Reassign to some other online CPU */
 230                 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
 231
 232                 re_ctrl.target_vp = hv_vp_index[new_cpu];
 233                 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
 234         }
 235
 236         return 0;
 237 }
 238
 239 /*
 240  * This function is to be invoked early in the boot sequence after the
 241  * hypervisor has been detected.
 242  *
 243  * 1. Setup the hypercall page.
 244  * 2. Register Hyper-V specific clocksource.
 245  * 3. Setup Hyper-V specific APIC entry points.
 246  */
 247 void __init hyperv_init(void)
 248 {
 249         u64 guest_id, required_msrs;
 250         union hv_x64_msr_hypercall_contents hypercall_msr;
 251         int cpuhp;
 252
 253         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 254                 return;
 255
 256         /* Absolutely required MSRs */
 257         required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
 258                 HV_X64_MSR_VP_INDEX_AVAILABLE;
 259
 260         if ((ms_hyperv.features & required_msrs) != required_msrs)
 261                 return;
 262
 263         /* Allocate percpu VP index */
 264         hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
 265                                     GFP_KERNEL);
 266         if (!hv_vp_index)
 267                 return;
 268
 269         hv_vp_assist_page = kcalloc(num_possible_cpus(),
 270                                     sizeof(*hv_vp_assist_page), GFP_KERNEL);
 271         if (!hv_vp_assist_page) {
 272                 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
 273                 goto free_vp_index;
 274         }
 275
 276         cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
 277                                   hv_cpu_init, hv_cpu_die);
 278         if (cpuhp < 0)
 279                 goto free_vp_assist_page;
 280
 281         /*
 282          * Setup the hypercall page and enable hypercalls.
 283          * 1. Register the guest ID
 284          * 2. Enable the hypercall and register the hypercall page
 285          */
 286         guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
 287         wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 288
 289         hv_hypercall_pg  = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
 290         if (hv_hypercall_pg == NULL) {
 291                 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 292                 goto remove_cpuhp_state;
 293         }
 294
 295         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 296         hypercall_msr.enable = 1;
 297         hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
 298         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 299
 300         hyper_alloc_mmu();
 301
 302         hv_apic_init();
 303
 304         /*
 305          * Register Hyper-V specific clocksource.
 306          */
 307 #ifdef CONFIG_HYPERV_TSCPAGE
 308         if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
 309                 union hv_x64_msr_hypercall_contents tsc_msr;
 310
 311                 tsc_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
 312                 if (!tsc_pg)
 313                         goto register_msr_cs;
 314
 315                 hyperv_cs = &hyperv_cs_tsc;
 316
 317                 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
 318
 319                 tsc_msr.enable = 1;
 320                 tsc_msr.guest_physical_address = vmalloc_to_pfn(tsc_pg);
 321
 322                 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
 323
 324                 hyperv_cs_tsc.archdata.vclock_mode = VCLOCK_HVCLOCK;
 325
 326                 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
 327                 return;
 328         }
 329 register_msr_cs:
 330 #endif
 331         /*
 332          * For 32 bit guests just use the MSR based mechanism for reading
 333          * the partition counter.
 334          */
 335
 336         hyperv_cs = &hyperv_cs_msr;
 337         if (ms_hyperv.features & HV_X64_MSR_TIME_REF_COUNT_AVAILABLE)
 338                 clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
 339
 340         return;
 341
 342 remove_cpuhp_state:
 343         cpuhp_remove_state(cpuhp);
 344 free_vp_assist_page:
 345         kfree(hv_vp_assist_page);
 346         hv_vp_assist_page = NULL;
 347 free_vp_index:
 348         kfree(hv_vp_index);
 349         hv_vp_index = NULL;
 350 }
 351
 352 /*
 353  * This routine is called before kexec/kdump, it does the required cleanup.
 354  */
 355 void hyperv_cleanup(void)
 356 {
 357         union hv_x64_msr_hypercall_contents hypercall_msr;
 358
 359         /* Reset our OS id */
 360         wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 361
 362         /* Reset the hypercall page */
 363         hypercall_msr.as_uint64 = 0;
 364         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 365
 366         /* Reset the TSC page */
 367         hypercall_msr.as_uint64 = 0;
 368         wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
 369 }
 370 EXPORT_SYMBOL_GPL(hyperv_cleanup);
 371
 372 void hyperv_report_panic(struct pt_regs *regs, long err)
 373 {
 374         static bool panic_reported;
 375         u64 guest_id;
 376
 377         /*
 378          * We prefer to report panic on 'die' chain as we have proper
 379          * registers to report, but if we miss it (e.g. on BUG()) we need
 380          * to report it on 'panic'.
 381          */
 382         if (panic_reported)
 383                 return;
 384         panic_reported = true;
 385
 386         rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
 387
 388         wrmsrl(HV_X64_MSR_CRASH_P0, err);
 389         wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
 390         wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
 391         wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
 392         wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
 393
 394         /*
 395          * Let Hyper-V know there is crash data available
 396          */
 397         wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
 398 }
 399 EXPORT_SYMBOL_GPL(hyperv_report_panic);
 400
 401 bool hv_is_hyperv_initialized(void)
 402 {
 403         union hv_x64_msr_hypercall_contents hypercall_msr;
 404
 405         /*
 406          * Ensure that we're really on Hyper-V, and not a KVM or Xen
 407          * emulation of Hyper-V
 408          */
 409         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
 410                 return false;
 411
 412         /*
 413          * Verify that earlier initialization succeeded by checking
 414          * that the hypercall page is setup
 415          */
 416         hypercall_msr.as_uint64 = 0;
 417         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 418
 419         return hypercall_msr.enable;
 420 }
 421 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);