arch/x86/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  10  *
  11  * Authors:
  12  *   Avi Kivity   <avi@qumranet.com>
  13  *   Yaniv Kamay  <yaniv@qumranet.com>
  14  *   Amit Shah    <amit.shah@qumranet.com>
  15  *   Ben-Ami Yassour <benami@il.ibm.com>
  16  *
  17  * This work is licensed under the terms of the GNU GPL, version 2.  See
  18  * the COPYING file in the top-level directory.
  19  *
  20  */
  21
  22 #include <linux/kvm_host.h>
  23 #include "irq.h"
  24 #include "mmu.h"
  25 #include "i8254.h"
  26 #include "tss.h"
  27 #include "kvm_cache_regs.h"
  28 #include "x86.h"
  29 #include "cpuid.h"
  30 #include "assigned-dev.h"
  31 #include "pmu.h"
  32 #include "hyperv.h"
  33
  34 #include <linux/clocksource.h>
  35 #include <linux/interrupt.h>
  36 #include <linux/kvm.h>
  37 #include <linux/fs.h>
  38 #include <linux/vmalloc.h>
  39 #include <linux/module.h>
  40 #include <linux/mman.h>
  41 #include <linux/highmem.h>
  42 #include <linux/iommu.h>
  43 #include <linux/intel-iommu.h>
  44 #include <linux/cpufreq.h>
  45 #include <linux/user-return-notifier.h>
  46 #include <linux/srcu.h>
  47 #include <linux/slab.h>
  48 #include <linux/perf_event.h>
  49 #include <linux/uaccess.h>
  50 #include <linux/hash.h>
  51 #include <linux/pci.h>
  52 #include <linux/timekeeper_internal.h>
  53 #include <linux/pvclock_gtod.h>
  54 #include <linux/kvm_irqfd.h>
  55 #include <linux/irqbypass.h>
  56 #include <trace/events/kvm.h>
  57
  58 #include <asm/debugreg.h>
  59 #include <asm/msr.h>
  60 #include <asm/desc.h>
  61 #include <asm/mce.h>
  62 #include <linux/kernel_stat.h>
  63 #include <asm/fpu/internal.h> /* Ugh! */
  64 #include <asm/pvclock.h>
  65 #include <asm/div64.h>
  66 #include <asm/irq_remapping.h>
  67
  68 #define CREATE_TRACE_POINTS
  69 #include "trace.h"
  70
  71 #define MAX_IO_MSRS 256
  72 #define KVM_MAX_MCE_BANKS 32
  73 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
  74
  75 #define emul_to_vcpu(ctxt) \
  76         container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
  77
  78 /* EFER defaults:
  79  * - enable syscall per default because its emulated by KVM
  80  * - enable LME and LMA per default on 64 bit KVM
  81  */
  82 #ifdef CONFIG_X86_64
  83 static
  84 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
  85 #else
  86 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
  87 #endif
  88
  89 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  90 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  91
  92 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  93 static void process_nmi(struct kvm_vcpu *vcpu);
  94 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
  95
  96 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
  97 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  98
  99 static bool __read_mostly ignore_msrs = 0;
 100 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
 101
 102 unsigned int min_timer_period_us = 500;
 103 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
 104
 105 static bool __read_mostly kvmclock_periodic_sync = true;
 106 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
 107
 108 bool __read_mostly kvm_has_tsc_control;
 109 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
 110 u32  __read_mostly kvm_max_guest_tsc_khz;
 111 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
 112 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
 113 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
 114 u64  __read_mostly kvm_max_tsc_scaling_ratio;
 115 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
 116 static u64 __read_mostly kvm_default_tsc_scaling_ratio;
 117
 118 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
 119 static u32 __read_mostly tsc_tolerance_ppm = 250;
 120 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
 121
 122 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
 123 unsigned int __read_mostly lapic_timer_advance_ns = 0;
 124 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
 125
 126 static bool __read_mostly vector_hashing = true;
 127 module_param(vector_hashing, bool, S_IRUGO);
 128
 129 static bool __read_mostly backwards_tsc_observed = false;
 130
 131 #define KVM_NR_SHARED_MSRS 16
 132
 133 struct kvm_shared_msrs_global {
 134         int nr;
 135         u32 msrs[KVM_NR_SHARED_MSRS];
 136 };
 137
 138 struct kvm_shared_msrs {
 139         struct user_return_notifier urn;
 140         bool registered;
 141         struct kvm_shared_msr_values {
 142                 u64 host;
 143                 u64 curr;
 144         } values[KVM_NR_SHARED_MSRS];
 145 };
 146
 147 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 148 static struct kvm_shared_msrs __percpu *shared_msrs;
 149
 150 struct kvm_stats_debugfs_item debugfs_entries[] = {
 151         { "pf_fixed", VCPU_STAT(pf_fixed) },
 152         { "pf_guest", VCPU_STAT(pf_guest) },
 153         { "tlb_flush", VCPU_STAT(tlb_flush) },
 154         { "invlpg", VCPU_STAT(invlpg) },
 155         { "exits", VCPU_STAT(exits) },
 156         { "io_exits", VCPU_STAT(io_exits) },
 157         { "mmio_exits", VCPU_STAT(mmio_exits) },
 158         { "signal_exits", VCPU_STAT(signal_exits) },
 159         { "irq_window", VCPU_STAT(irq_window_exits) },
 160         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 161         { "halt_exits", VCPU_STAT(halt_exits) },
 162         { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
 163         { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
 164         { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
 165         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 166         { "hypercalls", VCPU_STAT(hypercalls) },
 167         { "request_irq", VCPU_STAT(request_irq_exits) },
 168         { "irq_exits", VCPU_STAT(irq_exits) },
 169         { "host_state_reload", VCPU_STAT(host_state_reload) },
 170         { "efer_reload", VCPU_STAT(efer_reload) },
 171         { "fpu_reload", VCPU_STAT(fpu_reload) },
 172         { "insn_emulation", VCPU_STAT(insn_emulation) },
 173         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 174         { "irq_injections", VCPU_STAT(irq_injections) },
 175         { "nmi_injections", VCPU_STAT(nmi_injections) },
 176         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 177         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 178         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 179         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 180         { "mmu_flooded", VM_STAT(mmu_flooded) },
 181         { "mmu_recycled", VM_STAT(mmu_recycled) },
 182         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 183         { "mmu_unsync", VM_STAT(mmu_unsync) },
 184         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 185         { "largepages", VM_STAT(lpages) },
 186         { NULL }
 187 };
 188
 189 u64 __read_mostly host_xcr0;
 190
 191 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
 192
 193 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
 194 {
 195         int i;
 196         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
 197                 vcpu->arch.apf.gfns[i] = ~0;
 198 }
 199
 200 static void kvm_on_user_return(struct user_return_notifier *urn)
 201 {
 202         unsigned slot;
 203         struct kvm_shared_msrs *locals
 204                 = container_of(urn, struct kvm_shared_msrs, urn);
 205         struct kvm_shared_msr_values *values;
 206
 207         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 208                 values = &locals->values[slot];
 209                 if (values->host != values->curr) {
 210                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 211                         values->curr = values->host;
 212                 }
 213         }
 214         locals->registered = false;
 215         user_return_notifier_unregister(urn);
 216 }
 217
 218 static void shared_msr_update(unsigned slot, u32 msr)
 219 {
 220         u64 value;
 221         unsigned int cpu = smp_processor_id();
 222         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 223
 224         /* only read, and nobody should modify it at this time,
 225          * so don't need lock */
 226         if (slot >= shared_msrs_global.nr) {
 227                 printk(KERN_ERR "kvm: invalid MSR slot!");
 228                 return;
 229         }
 230         rdmsrl_safe(msr, &value);
 231         smsr->values[slot].host = value;
 232         smsr->values[slot].curr = value;
 233 }
 234
 235 void kvm_define_shared_msr(unsigned slot, u32 msr)
 236 {
 237         BUG_ON(slot >= KVM_NR_SHARED_MSRS);
 238         shared_msrs_global.msrs[slot] = msr;
 239         if (slot >= shared_msrs_global.nr)
 240                 shared_msrs_global.nr = slot + 1;
 241 }
 242 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 243
 244 static void kvm_shared_msr_cpu_online(void)
 245 {
 246         unsigned i;
 247
 248         for (i = 0; i < shared_msrs_global.nr; ++i)
 249                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 250 }
 251
 252 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 253 {
 254         unsigned int cpu = smp_processor_id();
 255         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 256         int err;
 257
 258         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 259                 return 0;
 260         smsr->values[slot].curr = value;
 261         err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
 262         if (err)
 263                 return 1;
 264
 265         if (!smsr->registered) {
 266                 smsr->urn.on_user_return = kvm_on_user_return;
 267                 user_return_notifier_register(&smsr->urn);
 268                 smsr->registered = true;
 269         }
 270         return 0;
 271 }
 272 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 273
 274 static void drop_user_return_notifiers(void)
 275 {
 276         unsigned int cpu = smp_processor_id();
 277         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 278
 279         if (smsr->registered)
 280                 kvm_on_user_return(&smsr->urn);
 281 }
 282
 283 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 284 {
 285         return vcpu->arch.apic_base;
 286 }
 287 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 288
 289 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 290 {
 291         u64 old_state = vcpu->arch.apic_base &
 292                 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
 293         u64 new_state = msr_info->data &
 294                 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
 295         u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
 296                 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
 297
 298         if (!msr_info->host_initiated &&
 299             ((msr_info->data & reserved_bits) != 0 ||
 300              new_state == X2APIC_ENABLE ||
 301              (new_state == MSR_IA32_APICBASE_ENABLE &&
 302               old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
 303              (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
 304               old_state == 0)))
 305                 return 1;
 306
 307         kvm_lapic_set_base(vcpu, msr_info->data);
 308         return 0;
 309 }
 310 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 311
 312 asmlinkage __visible void kvm_spurious_fault(void)
 313 {
 314         /* Fault while not rebooting.  We want the trace. */
 315         BUG();
 316 }
 317 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
 318
 319 #define EXCPT_BENIGN            0
 320 #define EXCPT_CONTRIBUTORY      1
 321 #define EXCPT_PF                2
 322
 323 static int exception_class(int vector)
 324 {
 325         switch (vector) {
 326         case PF_VECTOR:
 327                 return EXCPT_PF;
 328         case DE_VECTOR:
 329         case TS_VECTOR:
 330         case NP_VECTOR:
 331         case SS_VECTOR:
 332         case GP_VECTOR:
 333                 return EXCPT_CONTRIBUTORY;
 334         default:
 335                 break;
 336         }
 337         return EXCPT_BENIGN;
 338 }
 339
 340 #define EXCPT_FAULT             0
 341 #define EXCPT_TRAP              1
 342 #define EXCPT_ABORT             2
 343 #define EXCPT_INTERRUPT         3
 344
 345 static int exception_type(int vector)
 346 {
 347         unsigned int mask;
 348
 349         if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
 350                 return EXCPT_INTERRUPT;
 351
 352         mask = 1 << vector;
 353
 354         /* #DB is trap, as instruction watchpoints are handled elsewhere */
 355         if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
 356                 return EXCPT_TRAP;
 357
 358         if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
 359                 return EXCPT_ABORT;
 360
 361         /* Reserved exceptions will result in fault */
 362         return EXCPT_FAULT;
 363 }
 364
 365 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 366                 unsigned nr, bool has_error, u32 error_code,
 367                 bool reinject)
 368 {
 369         u32 prev_nr;
 370         int class1, class2;
 371
 372         kvm_make_request(KVM_REQ_EVENT, vcpu);
 373
 374         if (!vcpu->arch.exception.pending) {
 375         queue:
 376                 if (has_error && !is_protmode(vcpu))
 377                         has_error = false;
 378                 vcpu->arch.exception.pending = true;
 379                 vcpu->arch.exception.has_error_code = has_error;
 380                 vcpu->arch.exception.nr = nr;
 381                 vcpu->arch.exception.error_code = error_code;
 382                 vcpu->arch.exception.reinject = reinject;
 383                 return;
 384         }
 385
 386         /* to check exception */
 387         prev_nr = vcpu->arch.exception.nr;
 388         if (prev_nr == DF_VECTOR) {
 389                 /* triple fault -> shutdown */
 390                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
 391                 return;
 392         }
 393         class1 = exception_class(prev_nr);
 394         class2 = exception_class(nr);
 395         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 396                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 397                 /* generate double fault per SDM Table 5-5 */
 398                 vcpu->arch.exception.pending = true;
 399                 vcpu->arch.exception.has_error_code = true;
 400                 vcpu->arch.exception.nr = DF_VECTOR;
 401                 vcpu->arch.exception.error_code = 0;
 402         } else
 403                 /* replace previous exception with a new one in a hope
 404                    that instruction re-execution will regenerate lost
 405                    exception */
 406                 goto queue;
 407 }
 408
 409 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 410 {
 411         kvm_multiple_exception(vcpu, nr, false, 0, false);
 412 }
 413 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 414
 415 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 416 {
 417         kvm_multiple_exception(vcpu, nr, false, 0, true);
 418 }
 419 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 420
 421 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
 422 {
 423         if (err)
 424                 kvm_inject_gp(vcpu, 0);
 425         else
 426                 kvm_x86_ops->skip_emulated_instruction(vcpu);
 427 }
 428 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
 429
 430 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 431 {
 432         ++vcpu->stat.pf_guest;
 433         vcpu->arch.cr2 = fault->address;
 434         kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
 435 }
 436 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
 437
 438 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 439 {
 440         if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
 441                 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
 442         else
 443                 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
 444
 445         return fault->nested_page_fault;
 446 }
 447
 448 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 449 {
 450         atomic_inc(&vcpu->arch.nmi_queued);
 451         kvm_make_request(KVM_REQ_NMI, vcpu);
 452 }
 453 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 454
 455 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 456 {
 457         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 458 }
 459 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 460
 461 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 462 {
 463         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 464 }
 465 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 466
 467 /*
 468  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 469  * a #GP and return false.
 470  */
 471 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 472 {
 473         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 474                 return true;
 475         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 476         return false;
 477 }
 478 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 479
 480 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
 481 {
 482         if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 483                 return true;
 484
 485         kvm_queue_exception(vcpu, UD_VECTOR);
 486         return false;
 487 }
 488 EXPORT_SYMBOL_GPL(kvm_require_dr);
 489
 490 /*
 491  * This function will be used to read from the physical memory of the currently
 492  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
 493  * can read from guest physical or from the guest's guest physical memory.
 494  */
 495 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 496                             gfn_t ngfn, void *data, int offset, int len,
 497                             u32 access)
 498 {
 499         struct x86_exception exception;
 500         gfn_t real_gfn;
 501         gpa_t ngpa;
 502
 503         ngpa     = gfn_to_gpa(ngfn);
 504         real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
 505         if (real_gfn == UNMAPPED_GVA)
 506                 return -EFAULT;
 507
 508         real_gfn = gpa_to_gfn(real_gfn);
 509
 510         return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
 511 }
 512 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
 513
 514 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
 515                                void *data, int offset, int len, u32 access)
 516 {
 517         return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
 518                                        data, offset, len, access);
 519 }
 520
 521 /*
 522  * Load the pae pdptrs.  Return true is they are all valid.
 523  */
 524 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
 525 {
 526         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 527         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 528         int i;
 529         int ret;
 530         u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
 531
 532         ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
 533                                       offset * sizeof(u64), sizeof(pdpte),
 534                                       PFERR_USER_MASK|PFERR_WRITE_MASK);
 535         if (ret < 0) {
 536                 ret = 0;
 537                 goto out;
 538         }
 539         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 540                 if (is_present_gpte(pdpte[i]) &&
 541                     (pdpte[i] &
 542                      vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
 543                         ret = 0;
 544                         goto out;
 545                 }
 546         }
 547         ret = 1;
 548
 549         memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
 550         __set_bit(VCPU_EXREG_PDPTR,
 551                   (unsigned long *)&vcpu->arch.regs_avail);
 552         __set_bit(VCPU_EXREG_PDPTR,
 553                   (unsigned long *)&vcpu->arch.regs_dirty);
 554 out:
 555
 556         return ret;
 557 }
 558 EXPORT_SYMBOL_GPL(load_pdptrs);
 559
 560 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 561 {
 562         u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
 563         bool changed = true;
 564         int offset;
 565         gfn_t gfn;
 566         int r;
 567
 568         if (is_long_mode(vcpu) || !is_pae(vcpu))
 569                 return false;
 570
 571         if (!test_bit(VCPU_EXREG_PDPTR,
 572                       (unsigned long *)&vcpu->arch.regs_avail))
 573                 return true;
 574
 575         gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
 576         offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
 577         r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
 578                                        PFERR_USER_MASK | PFERR_WRITE_MASK);
 579         if (r < 0)
 580                 goto out;
 581         changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
 582 out:
 583
 584         return changed;
 585 }
 586
 587 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 588 {
 589         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 590         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
 591
 592         cr0 |= X86_CR0_ET;
 593
 594 #ifdef CONFIG_X86_64
 595         if (cr0 & 0xffffffff00000000UL)
 596                 return 1;
 597 #endif
 598
 599         cr0 &= ~CR0_RESERVED_BITS;
 600
 601         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 602                 return 1;
 603
 604         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 605                 return 1;
 606
 607         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 608 #ifdef CONFIG_X86_64
 609                 if ((vcpu->arch.efer & EFER_LME)) {
 610                         int cs_db, cs_l;
 611
 612                         if (!is_pae(vcpu))
 613                                 return 1;
 614                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 615                         if (cs_l)
 616                                 return 1;
 617                 } else
 618 #endif
 619                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 620                                                  kvm_read_cr3(vcpu)))
 621                         return 1;
 622         }
 623
 624         if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
 625                 return 1;
 626
 627         kvm_x86_ops->set_cr0(vcpu, cr0);
 628
 629         if ((cr0 ^ old_cr0) & X86_CR0_PG) {
 630                 kvm_clear_async_pf_completion_queue(vcpu);
 631                 kvm_async_pf_hash_reset(vcpu);
 632         }
 633
 634         if ((cr0 ^ old_cr0) & update_bits)
 635                 kvm_mmu_reset_context(vcpu);
 636
 637         if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
 638             kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
 639             !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
 640                 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
 641
 642         return 0;
 643 }
 644 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 645
 646 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 647 {
 648         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 649 }
 650 EXPORT_SYMBOL_GPL(kvm_lmsw);
 651
 652 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
 653 {
 654         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
 655                         !vcpu->guest_xcr0_loaded) {
 656                 /* kvm_set_xcr() also depends on this */
 657                 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
 658                 vcpu->guest_xcr0_loaded = 1;
 659         }
 660 }
 661
 662 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
 663 {
 664         if (vcpu->guest_xcr0_loaded) {
 665                 if (vcpu->arch.xcr0 != host_xcr0)
 666                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
 667                 vcpu->guest_xcr0_loaded = 0;
 668         }
 669 }
 670
 671 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 672 {
 673         u64 xcr0 = xcr;
 674         u64 old_xcr0 = vcpu->arch.xcr0;
 675         u64 valid_bits;
 676
 677         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
 678         if (index != XCR_XFEATURE_ENABLED_MASK)
 679                 return 1;
 680         if (!(xcr0 & XFEATURE_MASK_FP))
 681                 return 1;
 682         if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
 683                 return 1;
 684
 685         /*
 686          * Do not allow the guest to set bits that we do not support
 687          * saving.  However, xcr0 bit 0 is always set, even if the
 688          * emulated CPU does not support XSAVE (see fx_init).
 689          */
 690         valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
 691         if (xcr0 & ~valid_bits)
 692                 return 1;
 693
 694         if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
 695             (!(xcr0 & XFEATURE_MASK_BNDCSR)))
 696                 return 1;
 697
 698         if (xcr0 & XFEATURE_MASK_AVX512) {
 699                 if (!(xcr0 & XFEATURE_MASK_YMM))
 700                         return 1;
 701                 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
 702                         return 1;
 703         }
 704         vcpu->arch.xcr0 = xcr0;
 705
 706         if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
 707                 kvm_update_cpuid(vcpu);
 708         return 0;
 709 }
 710
 711 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 712 {
 713         if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
 714             __kvm_set_xcr(vcpu, index, xcr)) {
 715                 kvm_inject_gp(vcpu, 0);
 716                 return 1;
 717         }
 718         return 0;
 719 }
 720 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 721
 722 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 723 {
 724         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 725         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
 726                                    X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
 727
 728         if (cr4 & CR4_RESERVED_BITS)
 729                 return 1;
 730
 731         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
 732                 return 1;
 733
 734         if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
 735                 return 1;
 736
 737         if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
 738                 return 1;
 739
 740         if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
 741                 return 1;
 742
 743         if (!guest_cpuid_has_pku(vcpu) && (cr4 & X86_CR4_PKE))
 744                 return 1;
 745
 746         if (is_long_mode(vcpu)) {
 747                 if (!(cr4 & X86_CR4_PAE))
 748                         return 1;
 749         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 750                    && ((cr4 ^ old_cr4) & pdptr_bits)
 751                    && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 752                                    kvm_read_cr3(vcpu)))
 753                 return 1;
 754
 755         if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
 756                 if (!guest_cpuid_has_pcid(vcpu))
 757                         return 1;
 758
 759                 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
 760                 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
 761                         return 1;
 762         }
 763
 764         if (kvm_x86_ops->set_cr4(vcpu, cr4))
 765                 return 1;
 766
 767         if (((cr4 ^ old_cr4) & pdptr_bits) ||
 768             (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
 769                 kvm_mmu_reset_context(vcpu);
 770
 771         if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
 772                 kvm_update_cpuid(vcpu);
 773
 774         return 0;
 775 }
 776 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 777
 778 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 779 {
 780 #ifdef CONFIG_X86_64
 781         cr3 &= ~CR3_PCID_INVD;
 782 #endif
 783
 784         if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
 785                 kvm_mmu_sync_roots(vcpu);
 786                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 787                 return 0;
 788         }
 789
 790         if (is_long_mode(vcpu)) {
 791                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 792                         return 1;
 793         } else if (is_pae(vcpu) && is_paging(vcpu) &&
 794                    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 795                 return 1;
 796
 797         vcpu->arch.cr3 = cr3;
 798         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
 799         kvm_mmu_new_cr3(vcpu);
 800         return 0;
 801 }
 802 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 803
 804 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 805 {
 806         if (cr8 & CR8_RESERVED_BITS)
 807                 return 1;
 808         if (lapic_in_kernel(vcpu))
 809                 kvm_lapic_set_tpr(vcpu, cr8);
 810         else
 811                 vcpu->arch.cr8 = cr8;
 812         return 0;
 813 }
 814 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 815
 816 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 817 {
 818         if (lapic_in_kernel(vcpu))
 819                 return kvm_lapic_get_cr8(vcpu);
 820         else
 821                 return vcpu->arch.cr8;
 822 }
 823 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 824
 825 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
 826 {
 827         int i;
 828
 829         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 830                 for (i = 0; i < KVM_NR_DB_REGS; i++)
 831                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
 832                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
 833         }
 834 }
 835
 836 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
 837 {
 838         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 839                 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
 840 }
 841
 842 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
 843 {
 844         unsigned long dr7;
 845
 846         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 847                 dr7 = vcpu->arch.guest_debug_dr7;
 848         else
 849                 dr7 = vcpu->arch.dr7;
 850         kvm_x86_ops->set_dr7(vcpu, dr7);
 851         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
 852         if (dr7 & DR7_BP_EN_MASK)
 853                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
 854 }
 855
 856 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
 857 {
 858         u64 fixed = DR6_FIXED_1;
 859
 860         if (!guest_cpuid_has_rtm(vcpu))
 861                 fixed |= DR6_RTM;
 862         return fixed;
 863 }
 864
 865 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 866 {
 867         switch (dr) {
 868         case 0 ... 3:
 869                 vcpu->arch.db[dr] = val;
 870                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 871                         vcpu->arch.eff_db[dr] = val;
 872                 break;
 873         case 4:
 874                 /* fall through */
 875         case 6:
 876                 if (val & 0xffffffff00000000ULL)
 877                         return -1; /* #GP */
 878                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
 879                 kvm_update_dr6(vcpu);
 880                 break;
 881         case 5:
 882                 /* fall through */
 883         default: /* 7 */
 884                 if (val & 0xffffffff00000000ULL)
 885                         return -1; /* #GP */
 886                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 887                 kvm_update_dr7(vcpu);
 888                 break;
 889         }
 890
 891         return 0;
 892 }
 893
 894 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 895 {
 896         if (__kvm_set_dr(vcpu, dr, val)) {
 897                 kvm_inject_gp(vcpu, 0);
 898                 return 1;
 899         }
 900         return 0;
 901 }
 902 EXPORT_SYMBOL_GPL(kvm_set_dr);
 903
 904 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 905 {
 906         switch (dr) {
 907         case 0 ... 3:
 908                 *val = vcpu->arch.db[dr];
 909                 break;
 910         case 4:
 911                 /* fall through */
 912         case 6:
 913                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 914                         *val = vcpu->arch.dr6;
 915                 else
 916                         *val = kvm_x86_ops->get_dr6(vcpu);
 917                 break;
 918         case 5:
 919                 /* fall through */
 920         default: /* 7 */
 921                 *val = vcpu->arch.dr7;
 922                 break;
 923         }
 924         return 0;
 925 }
 926 EXPORT_SYMBOL_GPL(kvm_get_dr);
 927
 928 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
 929 {
 930         u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 931         u64 data;
 932         int err;
 933
 934         err = kvm_pmu_rdpmc(vcpu, ecx, &data);
 935         if (err)
 936                 return err;
 937         kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
 938         kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
 939         return err;
 940 }
 941 EXPORT_SYMBOL_GPL(kvm_rdpmc);
 942
 943 /*
 944  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 945  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 946  *
 947  * This list is modified at module load time to reflect the
 948  * capabilities of the host cpu. This capabilities test skips MSRs that are
 949  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
 950  * may depend on host virtualization features rather than host cpu features.
 951  */
 952
 953 static u32 msrs_to_save[] = {
 954         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 955         MSR_STAR,
 956 #ifdef CONFIG_X86_64
 957         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 958 #endif
 959         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
 960         MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
 961 };
 962
 963 static unsigned num_msrs_to_save;
 964
 965 static u32 emulated_msrs[] = {
 966         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 967         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 968         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 969         HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
 970         HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
 971         HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
 972         HV_X64_MSR_RESET,
 973         HV_X64_MSR_VP_INDEX,
 974         HV_X64_MSR_VP_RUNTIME,
 975         HV_X64_MSR_SCONTROL,
 976         HV_X64_MSR_STIMER0_CONFIG,
 977         HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
 978         MSR_KVM_PV_EOI_EN,
 979
 980         MSR_IA32_TSC_ADJUST,
 981         MSR_IA32_TSCDEADLINE,
 982         MSR_IA32_MISC_ENABLE,
 983         MSR_IA32_MCG_STATUS,
 984         MSR_IA32_MCG_CTL,
 985         MSR_IA32_SMBASE,
 986 };
 987
 988 static unsigned num_emulated_msrs;
 989
 990 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
 991 {
 992         if (efer & efer_reserved_bits)
 993                 return false;
 994
 995         if (efer & EFER_FFXSR) {
 996                 struct kvm_cpuid_entry2 *feat;
 997
 998                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 999                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
1000                         return false;
1001         }
1002
1003         if (efer & EFER_SVME) {
1004                 struct kvm_cpuid_entry2 *feat;
1005
1006                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1007                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1008                         return false;
1009         }
1010
1011         return true;
1012 }
1013 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1014
1015 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1016 {
1017         u64 old_efer = vcpu->arch.efer;
1018
1019         if (!kvm_valid_efer(vcpu, efer))
1020                 return 1;
1021
1022         if (is_paging(vcpu)
1023             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1024                 return 1;
1025
1026         efer &= ~EFER_LMA;
1027         efer |= vcpu->arch.efer & EFER_LMA;
1028
1029         kvm_x86_ops->set_efer(vcpu, efer);
1030
1031         /* Update reserved bits */
1032         if ((efer ^ old_efer) & EFER_NX)
1033                 kvm_mmu_reset_context(vcpu);
1034
1035         return 0;
1036 }
1037
1038 void kvm_enable_efer_bits(u64 mask)
1039 {
1040        efer_reserved_bits &= ~mask;
1041 }
1042 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1043
1044 /*
1045  * Writes msr value into into the appropriate "register".
1046  * Returns 0 on success, non-0 otherwise.
1047  * Assumes vcpu_load() was already called.
1048  */
1049 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1050 {
1051         switch (msr->index) {
1052         case MSR_FS_BASE:
1053         case MSR_GS_BASE:
1054         case MSR_KERNEL_GS_BASE:
1055         case MSR_CSTAR:
1056         case MSR_LSTAR:
1057                 if (is_noncanonical_address(msr->data))
1058                         return 1;
1059                 break;
1060         case MSR_IA32_SYSENTER_EIP:
1061         case MSR_IA32_SYSENTER_ESP:
1062                 /*
1063                  * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1064                  * non-canonical address is written on Intel but not on
1065                  * AMD (which ignores the top 32-bits, because it does
1066                  * not implement 64-bit SYSENTER).
1067                  *
1068                  * 64-bit code should hence be able to write a non-canonical
1069                  * value on AMD.  Making the address canonical ensures that
1070                  * vmentry does not fail on Intel after writing a non-canonical
1071                  * value, and that something deterministic happens if the guest
1072                  * invokes 64-bit SYSENTER.
1073                  */
1074                 msr->data = get_canonical(msr->data);
1075         }
1076         return kvm_x86_ops->set_msr(vcpu, msr);
1077 }
1078 EXPORT_SYMBOL_GPL(kvm_set_msr);
1079
1080 /*
1081  * Adapt set_msr() to msr_io()'s calling convention
1082  */
1083 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1084 {
1085         struct msr_data msr;
1086         int r;
1087
1088         msr.index = index;
1089         msr.host_initiated = true;
1090         r = kvm_get_msr(vcpu, &msr);
1091         if (r)
1092                 return r;
1093
1094         *data = msr.data;
1095         return 0;
1096 }
1097
1098 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1099 {
1100         struct msr_data msr;
1101
1102         msr.data = *data;
1103         msr.index = index;
1104         msr.host_initiated = true;
1105         return kvm_set_msr(vcpu, &msr);
1106 }
1107
1108 #ifdef CONFIG_X86_64
1109 struct pvclock_gtod_data {
1110         seqcount_t      seq;
1111
1112         struct { /* extract of a clocksource struct */
1113                 int vclock_mode;
1114                 cycle_t cycle_last;
1115                 cycle_t mask;
1116                 u32     mult;
1117                 u32     shift;
1118         } clock;
1119
1120         u64             boot_ns;
1121         u64             nsec_base;
1122 };
1123
1124 static struct pvclock_gtod_data pvclock_gtod_data;
1125
1126 static void update_pvclock_gtod(struct timekeeper *tk)
1127 {
1128         struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1129         u64 boot_ns;
1130
1131         boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1132
1133         write_seqcount_begin(&vdata->seq);
1134
1135         /* copy pvclock gtod data */
1136         vdata->clock.vclock_mode        = tk->tkr_mono.clock->archdata.vclock_mode;
1137         vdata->clock.cycle_last         = tk->tkr_mono.cycle_last;
1138         vdata->clock.mask               = tk->tkr_mono.mask;
1139         vdata->clock.mult               = tk->tkr_mono.mult;
1140         vdata->clock.shift              = tk->tkr_mono.shift;
1141
1142         vdata->boot_ns                  = boot_ns;
1143         vdata->nsec_base                = tk->tkr_mono.xtime_nsec;
1144
1145         write_seqcount_end(&vdata->seq);
1146 }
1147 #endif
1148
1149 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1150 {
1151         /*
1152          * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1153          * vcpu_enter_guest.  This function is only called from
1154          * the physical CPU that is running vcpu.
1155          */
1156         kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1157 }
1158
1159 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1160 {
1161         int version;
1162         int r;
1163         struct pvclock_wall_clock wc;
1164         struct timespec boot;
1165
1166         if (!wall_clock)
1167                 return;
1168
1169         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1170         if (r)
1171                 return;
1172
1173         if (version & 1)
1174                 ++version;  /* first time write, random junk */
1175
1176         ++version;
1177
1178         if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1179                 return;
1180
1181         /*
1182          * The guest calculates current wall clock time by adding
1183          * system time (updated by kvm_guest_time_update below) to the
1184          * wall clock specified here.  guest system time equals host
1185          * system time for us, thus we must fill in host boot time here.
1186          */
1187         getboottime(&boot);
1188
1189         if (kvm->arch.kvmclock_offset) {
1190                 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1191                 boot = timespec_sub(boot, ts);
1192         }
1193         wc.sec = boot.tv_sec;
1194         wc.nsec = boot.tv_nsec;
1195         wc.version = version;
1196
1197         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1198
1199         version++;
1200         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1201 }
1202
1203 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1204 {
1205         do_shl32_div32(dividend, divisor);
1206         return dividend;
1207 }
1208
1209 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1210                                s8 *pshift, u32 *pmultiplier)
1211 {
1212         uint64_t scaled64;
1213         int32_t  shift = 0;
1214         uint64_t tps64;
1215         uint32_t tps32;
1216
1217         tps64 = base_hz;
1218         scaled64 = scaled_hz;
1219         while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1220                 tps64 >>= 1;
1221                 shift--;
1222         }
1223
1224         tps32 = (uint32_t)tps64;
1225         while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1226                 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1227                         scaled64 >>= 1;
1228                 else
1229                         tps32 <<= 1;
1230                 shift++;
1231         }
1232
1233         *pshift = shift;
1234         *pmultiplier = div_frac(scaled64, tps32);
1235
1236         pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1237                  __func__, base_hz, scaled_hz, shift, *pmultiplier);
1238 }
1239
1240 #ifdef CONFIG_X86_64
1241 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1242 #endif
1243
1244 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1245 static unsigned long max_tsc_khz;
1246
1247 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1248 {
1249         u64 v = (u64)khz * (1000000 + ppm);
1250         do_div(v, 1000000);
1251         return v;
1252 }
1253
1254 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1255 {
1256         u64 ratio;
1257
1258         /* Guest TSC same frequency as host TSC? */
1259         if (!scale) {
1260                 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1261                 return 0;
1262         }
1263
1264         /* TSC scaling supported? */
1265         if (!kvm_has_tsc_control) {
1266                 if (user_tsc_khz > tsc_khz) {
1267                         vcpu->arch.tsc_catchup = 1;
1268                         vcpu->arch.tsc_always_catchup = 1;
1269                         return 0;
1270                 } else {
1271                         WARN(1, "user requested TSC rate below hardware speed\n");
1272                         return -1;
1273                 }
1274         }
1275
1276         /* TSC scaling required  - calculate ratio */
1277         ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1278                                 user_tsc_khz, tsc_khz);
1279
1280         if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1281                 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1282                           user_tsc_khz);
1283                 return -1;
1284         }
1285
1286         vcpu->arch.tsc_scaling_ratio = ratio;
1287         return 0;
1288 }
1289
1290 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1291 {
1292         u32 thresh_lo, thresh_hi;
1293         int use_scaling = 0;
1294
1295         /* tsc_khz can be zero if TSC calibration fails */
1296         if (user_tsc_khz == 0) {
1297                 /* set tsc_scaling_ratio to a safe value */
1298                 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1299                 return -1;
1300         }
1301
1302         /* Compute a scale to convert nanoseconds in TSC cycles */
1303         kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1304                            &vcpu->arch.virtual_tsc_shift,
1305                            &vcpu->arch.virtual_tsc_mult);
1306         vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1307
1308         /*
1309          * Compute the variation in TSC rate which is acceptable
1310          * within the range of tolerance and decide if the
1311          * rate being applied is within that bounds of the hardware
1312          * rate.  If so, no scaling or compensation need be done.
1313          */
1314         thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1315         thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1316         if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1317                 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1318                 use_scaling = 1;
1319         }
1320         return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1321 }
1322
1323 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1324 {
1325         u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1326                                       vcpu->arch.virtual_tsc_mult,
1327                                       vcpu->arch.virtual_tsc_shift);
1328         tsc += vcpu->arch.this_tsc_write;
1329         return tsc;
1330 }
1331
1332 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1333 {
1334 #ifdef CONFIG_X86_64
1335         bool vcpus_matched;
1336         struct kvm_arch *ka = &vcpu->kvm->arch;
1337         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1338
1339         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1340                          atomic_read(&vcpu->kvm->online_vcpus));
1341
1342         /*
1343          * Once the masterclock is enabled, always perform request in
1344          * order to update it.
1345          *
1346          * In order to enable masterclock, the host clocksource must be TSC
1347          * and the vcpus need to have matched TSCs.  When that happens,
1348          * perform request to enable masterclock.
1349          */
1350         if (ka->use_master_clock ||
1351             (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1352                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1353
1354         trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1355                             atomic_read(&vcpu->kvm->online_vcpus),
1356                             ka->use_master_clock, gtod->clock.vclock_mode);
1357 #endif
1358 }
1359
1360 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1361 {
1362         u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1363         vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1364 }
1365
1366 /*
1367  * Multiply tsc by a fixed point number represented by ratio.
1368  *
1369  * The most significant 64-N bits (mult) of ratio represent the
1370  * integral part of the fixed point number; the remaining N bits
1371  * (frac) represent the fractional part, ie. ratio represents a fixed
1372  * point number (mult + frac * 2^(-N)).
1373  *
1374  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1375  */
1376 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1377 {
1378         return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1379 }
1380
1381 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1382 {
1383         u64 _tsc = tsc;
1384         u64 ratio = vcpu->arch.tsc_scaling_ratio;
1385
1386         if (ratio != kvm_default_tsc_scaling_ratio)
1387                 _tsc = __scale_tsc(ratio, tsc);
1388
1389         return _tsc;
1390 }
1391 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1392
1393 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1394 {
1395         u64 tsc;
1396
1397         tsc = kvm_scale_tsc(vcpu, rdtsc());
1398
1399         return target_tsc - tsc;
1400 }
1401
1402 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1403 {
1404         return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc));
1405 }
1406 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1407
1408 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1409 {
1410         struct kvm *kvm = vcpu->kvm;
1411         u64 offset, ns, elapsed;
1412         unsigned long flags;
1413         s64 usdiff;
1414         bool matched;
1415         bool already_matched;
1416         u64 data = msr->data;
1417
1418         raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1419         offset = kvm_compute_tsc_offset(vcpu, data);
1420         ns = get_kernel_ns();
1421         elapsed = ns - kvm->arch.last_tsc_nsec;
1422
1423         if (vcpu->arch.virtual_tsc_khz) {
1424                 int faulted = 0;
1425
1426                 /* n.b - signed multiplication and division required */
1427                 usdiff = data - kvm->arch.last_tsc_write;
1428 #ifdef CONFIG_X86_64
1429                 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1430 #else
1431                 /* do_div() only does unsigned */
1432                 asm("1: idivl %[divisor]\n"
1433                     "2: xor %%edx, %%edx\n"
1434                     "   movl $0, %[faulted]\n"
1435                     "3:\n"
1436                     ".section .fixup,\"ax\"\n"
1437                     "4: movl $1, %[faulted]\n"
1438                     "   jmp  3b\n"
1439                     ".previous\n"
1440
1441                 _ASM_EXTABLE(1b, 4b)
1442
1443                 : "=A"(usdiff), [faulted] "=r" (faulted)
1444                 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1445
1446 #endif
1447                 do_div(elapsed, 1000);
1448                 usdiff -= elapsed;
1449                 if (usdiff < 0)
1450                         usdiff = -usdiff;
1451
1452                 /* idivl overflow => difference is larger than USEC_PER_SEC */
1453                 if (faulted)
1454                         usdiff = USEC_PER_SEC;
1455         } else
1456                 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1457
1458         /*
1459          * Special case: TSC write with a small delta (1 second) of virtual
1460          * cycle time against real time is interpreted as an attempt to
1461          * synchronize the CPU.
1462          *
1463          * For a reliable TSC, we can match TSC offsets, and for an unstable
1464          * TSC, we add elapsed time in this computation.  We could let the
1465          * compensation code attempt to catch up if we fall behind, but
1466          * it's better to try to match offsets from the beginning.
1467          */
1468         if (usdiff < USEC_PER_SEC &&
1469             vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1470                 if (!check_tsc_unstable()) {
1471                         offset = kvm->arch.cur_tsc_offset;
1472                         pr_debug("kvm: matched tsc offset for %llu\n", data);
1473                 } else {
1474                         u64 delta = nsec_to_cycles(vcpu, elapsed);
1475                         data += delta;
1476                         offset = kvm_compute_tsc_offset(vcpu, data);
1477                         pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1478                 }
1479                 matched = true;
1480                 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1481         } else {
1482                 /*
1483                  * We split periods of matched TSC writes into generations.
1484                  * For each generation, we track the original measured
1485                  * nanosecond time, offset, and write, so if TSCs are in
1486                  * sync, we can match exact offset, and if not, we can match
1487                  * exact software computation in compute_guest_tsc()
1488                  *
1489                  * These values are tracked in kvm->arch.cur_xxx variables.
1490                  */
1491                 kvm->arch.cur_tsc_generation++;
1492                 kvm->arch.cur_tsc_nsec = ns;
1493                 kvm->arch.cur_tsc_write = data;
1494                 kvm->arch.cur_tsc_offset = offset;
1495                 matched = false;
1496                 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1497                          kvm->arch.cur_tsc_generation, data);
1498         }
1499
1500         /*
1501          * We also track th most recent recorded KHZ, write and time to
1502          * allow the matching interval to be extended at each write.
1503          */
1504         kvm->arch.last_tsc_nsec = ns;
1505         kvm->arch.last_tsc_write = data;
1506         kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1507
1508         vcpu->arch.last_guest_tsc = data;
1509
1510         /* Keep track of which generation this VCPU has synchronized to */
1511         vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1512         vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1513         vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1514
1515         if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1516                 update_ia32_tsc_adjust_msr(vcpu, offset);
1517         kvm_x86_ops->write_tsc_offset(vcpu, offset);
1518         raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1519
1520         spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1521         if (!matched) {
1522                 kvm->arch.nr_vcpus_matched_tsc = 0;
1523         } else if (!already_matched) {
1524                 kvm->arch.nr_vcpus_matched_tsc++;
1525         }
1526
1527         kvm_track_tsc_matching(vcpu);
1528         spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1529 }
1530
1531 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1532
1533 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1534                                            s64 adjustment)
1535 {
1536         kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1537 }
1538
1539 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1540 {
1541         if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1542                 WARN_ON(adjustment < 0);
1543         adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1544         kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1545 }
1546
1547 #ifdef CONFIG_X86_64
1548
1549 static cycle_t read_tsc(void)
1550 {
1551         cycle_t ret = (cycle_t)rdtsc_ordered();
1552         u64 last = pvclock_gtod_data.clock.cycle_last;
1553
1554         if (likely(ret >= last))
1555                 return ret;
1556
1557         /*
1558          * GCC likes to generate cmov here, but this branch is extremely
1559          * predictable (it's just a function of time and the likely is
1560          * very likely) and there's a data dependence, so force GCC
1561          * to generate a branch instead.  I don't barrier() because
1562          * we don't actually need a barrier, and if this function
1563          * ever gets inlined it will generate worse code.
1564          */
1565         asm volatile ("");
1566         return last;
1567 }
1568
1569 static inline u64 vgettsc(cycle_t *cycle_now)
1570 {
1571         long v;
1572         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1573
1574         *cycle_now = read_tsc();
1575
1576         v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1577         return v * gtod->clock.mult;
1578 }
1579
1580 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1581 {
1582         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1583         unsigned long seq;
1584         int mode;
1585         u64 ns;
1586
1587         do {
1588                 seq = read_seqcount_begin(&gtod->seq);
1589                 mode = gtod->clock.vclock_mode;
1590                 ns = gtod->nsec_base;
1591                 ns += vgettsc(cycle_now);
1592                 ns >>= gtod->clock.shift;
1593                 ns += gtod->boot_ns;
1594         } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1595         *t = ns;
1596
1597         return mode;
1598 }
1599
1600 /* returns true if host is using tsc clocksource */
1601 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1602 {
1603         /* checked again under seqlock below */
1604         if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1605                 return false;
1606
1607         return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1608 }
1609 #endif
1610
1611 /*
1612  *
1613  * Assuming a stable TSC across physical CPUS, and a stable TSC
1614  * across virtual CPUs, the following condition is possible.
1615  * Each numbered line represents an event visible to both
1616  * CPUs at the next numbered event.
1617  *
1618  * "timespecX" represents host monotonic time. "tscX" represents
1619  * RDTSC value.
1620  *
1621  *              VCPU0 on CPU0           |       VCPU1 on CPU1
1622  *
1623  * 1.  read timespec0,tsc0
1624  * 2.                                   | timespec1 = timespec0 + N
1625  *                                      | tsc1 = tsc0 + M
1626  * 3. transition to guest               | transition to guest
1627  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1628  * 5.                                   | ret1 = timespec1 + (rdtsc - tsc1)
1629  *                                      | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1630  *
1631  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1632  *
1633  *      - ret0 < ret1
1634  *      - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1635  *              ...
1636  *      - 0 < N - M => M < N
1637  *
1638  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1639  * always the case (the difference between two distinct xtime instances
1640  * might be smaller then the difference between corresponding TSC reads,
1641  * when updating guest vcpus pvclock areas).
1642  *
1643  * To avoid that problem, do not allow visibility of distinct
1644  * system_timestamp/tsc_timestamp values simultaneously: use a master
1645  * copy of host monotonic time values. Update that master copy
1646  * in lockstep.
1647  *
1648  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1649  *
1650  */
1651
1652 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1653 {
1654 #ifdef CONFIG_X86_64
1655         struct kvm_arch *ka = &kvm->arch;
1656         int vclock_mode;
1657         bool host_tsc_clocksource, vcpus_matched;
1658
1659         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1660                         atomic_read(&kvm->online_vcpus));
1661
1662         /*
1663          * If the host uses TSC clock, then passthrough TSC as stable
1664          * to the guest.
1665          */
1666         host_tsc_clocksource = kvm_get_time_and_clockread(
1667                                         &ka->master_kernel_ns,
1668                                         &ka->master_cycle_now);
1669
1670         ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1671                                 && !backwards_tsc_observed
1672                                 && !ka->boot_vcpu_runs_old_kvmclock;
1673
1674         if (ka->use_master_clock)
1675                 atomic_set(&kvm_guest_has_master_clock, 1);
1676
1677         vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1678         trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1679                                         vcpus_matched);
1680 #endif
1681 }
1682
1683 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1684 {
1685         kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1686 }
1687
1688 static void kvm_gen_update_masterclock(struct kvm *kvm)
1689 {
1690 #ifdef CONFIG_X86_64
1691         int i;
1692         struct kvm_vcpu *vcpu;
1693         struct kvm_arch *ka = &kvm->arch;
1694
1695         spin_lock(&ka->pvclock_gtod_sync_lock);
1696         kvm_make_mclock_inprogress_request(kvm);
1697         /* no guest entries from this point */
1698         pvclock_update_vm_gtod_copy(kvm);
1699
1700         kvm_for_each_vcpu(i, vcpu, kvm)
1701                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1702
1703         /* guest entries allowed */
1704         kvm_for_each_vcpu(i, vcpu, kvm)
1705                 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1706
1707         spin_unlock(&ka->pvclock_gtod_sync_lock);
1708 #endif
1709 }
1710
1711 static int kvm_guest_time_update(struct kvm_vcpu *v)
1712 {
1713         unsigned long flags, tgt_tsc_khz;
1714         struct kvm_vcpu_arch *vcpu = &v->arch;
1715         struct kvm_arch *ka = &v->kvm->arch;
1716         s64 kernel_ns;
1717         u64 tsc_timestamp, host_tsc;
1718         struct pvclock_vcpu_time_info guest_hv_clock;
1719         u8 pvclock_flags;
1720         bool use_master_clock;
1721
1722         kernel_ns = 0;
1723         host_tsc = 0;
1724
1725         /*
1726          * If the host uses TSC clock, then passthrough TSC as stable
1727          * to the guest.
1728          */
1729         spin_lock(&ka->pvclock_gtod_sync_lock);
1730         use_master_clock = ka->use_master_clock;
1731         if (use_master_clock) {
1732                 host_tsc = ka->master_cycle_now;
1733                 kernel_ns = ka->master_kernel_ns;
1734         }
1735         spin_unlock(&ka->pvclock_gtod_sync_lock);
1736
1737         /* Keep irq disabled to prevent changes to the clock */
1738         local_irq_save(flags);
1739         tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1740         if (unlikely(tgt_tsc_khz == 0)) {
1741                 local_irq_restore(flags);
1742                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1743                 return 1;
1744         }
1745         if (!use_master_clock) {
1746                 host_tsc = rdtsc();
1747                 kernel_ns = get_kernel_ns();
1748         }
1749
1750         tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1751
1752         /*
1753          * We may have to catch up the TSC to match elapsed wall clock
1754          * time for two reasons, even if kvmclock is used.
1755          *   1) CPU could have been running below the maximum TSC rate
1756          *   2) Broken TSC compensation resets the base at each VCPU
1757          *      entry to avoid unknown leaps of TSC even when running
1758          *      again on the same CPU.  This may cause apparent elapsed
1759          *      time to disappear, and the guest to stand still or run
1760          *      very slowly.
1761          */
1762         if (vcpu->tsc_catchup) {
1763                 u64 tsc = compute_guest_tsc(v, kernel_ns);
1764                 if (tsc > tsc_timestamp) {
1765                         adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1766                         tsc_timestamp = tsc;
1767                 }
1768         }
1769
1770         local_irq_restore(flags);
1771
1772         if (!vcpu->pv_time_enabled)
1773                 return 0;
1774
1775         if (kvm_has_tsc_control)
1776                 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1777
1778         if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1779                 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1780                                    &vcpu->hv_clock.tsc_shift,
1781                                    &vcpu->hv_clock.tsc_to_system_mul);
1782                 vcpu->hw_tsc_khz = tgt_tsc_khz;
1783         }
1784
1785         /* With all the info we got, fill in the values */
1786         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1787         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1788         vcpu->last_guest_tsc = tsc_timestamp;
1789
1790         if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1791                 &guest_hv_clock, sizeof(guest_hv_clock))))
1792                 return 0;
1793
1794         /* This VCPU is paused, but it's legal for a guest to read another
1795          * VCPU's kvmclock, so we really have to follow the specification where
1796          * it says that version is odd if data is being modified, and even after
1797          * it is consistent.
1798          *
1799          * Version field updates must be kept separate.  This is because
1800          * kvm_write_guest_cached might use a "rep movs" instruction, and
1801          * writes within a string instruction are weakly ordered.  So there
1802          * are three writes overall.
1803          *
1804          * As a small optimization, only write the version field in the first
1805          * and third write.  The vcpu->pv_time cache is still valid, because the
1806          * version field is the first in the struct.
1807          */
1808         BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1809
1810         vcpu->hv_clock.version = guest_hv_clock.version + 1;
1811         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1812                                 &vcpu->hv_clock,
1813                                 sizeof(vcpu->hv_clock.version));
1814
1815         smp_wmb();
1816
1817         /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1818         pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1819
1820         if (vcpu->pvclock_set_guest_stopped_request) {
1821                 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1822                 vcpu->pvclock_set_guest_stopped_request = false;
1823         }
1824
1825         /* If the host uses TSC clocksource, then it is stable */
1826         if (use_master_clock)
1827                 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1828
1829         vcpu->hv_clock.flags = pvclock_flags;
1830
1831         trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1832
1833         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1834                                 &vcpu->hv_clock,
1835                                 sizeof(vcpu->hv_clock));
1836
1837         smp_wmb();
1838
1839         vcpu->hv_clock.version++;
1840         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1841                                 &vcpu->hv_clock,
1842                                 sizeof(vcpu->hv_clock.version));
1843         return 0;
1844 }
1845
1846 /*
1847  * kvmclock updates which are isolated to a given vcpu, such as
1848  * vcpu->cpu migration, should not allow system_timestamp from
1849  * the rest of the vcpus to remain static. Otherwise ntp frequency
1850  * correction applies to one vcpu's system_timestamp but not
1851  * the others.
1852  *
1853  * So in those cases, request a kvmclock update for all vcpus.
1854  * We need to rate-limit these requests though, as they can
1855  * considerably slow guests that have a large number of vcpus.
1856  * The time for a remote vcpu to update its kvmclock is bound
1857  * by the delay we use to rate-limit the updates.
1858  */
1859
1860 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1861
1862 static void kvmclock_update_fn(struct work_struct *work)
1863 {
1864         int i;
1865         struct delayed_work *dwork = to_delayed_work(work);
1866         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1867                                            kvmclock_update_work);
1868         struct kvm *kvm = container_of(ka, struct kvm, arch);
1869         struct kvm_vcpu *vcpu;
1870
1871         kvm_for_each_vcpu(i, vcpu, kvm) {
1872                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1873                 kvm_vcpu_kick(vcpu);
1874         }
1875 }
1876
1877 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1878 {
1879         struct kvm *kvm = v->kvm;
1880
1881         kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1882         schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1883                                         KVMCLOCK_UPDATE_DELAY);
1884 }
1885
1886 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1887
1888 static void kvmclock_sync_fn(struct work_struct *work)
1889 {
1890         struct delayed_work *dwork = to_delayed_work(work);
1891         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1892                                            kvmclock_sync_work);
1893         struct kvm *kvm = container_of(ka, struct kvm, arch);
1894
1895         if (!kvmclock_periodic_sync)
1896                 return;
1897
1898         schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1899         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1900                                         KVMCLOCK_SYNC_PERIOD);
1901 }
1902
1903 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1904 {
1905         u64 mcg_cap = vcpu->arch.mcg_cap;
1906         unsigned bank_num = mcg_cap & 0xff;
1907
1908         switch (msr) {
1909         case MSR_IA32_MCG_STATUS:
1910                 vcpu->arch.mcg_status = data;
1911                 break;
1912         case MSR_IA32_MCG_CTL:
1913                 if (!(mcg_cap & MCG_CTL_P))
1914                         return 1;
1915                 if (data != 0 && data != ~(u64)0)
1916                         return -1;
1917                 vcpu->arch.mcg_ctl = data;
1918                 break;
1919         default:
1920                 if (msr >= MSR_IA32_MC0_CTL &&
1921                     msr < MSR_IA32_MCx_CTL(bank_num)) {
1922                         u32 offset = msr - MSR_IA32_MC0_CTL;
1923                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1924                          * some Linux kernels though clear bit 10 in bank 4 to
1925                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1926                          * this to avoid an uncatched #GP in the guest
1927                          */
1928                         if ((offset & 0x3) == 0 &&
1929                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1930                                 return -1;
1931                         vcpu->arch.mce_banks[offset] = data;
1932                         break;
1933                 }
1934                 return 1;
1935         }
1936         return 0;
1937 }
1938
1939 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1940 {
1941         struct kvm *kvm = vcpu->kvm;
1942         int lm = is_long_mode(vcpu);
1943         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1944                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1945         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1946                 : kvm->arch.xen_hvm_config.blob_size_32;
1947         u32 page_num = data & ~PAGE_MASK;
1948         u64 page_addr = data & PAGE_MASK;
1949         u8 *page;
1950         int r;
1951
1952         r = -E2BIG;
1953         if (page_num >= blob_size)
1954                 goto out;
1955         r = -ENOMEM;
1956         page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1957         if (IS_ERR(page)) {
1958                 r = PTR_ERR(page);
1959                 goto out;
1960         }
1961         if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1962                 goto out_free;
1963         r = 0;
1964 out_free:
1965         kfree(page);
1966 out:
1967         return r;
1968 }
1969
1970 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1971 {
1972         gpa_t gpa = data & ~0x3f;
1973
1974         /* Bits 2:5 are reserved, Should be zero */
1975         if (data & 0x3c)
1976                 return 1;
1977
1978         vcpu->arch.apf.msr_val = data;
1979
1980         if (!(data & KVM_ASYNC_PF_ENABLED)) {
1981                 kvm_clear_async_pf_completion_queue(vcpu);
1982                 kvm_async_pf_hash_reset(vcpu);
1983                 return 0;
1984         }
1985
1986         if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1987                                         sizeof(u32)))
1988                 return 1;
1989
1990         vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1991         kvm_async_pf_wakeup_all(vcpu);
1992         return 0;
1993 }
1994
1995 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1996 {
1997         vcpu->arch.pv_time_enabled = false;
1998 }
1999
2000 static void record_steal_time(struct kvm_vcpu *vcpu)
2001 {
2002         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2003                 return;
2004
2005         if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2006                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2007                 return;
2008
2009         if (vcpu->arch.st.steal.version & 1)
2010                 vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2011
2012         vcpu->arch.st.steal.version += 1;
2013
2014         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2015                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2016
2017         smp_wmb();
2018
2019         vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2020                 vcpu->arch.st.last_steal;
2021         vcpu->arch.st.last_steal = current->sched_info.run_delay;
2022
2023         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2024                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2025
2026         smp_wmb();
2027
2028         vcpu->arch.st.steal.version += 1;
2029
2030         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2031                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2032 }
2033
2034 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2035 {
2036         bool pr = false;
2037         u32 msr = msr_info->index;
2038         u64 data = msr_info->data;
2039
2040         switch (msr) {
2041         case MSR_AMD64_NB_CFG:
2042         case MSR_IA32_UCODE_REV:
2043         case MSR_IA32_UCODE_WRITE:
2044         case MSR_VM_HSAVE_PA:
2045         case MSR_AMD64_PATCH_LOADER:
2046         case MSR_AMD64_BU_CFG2:
2047                 break;
2048
2049         case MSR_EFER:
2050                 return set_efer(vcpu, data);
2051         case MSR_K7_HWCR:
2052                 data &= ~(u64)0x40;     /* ignore flush filter disable */
2053                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
2054                 data &= ~(u64)0x8;      /* ignore TLB cache disable */
2055                 data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2056                 if (data != 0) {
2057                         vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2058                                     data);
2059                         return 1;
2060                 }
2061                 break;
2062         case MSR_FAM10H_MMIO_CONF_BASE:
2063                 if (data != 0) {
2064                         vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2065                                     "0x%llx\n", data);
2066                         return 1;
2067                 }
2068                 break;
2069         case MSR_IA32_DEBUGCTLMSR:
2070                 if (!data) {
2071                         /* We support the non-activated case already */
2072                         break;
2073                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2074                         /* Values other than LBR and BTF are vendor-specific,
2075                            thus reserved and should throw a #GP */
2076                         return 1;
2077                 }
2078                 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2079                             __func__, data);
2080                 break;
2081         case 0x200 ... 0x2ff:
2082                 return kvm_mtrr_set_msr(vcpu, msr, data);
2083         case MSR_IA32_APICBASE:
2084                 return kvm_set_apic_base(vcpu, msr_info);
2085         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2086                 return kvm_x2apic_msr_write(vcpu, msr, data);
2087         case MSR_IA32_TSCDEADLINE:
2088                 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2089                 break;
2090         case MSR_IA32_TSC_ADJUST:
2091                 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2092                         if (!msr_info->host_initiated) {
2093                                 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2094                                 adjust_tsc_offset_guest(vcpu, adj);
2095                         }
2096                         vcpu->arch.ia32_tsc_adjust_msr = data;
2097                 }
2098                 break;
2099         case MSR_IA32_MISC_ENABLE:
2100                 vcpu->arch.ia32_misc_enable_msr = data;
2101                 break;
2102         case MSR_IA32_SMBASE:
2103                 if (!msr_info->host_initiated)
2104                         return 1;
2105                 vcpu->arch.smbase = data;
2106                 break;
2107         case MSR_KVM_WALL_CLOCK_NEW:
2108         case MSR_KVM_WALL_CLOCK:
2109                 vcpu->kvm->arch.wall_clock = data;
2110                 kvm_write_wall_clock(vcpu->kvm, data);
2111                 break;
2112         case MSR_KVM_SYSTEM_TIME_NEW:
2113         case MSR_KVM_SYSTEM_TIME: {
2114                 u64 gpa_offset;
2115                 struct kvm_arch *ka = &vcpu->kvm->arch;
2116
2117                 kvmclock_reset(vcpu);
2118
2119                 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2120                         bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2121
2122                         if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2123                                 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2124                                         &vcpu->requests);
2125
2126                         ka->boot_vcpu_runs_old_kvmclock = tmp;
2127                 }
2128
2129                 vcpu->arch.time = data;
2130                 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2131
2132                 /* we verify if the enable bit is set... */
2133                 if (!(data & 1))
2134                         break;
2135
2136                 gpa_offset = data & ~(PAGE_MASK | 1);
2137
2138                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2139                      &vcpu->arch.pv_time, data & ~1ULL,
2140                      sizeof(struct pvclock_vcpu_time_info)))
2141                         vcpu->arch.pv_time_enabled = false;
2142                 else
2143                         vcpu->arch.pv_time_enabled = true;
2144
2145                 break;
2146         }
2147         case MSR_KVM_ASYNC_PF_EN:
2148                 if (kvm_pv_enable_async_pf(vcpu, data))
2149                         return 1;
2150                 break;
2151         case MSR_KVM_STEAL_TIME:
2152
2153                 if (unlikely(!sched_info_on()))
2154                         return 1;
2155
2156                 if (data & KVM_STEAL_RESERVED_MASK)
2157                         return 1;
2158
2159                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2160                                                 data & KVM_STEAL_VALID_BITS,
2161                                                 sizeof(struct kvm_steal_time)))
2162                         return 1;
2163
2164                 vcpu->arch.st.msr_val = data;
2165
2166                 if (!(data & KVM_MSR_ENABLED))
2167                         break;
2168
2169                 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2170
2171                 break;
2172         case MSR_KVM_PV_EOI_EN:
2173                 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2174                         return 1;
2175                 break;
2176
2177         case MSR_IA32_MCG_CTL:
2178         case MSR_IA32_MCG_STATUS:
2179         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2180                 return set_msr_mce(vcpu, msr, data);
2181
2182         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2183         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2184                 pr = true; /* fall through */
2185         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2186         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2187                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2188                         return kvm_pmu_set_msr(vcpu, msr_info);
2189
2190                 if (pr || data != 0)
2191                         vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2192                                     "0x%x data 0x%llx\n", msr, data);
2193                 break;
2194         case MSR_K7_CLK_CTL:
2195                 /*
2196                  * Ignore all writes to this no longer documented MSR.
2197                  * Writes are only relevant for old K7 processors,
2198                  * all pre-dating SVM, but a recommended workaround from
2199                  * AMD for these chips. It is possible to specify the
2200                  * affected processor models on the command line, hence
2201                  * the need to ignore the workaround.
2202                  */
2203                 break;
2204         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2205         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2206         case HV_X64_MSR_CRASH_CTL:
2207         case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2208                 return kvm_hv_set_msr_common(vcpu, msr, data,
2209                                              msr_info->host_initiated);
2210         case MSR_IA32_BBL_CR_CTL3:
2211                 /* Drop writes to this legacy MSR -- see rdmsr
2212                  * counterpart for further detail.
2213                  */
2214                 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2215                 break;
2216         case MSR_AMD64_OSVW_ID_LENGTH:
2217                 if (!guest_cpuid_has_osvw(vcpu))
2218                         return 1;
2219                 vcpu->arch.osvw.length = data;
2220                 break;
2221         case MSR_AMD64_OSVW_STATUS:
2222                 if (!guest_cpuid_has_osvw(vcpu))
2223                         return 1;
2224                 vcpu->arch.osvw.status = data;
2225                 break;
2226         default:
2227                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2228                         return xen_hvm_config(vcpu, data);
2229                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2230                         return kvm_pmu_set_msr(vcpu, msr_info);
2231                 if (!ignore_msrs) {
2232                         vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2233                                     msr, data);
2234                         return 1;
2235                 } else {
2236                         vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2237                                     msr, data);
2238                         break;
2239                 }
2240         }
2241         return 0;
2242 }
2243 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2244
2245
2246 /*
2247  * Reads an msr value (of 'msr_index') into 'pdata'.
2248  * Returns 0 on success, non-0 otherwise.
2249  * Assumes vcpu_load() was already called.
2250  */
2251 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2252 {
2253         return kvm_x86_ops->get_msr(vcpu, msr);
2254 }
2255 EXPORT_SYMBOL_GPL(kvm_get_msr);
2256
2257 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2258 {
2259         u64 data;
2260         u64 mcg_cap = vcpu->arch.mcg_cap;
2261         unsigned bank_num = mcg_cap & 0xff;
2262
2263         switch (msr) {
2264         case MSR_IA32_P5_MC_ADDR:
2265         case MSR_IA32_P5_MC_TYPE:
2266                 data = 0;
2267                 break;
2268         case MSR_IA32_MCG_CAP:
2269                 data = vcpu->arch.mcg_cap;
2270                 break;
2271         case MSR_IA32_MCG_CTL:
2272                 if (!(mcg_cap & MCG_CTL_P))
2273                         return 1;
2274                 data = vcpu->arch.mcg_ctl;
2275                 break;
2276         case MSR_IA32_MCG_STATUS:
2277                 data = vcpu->arch.mcg_status;
2278                 break;
2279         default:
2280                 if (msr >= MSR_IA32_MC0_CTL &&
2281                     msr < MSR_IA32_MCx_CTL(bank_num)) {
2282                         u32 offset = msr - MSR_IA32_MC0_CTL;
2283                         data = vcpu->arch.mce_banks[offset];
2284                         break;
2285                 }
2286                 return 1;
2287         }
2288         *pdata = data;
2289         return 0;
2290 }
2291
2292 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2293 {
2294         switch (msr_info->index) {
2295         case MSR_IA32_PLATFORM_ID:
2296         case MSR_IA32_EBL_CR_POWERON:
2297         case MSR_IA32_DEBUGCTLMSR:
2298         case MSR_IA32_LASTBRANCHFROMIP:
2299         case MSR_IA32_LASTBRANCHTOIP:
2300         case MSR_IA32_LASTINTFROMIP:
2301         case MSR_IA32_LASTINTTOIP:
2302         case MSR_K8_SYSCFG:
2303         case MSR_K8_TSEG_ADDR:
2304         case MSR_K8_TSEG_MASK:
2305         case MSR_K7_HWCR:
2306         case MSR_VM_HSAVE_PA:
2307         case MSR_K8_INT_PENDING_MSG:
2308         case MSR_AMD64_NB_CFG:
2309         case MSR_FAM10H_MMIO_CONF_BASE:
2310         case MSR_AMD64_BU_CFG2:
2311         case MSR_IA32_PERF_CTL:
2312                 msr_info->data = 0;
2313                 break;
2314         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2315         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2316         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2317         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2318                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2319                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2320                 msr_info->data = 0;
2321                 break;
2322         case MSR_IA32_UCODE_REV:
2323                 msr_info->data = 0x100000000ULL;
2324                 break;
2325         case MSR_MTRRcap:
2326         case 0x200 ... 0x2ff:
2327                 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2328         case 0xcd: /* fsb frequency */
2329                 msr_info->data = 3;
2330                 break;
2331                 /*
2332                  * MSR_EBC_FREQUENCY_ID
2333                  * Conservative value valid for even the basic CPU models.
2334                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2335                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2336                  * and 266MHz for model 3, or 4. Set Core Clock
2337                  * Frequency to System Bus Frequency Ratio to 1 (bits
2338                  * 31:24) even though these are only valid for CPU
2339                  * models > 2, however guests may end up dividing or
2340                  * multiplying by zero otherwise.
2341                  */
2342         case MSR_EBC_FREQUENCY_ID:
2343                 msr_info->data = 1 << 24;
2344                 break;
2345         case MSR_IA32_APICBASE:
2346                 msr_info->data = kvm_get_apic_base(vcpu);
2347                 break;
2348         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2349                 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2350                 break;
2351         case MSR_IA32_TSCDEADLINE:
2352                 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2353                 break;
2354         case MSR_IA32_TSC_ADJUST:
2355                 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2356                 break;
2357         case MSR_IA32_MISC_ENABLE:
2358                 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2359                 break;
2360         case MSR_IA32_SMBASE:
2361                 if (!msr_info->host_initiated)
2362                         return 1;
2363                 msr_info->data = vcpu->arch.smbase;
2364                 break;
2365         case MSR_IA32_PERF_STATUS:
2366                 /* TSC increment by tick */
2367                 msr_info->data = 1000ULL;
2368                 /* CPU multiplier */
2369                 msr_info->data |= (((uint64_t)4ULL) << 40);
2370                 break;
2371         case MSR_EFER:
2372                 msr_info->data = vcpu->arch.efer;
2373                 break;
2374         case MSR_KVM_WALL_CLOCK:
2375         case MSR_KVM_WALL_CLOCK_NEW:
2376                 msr_info->data = vcpu->kvm->arch.wall_clock;
2377                 break;
2378         case MSR_KVM_SYSTEM_TIME:
2379         case MSR_KVM_SYSTEM_TIME_NEW:
2380                 msr_info->data = vcpu->arch.time;
2381                 break;
2382         case MSR_KVM_ASYNC_PF_EN:
2383                 msr_info->data = vcpu->arch.apf.msr_val;
2384                 break;
2385         case MSR_KVM_STEAL_TIME:
2386                 msr_info->data = vcpu->arch.st.msr_val;
2387                 break;
2388         case MSR_KVM_PV_EOI_EN:
2389                 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2390                 break;
2391         case MSR_IA32_P5_MC_ADDR:
2392         case MSR_IA32_P5_MC_TYPE:
2393         case MSR_IA32_MCG_CAP:
2394         case MSR_IA32_MCG_CTL:
2395         case MSR_IA32_MCG_STATUS:
2396         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2397                 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2398         case MSR_K7_CLK_CTL:
2399                 /*
2400                  * Provide expected ramp-up count for K7. All other
2401                  * are set to zero, indicating minimum divisors for
2402                  * every field.
2403                  *
2404                  * This prevents guest kernels on AMD host with CPU
2405                  * type 6, model 8 and higher from exploding due to
2406                  * the rdmsr failing.
2407                  */
2408                 msr_info->data = 0x20000000;
2409                 break;
2410         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2411         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2412         case HV_X64_MSR_CRASH_CTL:
2413         case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2414                 return kvm_hv_get_msr_common(vcpu,
2415                                              msr_info->index, &msr_info->data);
2416                 break;
2417         case MSR_IA32_BBL_CR_CTL3:
2418                 /* This legacy MSR exists but isn't fully documented in current
2419                  * silicon.  It is however accessed by winxp in very narrow
2420                  * scenarios where it sets bit #19, itself documented as
2421                  * a "reserved" bit.  Best effort attempt to source coherent
2422                  * read data here should the balance of the register be
2423                  * interpreted by the guest:
2424                  *
2425                  * L2 cache control register 3: 64GB range, 256KB size,
2426                  * enabled, latency 0x1, configured
2427                  */
2428                 msr_info->data = 0xbe702111;
2429                 break;
2430         case MSR_AMD64_OSVW_ID_LENGTH:
2431                 if (!guest_cpuid_has_osvw(vcpu))
2432                         return 1;
2433                 msr_info->data = vcpu->arch.osvw.length;
2434                 break;
2435         case MSR_AMD64_OSVW_STATUS:
2436                 if (!guest_cpuid_has_osvw(vcpu))
2437                         return 1;
2438                 msr_info->data = vcpu->arch.osvw.status;
2439                 break;
2440         default:
2441                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2442                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2443                 if (!ignore_msrs) {
2444                         vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2445                         return 1;
2446                 } else {
2447                         vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2448                         msr_info->data = 0;
2449                 }
2450                 break;
2451         }
2452         return 0;
2453 }
2454 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2455
2456 /*
2457  * Read or write a bunch of msrs. All parameters are kernel addresses.
2458  *
2459  * @return number of msrs set successfully.
2460  */
2461 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2462                     struct kvm_msr_entry *entries,
2463                     int (*do_msr)(struct kvm_vcpu *vcpu,
2464                                   unsigned index, u64 *data))
2465 {
2466         int i, idx;
2467
2468         idx = srcu_read_lock(&vcpu->kvm->srcu);
2469         for (i = 0; i < msrs->nmsrs; ++i)
2470                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2471                         break;
2472         srcu_read_unlock(&vcpu->kvm->srcu, idx);
2473
2474         return i;
2475 }
2476
2477 /*
2478  * Read or write a bunch of msrs. Parameters are user addresses.
2479  *
2480  * @return number of msrs set successfully.
2481  */
2482 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2483                   int (*do_msr)(struct kvm_vcpu *vcpu,
2484                                 unsigned index, u64 *data),
2485                   int writeback)
2486 {
2487         struct kvm_msrs msrs;
2488         struct kvm_msr_entry *entries;
2489         int r, n;
2490         unsigned size;
2491
2492         r = -EFAULT;
2493         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2494                 goto out;
2495
2496         r = -E2BIG;
2497         if (msrs.nmsrs >= MAX_IO_MSRS)
2498                 goto out;
2499
2500         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2501         entries = memdup_user(user_msrs->entries, size);
2502         if (IS_ERR(entries)) {
2503                 r = PTR_ERR(entries);
2504                 goto out;
2505         }
2506
2507         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2508         if (r < 0)
2509                 goto out_free;
2510
2511         r = -EFAULT;
2512         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2513                 goto out_free;
2514
2515         r = n;
2516
2517 out_free:
2518         kfree(entries);
2519 out:
2520         return r;
2521 }
2522
2523 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2524 {
2525         int r;
2526
2527         switch (ext) {
2528         case KVM_CAP_IRQCHIP:
2529         case KVM_CAP_HLT:
2530         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2531         case KVM_CAP_SET_TSS_ADDR:
2532         case KVM_CAP_EXT_CPUID:
2533         case KVM_CAP_EXT_EMUL_CPUID:
2534         case KVM_CAP_CLOCKSOURCE:
2535         case KVM_CAP_PIT:
2536         case KVM_CAP_NOP_IO_DELAY:
2537         case KVM_CAP_MP_STATE:
2538         case KVM_CAP_SYNC_MMU:
2539         case KVM_CAP_USER_NMI:
2540         case KVM_CAP_REINJECT_CONTROL:
2541         case KVM_CAP_IRQ_INJECT_STATUS:
2542         case KVM_CAP_IOEVENTFD:
2543         case KVM_CAP_IOEVENTFD_NO_LENGTH:
2544         case KVM_CAP_PIT2:
2545         case KVM_CAP_PIT_STATE2:
2546         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2547         case KVM_CAP_XEN_HVM:
2548         case KVM_CAP_ADJUST_CLOCK:
2549         case KVM_CAP_VCPU_EVENTS:
2550         case KVM_CAP_HYPERV:
2551         case KVM_CAP_HYPERV_VAPIC:
2552         case KVM_CAP_HYPERV_SPIN:
2553         case KVM_CAP_HYPERV_SYNIC:
2554         case KVM_CAP_PCI_SEGMENT:
2555         case KVM_CAP_DEBUGREGS:
2556         case KVM_CAP_X86_ROBUST_SINGLESTEP:
2557         case KVM_CAP_XSAVE:
2558         case KVM_CAP_ASYNC_PF:
2559         case KVM_CAP_GET_TSC_KHZ:
2560         case KVM_CAP_KVMCLOCK_CTRL:
2561         case KVM_CAP_READONLY_MEM:
2562         case KVM_CAP_HYPERV_TIME:
2563         case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2564         case KVM_CAP_TSC_DEADLINE_TIMER:
2565         case KVM_CAP_ENABLE_CAP_VM:
2566         case KVM_CAP_DISABLE_QUIRKS:
2567         case KVM_CAP_SET_BOOT_CPU_ID:
2568         case KVM_CAP_SPLIT_IRQCHIP:
2569 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2570         case KVM_CAP_ASSIGN_DEV_IRQ:
2571         case KVM_CAP_PCI_2_3:
2572 #endif
2573                 r = 1;
2574                 break;
2575         case KVM_CAP_X86_SMM:
2576                 /* SMBASE is usually relocated above 1M on modern chipsets,
2577                  * and SMM handlers might indeed rely on 4G segment limits,
2578                  * so do not report SMM to be available if real mode is
2579                  * emulated via vm86 mode.  Still, do not go to great lengths
2580                  * to avoid userspace's usage of the feature, because it is a
2581                  * fringe case that is not enabled except via specific settings
2582                  * of the module parameters.
2583                  */
2584                 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2585                 break;
2586         case KVM_CAP_COALESCED_MMIO:
2587                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2588                 break;
2589         case KVM_CAP_VAPIC:
2590                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2591                 break;
2592         case KVM_CAP_NR_VCPUS:
2593                 r = KVM_SOFT_MAX_VCPUS;
2594                 break;
2595         case KVM_CAP_MAX_VCPUS:
2596                 r = KVM_MAX_VCPUS;
2597                 break;
2598         case KVM_CAP_NR_MEMSLOTS:
2599                 r = KVM_USER_MEM_SLOTS;
2600                 break;
2601         case KVM_CAP_PV_MMU:    /* obsolete */
2602                 r = 0;
2603                 break;
2604 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2605         case KVM_CAP_IOMMU:
2606                 r = iommu_present(&pci_bus_type);
2607                 break;
2608 #endif
2609         case KVM_CAP_MCE:
2610                 r = KVM_MAX_MCE_BANKS;
2611                 break;
2612         case KVM_CAP_XCRS:
2613                 r = boot_cpu_has(X86_FEATURE_XSAVE);
2614                 break;
2615         case KVM_CAP_TSC_CONTROL:
2616                 r = kvm_has_tsc_control;
2617                 break;
2618         default:
2619                 r = 0;
2620                 break;
2621         }
2622         return r;
2623
2624 }
2625
2626 long kvm_arch_dev_ioctl(struct file *filp,
2627                         unsigned int ioctl, unsigned long arg)
2628 {
2629         void __user *argp = (void __user *)arg;
2630         long r;
2631
2632         switch (ioctl) {
2633         case KVM_GET_MSR_INDEX_LIST: {
2634                 struct kvm_msr_list __user *user_msr_list = argp;
2635                 struct kvm_msr_list msr_list;
2636                 unsigned n;
2637
2638                 r = -EFAULT;
2639                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2640                         goto out;
2641                 n = msr_list.nmsrs;
2642                 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2643                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2644                         goto out;
2645                 r = -E2BIG;
2646                 if (n < msr_list.nmsrs)
2647                         goto out;
2648                 r = -EFAULT;
2649                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2650                                  num_msrs_to_save * sizeof(u32)))
2651                         goto out;
2652                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2653                                  &emulated_msrs,
2654                                  num_emulated_msrs * sizeof(u32)))
2655                         goto out;
2656                 r = 0;
2657                 break;
2658         }
2659         case KVM_GET_SUPPORTED_CPUID:
2660         case KVM_GET_EMULATED_CPUID: {
2661                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2662                 struct kvm_cpuid2 cpuid;
2663
2664                 r = -EFAULT;
2665                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2666                         goto out;
2667
2668                 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2669                                             ioctl);
2670                 if (r)
2671                         goto out;
2672
2673                 r = -EFAULT;
2674                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2675                         goto out;
2676                 r = 0;
2677                 break;
2678         }
2679         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2680                 u64 mce_cap;
2681
2682                 mce_cap = KVM_MCE_CAP_SUPPORTED;
2683                 r = -EFAULT;
2684                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2685                         goto out;
2686                 r = 0;
2687                 break;
2688         }
2689         default:
2690                 r = -EINVAL;
2691         }
2692 out:
2693         return r;
2694 }
2695
2696 static void wbinvd_ipi(void *garbage)
2697 {
2698         wbinvd();
2699 }
2700
2701 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2702 {
2703         return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2704 }
2705
2706 static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu)
2707 {
2708         set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
2709 }
2710
2711 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2712 {
2713         /* Address WBINVD may be executed by guest */
2714         if (need_emulate_wbinvd(vcpu)) {
2715                 if (kvm_x86_ops->has_wbinvd_exit())
2716                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2717                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2718                         smp_call_function_single(vcpu->cpu,
2719                                         wbinvd_ipi, NULL, 1);
2720         }
2721
2722         kvm_x86_ops->vcpu_load(vcpu, cpu);
2723
2724         /* Apply any externally detected TSC adjustments (due to suspend) */
2725         if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2726                 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2727                 vcpu->arch.tsc_offset_adjustment = 0;
2728                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2729         }
2730
2731         if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2732                 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2733                                 rdtsc() - vcpu->arch.last_host_tsc;
2734                 if (tsc_delta < 0)
2735                         mark_tsc_unstable("KVM discovered backwards TSC");
2736                 if (check_tsc_unstable()) {
2737                         u64 offset = kvm_compute_tsc_offset(vcpu,
2738                                                 vcpu->arch.last_guest_tsc);
2739                         kvm_x86_ops->write_tsc_offset(vcpu, offset);
2740                         vcpu->arch.tsc_catchup = 1;
2741                 }
2742                 /*
2743                  * On a host with synchronized TSC, there is no need to update
2744                  * kvmclock on vcpu->cpu migration
2745                  */
2746                 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2747                         kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2748                 if (vcpu->cpu != cpu)
2749                         kvm_migrate_timers(vcpu);
2750                 vcpu->cpu = cpu;
2751         }
2752
2753         kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2754 }
2755
2756 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2757 {
2758         kvm_x86_ops->vcpu_put(vcpu);
2759         kvm_put_guest_fpu(vcpu);
2760         vcpu->arch.last_host_tsc = rdtsc();
2761 }
2762
2763 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2764                                     struct kvm_lapic_state *s)
2765 {
2766         if (vcpu->arch.apicv_active)
2767                 kvm_x86_ops->sync_pir_to_irr(vcpu);
2768
2769         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2770
2771         return 0;
2772 }
2773
2774 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2775                                     struct kvm_lapic_state *s)
2776 {
2777         kvm_apic_post_state_restore(vcpu, s);
2778         update_cr8_intercept(vcpu);
2779
2780         return 0;
2781 }
2782
2783 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2784 {
2785         return (!lapic_in_kernel(vcpu) ||
2786                 kvm_apic_accept_pic_intr(vcpu));
2787 }
2788
2789 /*
2790  * if userspace requested an interrupt window, check that the
2791  * interrupt window is open.
2792  *
2793  * No need to exit to userspace if we already have an interrupt queued.
2794  */
2795 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2796 {
2797         return kvm_arch_interrupt_allowed(vcpu) &&
2798                 !kvm_cpu_has_interrupt(vcpu) &&
2799                 !kvm_event_needs_reinjection(vcpu) &&
2800                 kvm_cpu_accept_dm_intr(vcpu);
2801 }
2802
2803 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2804                                     struct kvm_interrupt *irq)
2805 {
2806         if (irq->irq >= KVM_NR_INTERRUPTS)
2807                 return -EINVAL;
2808
2809         if (!irqchip_in_kernel(vcpu->kvm)) {
2810                 kvm_queue_interrupt(vcpu, irq->irq, false);
2811                 kvm_make_request(KVM_REQ_EVENT, vcpu);
2812                 return 0;
2813         }
2814
2815         /*
2816          * With in-kernel LAPIC, we only use this to inject EXTINT, so
2817          * fail for in-kernel 8259.
2818          */
2819         if (pic_in_kernel(vcpu->kvm))
2820                 return -ENXIO;
2821
2822         if (vcpu->arch.pending_external_vector != -1)
2823                 return -EEXIST;
2824
2825         vcpu->arch.pending_external_vector = irq->irq;
2826         kvm_make_request(KVM_REQ_EVENT, vcpu);
2827         return 0;
2828 }
2829
2830 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2831 {
2832         kvm_inject_nmi(vcpu);
2833
2834         return 0;
2835 }
2836
2837 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2838 {
2839         kvm_make_request(KVM_REQ_SMI, vcpu);
2840
2841         return 0;
2842 }
2843
2844 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2845                                            struct kvm_tpr_access_ctl *tac)
2846 {
2847         if (tac->flags)
2848                 return -EINVAL;
2849         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2850         return 0;
2851 }
2852
2853 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2854                                         u64 mcg_cap)
2855 {
2856         int r;
2857         unsigned bank_num = mcg_cap & 0xff, bank;
2858
2859         r = -EINVAL;
2860         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2861                 goto out;
2862         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2863                 goto out;
2864         r = 0;
2865         vcpu->arch.mcg_cap = mcg_cap;
2866         /* Init IA32_MCG_CTL to all 1s */
2867         if (mcg_cap & MCG_CTL_P)
2868                 vcpu->arch.mcg_ctl = ~(u64)0;
2869         /* Init IA32_MCi_CTL to all 1s */
2870         for (bank = 0; bank < bank_num; bank++)
2871                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2872 out:
2873         return r;
2874 }
2875
2876 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2877                                       struct kvm_x86_mce *mce)
2878 {
2879         u64 mcg_cap = vcpu->arch.mcg_cap;
2880         unsigned bank_num = mcg_cap & 0xff;
2881         u64 *banks = vcpu->arch.mce_banks;
2882
2883         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2884                 return -EINVAL;
2885         /*
2886          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2887          * reporting is disabled
2888          */
2889         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2890             vcpu->arch.mcg_ctl != ~(u64)0)
2891                 return 0;
2892         banks += 4 * mce->bank;
2893         /*
2894          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2895          * reporting is disabled for the bank
2896          */
2897         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2898                 return 0;
2899         if (mce->status & MCI_STATUS_UC) {
2900                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2901                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2902                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2903                         return 0;
2904                 }
2905                 if (banks[1] & MCI_STATUS_VAL)
2906                         mce->status |= MCI_STATUS_OVER;
2907                 banks[2] = mce->addr;
2908                 banks[3] = mce->misc;
2909                 vcpu->arch.mcg_status = mce->mcg_status;
2910                 banks[1] = mce->status;
2911                 kvm_queue_exception(vcpu, MC_VECTOR);
2912         } else if (!(banks[1] & MCI_STATUS_VAL)
2913                    || !(banks[1] & MCI_STATUS_UC)) {
2914                 if (banks[1] & MCI_STATUS_VAL)
2915                         mce->status |= MCI_STATUS_OVER;
2916                 banks[2] = mce->addr;
2917                 banks[3] = mce->misc;
2918                 banks[1] = mce->status;
2919         } else
2920                 banks[1] |= MCI_STATUS_OVER;
2921         return 0;
2922 }
2923
2924 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2925                                                struct kvm_vcpu_events *events)
2926 {
2927         process_nmi(vcpu);
2928         events->exception.injected =
2929                 vcpu->arch.exception.pending &&
2930                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2931         events->exception.nr = vcpu->arch.exception.nr;
2932         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2933         events->exception.pad = 0;
2934         events->exception.error_code = vcpu->arch.exception.error_code;
2935
2936         events->interrupt.injected =
2937                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2938         events->interrupt.nr = vcpu->arch.interrupt.nr;
2939         events->interrupt.soft = 0;
2940         events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2941
2942         events->nmi.injected = vcpu->arch.nmi_injected;
2943         events->nmi.pending = vcpu->arch.nmi_pending != 0;
2944         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2945         events->nmi.pad = 0;
2946
2947         events->sipi_vector = 0; /* never valid when reporting to user space */
2948
2949         events->smi.smm = is_smm(vcpu);
2950         events->smi.pending = vcpu->arch.smi_pending;
2951         events->smi.smm_inside_nmi =
2952                 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2953         events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2954
2955         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2956                          | KVM_VCPUEVENT_VALID_SHADOW
2957                          | KVM_VCPUEVENT_VALID_SMM);
2958         memset(&events->reserved, 0, sizeof(events->reserved));
2959 }
2960
2961 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2962                                               struct kvm_vcpu_events *events)
2963 {
2964         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2965                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2966                               | KVM_VCPUEVENT_VALID_SHADOW
2967                               | KVM_VCPUEVENT_VALID_SMM))
2968                 return -EINVAL;
2969
2970         if (events->exception.injected &&
2971             (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
2972                 return -EINVAL;
2973
2974         process_nmi(vcpu);
2975         vcpu->arch.exception.pending = events->exception.injected;
2976         vcpu->arch.exception.nr = events->exception.nr;
2977         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2978         vcpu->arch.exception.error_code = events->exception.error_code;
2979
2980         vcpu->arch.interrupt.pending = events->interrupt.injected;
2981         vcpu->arch.interrupt.nr = events->interrupt.nr;
2982         vcpu->arch.interrupt.soft = events->interrupt.soft;
2983         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2984                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2985                                                   events->interrupt.shadow);
2986
2987         vcpu->arch.nmi_injected = events->nmi.injected;
2988         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2989                 vcpu->arch.nmi_pending = events->nmi.pending;
2990         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2991
2992         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2993             lapic_in_kernel(vcpu))
2994                 vcpu->arch.apic->sipi_vector = events->sipi_vector;
2995
2996         if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
2997                 if (events->smi.smm)
2998                         vcpu->arch.hflags |= HF_SMM_MASK;
2999                 else
3000                         vcpu->arch.hflags &= ~HF_SMM_MASK;
3001                 vcpu->arch.smi_pending = events->smi.pending;
3002                 if (events->smi.smm_inside_nmi)
3003                         vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3004                 else
3005                         vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3006                 if (lapic_in_kernel(vcpu)) {
3007                         if (events->smi.latched_init)
3008                                 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3009                         else
3010                                 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3011                 }
3012         }
3013
3014         kvm_make_request(KVM_REQ_EVENT, vcpu);
3015
3016         return 0;
3017 }
3018
3019 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3020                                              struct kvm_debugregs *dbgregs)
3021 {
3022         unsigned long val;
3023
3024         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3025         kvm_get_dr(vcpu, 6, &val);
3026         dbgregs->dr6 = val;
3027         dbgregs->dr7 = vcpu->arch.dr7;
3028         dbgregs->flags = 0;
3029         memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3030 }
3031
3032 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3033                                             struct kvm_debugregs *dbgregs)
3034 {
3035         if (dbgregs->flags)
3036                 return -EINVAL;
3037
3038         if (dbgregs->dr6 & ~0xffffffffull)
3039                 return -EINVAL;
3040         if (dbgregs->dr7 & ~0xffffffffull)
3041                 return -EINVAL;
3042
3043         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3044         kvm_update_dr0123(vcpu);
3045         vcpu->arch.dr6 = dbgregs->dr6;
3046         kvm_update_dr6(vcpu);
3047         vcpu->arch.dr7 = dbgregs->dr7;
3048         kvm_update_dr7(vcpu);
3049
3050         return 0;
3051 }
3052
3053 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3054
3055 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3056 {
3057         struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3058         u64 xstate_bv = xsave->header.xfeatures;
3059         u64 valid;
3060
3061         /*
3062          * Copy legacy XSAVE area, to avoid complications with CPUID
3063          * leaves 0 and 1 in the loop below.
3064          */
3065         memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3066
3067         /* Set XSTATE_BV */
3068         *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3069
3070         /*
3071          * Copy each region from the possibly compacted offset to the
3072          * non-compacted offset.
3073          */
3074         valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3075         while (valid) {
3076                 u64 feature = valid & -valid;
3077                 int index = fls64(feature) - 1;
3078                 void *src = get_xsave_addr(xsave, feature);
3079
3080                 if (src) {
3081                         u32 size, offset, ecx, edx;
3082                         cpuid_count(XSTATE_CPUID, index,
3083                                     &size, &offset, &ecx, &edx);
3084                         memcpy(dest + offset, src, size);
3085                 }
3086
3087                 valid -= feature;
3088         }
3089 }
3090
3091 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3092 {
3093         struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3094         u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3095         u64 valid;
3096
3097         /*
3098          * Copy legacy XSAVE area, to avoid complications with CPUID
3099          * leaves 0 and 1 in the loop below.
3100          */
3101         memcpy(xsave, src, XSAVE_HDR_OFFSET);
3102
3103         /* Set XSTATE_BV and possibly XCOMP_BV.  */
3104         xsave->header.xfeatures = xstate_bv;
3105         if (boot_cpu_has(X86_FEATURE_XSAVES))
3106                 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3107
3108         /*
3109          * Copy each region from the non-compacted offset to the
3110          * possibly compacted offset.
3111          */
3112         valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3113         while (valid) {
3114                 u64 feature = valid & -valid;
3115                 int index = fls64(feature) - 1;
3116                 void *dest = get_xsave_addr(xsave, feature);
3117
3118                 if (dest) {
3119                         u32 size, offset, ecx, edx;
3120                         cpuid_count(XSTATE_CPUID, index,
3121                                     &size, &offset, &ecx, &edx);
3122                         memcpy(dest, src + offset, size);
3123                 }
3124
3125                 valid -= feature;
3126         }
3127 }
3128
3129 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3130                                          struct kvm_xsave *guest_xsave)
3131 {
3132         if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3133                 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3134                 fill_xsave((u8 *) guest_xsave->region, vcpu);
3135         } else {
3136                 memcpy(guest_xsave->region,
3137                         &vcpu->arch.guest_fpu.state.fxsave,
3138                         sizeof(struct fxregs_state));
3139                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3140                         XFEATURE_MASK_FPSSE;
3141         }
3142 }
3143
3144 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3145                                         struct kvm_xsave *guest_xsave)
3146 {
3147         u64 xstate_bv =
3148                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3149
3150         if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3151                 /*
3152                  * Here we allow setting states that are not present in
3153                  * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3154                  * with old userspace.
3155                  */
3156                 if (xstate_bv & ~kvm_supported_xcr0())
3157                         return -EINVAL;
3158                 load_xsave(vcpu, (u8 *)guest_xsave->region);
3159         } else {
3160                 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3161                         return -EINVAL;
3162                 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3163                         guest_xsave->region, sizeof(struct fxregs_state));
3164         }
3165         return 0;
3166 }
3167
3168 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3169                                         struct kvm_xcrs *guest_xcrs)
3170 {
3171         if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3172                 guest_xcrs->nr_xcrs = 0;
3173                 return;
3174         }
3175
3176         guest_xcrs->nr_xcrs = 1;
3177         guest_xcrs->flags = 0;
3178         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3179         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3180 }
3181
3182 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3183                                        struct kvm_xcrs *guest_xcrs)
3184 {
3185         int i, r = 0;
3186
3187         if (!boot_cpu_has(X86_FEATURE_XSAVE))
3188                 return -EINVAL;
3189
3190         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3191                 return -EINVAL;
3192
3193         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3194                 /* Only support XCR0 currently */
3195                 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3196                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3197                                 guest_xcrs->xcrs[i].value);
3198                         break;
3199                 }
3200         if (r)
3201                 r = -EINVAL;
3202         return r;
3203 }
3204
3205 /*
3206  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3207  * stopped by the hypervisor.  This function will be called from the host only.
3208  * EINVAL is returned when the host attempts to set the flag for a guest that
3209  * does not support pv clocks.
3210  */
3211 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3212 {
3213         if (!vcpu->arch.pv_time_enabled)
3214                 return -EINVAL;
3215         vcpu->arch.pvclock_set_guest_stopped_request = true;
3216         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3217         return 0;
3218 }
3219
3220 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3221                                      struct kvm_enable_cap *cap)
3222 {
3223         if (cap->flags)
3224                 return -EINVAL;
3225
3226         switch (cap->cap) {
3227         case KVM_CAP_HYPERV_SYNIC:
3228                 return kvm_hv_activate_synic(vcpu);
3229         default:
3230                 return -EINVAL;
3231         }
3232 }
3233
3234 long kvm_arch_vcpu_ioctl(struct file *filp,
3235                          unsigned int ioctl, unsigned long arg)
3236 {
3237         struct kvm_vcpu *vcpu = filp->private_data;
3238         void __user *argp = (void __user *)arg;
3239         int r;
3240         union {
3241                 struct kvm_lapic_state *lapic;
3242                 struct kvm_xsave *xsave;
3243                 struct kvm_xcrs *xcrs;
3244                 void *buffer;
3245         } u;
3246
3247         u.buffer = NULL;
3248         switch (ioctl) {
3249         case KVM_GET_LAPIC: {
3250                 r = -EINVAL;
3251                 if (!lapic_in_kernel(vcpu))
3252                         goto out;
3253                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3254
3255                 r = -ENOMEM;
3256                 if (!u.lapic)
3257                         goto out;
3258                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3259                 if (r)
3260                         goto out;
3261                 r = -EFAULT;
3262                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3263                         goto out;
3264                 r = 0;
3265                 break;
3266         }
3267         case KVM_SET_LAPIC: {
3268                 r = -EINVAL;
3269                 if (!lapic_in_kernel(vcpu))
3270                         goto out;
3271                 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3272                 if (IS_ERR(u.lapic))
3273                         return PTR_ERR(u.lapic);
3274
3275                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3276                 break;
3277         }
3278         case KVM_INTERRUPT: {
3279                 struct kvm_interrupt irq;
3280
3281                 r = -EFAULT;
3282                 if (copy_from_user(&irq, argp, sizeof irq))
3283                         goto out;
3284                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3285                 break;
3286         }
3287         case KVM_NMI: {
3288                 r = kvm_vcpu_ioctl_nmi(vcpu);
3289                 break;
3290         }
3291         case KVM_SMI: {
3292                 r = kvm_vcpu_ioctl_smi(vcpu);
3293                 break;
3294         }
3295         case KVM_SET_CPUID: {
3296                 struct kvm_cpuid __user *cpuid_arg = argp;
3297                 struct kvm_cpuid cpuid;
3298
3299                 r = -EFAULT;
3300                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3301                         goto out;
3302                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3303                 break;
3304         }
3305         case KVM_SET_CPUID2: {
3306                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3307                 struct kvm_cpuid2 cpuid;
3308
3309                 r = -EFAULT;
3310                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3311                         goto out;
3312                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3313                                               cpuid_arg->entries);
3314                 break;
3315         }
3316         case KVM_GET_CPUID2: {
3317                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3318                 struct kvm_cpuid2 cpuid;
3319
3320                 r = -EFAULT;
3321                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3322                         goto out;
3323                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3324                                               cpuid_arg->entries);
3325                 if (r)
3326                         goto out;
3327                 r = -EFAULT;
3328                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3329                         goto out;
3330                 r = 0;
3331                 break;
3332         }
3333         case KVM_GET_MSRS:
3334                 r = msr_io(vcpu, argp, do_get_msr, 1);
3335                 break;
3336         case KVM_SET_MSRS:
3337                 r = msr_io(vcpu, argp, do_set_msr, 0);
3338                 break;
3339         case KVM_TPR_ACCESS_REPORTING: {
3340                 struct kvm_tpr_access_ctl tac;
3341
3342                 r = -EFAULT;
3343                 if (copy_from_user(&tac, argp, sizeof tac))
3344                         goto out;
3345                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3346                 if (r)
3347                         goto out;
3348                 r = -EFAULT;
3349                 if (copy_to_user(argp, &tac, sizeof tac))
3350                         goto out;
3351                 r = 0;
3352                 break;
3353         };
3354         case KVM_SET_VAPIC_ADDR: {
3355                 struct kvm_vapic_addr va;
3356
3357                 r = -EINVAL;
3358                 if (!lapic_in_kernel(vcpu))
3359                         goto out;
3360                 r = -EFAULT;
3361                 if (copy_from_user(&va, argp, sizeof va))
3362                         goto out;
3363                 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3364                 break;
3365         }
3366         case KVM_X86_SETUP_MCE: {
3367                 u64 mcg_cap;
3368
3369                 r = -EFAULT;
3370                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3371                         goto out;
3372                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3373                 break;
3374         }
3375         case KVM_X86_SET_MCE: {
3376                 struct kvm_x86_mce mce;
3377
3378                 r = -EFAULT;
3379                 if (copy_from_user(&mce, argp, sizeof mce))
3380                         goto out;
3381                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3382                 break;
3383         }
3384         case KVM_GET_VCPU_EVENTS: {
3385                 struct kvm_vcpu_events events;
3386
3387                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3388
3389                 r = -EFAULT;
3390                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3391                         break;
3392                 r = 0;
3393                 break;
3394         }
3395         case KVM_SET_VCPU_EVENTS: {
3396                 struct kvm_vcpu_events events;
3397
3398                 r = -EFAULT;
3399                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3400                         break;
3401
3402                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3403                 break;
3404         }
3405         case KVM_GET_DEBUGREGS: {
3406                 struct kvm_debugregs dbgregs;
3407
3408                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3409
3410                 r = -EFAULT;
3411                 if (copy_to_user(argp, &dbgregs,
3412                                  sizeof(struct kvm_debugregs)))
3413                         break;
3414                 r = 0;
3415                 break;
3416         }
3417         case KVM_SET_DEBUGREGS: {
3418                 struct kvm_debugregs dbgregs;
3419
3420                 r = -EFAULT;
3421                 if (copy_from_user(&dbgregs, argp,
3422                                    sizeof(struct kvm_debugregs)))
3423                         break;
3424
3425                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3426                 break;
3427         }
3428         case KVM_GET_XSAVE: {
3429                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3430                 r = -ENOMEM;
3431                 if (!u.xsave)
3432                         break;
3433
3434                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3435
3436                 r = -EFAULT;
3437                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3438                         break;
3439                 r = 0;
3440                 break;
3441         }
3442         case KVM_SET_XSAVE: {
3443                 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3444                 if (IS_ERR(u.xsave))
3445                         return PTR_ERR(u.xsave);
3446
3447                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3448                 break;
3449         }
3450         case KVM_GET_XCRS: {
3451                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3452                 r = -ENOMEM;
3453                 if (!u.xcrs)
3454                         break;
3455
3456                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3457
3458                 r = -EFAULT;
3459                 if (copy_to_user(argp, u.xcrs,
3460                                  sizeof(struct kvm_xcrs)))
3461                         break;
3462                 r = 0;
3463                 break;
3464         }
3465         case KVM_SET_XCRS: {
3466                 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3467                 if (IS_ERR(u.xcrs))
3468                         return PTR_ERR(u.xcrs);
3469
3470                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3471                 break;
3472         }
3473         case KVM_SET_TSC_KHZ: {
3474                 u32 user_tsc_khz;
3475
3476                 r = -EINVAL;
3477                 user_tsc_khz = (u32)arg;
3478
3479                 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3480                         goto out;
3481
3482                 if (user_tsc_khz == 0)
3483                         user_tsc_khz = tsc_khz;
3484
3485                 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3486                         r = 0;
3487
3488                 goto out;
3489         }
3490         case KVM_GET_TSC_KHZ: {
3491                 r = vcpu->arch.virtual_tsc_khz;
3492                 goto out;
3493         }
3494         case KVM_KVMCLOCK_CTRL: {
3495                 r = kvm_set_guest_paused(vcpu);
3496                 goto out;
3497         }
3498         case KVM_ENABLE_CAP: {
3499                 struct kvm_enable_cap cap;
3500
3501                 r = -EFAULT;
3502                 if (copy_from_user(&cap, argp, sizeof(cap)))
3503                         goto out;
3504                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3505                 break;
3506         }
3507         default:
3508                 r = -EINVAL;
3509         }
3510 out:
3511         kfree(u.buffer);
3512         return r;
3513 }
3514
3515 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3516 {
3517         return VM_FAULT_SIGBUS;
3518 }
3519
3520 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3521 {
3522         int ret;
3523
3524         if (addr > (unsigned int)(-3 * PAGE_SIZE))
3525                 return -EINVAL;
3526         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3527         return ret;
3528 }
3529
3530 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3531                                               u64 ident_addr)
3532 {
3533         kvm->arch.ept_identity_map_addr = ident_addr;
3534         return 0;
3535 }
3536
3537 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3538                                           u32 kvm_nr_mmu_pages)
3539 {
3540         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3541                 return -EINVAL;
3542
3543         mutex_lock(&kvm->slots_lock);
3544
3545         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3546         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3547
3548         mutex_unlock(&kvm->slots_lock);
3549         return 0;
3550 }
3551
3552 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3553 {
3554         return kvm->arch.n_max_mmu_pages;
3555 }
3556
3557 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3558 {
3559         int r;
3560
3561         r = 0;
3562         switch (chip->chip_id) {
3563         case KVM_IRQCHIP_PIC_MASTER:
3564                 memcpy(&chip->chip.pic,
3565                         &pic_irqchip(kvm)->pics[0],
3566                         sizeof(struct kvm_pic_state));
3567                 break;
3568         case KVM_IRQCHIP_PIC_SLAVE:
3569                 memcpy(&chip->chip.pic,
3570                         &pic_irqchip(kvm)->pics[1],
3571                         sizeof(struct kvm_pic_state));
3572                 break;
3573         case KVM_IRQCHIP_IOAPIC:
3574                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3575                 break;
3576         default:
3577                 r = -EINVAL;
3578                 break;
3579         }
3580         return r;
3581 }
3582
3583 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3584 {
3585         int r;
3586
3587         r = 0;
3588         switch (chip->chip_id) {
3589         case KVM_IRQCHIP_PIC_MASTER:
3590                 spin_lock(&pic_irqchip(kvm)->lock);
3591                 memcpy(&pic_irqchip(kvm)->pics[0],
3592                         &chip->chip.pic,
3593                         sizeof(struct kvm_pic_state));
3594                 spin_unlock(&pic_irqchip(kvm)->lock);
3595                 break;
3596         case KVM_IRQCHIP_PIC_SLAVE:
3597                 spin_lock(&pic_irqchip(kvm)->lock);
3598                 memcpy(&pic_irqchip(kvm)->pics[1],
3599                         &chip->chip.pic,
3600                         sizeof(struct kvm_pic_state));
3601                 spin_unlock(&pic_irqchip(kvm)->lock);
3602                 break;
3603         case KVM_IRQCHIP_IOAPIC:
3604                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3605                 break;
3606         default:
3607                 r = -EINVAL;
3608                 break;
3609         }
3610         kvm_pic_update_irq(pic_irqchip(kvm));
3611         return r;
3612 }
3613
3614 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3615 {
3616         struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3617
3618         BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3619
3620         mutex_lock(&kps->lock);
3621         memcpy(ps, &kps->channels, sizeof(*ps));
3622         mutex_unlock(&kps->lock);
3623         return 0;
3624 }
3625
3626 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3627 {
3628         int i;
3629         struct kvm_pit *pit = kvm->arch.vpit;
3630
3631         mutex_lock(&pit->pit_state.lock);
3632         memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3633         for (i = 0; i < 3; i++)
3634                 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3635         mutex_unlock(&pit->pit_state.lock);
3636         return 0;
3637 }
3638
3639 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3640 {
3641         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3642         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3643                 sizeof(ps->channels));
3644         ps->flags = kvm->arch.vpit->pit_state.flags;
3645         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3646         memset(&ps->reserved, 0, sizeof(ps->reserved));
3647         return 0;
3648 }
3649
3650 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3651 {
3652         int start = 0;
3653         int i;
3654         u32 prev_legacy, cur_legacy;
3655         struct kvm_pit *pit = kvm->arch.vpit;
3656
3657         mutex_lock(&pit->pit_state.lock);
3658         prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3659         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3660         if (!prev_legacy && cur_legacy)
3661                 start = 1;
3662         memcpy(&pit->pit_state.channels, &ps->channels,
3663                sizeof(pit->pit_state.channels));
3664         pit->pit_state.flags = ps->flags;
3665         for (i = 0; i < 3; i++)
3666                 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3667                                    start && i == 0);
3668         mutex_unlock(&pit->pit_state.lock);
3669         return 0;
3670 }
3671
3672 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3673                                  struct kvm_reinject_control *control)
3674 {
3675         struct kvm_pit *pit = kvm->arch.vpit;
3676
3677         if (!pit)
3678                 return -ENXIO;
3679
3680         /* pit->pit_state.lock was overloaded to prevent userspace from getting
3681          * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3682          * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
3683          */
3684         mutex_lock(&pit->pit_state.lock);
3685         kvm_pit_set_reinject(pit, control->pit_reinject);
3686         mutex_unlock(&pit->pit_state.lock);
3687
3688         return 0;
3689 }
3690
3691 /**
3692  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3693  * @kvm: kvm instance
3694  * @log: slot id and address to which we copy the log
3695  *
3696  * Steps 1-4 below provide general overview of dirty page logging. See
3697  * kvm_get_dirty_log_protect() function description for additional details.
3698  *
3699  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3700  * always flush the TLB (step 4) even if previous step failed  and the dirty
3701  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3702  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3703  * writes will be marked dirty for next log read.
3704  *
3705  *   1. Take a snapshot of the bit and clear it if needed.
3706  *   2. Write protect the corresponding page.
3707  *   3. Copy the snapshot to the userspace.
3708  *   4. Flush TLB's if needed.
3709  */
3710 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3711 {
3712         bool is_dirty = false;
3713         int r;
3714
3715         mutex_lock(&kvm->slots_lock);
3716
3717         /*
3718          * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3719          */
3720         if (kvm_x86_ops->flush_log_dirty)
3721                 kvm_x86_ops->flush_log_dirty(kvm);
3722
3723         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3724
3725         /*
3726          * All the TLBs can be flushed out of mmu lock, see the comments in
3727          * kvm_mmu_slot_remove_write_access().
3728          */
3729         lockdep_assert_held(&kvm->slots_lock);
3730         if (is_dirty)
3731                 kvm_flush_remote_tlbs(kvm);
3732
3733         mutex_unlock(&kvm->slots_lock);
3734         return r;
3735 }
3736
3737 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3738                         bool line_status)
3739 {
3740         if (!irqchip_in_kernel(kvm))
3741                 return -ENXIO;
3742
3743         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3744                                         irq_event->irq, irq_event->level,
3745                                         line_status);
3746         return 0;
3747 }
3748
3749 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3750                                    struct kvm_enable_cap *cap)
3751 {
3752         int r;
3753
3754         if (cap->flags)
3755                 return -EINVAL;
3756
3757         switch (cap->cap) {
3758         case KVM_CAP_DISABLE_QUIRKS:
3759                 kvm->arch.disabled_quirks = cap->args[0];
3760                 r = 0;
3761                 break;
3762         case KVM_CAP_SPLIT_IRQCHIP: {
3763                 mutex_lock(&kvm->lock);
3764                 r = -EINVAL;
3765                 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3766                         goto split_irqchip_unlock;
3767                 r = -EEXIST;
3768                 if (irqchip_in_kernel(kvm))
3769                         goto split_irqchip_unlock;
3770                 if (atomic_read(&kvm->online_vcpus))
3771                         goto split_irqchip_unlock;
3772                 r = kvm_setup_empty_irq_routing(kvm);
3773                 if (r)
3774                         goto split_irqchip_unlock;
3775                 /* Pairs with irqchip_in_kernel. */
3776                 smp_wmb();
3777                 kvm->arch.irqchip_split = true;
3778                 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3779                 r = 0;
3780 split_irqchip_unlock:
3781                 mutex_unlock(&kvm->lock);
3782                 break;
3783         }
3784         default:
3785                 r = -EINVAL;
3786                 break;
3787         }
3788         return r;
3789 }
3790
3791 long kvm_arch_vm_ioctl(struct file *filp,
3792                        unsigned int ioctl, unsigned long arg)
3793 {
3794         struct kvm *kvm = filp->private_data;
3795         void __user *argp = (void __user *)arg;
3796         int r = -ENOTTY;
3797         /*
3798          * This union makes it completely explicit to gcc-3.x
3799          * that these two variables' stack usage should be
3800          * combined, not added together.
3801          */
3802         union {
3803                 struct kvm_pit_state ps;
3804                 struct kvm_pit_state2 ps2;
3805                 struct kvm_pit_config pit_config;
3806         } u;
3807
3808         switch (ioctl) {
3809         case KVM_SET_TSS_ADDR:
3810                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3811                 break;
3812         case KVM_SET_IDENTITY_MAP_ADDR: {
3813                 u64 ident_addr;
3814
3815                 r = -EFAULT;
3816                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3817                         goto out;
3818                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3819                 break;
3820         }
3821         case KVM_SET_NR_MMU_PAGES:
3822                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3823                 break;
3824         case KVM_GET_NR_MMU_PAGES:
3825                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3826                 break;
3827         case KVM_CREATE_IRQCHIP: {
3828                 struct kvm_pic *vpic;
3829
3830                 mutex_lock(&kvm->lock);
3831                 r = -EEXIST;
3832                 if (kvm->arch.vpic)
3833                         goto create_irqchip_unlock;
3834                 r = -EINVAL;
3835                 if (atomic_read(&kvm->online_vcpus))
3836                         goto create_irqchip_unlock;
3837                 r = -ENOMEM;
3838                 vpic = kvm_create_pic(kvm);
3839                 if (vpic) {
3840                         r = kvm_ioapic_init(kvm);
3841                         if (r) {
3842                                 mutex_lock(&kvm->slots_lock);
3843                                 kvm_destroy_pic(vpic);
3844                                 mutex_unlock(&kvm->slots_lock);
3845                                 goto create_irqchip_unlock;
3846                         }
3847                 } else
3848                         goto create_irqchip_unlock;
3849                 r = kvm_setup_default_irq_routing(kvm);
3850                 if (r) {
3851                         mutex_lock(&kvm->slots_lock);
3852                         mutex_lock(&kvm->irq_lock);
3853                         kvm_ioapic_destroy(kvm);
3854                         kvm_destroy_pic(vpic);
3855                         mutex_unlock(&kvm->irq_lock);
3856                         mutex_unlock(&kvm->slots_lock);
3857                         goto create_irqchip_unlock;
3858                 }
3859                 /* Write kvm->irq_routing before kvm->arch.vpic.  */
3860                 smp_wmb();
3861                 kvm->arch.vpic = vpic;
3862         create_irqchip_unlock:
3863                 mutex_unlock(&kvm->lock);
3864                 break;
3865         }
3866         case KVM_CREATE_PIT:
3867                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3868                 goto create_pit;
3869         case KVM_CREATE_PIT2:
3870                 r = -EFAULT;
3871                 if (copy_from_user(&u.pit_config, argp,
3872                                    sizeof(struct kvm_pit_config)))
3873                         goto out;
3874         create_pit:
3875                 mutex_lock(&kvm->slots_lock);
3876                 r = -EEXIST;
3877                 if (kvm->arch.vpit)
3878                         goto create_pit_unlock;
3879                 r = -ENOMEM;
3880                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3881                 if (kvm->arch.vpit)
3882                         r = 0;
3883         create_pit_unlock:
3884                 mutex_unlock(&kvm->slots_lock);
3885                 break;
3886         case KVM_GET_IRQCHIP: {
3887                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3888                 struct kvm_irqchip *chip;
3889
3890                 chip = memdup_user(argp, sizeof(*chip));
3891                 if (IS_ERR(chip)) {
3892                         r = PTR_ERR(chip);
3893                         goto out;
3894                 }
3895
3896                 r = -ENXIO;
3897                 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3898                         goto get_irqchip_out;
3899                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3900                 if (r)
3901                         goto get_irqchip_out;
3902                 r = -EFAULT;
3903                 if (copy_to_user(argp, chip, sizeof *chip))
3904                         goto get_irqchip_out;
3905                 r = 0;
3906         get_irqchip_out:
3907                 kfree(chip);
3908                 break;
3909         }
3910         case KVM_SET_IRQCHIP: {
3911                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3912                 struct kvm_irqchip *chip;
3913
3914                 chip = memdup_user(argp, sizeof(*chip));
3915                 if (IS_ERR(chip)) {
3916                         r = PTR_ERR(chip);
3917                         goto out;
3918                 }
3919
3920                 r = -ENXIO;
3921                 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3922                         goto set_irqchip_out;
3923                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3924                 if (r)
3925                         goto set_irqchip_out;
3926                 r = 0;
3927         set_irqchip_out:
3928                 kfree(chip);
3929                 break;
3930         }
3931         case KVM_GET_PIT: {
3932                 r = -EFAULT;
3933                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3934                         goto out;
3935                 r = -ENXIO;
3936                 if (!kvm->arch.vpit)
3937                         goto out;
3938                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3939                 if (r)
3940                         goto out;
3941                 r = -EFAULT;
3942                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3943                         goto out;
3944                 r = 0;
3945                 break;
3946         }
3947         case KVM_SET_PIT: {
3948                 r = -EFAULT;
3949                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3950                         goto out;
3951                 r = -ENXIO;
3952                 if (!kvm->arch.vpit)
3953                         goto out;
3954                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3955                 break;
3956         }
3957         case KVM_GET_PIT2: {
3958                 r = -ENXIO;
3959                 if (!kvm->arch.vpit)
3960                         goto out;
3961                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3962                 if (r)
3963                         goto out;
3964                 r = -EFAULT;
3965                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3966                         goto out;
3967                 r = 0;
3968                 break;
3969         }
3970         case KVM_SET_PIT2: {
3971                 r = -EFAULT;
3972                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3973                         goto out;
3974                 r = -ENXIO;
3975                 if (!kvm->arch.vpit)
3976                         goto out;
3977                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3978                 break;
3979         }
3980         case KVM_REINJECT_CONTROL: {
3981                 struct kvm_reinject_control control;
3982                 r =  -EFAULT;
3983                 if (copy_from_user(&control, argp, sizeof(control)))
3984                         goto out;
3985                 r = kvm_vm_ioctl_reinject(kvm, &control);
3986                 break;
3987         }
3988         case KVM_SET_BOOT_CPU_ID:
3989                 r = 0;
3990                 mutex_lock(&kvm->lock);
3991                 if (atomic_read(&kvm->online_vcpus) != 0)
3992                         r = -EBUSY;
3993                 else
3994                         kvm->arch.bsp_vcpu_id = arg;
3995                 mutex_unlock(&kvm->lock);
3996                 break;
3997         case KVM_XEN_HVM_CONFIG: {
3998                 r = -EFAULT;
3999                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4000                                    sizeof(struct kvm_xen_hvm_config)))
4001                         goto out;
4002                 r = -EINVAL;
4003                 if (kvm->arch.xen_hvm_config.flags)
4004                         goto out;
4005                 r = 0;
4006                 break;
4007         }
4008         case KVM_SET_CLOCK: {
4009                 struct kvm_clock_data user_ns;
4010                 u64 now_ns;
4011                 s64 delta;
4012
4013                 r = -EFAULT;
4014                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4015                         goto out;
4016
4017                 r = -EINVAL;
4018                 if (user_ns.flags)
4019                         goto out;
4020
4021                 r = 0;
4022                 local_irq_disable();
4023                 now_ns = get_kernel_ns();
4024                 delta = user_ns.clock - now_ns;
4025                 local_irq_enable();
4026                 kvm->arch.kvmclock_offset = delta;
4027                 kvm_gen_update_masterclock(kvm);
4028                 break;
4029         }
4030         case KVM_GET_CLOCK: {
4031                 struct kvm_clock_data user_ns;
4032                 u64 now_ns;
4033
4034                 local_irq_disable();
4035                 now_ns = get_kernel_ns();
4036                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4037                 local_irq_enable();
4038                 user_ns.flags = 0;
4039                 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4040
4041                 r = -EFAULT;
4042                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4043                         goto out;
4044                 r = 0;
4045                 break;
4046         }
4047         case KVM_ENABLE_CAP: {
4048                 struct kvm_enable_cap cap;
4049
4050                 r = -EFAULT;
4051                 if (copy_from_user(&cap, argp, sizeof(cap)))
4052                         goto out;
4053                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4054                 break;
4055         }
4056         default:
4057                 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4058         }
4059 out:
4060         return r;
4061 }
4062
4063 static void kvm_init_msr_list(void)
4064 {
4065         u32 dummy[2];
4066         unsigned i, j;
4067
4068         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4069                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4070                         continue;
4071
4072                 /*
4073                  * Even MSRs that are valid in the host may not be exposed
4074                  * to the guests in some cases.
4075                  */
4076                 switch (msrs_to_save[i]) {
4077                 case MSR_IA32_BNDCFGS:
4078                         if (!kvm_x86_ops->mpx_supported())
4079                                 continue;
4080                         break;
4081                 case MSR_TSC_AUX:
4082                         if (!kvm_x86_ops->rdtscp_supported())
4083                                 continue;
4084                         break;
4085                 default:
4086                         break;
4087                 }
4088
4089                 if (j < i)
4090                         msrs_to_save[j] = msrs_to_save[i];
4091                 j++;
4092         }
4093         num_msrs_to_save = j;
4094
4095         for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4096                 switch (emulated_msrs[i]) {
4097                 case MSR_IA32_SMBASE:
4098                         if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4099                                 continue;
4100                         break;
4101                 default:
4102                         break;
4103                 }
4104
4105                 if (j < i)
4106                         emulated_msrs[j] = emulated_msrs[i];
4107                 j++;
4108         }
4109         num_emulated_msrs = j;
4110 }
4111
4112 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4113                            const void *v)
4114 {
4115         int handled = 0;
4116         int n;
4117
4118         do {
4119                 n = min(len, 8);
4120                 if (!(lapic_in_kernel(vcpu) &&
4121                       !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4122                     && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4123                         break;
4124                 handled += n;
4125                 addr += n;
4126                 len -= n;
4127                 v += n;
4128         } while (len);
4129
4130         return handled;
4131 }
4132
4133 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4134 {
4135         int handled = 0;
4136         int n;
4137
4138         do {
4139                 n = min(len, 8);
4140                 if (!(lapic_in_kernel(vcpu) &&
4141                       !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4142                                          addr, n, v))
4143                     && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4144                         break;
4145                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4146                 handled += n;
4147                 addr += n;
4148                 len -= n;
4149                 v += n;
4150         } while (len);
4151
4152         return handled;
4153 }
4154
4155 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4156                         struct kvm_segment *var, int seg)
4157 {
4158         kvm_x86_ops->set_segment(vcpu, var, seg);
4159 }
4160
4161 void kvm_get_segment(struct kvm_vcpu *vcpu,
4162                      struct kvm_segment *var, int seg)
4163 {
4164         kvm_x86_ops->get_segment(vcpu, var, seg);
4165 }
4166
4167 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4168                            struct x86_exception *exception)
4169 {
4170         gpa_t t_gpa;
4171
4172         BUG_ON(!mmu_is_nested(vcpu));
4173
4174         /* NPT walks are always user-walks */
4175         access |= PFERR_USER_MASK;
4176         t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4177
4178         return t_gpa;
4179 }
4180
4181 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4182                               struct x86_exception *exception)
4183 {
4184         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4185         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4186 }
4187
4188  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4189                                 struct x86_exception *exception)
4190 {
4191         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4192         access |= PFERR_FETCH_MASK;
4193         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4194 }
4195
4196 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4197                                struct x86_exception *exception)
4198 {
4199         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4200         access |= PFERR_WRITE_MASK;
4201         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4202 }
4203
4204 /* uses this to access any guest's mapped memory without checking CPL */
4205 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4206                                 struct x86_exception *exception)
4207 {
4208         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4209 }
4210
4211 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4212                                       struct kvm_vcpu *vcpu, u32 access,
4213                                       struct x86_exception *exception)
4214 {
4215         void *data = val;
4216         int r = X86EMUL_CONTINUE;
4217
4218         while (bytes) {
4219                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4220                                                             exception);
4221                 unsigned offset = addr & (PAGE_SIZE-1);
4222                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4223                 int ret;
4224
4225                 if (gpa == UNMAPPED_GVA)
4226                         return X86EMUL_PROPAGATE_FAULT;
4227                 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4228                                                offset, toread);
4229                 if (ret < 0) {
4230                         r = X86EMUL_IO_NEEDED;
4231                         goto out;
4232                 }
4233
4234                 bytes -= toread;
4235                 data += toread;
4236                 addr += toread;
4237         }
4238 out:
4239         return r;
4240 }
4241
4242 /* used for instruction fetching */
4243 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4244                                 gva_t addr, void *val, unsigned int bytes,
4245                                 struct x86_exception *exception)
4246 {
4247         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4248         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4249         unsigned offset;
4250         int ret;
4251
4252         /* Inline kvm_read_guest_virt_helper for speed.  */
4253         gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4254                                                     exception);
4255         if (unlikely(gpa == UNMAPPED_GVA))
4256                 return X86EMUL_PROPAGATE_FAULT;
4257
4258         offset = addr & (PAGE_SIZE-1);
4259         if (WARN_ON(offset + bytes > PAGE_SIZE))
4260                 bytes = (unsigned)PAGE_SIZE - offset;
4261         ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4262                                        offset, bytes);
4263         if (unlikely(ret < 0))
4264                 return X86EMUL_IO_NEEDED;
4265
4266         return X86EMUL_CONTINUE;
4267 }
4268
4269 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4270                                gva_t addr, void *val, unsigned int bytes,
4271                                struct x86_exception *exception)
4272 {
4273         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4274         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4275
4276         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4277                                           exception);
4278 }
4279 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4280
4281 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4282                                       gva_t addr, void *val, unsigned int bytes,
4283                                       struct x86_exception *exception)
4284 {
4285         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4286         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4287 }
4288
4289 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4290                 unsigned long addr, void *val, unsigned int bytes)
4291 {
4292         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4293         int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4294
4295         return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4296 }
4297
4298 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4299                                        gva_t addr, void *val,
4300                                        unsigned int bytes,
4301                                        struct x86_exception *exception)
4302 {
4303         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4304         void *data = val;
4305         int r = X86EMUL_CONTINUE;
4306
4307         while (bytes) {
4308                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4309                                                              PFERR_WRITE_MASK,
4310                                                              exception);
4311                 unsigned offset = addr & (PAGE_SIZE-1);
4312                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4313                 int ret;
4314
4315                 if (gpa == UNMAPPED_GVA)
4316                         return X86EMUL_PROPAGATE_FAULT;
4317                 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4318                 if (ret < 0) {
4319                         r = X86EMUL_IO_NEEDED;
4320                         goto out;
4321                 }
4322
4323                 bytes -= towrite;
4324                 data += towrite;
4325                 addr += towrite;
4326         }
4327 out:
4328         return r;
4329 }
4330 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4331
4332 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4333                                 gpa_t *gpa, struct x86_exception *exception,
4334                                 bool write)
4335 {
4336         u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4337                 | (write ? PFERR_WRITE_MASK : 0);
4338
4339         /*
4340          * currently PKRU is only applied to ept enabled guest so
4341          * there is no pkey in EPT page table for L1 guest or EPT
4342          * shadow page table for L2 guest.
4343          */
4344         if (vcpu_match_mmio_gva(vcpu, gva)
4345             && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4346                                  vcpu->arch.access, 0, access)) {
4347                 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4348                                         (gva & (PAGE_SIZE - 1));
4349                 trace_vcpu_match_mmio(gva, *gpa, write, false);
4350                 return 1;
4351         }
4352
4353         *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4354
4355         if (*gpa == UNMAPPED_GVA)
4356                 return -1;
4357
4358         /* For APIC access vmexit */
4359         if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4360                 return 1;
4361
4362         if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4363                 trace_vcpu_match_mmio(gva, *gpa, write, true);
4364                 return 1;
4365         }
4366
4367         return 0;
4368 }
4369
4370 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4371                         const void *val, int bytes)
4372 {
4373         int ret;
4374
4375         ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4376         if (ret < 0)
4377                 return 0;
4378         kvm_page_track_write(vcpu, gpa, val, bytes);
4379         return 1;
4380 }
4381
4382 struct read_write_emulator_ops {
4383         int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4384                                   int bytes);
4385         int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4386                                   void *val, int bytes);
4387         int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4388                                int bytes, void *val);
4389         int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4390                                     void *val, int bytes);
4391         bool write;
4392 };
4393
4394 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4395 {
4396         if (vcpu->mmio_read_completed) {
4397                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4398                                vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4399                 vcpu->mmio_read_completed = 0;
4400                 return 1;
4401         }
4402
4403         return 0;
4404 }
4405
4406 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4407                         void *val, int bytes)
4408 {
4409         return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4410 }
4411
4412 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4413                          void *val, int bytes)
4414 {
4415         return emulator_write_phys(vcpu, gpa, val, bytes);
4416 }
4417
4418 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4419 {
4420         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4421         return vcpu_mmio_write(vcpu, gpa, bytes, val);
4422 }
4423
4424 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4425                           void *val, int bytes)
4426 {
4427         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4428         return X86EMUL_IO_NEEDED;
4429 }
4430
4431 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4432                            void *val, int bytes)
4433 {
4434         struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4435
4436         memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4437         return X86EMUL_CONTINUE;
4438 }
4439
4440 static const struct read_write_emulator_ops read_emultor = {
4441         .read_write_prepare = read_prepare,
4442         .read_write_emulate = read_emulate,
4443         .read_write_mmio = vcpu_mmio_read,
4444         .read_write_exit_mmio = read_exit_mmio,
4445 };
4446
4447 static const struct read_write_emulator_ops write_emultor = {
4448         .read_write_emulate = write_emulate,
4449         .read_write_mmio = write_mmio,
4450         .read_write_exit_mmio = write_exit_mmio,
4451         .write = true,
4452 };
4453
4454 static int emulator_read_write_onepage(unsigned long addr, void *val,
4455                                        unsigned int bytes,
4456                                        struct x86_exception *exception,
4457                                        struct kvm_vcpu *vcpu,
4458                                        const struct read_write_emulator_ops *ops)
4459 {
4460         gpa_t gpa;
4461         int handled, ret;
4462         bool write = ops->write;
4463         struct kvm_mmio_fragment *frag;
4464
4465         ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4466
4467         if (ret < 0)
4468                 return X86EMUL_PROPAGATE_FAULT;
4469
4470         /* For APIC access vmexit */
4471         if (ret)
4472                 goto mmio;
4473
4474         if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4475                 return X86EMUL_CONTINUE;
4476
4477 mmio:
4478         /*
4479          * Is this MMIO handled locally?
4480          */
4481         handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4482         if (handled == bytes)
4483                 return X86EMUL_CONTINUE;
4484
4485         gpa += handled;
4486         bytes -= handled;
4487         val += handled;
4488
4489         WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4490         frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4491         frag->gpa = gpa;
4492         frag->data = val;
4493         frag->len = bytes;
4494         return X86EMUL_CONTINUE;
4495 }
4496
4497 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4498                         unsigned long addr,
4499                         void *val, unsigned int bytes,
4500                         struct x86_exception *exception,
4501                         const struct read_write_emulator_ops *ops)
4502 {
4503         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4504         gpa_t gpa;
4505         int rc;
4506
4507         if (ops->read_write_prepare &&
4508                   ops->read_write_prepare(vcpu, val, bytes))
4509                 return X86EMUL_CONTINUE;
4510
4511         vcpu->mmio_nr_fragments = 0;
4512
4513         /* Crossing a page boundary? */
4514         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4515                 int now;
4516
4517                 now = -addr & ~PAGE_MASK;
4518                 rc = emulator_read_write_onepage(addr, val, now, exception,
4519                                                  vcpu, ops);
4520
4521                 if (rc != X86EMUL_CONTINUE)
4522                         return rc;
4523                 addr += now;
4524                 if (ctxt->mode != X86EMUL_MODE_PROT64)
4525                         addr = (u32)addr;
4526                 val += now;
4527                 bytes -= now;
4528         }
4529
4530         rc = emulator_read_write_onepage(addr, val, bytes, exception,
4531                                          vcpu, ops);
4532         if (rc != X86EMUL_CONTINUE)
4533                 return rc;
4534
4535         if (!vcpu->mmio_nr_fragments)
4536                 return rc;
4537
4538         gpa = vcpu->mmio_fragments[0].gpa;
4539
4540         vcpu->mmio_needed = 1;
4541         vcpu->mmio_cur_fragment = 0;
4542
4543         vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4544         vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4545         vcpu->run->exit_reason = KVM_EXIT_MMIO;
4546         vcpu->run->mmio.phys_addr = gpa;
4547
4548         return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4549 }
4550
4551 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4552                                   unsigned long addr,
4553                                   void *val,
4554                                   unsigned int bytes,
4555                                   struct x86_exception *exception)
4556 {
4557         return emulator_read_write(ctxt, addr, val, bytes,
4558                                    exception, &read_emultor);
4559 }
4560
4561 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4562                             unsigned long addr,
4563                             const void *val,
4564                             unsigned int bytes,
4565                             struct x86_exception *exception)
4566 {
4567         return emulator_read_write(ctxt, addr, (void *)val, bytes,
4568                                    exception, &write_emultor);
4569 }
4570
4571 #define CMPXCHG_TYPE(t, ptr, old, new) \
4572         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4573
4574 #ifdef CONFIG_X86_64
4575 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4576 #else
4577 #  define CMPXCHG64(ptr, old, new) \
4578         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4579 #endif
4580
4581 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4582                                      unsigned long addr,
4583                                      const void *old,
4584                                      const void *new,
4585                                      unsigned int bytes,
4586                                      struct x86_exception *exception)
4587 {
4588         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4589         gpa_t gpa;
4590         struct page *page;
4591         char *kaddr;
4592         bool exchanged;
4593
4594         /* guests cmpxchg8b have to be emulated atomically */
4595         if (bytes > 8 || (bytes & (bytes - 1)))
4596                 goto emul_write;
4597
4598         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4599
4600         if (gpa == UNMAPPED_GVA ||
4601             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4602                 goto emul_write;
4603
4604         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4605                 goto emul_write;
4606
4607         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4608         if (is_error_page(page))
4609                 goto emul_write;
4610
4611         kaddr = kmap_atomic(page);
4612         kaddr += offset_in_page(gpa);
4613         switch (bytes) {
4614         case 1:
4615                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4616                 break;
4617         case 2:
4618                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4619                 break;
4620         case 4:
4621                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4622                 break;
4623         case 8:
4624                 exchanged = CMPXCHG64(kaddr, old, new);
4625                 break;
4626         default:
4627                 BUG();
4628         }
4629         kunmap_atomic(kaddr);
4630         kvm_release_page_dirty(page);
4631
4632         if (!exchanged)
4633                 return X86EMUL_CMPXCHG_FAILED;
4634
4635         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4636         kvm_page_track_write(vcpu, gpa, new, bytes);
4637
4638         return X86EMUL_CONTINUE;
4639
4640 emul_write:
4641         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4642
4643         return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4644 }
4645
4646 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4647 {
4648         /* TODO: String I/O for in kernel device */
4649         int r;
4650
4651         if (vcpu->arch.pio.in)
4652                 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4653                                     vcpu->arch.pio.size, pd);
4654         else
4655                 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4656                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
4657                                      pd);
4658         return r;
4659 }
4660
4661 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4662                                unsigned short port, void *val,
4663                                unsigned int count, bool in)
4664 {
4665         vcpu->arch.pio.port = port;
4666         vcpu->arch.pio.in = in;
4667         vcpu->arch.pio.count  = count;
4668         vcpu->arch.pio.size = size;
4669
4670         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4671                 vcpu->arch.pio.count = 0;
4672                 return 1;
4673         }
4674
4675         vcpu->run->exit_reason = KVM_EXIT_IO;
4676         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4677         vcpu->run->io.size = size;
4678         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4679         vcpu->run->io.count = count;
4680         vcpu->run->io.port = port;
4681
4682         return 0;
4683 }
4684
4685 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4686                                     int size, unsigned short port, void *val,
4687                                     unsigned int count)
4688 {
4689         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4690         int ret;
4691
4692         if (vcpu->arch.pio.count)
4693                 goto data_avail;
4694
4695         ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4696         if (ret) {
4697 data_avail:
4698                 memcpy(val, vcpu->arch.pio_data, size * count);
4699                 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4700                 vcpu->arch.pio.count = 0;
4701                 return 1;
4702         }
4703
4704         return 0;
4705 }
4706
4707 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4708                                      int size, unsigned short port,
4709                                      const void *val, unsigned int count)
4710 {
4711         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4712
4713         memcpy(vcpu->arch.pio_data, val, size * count);
4714         trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4715         return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4716 }
4717
4718 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4719 {
4720         return kvm_x86_ops->get_segment_base(vcpu, seg);
4721 }
4722
4723 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4724 {
4725         kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4726 }
4727
4728 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4729 {
4730         if (!need_emulate_wbinvd(vcpu))
4731                 return X86EMUL_CONTINUE;
4732
4733         if (kvm_x86_ops->has_wbinvd_exit()) {
4734                 int cpu = get_cpu();
4735
4736                 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4737                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4738                                 wbinvd_ipi, NULL, 1);
4739                 put_cpu();
4740                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4741         } else
4742                 wbinvd();
4743         return X86EMUL_CONTINUE;
4744 }
4745
4746 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4747 {
4748         kvm_x86_ops->skip_emulated_instruction(vcpu);
4749         return kvm_emulate_wbinvd_noskip(vcpu);
4750 }
4751 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4752
4753
4754
4755 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4756 {
4757         kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4758 }
4759
4760 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4761                            unsigned long *dest)
4762 {
4763         return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4764 }
4765
4766 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4767                            unsigned long value)
4768 {
4769
4770         return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4771 }
4772
4773 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4774 {
4775         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4776 }
4777
4778 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4779 {
4780         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4781         unsigned long value;
4782
4783         switch (cr) {
4784         case 0:
4785                 value = kvm_read_cr0(vcpu);
4786                 break;
4787         case 2:
4788                 value = vcpu->arch.cr2;
4789                 break;
4790         case 3:
4791                 value = kvm_read_cr3(vcpu);
4792                 break;
4793         case 4:
4794                 value = kvm_read_cr4(vcpu);
4795                 break;
4796         case 8:
4797                 value = kvm_get_cr8(vcpu);
4798                 break;
4799         default:
4800                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4801                 return 0;
4802         }
4803
4804         return value;
4805 }
4806
4807 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4808 {
4809         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4810         int res = 0;
4811
4812         switch (cr) {
4813         case 0:
4814                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4815                 break;
4816         case 2:
4817                 vcpu->arch.cr2 = val;
4818                 break;
4819         case 3:
4820                 res = kvm_set_cr3(vcpu, val);
4821                 break;
4822         case 4:
4823                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4824                 break;
4825         case 8:
4826                 res = kvm_set_cr8(vcpu, val);
4827                 break;
4828         default:
4829                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4830                 res = -1;
4831         }
4832
4833         return res;
4834 }
4835
4836 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4837 {
4838         return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4839 }
4840
4841 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4842 {
4843         kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4844 }
4845
4846 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4847 {
4848         kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4849 }
4850
4851 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4852 {
4853         kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4854 }
4855
4856 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4857 {
4858         kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4859 }
4860
4861 static unsigned long emulator_get_cached_segment_base(
4862         struct x86_emulate_ctxt *ctxt, int seg)
4863 {
4864         return get_segment_base(emul_to_vcpu(ctxt), seg);
4865 }
4866
4867 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4868                                  struct desc_struct *desc, u32 *base3,
4869                                  int seg)
4870 {
4871         struct kvm_segment var;
4872
4873         kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4874         *selector = var.selector;
4875
4876         if (var.unusable) {
4877                 memset(desc, 0, sizeof(*desc));
4878                 return false;
4879         }
4880
4881         if (var.g)
4882                 var.limit >>= 12;
4883         set_desc_limit(desc, var.limit);
4884         set_desc_base(desc, (unsigned long)var.base);
4885 #ifdef CONFIG_X86_64
4886         if (base3)
4887                 *base3 = var.base >> 32;
4888 #endif
4889         desc->type = var.type;
4890         desc->s = var.s;
4891         desc->dpl = var.dpl;
4892         desc->p = var.present;
4893         desc->avl = var.avl;
4894         desc->l = var.l;
4895         desc->d = var.db;
4896         desc->g = var.g;
4897
4898         return true;
4899 }
4900
4901 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4902                                  struct desc_struct *desc, u32 base3,
4903                                  int seg)
4904 {
4905         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4906         struct kvm_segment var;
4907
4908         var.selector = selector;
4909         var.base = get_desc_base(desc);
4910 #ifdef CONFIG_X86_64
4911         var.base |= ((u64)base3) << 32;
4912 #endif
4913         var.limit = get_desc_limit(desc);
4914         if (desc->g)
4915                 var.limit = (var.limit << 12) | 0xfff;
4916         var.type = desc->type;
4917         var.dpl = desc->dpl;
4918         var.db = desc->d;
4919         var.s = desc->s;
4920         var.l = desc->l;
4921         var.g = desc->g;
4922         var.avl = desc->avl;
4923         var.present = desc->p;
4924         var.unusable = !var.present;
4925         var.padding = 0;
4926
4927         kvm_set_segment(vcpu, &var, seg);
4928         return;
4929 }
4930
4931 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4932                             u32 msr_index, u64 *pdata)
4933 {
4934         struct msr_data msr;
4935         int r;
4936
4937         msr.index = msr_index;
4938         msr.host_initiated = false;
4939         r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4940         if (r)
4941                 return r;
4942
4943         *pdata = msr.data;
4944         return 0;
4945 }
4946
4947 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4948                             u32 msr_index, u64 data)
4949 {
4950         struct msr_data msr;
4951
4952         msr.data = data;
4953         msr.index = msr_index;
4954         msr.host_initiated = false;
4955         return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4956 }
4957
4958 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4959 {
4960         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4961
4962         return vcpu->arch.smbase;
4963 }
4964
4965 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4966 {
4967         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4968
4969         vcpu->arch.smbase = smbase;
4970 }
4971
4972 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4973                               u32 pmc)
4974 {
4975         return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4976 }
4977
4978 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4979                              u32 pmc, u64 *pdata)
4980 {
4981         return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4982 }
4983
4984 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4985 {
4986         emul_to_vcpu(ctxt)->arch.halt_request = 1;
4987 }
4988
4989 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4990 {
4991         preempt_disable();
4992         kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4993         /*
4994          * CR0.TS may reference the host fpu state, not the guest fpu state,
4995          * so it may be clear at this point.
4996          */
4997         clts();
4998 }
4999
5000 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5001 {
5002         preempt_enable();
5003 }
5004
5005 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5006                               struct x86_instruction_info *info,
5007                               enum x86_intercept_stage stage)
5008 {
5009         return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5010 }
5011
5012 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5013                                u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5014 {
5015         kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5016 }
5017
5018 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5019 {
5020         return kvm_register_read(emul_to_vcpu(ctxt), reg);
5021 }
5022
5023 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5024 {
5025         kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5026 }
5027
5028 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5029 {
5030         kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5031 }
5032
5033 static const struct x86_emulate_ops emulate_ops = {
5034         .read_gpr            = emulator_read_gpr,
5035         .write_gpr           = emulator_write_gpr,
5036         .read_std            = kvm_read_guest_virt_system,
5037         .write_std           = kvm_write_guest_virt_system,
5038         .read_phys           = kvm_read_guest_phys_system,
5039         .fetch               = kvm_fetch_guest_virt,
5040         .read_emulated       = emulator_read_emulated,
5041         .write_emulated      = emulator_write_emulated,
5042         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
5043         .invlpg              = emulator_invlpg,
5044         .pio_in_emulated     = emulator_pio_in_emulated,
5045         .pio_out_emulated    = emulator_pio_out_emulated,
5046         .get_segment         = emulator_get_segment,
5047         .set_segment         = emulator_set_segment,
5048         .get_cached_segment_base = emulator_get_cached_segment_base,
5049         .get_gdt             = emulator_get_gdt,
5050         .get_idt             = emulator_get_idt,
5051         .set_gdt             = emulator_set_gdt,
5052         .set_idt             = emulator_set_idt,
5053         .get_cr              = emulator_get_cr,
5054         .set_cr              = emulator_set_cr,
5055         .cpl                 = emulator_get_cpl,
5056         .get_dr              = emulator_get_dr,
5057         .set_dr              = emulator_set_dr,
5058         .get_smbase          = emulator_get_smbase,
5059         .set_smbase          = emulator_set_smbase,
5060         .set_msr             = emulator_set_msr,
5061         .get_msr             = emulator_get_msr,
5062         .check_pmc           = emulator_check_pmc,
5063         .read_pmc            = emulator_read_pmc,
5064         .halt                = emulator_halt,
5065         .wbinvd              = emulator_wbinvd,
5066         .fix_hypercall       = emulator_fix_hypercall,
5067         .get_fpu             = emulator_get_fpu,
5068         .put_fpu             = emulator_put_fpu,
5069         .intercept           = emulator_intercept,
5070         .get_cpuid           = emulator_get_cpuid,
5071         .set_nmi_mask        = emulator_set_nmi_mask,
5072 };
5073
5074 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5075 {
5076         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5077         /*
5078          * an sti; sti; sequence only disable interrupts for the first
5079          * instruction. So, if the last instruction, be it emulated or
5080          * not, left the system with the INT_STI flag enabled, it
5081          * means that the last instruction is an sti. We should not
5082          * leave the flag on in this case. The same goes for mov ss
5083          */
5084         if (int_shadow & mask)
5085                 mask = 0;
5086         if (unlikely(int_shadow || mask)) {
5087                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5088                 if (!mask)
5089                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5090         }
5091 }
5092
5093 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5094 {
5095         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5096         if (ctxt->exception.vector == PF_VECTOR)
5097                 return kvm_propagate_fault(vcpu, &ctxt->exception);
5098
5099         if (ctxt->exception.error_code_valid)
5100                 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5101                                       ctxt->exception.error_code);
5102         else
5103                 kvm_queue_exception(vcpu, ctxt->exception.vector);
5104         return false;
5105 }
5106
5107 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5108 {
5109         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5110         int cs_db, cs_l;
5111
5112         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5113
5114         ctxt->eflags = kvm_get_rflags(vcpu);
5115         ctxt->eip = kvm_rip_read(vcpu);
5116         ctxt->mode = (!is_protmode(vcpu))               ? X86EMUL_MODE_REAL :
5117                      (ctxt->eflags & X86_EFLAGS_VM)     ? X86EMUL_MODE_VM86 :
5118                      (cs_l && is_long_mode(vcpu))       ? X86EMUL_MODE_PROT64 :
5119                      cs_db                              ? X86EMUL_MODE_PROT32 :
5120                                                           X86EMUL_MODE_PROT16;
5121         BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5122         BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5123         BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5124         ctxt->emul_flags = vcpu->arch.hflags;
5125
5126         init_decode_cache(ctxt);
5127         vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5128 }
5129
5130 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5131 {
5132         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5133         int ret;
5134
5135         init_emulate_ctxt(vcpu);
5136
5137         ctxt->op_bytes = 2;
5138         ctxt->ad_bytes = 2;
5139         ctxt->_eip = ctxt->eip + inc_eip;
5140         ret = emulate_int_real(ctxt, irq);
5141
5142         if (ret != X86EMUL_CONTINUE)
5143                 return EMULATE_FAIL;
5144
5145         ctxt->eip = ctxt->_eip;
5146         kvm_rip_write(vcpu, ctxt->eip);
5147         kvm_set_rflags(vcpu, ctxt->eflags);
5148
5149         if (irq == NMI_VECTOR)
5150                 vcpu->arch.nmi_pending = 0;
5151         else
5152                 vcpu->arch.interrupt.pending = false;
5153
5154         return EMULATE_DONE;
5155 }
5156 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5157
5158 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5159 {
5160         int r = EMULATE_DONE;
5161
5162         ++vcpu->stat.insn_emulation_fail;
5163         trace_kvm_emulate_insn_failed(vcpu);
5164         if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5165                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5166                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5167                 vcpu->run->internal.ndata = 0;
5168                 r = EMULATE_FAIL;
5169         }
5170         kvm_queue_exception(vcpu, UD_VECTOR);
5171
5172         return r;
5173 }
5174
5175 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5176                                   bool write_fault_to_shadow_pgtable,
5177                                   int emulation_type)
5178 {
5179         gpa_t gpa = cr2;
5180         kvm_pfn_t pfn;
5181
5182         if (emulation_type & EMULTYPE_NO_REEXECUTE)
5183                 return false;
5184
5185         if (!vcpu->arch.mmu.direct_map) {
5186                 /*
5187                  * Write permission should be allowed since only
5188                  * write access need to be emulated.
5189                  */
5190                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5191
5192                 /*
5193                  * If the mapping is invalid in guest, let cpu retry
5194                  * it to generate fault.
5195                  */
5196                 if (gpa == UNMAPPED_GVA)
5197                         return true;
5198         }
5199
5200         /*
5201          * Do not retry the unhandleable instruction if it faults on the
5202          * readonly host memory, otherwise it will goto a infinite loop:
5203          * retry instruction -> write #PF -> emulation fail -> retry
5204          * instruction -> ...
5205          */
5206         pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5207
5208         /*
5209          * If the instruction failed on the error pfn, it can not be fixed,
5210          * report the error to userspace.
5211          */
5212         if (is_error_noslot_pfn(pfn))
5213                 return false;
5214
5215         kvm_release_pfn_clean(pfn);
5216
5217         /* The instructions are well-emulated on direct mmu. */
5218         if (vcpu->arch.mmu.direct_map) {
5219                 unsigned int indirect_shadow_pages;
5220
5221                 spin_lock(&vcpu->kvm->mmu_lock);
5222                 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5223                 spin_unlock(&vcpu->kvm->mmu_lock);
5224
5225                 if (indirect_shadow_pages)
5226                         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5227
5228                 return true;
5229         }
5230
5231         /*
5232          * if emulation was due to access to shadowed page table
5233          * and it failed try to unshadow page and re-enter the
5234          * guest to let CPU execute the instruction.
5235          */
5236         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5237
5238         /*
5239          * If the access faults on its page table, it can not
5240          * be fixed by unprotecting shadow page and it should
5241          * be reported to userspace.
5242          */
5243         return !write_fault_to_shadow_pgtable;
5244 }
5245
5246 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5247                               unsigned long cr2,  int emulation_type)
5248 {
5249         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5250         unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5251
5252         last_retry_eip = vcpu->arch.last_retry_eip;
5253         last_retry_addr = vcpu->arch.last_retry_addr;
5254
5255         /*
5256          * If the emulation is caused by #PF and it is non-page_table
5257          * writing instruction, it means the VM-EXIT is caused by shadow
5258          * page protected, we can zap the shadow page and retry this
5259          * instruction directly.
5260          *
5261          * Note: if the guest uses a non-page-table modifying instruction
5262          * on the PDE that points to the instruction, then we will unmap
5263          * the instruction and go to an infinite loop. So, we cache the
5264          * last retried eip and the last fault address, if we meet the eip
5265          * and the address again, we can break out of the potential infinite
5266          * loop.
5267          */
5268         vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5269
5270         if (!(emulation_type & EMULTYPE_RETRY))
5271                 return false;
5272
5273         if (x86_page_table_writing_insn(ctxt))
5274                 return false;
5275
5276         if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5277                 return false;
5278
5279         vcpu->arch.last_retry_eip = ctxt->eip;
5280         vcpu->arch.last_retry_addr = cr2;
5281
5282         if (!vcpu->arch.mmu.direct_map)
5283                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5284
5285         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5286
5287         return true;
5288 }
5289
5290 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5291 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5292
5293 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5294 {
5295         if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5296                 /* This is a good place to trace that we are exiting SMM.  */
5297                 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5298
5299                 if (unlikely(vcpu->arch.smi_pending)) {
5300                         kvm_make_request(KVM_REQ_SMI, vcpu);
5301                         vcpu->arch.smi_pending = 0;
5302                 } else {
5303                         /* Process a latched INIT, if any.  */
5304                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5305                 }
5306         }
5307
5308         kvm_mmu_reset_context(vcpu);
5309 }
5310
5311 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5312 {
5313         unsigned changed = vcpu->arch.hflags ^ emul_flags;
5314
5315         vcpu->arch.hflags = emul_flags;
5316
5317         if (changed & HF_SMM_MASK)
5318                 kvm_smm_changed(vcpu);
5319 }
5320
5321 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5322                                 unsigned long *db)
5323 {
5324         u32 dr6 = 0;
5325         int i;
5326         u32 enable, rwlen;
5327
5328         enable = dr7;
5329         rwlen = dr7 >> 16;
5330         for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5331                 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5332                         dr6 |= (1 << i);
5333         return dr6;
5334 }
5335
5336 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5337 {
5338         struct kvm_run *kvm_run = vcpu->run;
5339
5340         /*
5341          * rflags is the old, "raw" value of the flags.  The new value has
5342          * not been saved yet.
5343          *
5344          * This is correct even for TF set by the guest, because "the
5345          * processor will not generate this exception after the instruction
5346          * that sets the TF flag".
5347          */
5348         if (unlikely(rflags & X86_EFLAGS_TF)) {
5349                 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5350                         kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5351                                                   DR6_RTM;
5352                         kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5353                         kvm_run->debug.arch.exception = DB_VECTOR;
5354                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
5355                         *r = EMULATE_USER_EXIT;
5356                 } else {
5357                         vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5358                         /*
5359                          * "Certain debug exceptions may clear bit 0-3.  The
5360                          * remaining contents of the DR6 register are never
5361                          * cleared by the processor".
5362                          */
5363                         vcpu->arch.dr6 &= ~15;
5364                         vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5365                         kvm_queue_exception(vcpu, DB_VECTOR);
5366                 }
5367         }
5368 }
5369
5370 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5371 {
5372         if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5373             (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5374                 struct kvm_run *kvm_run = vcpu->run;
5375                 unsigned long eip = kvm_get_linear_rip(vcpu);
5376                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5377                                            vcpu->arch.guest_debug_dr7,
5378                                            vcpu->arch.eff_db);
5379
5380                 if (dr6 != 0) {
5381                         kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5382                         kvm_run->debug.arch.pc = eip;
5383                         kvm_run->debug.arch.exception = DB_VECTOR;
5384                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
5385                         *r = EMULATE_USER_EXIT;
5386                         return true;
5387                 }
5388         }
5389
5390         if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5391             !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5392                 unsigned long eip = kvm_get_linear_rip(vcpu);
5393                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5394                                            vcpu->arch.dr7,
5395                                            vcpu->arch.db);
5396
5397                 if (dr6 != 0) {
5398                         vcpu->arch.dr6 &= ~15;
5399                         vcpu->arch.dr6 |= dr6 | DR6_RTM;
5400                         kvm_queue_exception(vcpu, DB_VECTOR);
5401                         *r = EMULATE_DONE;
5402                         return true;
5403                 }
5404         }
5405
5406         return false;
5407 }
5408
5409 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5410                             unsigned long cr2,
5411                             int emulation_type,
5412                             void *insn,
5413                             int insn_len)
5414 {
5415         int r;
5416         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5417         bool writeback = true;
5418         bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5419
5420         /*
5421          * Clear write_fault_to_shadow_pgtable here to ensure it is
5422          * never reused.
5423          */
5424         vcpu->arch.write_fault_to_shadow_pgtable = false;
5425         kvm_clear_exception_queue(vcpu);
5426
5427         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5428                 init_emulate_ctxt(vcpu);
5429
5430                 /*
5431                  * We will reenter on the same instruction since
5432                  * we do not set complete_userspace_io.  This does not
5433                  * handle watchpoints yet, those would be handled in
5434                  * the emulate_ops.
5435                  */
5436                 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5437                         return r;
5438
5439                 ctxt->interruptibility = 0;
5440                 ctxt->have_exception = false;
5441                 ctxt->exception.vector = -1;
5442                 ctxt->perm_ok = false;
5443
5444                 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5445
5446                 r = x86_decode_insn(ctxt, insn, insn_len);
5447
5448                 trace_kvm_emulate_insn_start(vcpu);
5449                 ++vcpu->stat.insn_emulation;
5450                 if (r != EMULATION_OK)  {
5451                         if (emulation_type & EMULTYPE_TRAP_UD)
5452                                 return EMULATE_FAIL;
5453                         if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5454                                                 emulation_type))
5455                                 return EMULATE_DONE;
5456                         if (emulation_type & EMULTYPE_SKIP)
5457                                 return EMULATE_FAIL;
5458                         return handle_emulation_failure(vcpu);
5459                 }
5460         }
5461
5462         if (emulation_type & EMULTYPE_SKIP) {
5463                 kvm_rip_write(vcpu, ctxt->_eip);
5464                 if (ctxt->eflags & X86_EFLAGS_RF)
5465                         kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5466                 return EMULATE_DONE;
5467         }
5468
5469         if (retry_instruction(ctxt, cr2, emulation_type))
5470                 return EMULATE_DONE;
5471
5472         /* this is needed for vmware backdoor interface to work since it
5473            changes registers values  during IO operation */
5474         if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5475                 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5476                 emulator_invalidate_register_cache(ctxt);
5477         }
5478
5479 restart:
5480         r = x86_emulate_insn(ctxt);
5481
5482         if (r == EMULATION_INTERCEPTED)
5483                 return EMULATE_DONE;
5484
5485         if (r == EMULATION_FAILED) {
5486                 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5487                                         emulation_type))
5488                         return EMULATE_DONE;
5489
5490                 return handle_emulation_failure(vcpu);
5491         }
5492
5493         if (ctxt->have_exception) {
5494                 r = EMULATE_DONE;
5495                 if (inject_emulated_exception(vcpu))
5496                         return r;
5497         } else if (vcpu->arch.pio.count) {
5498                 if (!vcpu->arch.pio.in) {
5499                         /* FIXME: return into emulator if single-stepping.  */
5500                         vcpu->arch.pio.count = 0;
5501                 } else {
5502                         writeback = false;
5503                         vcpu->arch.complete_userspace_io = complete_emulated_pio;
5504                 }
5505                 r = EMULATE_USER_EXIT;
5506         } else if (vcpu->mmio_needed) {
5507                 if (!vcpu->mmio_is_write)
5508                         writeback = false;
5509                 r = EMULATE_USER_EXIT;
5510                 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5511         } else if (r == EMULATION_RESTART)
5512                 goto restart;
5513         else
5514                 r = EMULATE_DONE;
5515
5516         if (writeback) {
5517                 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5518                 toggle_interruptibility(vcpu, ctxt->interruptibility);
5519                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5520                 if (vcpu->arch.hflags != ctxt->emul_flags)
5521                         kvm_set_hflags(vcpu, ctxt->emul_flags);
5522                 kvm_rip_write(vcpu, ctxt->eip);
5523                 if (r == EMULATE_DONE)
5524                         kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5525                 if (!ctxt->have_exception ||
5526                     exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5527                         __kvm_set_rflags(vcpu, ctxt->eflags);
5528
5529                 /*
5530                  * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5531                  * do nothing, and it will be requested again as soon as
5532                  * the shadow expires.  But we still need to check here,
5533                  * because POPF has no interrupt shadow.
5534                  */
5535                 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5536                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5537         } else
5538                 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5539
5540         return r;
5541 }
5542 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5543
5544 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5545 {
5546         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5547         int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5548                                             size, port, &val, 1);
5549         /* do not return to emulator after return from userspace */
5550         vcpu->arch.pio.count = 0;
5551         return ret;
5552 }
5553 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5554
5555 static void tsc_bad(void *info)
5556 {
5557         __this_cpu_write(cpu_tsc_khz, 0);
5558 }
5559
5560 static void tsc_khz_changed(void *data)
5561 {
5562         struct cpufreq_freqs *freq = data;
5563         unsigned long khz = 0;
5564
5565         if (data)
5566                 khz = freq->new;
5567         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5568                 khz = cpufreq_quick_get(raw_smp_processor_id());
5569         if (!khz)
5570                 khz = tsc_khz;
5571         __this_cpu_write(cpu_tsc_khz, khz);
5572 }
5573
5574 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5575                                      void *data)
5576 {
5577         struct cpufreq_freqs *freq = data;
5578         struct kvm *kvm;
5579         struct kvm_vcpu *vcpu;
5580         int i, send_ipi = 0;
5581
5582         /*
5583          * We allow guests to temporarily run on slowing clocks,
5584          * provided we notify them after, or to run on accelerating
5585          * clocks, provided we notify them before.  Thus time never
5586          * goes backwards.
5587          *
5588          * However, we have a problem.  We can't atomically update
5589          * the frequency of a given CPU from this function; it is
5590          * merely a notifier, which can be called from any CPU.
5591          * Changing the TSC frequency at arbitrary points in time
5592          * requires a recomputation of local variables related to
5593          * the TSC for each VCPU.  We must flag these local variables
5594          * to be updated and be sure the update takes place with the
5595          * new frequency before any guests proceed.
5596          *
5597          * Unfortunately, the combination of hotplug CPU and frequency
5598          * change creates an intractable locking scenario; the order
5599          * of when these callouts happen is undefined with respect to
5600          * CPU hotplug, and they can race with each other.  As such,
5601          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5602          * undefined; you can actually have a CPU frequency change take
5603          * place in between the computation of X and the setting of the
5604          * variable.  To protect against this problem, all updates of
5605          * the per_cpu tsc_khz variable are done in an interrupt
5606          * protected IPI, and all callers wishing to update the value
5607          * must wait for a synchronous IPI to complete (which is trivial
5608          * if the caller is on the CPU already).  This establishes the
5609          * necessary total order on variable updates.
5610          *
5611          * Note that because a guest time update may take place
5612          * anytime after the setting of the VCPU's request bit, the
5613          * correct TSC value must be set before the request.  However,
5614          * to ensure the update actually makes it to any guest which
5615          * starts running in hardware virtualization between the set
5616          * and the acquisition of the spinlock, we must also ping the
5617          * CPU after setting the request bit.
5618          *
5619          */
5620
5621         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5622                 return 0;
5623         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5624                 return 0;
5625
5626         smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5627
5628         spin_lock(&kvm_lock);
5629         list_for_each_entry(kvm, &vm_list, vm_list) {
5630                 kvm_for_each_vcpu(i, vcpu, kvm) {
5631                         if (vcpu->cpu != freq->cpu)
5632                                 continue;
5633                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5634                         if (vcpu->cpu != smp_processor_id())
5635                                 send_ipi = 1;
5636                 }
5637         }
5638         spin_unlock(&kvm_lock);
5639
5640         if (freq->old < freq->new && send_ipi) {
5641                 /*
5642                  * We upscale the frequency.  Must make the guest
5643                  * doesn't see old kvmclock values while running with
5644                  * the new frequency, otherwise we risk the guest sees
5645                  * time go backwards.
5646                  *
5647                  * In case we update the frequency for another cpu
5648                  * (which might be in guest context) send an interrupt
5649                  * to kick the cpu out of guest context.  Next time
5650                  * guest context is entered kvmclock will be updated,
5651                  * so the guest will not see stale values.
5652                  */
5653                 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5654         }
5655         return 0;
5656 }
5657
5658 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5659         .notifier_call  = kvmclock_cpufreq_notifier
5660 };
5661
5662 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5663                                         unsigned long action, void *hcpu)
5664 {
5665         unsigned int cpu = (unsigned long)hcpu;
5666
5667         switch (action) {
5668                 case CPU_ONLINE:
5669                 case CPU_DOWN_FAILED:
5670                         smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5671                         break;
5672                 case CPU_DOWN_PREPARE:
5673                         smp_call_function_single(cpu, tsc_bad, NULL, 1);
5674                         break;
5675         }
5676         return NOTIFY_OK;
5677 }
5678
5679 static struct notifier_block kvmclock_cpu_notifier_block = {
5680         .notifier_call  = kvmclock_cpu_notifier,
5681         .priority = -INT_MAX
5682 };
5683
5684 static void kvm_timer_init(void)
5685 {
5686         int cpu;
5687
5688         max_tsc_khz = tsc_khz;
5689
5690         cpu_notifier_register_begin();
5691         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5692 #ifdef CONFIG_CPU_FREQ
5693                 struct cpufreq_policy policy;
5694                 memset(&policy, 0, sizeof(policy));
5695                 cpu = get_cpu();
5696                 cpufreq_get_policy(&policy, cpu);
5697                 if (policy.cpuinfo.max_freq)
5698                         max_tsc_khz = policy.cpuinfo.max_freq;
5699                 put_cpu();
5700 #endif
5701                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5702                                           CPUFREQ_TRANSITION_NOTIFIER);
5703         }
5704         pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5705         for_each_online_cpu(cpu)
5706                 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5707
5708         __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5709         cpu_notifier_register_done();
5710
5711 }
5712
5713 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5714
5715 int kvm_is_in_guest(void)
5716 {
5717         return __this_cpu_read(current_vcpu) != NULL;
5718 }
5719
5720 static int kvm_is_user_mode(void)
5721 {
5722         int user_mode = 3;
5723
5724         if (__this_cpu_read(current_vcpu))
5725                 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5726
5727         return user_mode != 0;
5728 }
5729
5730 static unsigned long kvm_get_guest_ip(void)
5731 {
5732         unsigned long ip = 0;
5733
5734         if (__this_cpu_read(current_vcpu))
5735                 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5736
5737         return ip;
5738 }
5739
5740 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5741         .is_in_guest            = kvm_is_in_guest,
5742         .is_user_mode           = kvm_is_user_mode,
5743         .get_guest_ip           = kvm_get_guest_ip,
5744 };
5745
5746 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5747 {
5748         __this_cpu_write(current_vcpu, vcpu);
5749 }
5750 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5751
5752 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5753 {
5754         __this_cpu_write(current_vcpu, NULL);
5755 }
5756 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5757
5758 static void kvm_set_mmio_spte_mask(void)
5759 {
5760         u64 mask;
5761         int maxphyaddr = boot_cpu_data.x86_phys_bits;
5762
5763         /*
5764          * Set the reserved bits and the present bit of an paging-structure
5765          * entry to generate page fault with PFER.RSV = 1.
5766          */
5767          /* Mask the reserved physical address bits. */
5768         mask = rsvd_bits(maxphyaddr, 51);
5769
5770         /* Bit 62 is always reserved for 32bit host. */
5771         mask |= 0x3ull << 62;
5772
5773         /* Set the present bit. */
5774         mask |= 1ull;
5775
5776 #ifdef CONFIG_X86_64
5777         /*
5778          * If reserved bit is not supported, clear the present bit to disable
5779          * mmio page fault.
5780          */
5781         if (maxphyaddr == 52)
5782                 mask &= ~1ull;
5783 #endif
5784
5785         kvm_mmu_set_mmio_spte_mask(mask);
5786 }
5787
5788 #ifdef CONFIG_X86_64
5789 static void pvclock_gtod_update_fn(struct work_struct *work)
5790 {
5791         struct kvm *kvm;
5792
5793         struct kvm_vcpu *vcpu;
5794         int i;
5795
5796         spin_lock(&kvm_lock);
5797         list_for_each_entry(kvm, &vm_list, vm_list)
5798                 kvm_for_each_vcpu(i, vcpu, kvm)
5799                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5800         atomic_set(&kvm_guest_has_master_clock, 0);
5801         spin_unlock(&kvm_lock);
5802 }
5803
5804 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5805
5806 /*
5807  * Notification about pvclock gtod data update.
5808  */
5809 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5810                                void *priv)
5811 {
5812         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5813         struct timekeeper *tk = priv;
5814
5815         update_pvclock_gtod(tk);
5816
5817         /* disable master clock if host does not trust, or does not
5818          * use, TSC clocksource
5819          */
5820         if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5821             atomic_read(&kvm_guest_has_master_clock) != 0)
5822                 queue_work(system_long_wq, &pvclock_gtod_work);
5823
5824         return 0;
5825 }
5826
5827 static struct notifier_block pvclock_gtod_notifier = {
5828         .notifier_call = pvclock_gtod_notify,
5829 };
5830 #endif
5831
5832 int kvm_arch_init(void *opaque)
5833 {
5834         int r;
5835         struct kvm_x86_ops *ops = opaque;
5836
5837         if (kvm_x86_ops) {
5838                 printk(KERN_ERR "kvm: already loaded the other module\n");
5839                 r = -EEXIST;
5840                 goto out;
5841         }
5842
5843         if (!ops->cpu_has_kvm_support()) {
5844                 printk(KERN_ERR "kvm: no hardware support\n");
5845                 r = -EOPNOTSUPP;
5846                 goto out;
5847         }
5848         if (ops->disabled_by_bios()) {
5849                 printk(KERN_ERR "kvm: disabled by bios\n");
5850                 r = -EOPNOTSUPP;
5851                 goto out;
5852         }
5853
5854         r = -ENOMEM;
5855         shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5856         if (!shared_msrs) {
5857                 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5858                 goto out;
5859         }
5860
5861         r = kvm_mmu_module_init();
5862         if (r)
5863                 goto out_free_percpu;
5864
5865         kvm_set_mmio_spte_mask();
5866
5867         kvm_x86_ops = ops;
5868
5869         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5870                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
5871
5872         kvm_timer_init();
5873
5874         perf_register_guest_info_callbacks(&kvm_guest_cbs);
5875
5876         if (boot_cpu_has(X86_FEATURE_XSAVE))
5877                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5878
5879         kvm_lapic_init();
5880 #ifdef CONFIG_X86_64
5881         pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5882 #endif
5883
5884         return 0;
5885
5886 out_free_percpu:
5887         free_percpu(shared_msrs);
5888 out:
5889         return r;
5890 }
5891
5892 void kvm_arch_exit(void)
5893 {
5894         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5895
5896         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5897                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5898                                             CPUFREQ_TRANSITION_NOTIFIER);
5899         unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5900 #ifdef CONFIG_X86_64
5901         pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5902 #endif
5903         kvm_x86_ops = NULL;
5904         kvm_mmu_module_exit();
5905         free_percpu(shared_msrs);
5906 }
5907
5908 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5909 {
5910         ++vcpu->stat.halt_exits;
5911         if (lapic_in_kernel(vcpu)) {
5912                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5913                 return 1;
5914         } else {
5915                 vcpu->run->exit_reason = KVM_EXIT_HLT;
5916                 return 0;
5917         }
5918 }
5919 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5920
5921 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5922 {
5923         kvm_x86_ops->skip_emulated_instruction(vcpu);
5924         return kvm_vcpu_halt(vcpu);
5925 }
5926 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5927
5928 /*
5929  * kvm_pv_kick_cpu_op:  Kick a vcpu.
5930  *
5931  * @apicid - apicid of vcpu to be kicked.
5932  */
5933 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5934 {
5935         struct kvm_lapic_irq lapic_irq;
5936
5937         lapic_irq.shorthand = 0;
5938         lapic_irq.dest_mode = 0;
5939         lapic_irq.dest_id = apicid;
5940         lapic_irq.msi_redir_hint = false;
5941
5942         lapic_irq.delivery_mode = APIC_DM_REMRD;
5943         kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5944 }
5945
5946 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
5947 {
5948         vcpu->arch.apicv_active = false;
5949         kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
5950 }
5951
5952 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5953 {
5954         unsigned long nr, a0, a1, a2, a3, ret;
5955         int op_64_bit, r = 1;
5956
5957         kvm_x86_ops->skip_emulated_instruction(vcpu);
5958
5959         if (kvm_hv_hypercall_enabled(vcpu->kvm))
5960                 return kvm_hv_hypercall(vcpu);
5961
5962         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5963         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5964         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5965         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5966         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5967
5968         trace_kvm_hypercall(nr, a0, a1, a2, a3);
5969
5970         op_64_bit = is_64_bit_mode(vcpu);
5971         if (!op_64_bit) {
5972                 nr &= 0xFFFFFFFF;
5973                 a0 &= 0xFFFFFFFF;
5974                 a1 &= 0xFFFFFFFF;
5975                 a2 &= 0xFFFFFFFF;
5976                 a3 &= 0xFFFFFFFF;
5977         }
5978
5979         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5980                 ret = -KVM_EPERM;
5981                 goto out;
5982         }
5983
5984         switch (nr) {
5985         case KVM_HC_VAPIC_POLL_IRQ:
5986                 ret = 0;
5987                 break;
5988         case KVM_HC_KICK_CPU:
5989                 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5990                 ret = 0;
5991                 break;
5992         default:
5993                 ret = -KVM_ENOSYS;
5994                 break;
5995         }
5996 out:
5997         if (!op_64_bit)
5998                 ret = (u32)ret;
5999         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6000         ++vcpu->stat.hypercalls;
6001         return r;
6002 }
6003 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6004
6005 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6006 {
6007         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6008         char instruction[3];
6009         unsigned long rip = kvm_rip_read(vcpu);
6010
6011         kvm_x86_ops->patch_hypercall(vcpu, instruction);
6012
6013         return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
6014 }
6015
6016 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6017 {
6018         return vcpu->run->request_interrupt_window &&
6019                 likely(!pic_in_kernel(vcpu->kvm));
6020 }
6021
6022 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6023 {
6024         struct kvm_run *kvm_run = vcpu->run;
6025
6026         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6027         kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6028         kvm_run->cr8 = kvm_get_cr8(vcpu);
6029         kvm_run->apic_base = kvm_get_apic_base(vcpu);
6030         kvm_run->ready_for_interrupt_injection =
6031                 pic_in_kernel(vcpu->kvm) ||
6032                 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6033 }
6034
6035 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6036 {
6037         int max_irr, tpr;
6038
6039         if (!kvm_x86_ops->update_cr8_intercept)
6040                 return;
6041
6042         if (!lapic_in_kernel(vcpu))
6043                 return;
6044
6045         if (vcpu->arch.apicv_active)
6046                 return;
6047
6048         if (!vcpu->arch.apic->vapic_addr)
6049                 max_irr = kvm_lapic_find_highest_irr(vcpu);
6050         else
6051                 max_irr = -1;
6052
6053         if (max_irr != -1)
6054                 max_irr >>= 4;
6055
6056         tpr = kvm_lapic_get_cr8(vcpu);
6057
6058         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6059 }
6060
6061 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6062 {
6063         int r;
6064
6065         /* try to reinject previous events if any */
6066         if (vcpu->arch.exception.pending) {
6067                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6068                                         vcpu->arch.exception.has_error_code,
6069                                         vcpu->arch.exception.error_code);
6070
6071                 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6072                         __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6073                                              X86_EFLAGS_RF);
6074
6075                 if (vcpu->arch.exception.nr == DB_VECTOR &&
6076                     (vcpu->arch.dr7 & DR7_GD)) {
6077                         vcpu->arch.dr7 &= ~DR7_GD;
6078                         kvm_update_dr7(vcpu);
6079                 }
6080
6081                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6082                                           vcpu->arch.exception.has_error_code,
6083                                           vcpu->arch.exception.error_code,
6084                                           vcpu->arch.exception.reinject);
6085                 return 0;
6086         }
6087
6088         if (vcpu->arch.nmi_injected) {
6089                 kvm_x86_ops->set_nmi(vcpu);
6090                 return 0;
6091         }
6092
6093         if (vcpu->arch.interrupt.pending) {
6094                 kvm_x86_ops->set_irq(vcpu);
6095                 return 0;
6096         }
6097
6098         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6099                 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6100                 if (r != 0)
6101                         return r;
6102         }
6103
6104         /* try to inject new event if pending */
6105         if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6106                 --vcpu->arch.nmi_pending;
6107                 vcpu->arch.nmi_injected = true;
6108                 kvm_x86_ops->set_nmi(vcpu);
6109         } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6110                 /*
6111                  * Because interrupts can be injected asynchronously, we are
6112                  * calling check_nested_events again here to avoid a race condition.
6113                  * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6114                  * proposal and current concerns.  Perhaps we should be setting
6115                  * KVM_REQ_EVENT only on certain events and not unconditionally?
6116                  */
6117                 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6118                         r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6119                         if (r != 0)
6120                                 return r;
6121                 }
6122                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6123                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6124                                             false);
6125                         kvm_x86_ops->set_irq(vcpu);
6126                 }
6127         }
6128         return 0;
6129 }
6130
6131 static void process_nmi(struct kvm_vcpu *vcpu)
6132 {
6133         unsigned limit = 2;
6134
6135         /*
6136          * x86 is limited to one NMI running, and one NMI pending after it.
6137          * If an NMI is already in progress, limit further NMIs to just one.
6138          * Otherwise, allow two (and we'll inject the first one immediately).
6139          */
6140         if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6141                 limit = 1;
6142
6143         vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6144         vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6145         kvm_make_request(KVM_REQ_EVENT, vcpu);
6146 }
6147
6148 #define put_smstate(type, buf, offset, val)                       \
6149         *(type *)((buf) + (offset) - 0x7e00) = val
6150
6151 static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
6152 {
6153         u32 flags = 0;
6154         flags |= seg->g       << 23;
6155         flags |= seg->db      << 22;
6156         flags |= seg->l       << 21;
6157         flags |= seg->avl     << 20;
6158         flags |= seg->present << 15;
6159         flags |= seg->dpl     << 13;
6160         flags |= seg->s       << 12;
6161         flags |= seg->type    << 8;
6162         return flags;
6163 }
6164
6165 static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6166 {
6167         struct kvm_segment seg;
6168         int offset;
6169
6170         kvm_get_segment(vcpu, &seg, n);
6171         put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6172
6173         if (n < 3)
6174                 offset = 0x7f84 + n * 12;
6175         else
6176                 offset = 0x7f2c + (n - 3) * 12;
6177
6178         put_smstate(u32, buf, offset + 8, seg.base);
6179         put_smstate(u32, buf, offset + 4, seg.limit);
6180         put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
6181 }
6182
6183 #ifdef CONFIG_X86_64
6184 static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6185 {
6186         struct kvm_segment seg;
6187         int offset;
6188         u16 flags;
6189
6190         kvm_get_segment(vcpu, &seg, n);
6191         offset = 0x7e00 + n * 16;
6192
6193         flags = process_smi_get_segment_flags(&seg) >> 8;
6194         put_smstate(u16, buf, offset, seg.selector);
6195         put_smstate(u16, buf, offset + 2, flags);
6196         put_smstate(u32, buf, offset + 4, seg.limit);
6197         put_smstate(u64, buf, offset + 8, seg.base);
6198 }
6199 #endif
6200
6201 static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6202 {
6203         struct desc_ptr dt;
6204         struct kvm_segment seg;
6205         unsigned long val;
6206         int i;
6207
6208         put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6209         put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6210         put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6211         put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6212
6213         for (i = 0; i < 8; i++)
6214                 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6215
6216         kvm_get_dr(vcpu, 6, &val);
6217         put_smstate(u32, buf, 0x7fcc, (u32)val);
6218         kvm_get_dr(vcpu, 7, &val);
6219         put_smstate(u32, buf, 0x7fc8, (u32)val);
6220
6221         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6222         put_smstate(u32, buf, 0x7fc4, seg.selector);
6223         put_smstate(u32, buf, 0x7f64, seg.base);
6224         put_smstate(u32, buf, 0x7f60, seg.limit);
6225         put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
6226
6227         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6228         put_smstate(u32, buf, 0x7fc0, seg.selector);
6229         put_smstate(u32, buf, 0x7f80, seg.base);
6230         put_smstate(u32, buf, 0x7f7c, seg.limit);
6231         put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
6232
6233         kvm_x86_ops->get_gdt(vcpu, &dt);
6234         put_smstate(u32, buf, 0x7f74, dt.address);
6235         put_smstate(u32, buf, 0x7f70, dt.size);
6236
6237         kvm_x86_ops->get_idt(vcpu, &dt);
6238         put_smstate(u32, buf, 0x7f58, dt.address);
6239         put_smstate(u32, buf, 0x7f54, dt.size);
6240
6241         for (i = 0; i < 6; i++)
6242                 process_smi_save_seg_32(vcpu, buf, i);
6243
6244         put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6245
6246         /* revision id */
6247         put_smstate(u32, buf, 0x7efc, 0x00020000);
6248         put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6249 }
6250
6251 static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6252 {
6253 #ifdef CONFIG_X86_64
6254         struct desc_ptr dt;
6255         struct kvm_segment seg;
6256         unsigned long val;
6257         int i;
6258
6259         for (i = 0; i < 16; i++)
6260                 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6261
6262         put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6263         put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6264
6265         kvm_get_dr(vcpu, 6, &val);
6266         put_smstate(u64, buf, 0x7f68, val);
6267         kvm_get_dr(vcpu, 7, &val);
6268         put_smstate(u64, buf, 0x7f60, val);
6269
6270         put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6271         put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6272         put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6273
6274         put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6275
6276         /* revision id */
6277         put_smstate(u32, buf, 0x7efc, 0x00020064);
6278
6279         put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6280
6281         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6282         put_smstate(u16, buf, 0x7e90, seg.selector);
6283         put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6284         put_smstate(u32, buf, 0x7e94, seg.limit);
6285         put_smstate(u64, buf, 0x7e98, seg.base);
6286
6287         kvm_x86_ops->get_idt(vcpu, &dt);
6288         put_smstate(u32, buf, 0x7e84, dt.size);
6289         put_smstate(u64, buf, 0x7e88, dt.address);
6290
6291         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6292         put_smstate(u16, buf, 0x7e70, seg.selector);
6293         put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6294         put_smstate(u32, buf, 0x7e74, seg.limit);
6295         put_smstate(u64, buf, 0x7e78, seg.base);
6296
6297         kvm_x86_ops->get_gdt(vcpu, &dt);
6298         put_smstate(u32, buf, 0x7e64, dt.size);
6299         put_smstate(u64, buf, 0x7e68, dt.address);
6300
6301         for (i = 0; i < 6; i++)
6302                 process_smi_save_seg_64(vcpu, buf, i);
6303 #else
6304         WARN_ON_ONCE(1);
6305 #endif
6306 }
6307
6308 static void process_smi(struct kvm_vcpu *vcpu)
6309 {
6310         struct kvm_segment cs, ds;
6311         struct desc_ptr dt;
6312         char buf[512];
6313         u32 cr0;
6314
6315         if (is_smm(vcpu)) {
6316                 vcpu->arch.smi_pending = true;
6317                 return;
6318         }
6319
6320         trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6321         vcpu->arch.hflags |= HF_SMM_MASK;
6322         memset(buf, 0, 512);
6323         if (guest_cpuid_has_longmode(vcpu))
6324                 process_smi_save_state_64(vcpu, buf);
6325         else
6326                 process_smi_save_state_32(vcpu, buf);
6327
6328         kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6329
6330         if (kvm_x86_ops->get_nmi_mask(vcpu))
6331                 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6332         else
6333                 kvm_x86_ops->set_nmi_mask(vcpu, true);
6334
6335         kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6336         kvm_rip_write(vcpu, 0x8000);
6337
6338         cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6339         kvm_x86_ops->set_cr0(vcpu, cr0);
6340         vcpu->arch.cr0 = cr0;
6341
6342         kvm_x86_ops->set_cr4(vcpu, 0);
6343
6344         /* Undocumented: IDT limit is set to zero on entry to SMM.  */
6345         dt.address = dt.size = 0;
6346         kvm_x86_ops->set_idt(vcpu, &dt);
6347
6348         __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6349
6350         cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6351         cs.base = vcpu->arch.smbase;
6352
6353         ds.selector = 0;
6354         ds.base = 0;
6355
6356         cs.limit    = ds.limit = 0xffffffff;
6357         cs.type     = ds.type = 0x3;
6358         cs.dpl      = ds.dpl = 0;
6359         cs.db       = ds.db = 0;
6360         cs.s        = ds.s = 1;
6361         cs.l        = ds.l = 0;
6362         cs.g        = ds.g = 1;
6363         cs.avl      = ds.avl = 0;
6364         cs.present  = ds.present = 1;
6365         cs.unusable = ds.unusable = 0;
6366         cs.padding  = ds.padding = 0;
6367
6368         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6369         kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6370         kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6371         kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6372         kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6373         kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6374
6375         if (guest_cpuid_has_longmode(vcpu))
6376                 kvm_x86_ops->set_efer(vcpu, 0);
6377
6378         kvm_update_cpuid(vcpu);
6379         kvm_mmu_reset_context(vcpu);
6380 }
6381
6382 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6383 {
6384         kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6385 }
6386
6387 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6388 {
6389         u64 eoi_exit_bitmap[4];
6390
6391         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6392                 return;
6393
6394         bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6395
6396         if (irqchip_split(vcpu->kvm))
6397                 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6398         else {
6399                 if (vcpu->arch.apicv_active)
6400                         kvm_x86_ops->sync_pir_to_irr(vcpu);
6401                 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6402         }
6403         bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6404                   vcpu_to_synic(vcpu)->vec_bitmap, 256);
6405         kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6406 }
6407
6408 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6409 {
6410         ++vcpu->stat.tlb_flush;
6411         kvm_x86_ops->tlb_flush(vcpu);
6412 }
6413
6414 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6415 {
6416         struct page *page = NULL;
6417
6418         if (!lapic_in_kernel(vcpu))
6419                 return;
6420
6421         if (!kvm_x86_ops->set_apic_access_page_addr)
6422                 return;
6423
6424         page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6425         if (is_error_page(page))
6426                 return;
6427         kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6428
6429         /*
6430          * Do not pin apic access page in memory, the MMU notifier
6431          * will call us again if it is migrated or swapped out.
6432          */
6433         put_page(page);
6434 }
6435 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6436
6437 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6438                                            unsigned long address)
6439 {
6440         /*
6441          * The physical address of apic access page is stored in the VMCS.
6442          * Update it when it becomes invalid.
6443          */
6444         if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6445                 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6446 }
6447
6448 /*
6449  * Returns 1 to let vcpu_run() continue the guest execution loop without
6450  * exiting to the userspace.  Otherwise, the value will be returned to the
6451  * userspace.
6452  */
6453 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6454 {
6455         int r;
6456         bool req_int_win =
6457                 dm_request_for_irq_injection(vcpu) &&
6458                 kvm_cpu_accept_dm_intr(vcpu);
6459
6460         bool req_immediate_exit = false;
6461
6462         if (vcpu->requests) {
6463                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6464                         kvm_mmu_unload(vcpu);
6465                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6466                         __kvm_migrate_timers(vcpu);
6467                 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6468                         kvm_gen_update_masterclock(vcpu->kvm);
6469                 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6470                         kvm_gen_kvmclock_update(vcpu);
6471                 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6472                         r = kvm_guest_time_update(vcpu);
6473                         if (unlikely(r))
6474                                 goto out;
6475                 }
6476                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6477                         kvm_mmu_sync_roots(vcpu);
6478                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6479                         kvm_vcpu_flush_tlb(vcpu);
6480                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6481                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6482                         r = 0;
6483                         goto out;
6484                 }
6485                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6486                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6487                         r = 0;
6488                         goto out;
6489                 }
6490                 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6491                         vcpu->fpu_active = 0;
6492                         kvm_x86_ops->fpu_deactivate(vcpu);
6493                 }
6494                 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6495                         /* Page is swapped out. Do synthetic halt */
6496                         vcpu->arch.apf.halted = true;
6497                         r = 1;
6498                         goto out;
6499                 }
6500                 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6501                         record_steal_time(vcpu);
6502                 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6503                         process_smi(vcpu);
6504                 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6505                         process_nmi(vcpu);
6506                 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6507                         kvm_pmu_handle_event(vcpu);
6508                 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6509                         kvm_pmu_deliver_pmi(vcpu);
6510                 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6511                         BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6512                         if (test_bit(vcpu->arch.pending_ioapic_eoi,
6513                                      vcpu->arch.ioapic_handled_vectors)) {
6514                                 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6515                                 vcpu->run->eoi.vector =
6516                                                 vcpu->arch.pending_ioapic_eoi;
6517                                 r = 0;
6518                                 goto out;
6519                         }
6520                 }
6521                 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6522                         vcpu_scan_ioapic(vcpu);
6523                 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6524                         kvm_vcpu_reload_apic_access_page(vcpu);
6525                 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6526                         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6527                         vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6528                         r = 0;
6529                         goto out;
6530                 }
6531                 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6532                         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6533                         vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6534                         r = 0;
6535                         goto out;
6536                 }
6537                 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6538                         vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6539                         vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6540                         r = 0;
6541                         goto out;
6542                 }
6543
6544                 /*
6545                  * KVM_REQ_HV_STIMER has to be processed after
6546                  * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6547                  * depend on the guest clock being up-to-date
6548                  */
6549                 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6550                         kvm_hv_process_stimers(vcpu);
6551         }
6552
6553         /*
6554          * KVM_REQ_EVENT is not set when posted interrupts are set by
6555          * VT-d hardware, so we have to update RVI unconditionally.
6556          */
6557         if (kvm_lapic_enabled(vcpu)) {
6558                 /*
6559                  * Update architecture specific hints for APIC
6560                  * virtual interrupt delivery.
6561                  */
6562                 if (vcpu->arch.apicv_active)
6563                         kvm_x86_ops->hwapic_irr_update(vcpu,
6564                                 kvm_lapic_find_highest_irr(vcpu));
6565         }
6566
6567         if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6568                 kvm_apic_accept_events(vcpu);
6569                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6570                         r = 1;
6571                         goto out;
6572                 }
6573
6574                 if (inject_pending_event(vcpu, req_int_win) != 0)
6575                         req_immediate_exit = true;
6576                 /* enable NMI/IRQ window open exits if needed */
6577                 else {
6578                         if (vcpu->arch.nmi_pending)
6579                                 kvm_x86_ops->enable_nmi_window(vcpu);
6580                         if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6581                                 kvm_x86_ops->enable_irq_window(vcpu);
6582                 }
6583
6584                 if (kvm_lapic_enabled(vcpu)) {
6585                         update_cr8_intercept(vcpu);
6586                         kvm_lapic_sync_to_vapic(vcpu);
6587                 }
6588         }
6589
6590         r = kvm_mmu_reload(vcpu);
6591         if (unlikely(r)) {
6592                 goto cancel_injection;
6593         }
6594
6595         preempt_disable();
6596
6597         kvm_x86_ops->prepare_guest_switch(vcpu);
6598         if (vcpu->fpu_active)
6599                 kvm_load_guest_fpu(vcpu);
6600         vcpu->mode = IN_GUEST_MODE;
6601
6602         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6603
6604         /*
6605          * We should set ->mode before check ->requests,
6606          * Please see the comment in kvm_make_all_cpus_request.
6607          * This also orders the write to mode from any reads
6608          * to the page tables done while the VCPU is running.
6609          * Please see the comment in kvm_flush_remote_tlbs.
6610          */
6611         smp_mb__after_srcu_read_unlock();
6612
6613         local_irq_disable();
6614
6615         if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6616             || need_resched() || signal_pending(current)) {
6617                 vcpu->mode = OUTSIDE_GUEST_MODE;
6618                 smp_wmb();
6619                 local_irq_enable();
6620                 preempt_enable();
6621                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6622                 r = 1;
6623                 goto cancel_injection;
6624         }
6625
6626         kvm_load_guest_xcr0(vcpu);
6627
6628         if (req_immediate_exit)
6629                 smp_send_reschedule(vcpu->cpu);
6630
6631         trace_kvm_entry(vcpu->vcpu_id);
6632         wait_lapic_expire(vcpu);
6633         __kvm_guest_enter();
6634
6635         if (unlikely(vcpu->arch.switch_db_regs)) {
6636                 set_debugreg(0, 7);
6637                 set_debugreg(vcpu->arch.eff_db[0], 0);
6638                 set_debugreg(vcpu->arch.eff_db[1], 1);
6639                 set_debugreg(vcpu->arch.eff_db[2], 2);
6640                 set_debugreg(vcpu->arch.eff_db[3], 3);
6641                 set_debugreg(vcpu->arch.dr6, 6);
6642                 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6643         }
6644
6645         kvm_x86_ops->run(vcpu);
6646
6647         /*
6648          * Do this here before restoring debug registers on the host.  And
6649          * since we do this before handling the vmexit, a DR access vmexit
6650          * can (a) read the correct value of the debug registers, (b) set
6651          * KVM_DEBUGREG_WONT_EXIT again.
6652          */
6653         if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6654                 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6655                 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6656                 kvm_update_dr0123(vcpu);
6657                 kvm_update_dr6(vcpu);
6658                 kvm_update_dr7(vcpu);
6659                 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6660         }
6661
6662         /*
6663          * If the guest has used debug registers, at least dr7
6664          * will be disabled while returning to the host.
6665          * If we don't have active breakpoints in the host, we don't
6666          * care about the messed up debug address registers. But if
6667          * we have some of them active, restore the old state.
6668          */
6669         if (hw_breakpoint_active())
6670                 hw_breakpoint_restore();
6671
6672         vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6673
6674         vcpu->mode = OUTSIDE_GUEST_MODE;
6675         smp_wmb();
6676
6677         kvm_put_guest_xcr0(vcpu);
6678
6679         /* Interrupt is enabled by handle_external_intr() */
6680         kvm_x86_ops->handle_external_intr(vcpu);
6681
6682         ++vcpu->stat.exits;
6683
6684         /*
6685          * We must have an instruction between local_irq_enable() and
6686          * kvm_guest_exit(), so the timer interrupt isn't delayed by
6687          * the interrupt shadow.  The stat.exits increment will do nicely.
6688          * But we need to prevent reordering, hence this barrier():
6689          */
6690         barrier();
6691
6692         kvm_guest_exit();
6693
6694         preempt_enable();
6695
6696         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6697
6698         /*
6699          * Profile KVM exit RIPs:
6700          */
6701         if (unlikely(prof_on == KVM_PROFILING)) {
6702                 unsigned long rip = kvm_rip_read(vcpu);
6703                 profile_hit(KVM_PROFILING, (void *)rip);
6704         }
6705
6706         if (unlikely(vcpu->arch.tsc_always_catchup))
6707                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6708
6709         if (vcpu->arch.apic_attention)
6710                 kvm_lapic_sync_from_vapic(vcpu);
6711
6712         r = kvm_x86_ops->handle_exit(vcpu);
6713         return r;
6714
6715 cancel_injection:
6716         kvm_x86_ops->cancel_injection(vcpu);
6717         if (unlikely(vcpu->arch.apic_attention))
6718                 kvm_lapic_sync_from_vapic(vcpu);
6719 out:
6720         return r;
6721 }
6722
6723 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6724 {
6725         if (!kvm_arch_vcpu_runnable(vcpu) &&
6726             (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6727                 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6728                 kvm_vcpu_block(vcpu);
6729                 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6730
6731                 if (kvm_x86_ops->post_block)
6732                         kvm_x86_ops->post_block(vcpu);
6733
6734                 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6735                         return 1;
6736         }
6737
6738         kvm_apic_accept_events(vcpu);
6739         switch(vcpu->arch.mp_state) {
6740         case KVM_MP_STATE_HALTED:
6741                 vcpu->arch.pv.pv_unhalted = false;
6742                 vcpu->arch.mp_state =
6743                         KVM_MP_STATE_RUNNABLE;
6744         case KVM_MP_STATE_RUNNABLE:
6745                 vcpu->arch.apf.halted = false;
6746                 break;
6747         case KVM_MP_STATE_INIT_RECEIVED:
6748                 break;
6749         default:
6750                 return -EINTR;
6751                 break;
6752         }
6753         return 1;
6754 }
6755
6756 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6757 {
6758         return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6759                 !vcpu->arch.apf.halted);
6760 }
6761
6762 static int vcpu_run(struct kvm_vcpu *vcpu)
6763 {
6764         int r;
6765         struct kvm *kvm = vcpu->kvm;
6766
6767         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6768
6769         for (;;) {
6770                 if (kvm_vcpu_running(vcpu)) {
6771                         r = vcpu_enter_guest(vcpu);
6772                 } else {
6773                         r = vcpu_block(kvm, vcpu);
6774                 }
6775
6776                 if (r <= 0)
6777                         break;
6778
6779                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6780                 if (kvm_cpu_has_pending_timer(vcpu))
6781                         kvm_inject_pending_timer_irqs(vcpu);
6782
6783                 if (dm_request_for_irq_injection(vcpu) &&
6784                         kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
6785                         r = 0;
6786                         vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
6787                         ++vcpu->stat.request_irq_exits;
6788                         break;
6789                 }
6790
6791                 kvm_check_async_pf_completion(vcpu);
6792
6793                 if (signal_pending(current)) {
6794                         r = -EINTR;
6795                         vcpu->run->exit_reason = KVM_EXIT_INTR;
6796                         ++vcpu->stat.signal_exits;
6797                         break;
6798                 }
6799                 if (need_resched()) {
6800                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6801                         cond_resched();
6802                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6803                 }
6804         }
6805
6806         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6807
6808         return r;
6809 }
6810
6811 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6812 {
6813         int r;
6814         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6815         r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6816         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6817         if (r != EMULATE_DONE)
6818                 return 0;
6819         return 1;
6820 }
6821
6822 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6823 {
6824         BUG_ON(!vcpu->arch.pio.count);
6825
6826         return complete_emulated_io(vcpu);
6827 }
6828
6829 /*
6830  * Implements the following, as a state machine:
6831  *
6832  * read:
6833  *   for each fragment
6834  *     for each mmio piece in the fragment
6835  *       write gpa, len
6836  *       exit
6837  *       copy data
6838  *   execute insn
6839  *
6840  * write:
6841  *   for each fragment
6842  *     for each mmio piece in the fragment
6843  *       write gpa, len
6844  *       copy data
6845  *       exit
6846  */
6847 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6848 {
6849         struct kvm_run *run = vcpu->run;
6850         struct kvm_mmio_fragment *frag;
6851         unsigned len;
6852
6853         BUG_ON(!vcpu->mmio_needed);
6854
6855         /* Complete previous fragment */
6856         frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6857         len = min(8u, frag->len);
6858         if (!vcpu->mmio_is_write)
6859                 memcpy(frag->data, run->mmio.data, len);
6860
6861         if (frag->len <= 8) {
6862                 /* Switch to the next fragment. */
6863                 frag++;
6864                 vcpu->mmio_cur_fragment++;
6865         } else {
6866                 /* Go forward to the next mmio piece. */
6867                 frag->data += len;
6868                 frag->gpa += len;
6869                 frag->len -= len;
6870         }
6871
6872         if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6873                 vcpu->mmio_needed = 0;
6874
6875                 /* FIXME: return into emulator if single-stepping.  */
6876                 if (vcpu->mmio_is_write)
6877                         return 1;
6878                 vcpu->mmio_read_completed = 1;
6879                 return complete_emulated_io(vcpu);
6880         }
6881
6882         run->exit_reason = KVM_EXIT_MMIO;
6883         run->mmio.phys_addr = frag->gpa;
6884         if (vcpu->mmio_is_write)
6885                 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6886         run->mmio.len = min(8u, frag->len);
6887         run->mmio.is_write = vcpu->mmio_is_write;
6888         vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6889         return 0;
6890 }
6891
6892
6893 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6894 {
6895         struct fpu *fpu = &current->thread.fpu;
6896         int r;
6897         sigset_t sigsaved;
6898
6899         fpu__activate_curr(fpu);
6900
6901         if (vcpu->sigset_active)
6902                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6903
6904         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6905                 kvm_vcpu_block(vcpu);
6906                 kvm_apic_accept_events(vcpu);
6907                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6908                 r = -EAGAIN;
6909                 goto out;
6910         }
6911
6912         /* re-sync apic's tpr */
6913         if (!lapic_in_kernel(vcpu)) {
6914                 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6915                         r = -EINVAL;
6916                         goto out;
6917                 }
6918         }
6919
6920         if (unlikely(vcpu->arch.complete_userspace_io)) {
6921                 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6922                 vcpu->arch.complete_userspace_io = NULL;
6923                 r = cui(vcpu);
6924                 if (r <= 0)
6925                         goto out;
6926         } else
6927                 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6928
6929         r = vcpu_run(vcpu);
6930
6931 out:
6932         post_kvm_run_save(vcpu);
6933         if (vcpu->sigset_active)
6934                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6935
6936         return r;
6937 }
6938
6939 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6940 {
6941         if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6942                 /*
6943                  * We are here if userspace calls get_regs() in the middle of
6944                  * instruction emulation. Registers state needs to be copied
6945                  * back from emulation context to vcpu. Userspace shouldn't do
6946                  * that usually, but some bad designed PV devices (vmware
6947                  * backdoor interface) need this to work
6948                  */
6949                 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6950                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6951         }
6952         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6953         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6954         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6955         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6956         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6957         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6958         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6959         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6960 #ifdef CONFIG_X86_64
6961         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6962         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6963         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6964         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6965         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6966         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6967         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6968         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6969 #endif
6970
6971         regs->rip = kvm_rip_read(vcpu);
6972         regs->rflags = kvm_get_rflags(vcpu);
6973
6974         return 0;
6975 }
6976
6977 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6978 {
6979         vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6980         vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6981
6982         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6983         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6984         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6985         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6986         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6987         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6988         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6989         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6990 #ifdef CONFIG_X86_64
6991         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6992         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6993         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6994         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6995         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6996         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6997         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6998         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6999 #endif
7000
7001         kvm_rip_write(vcpu, regs->rip);
7002         kvm_set_rflags(vcpu, regs->rflags);
7003
7004         vcpu->arch.exception.pending = false;
7005
7006         kvm_make_request(KVM_REQ_EVENT, vcpu);
7007
7008         return 0;
7009 }
7010
7011 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7012 {
7013         struct kvm_segment cs;
7014
7015         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7016         *db = cs.db;
7017         *l = cs.l;
7018 }
7019 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7020
7021 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7022                                   struct kvm_sregs *sregs)
7023 {
7024         struct desc_ptr dt;
7025
7026         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7027         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7028         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7029         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7030         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7031         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7032
7033         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7034         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7035
7036         kvm_x86_ops->get_idt(vcpu, &dt);
7037         sregs->idt.limit = dt.size;
7038         sregs->idt.base = dt.address;
7039         kvm_x86_ops->get_gdt(vcpu, &dt);
7040         sregs->gdt.limit = dt.size;
7041         sregs->gdt.base = dt.address;
7042
7043         sregs->cr0 = kvm_read_cr0(vcpu);
7044         sregs->cr2 = vcpu->arch.cr2;
7045         sregs->cr3 = kvm_read_cr3(vcpu);
7046         sregs->cr4 = kvm_read_cr4(vcpu);
7047         sregs->cr8 = kvm_get_cr8(vcpu);
7048         sregs->efer = vcpu->arch.efer;
7049         sregs->apic_base = kvm_get_apic_base(vcpu);
7050
7051         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7052
7053         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7054                 set_bit(vcpu->arch.interrupt.nr,
7055                         (unsigned long *)sregs->interrupt_bitmap);
7056
7057         return 0;
7058 }
7059
7060 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7061                                     struct kvm_mp_state *mp_state)
7062 {
7063         kvm_apic_accept_events(vcpu);
7064         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7065                                         vcpu->arch.pv.pv_unhalted)
7066                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7067         else
7068                 mp_state->mp_state = vcpu->arch.mp_state;
7069
7070         return 0;
7071 }
7072
7073 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7074                                     struct kvm_mp_state *mp_state)
7075 {
7076         if (!lapic_in_kernel(vcpu) &&
7077             mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7078                 return -EINVAL;
7079
7080         if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7081                 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7082                 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7083         } else
7084                 vcpu->arch.mp_state = mp_state->mp_state;
7085         kvm_make_request(KVM_REQ_EVENT, vcpu);
7086         return 0;
7087 }
7088
7089 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7090                     int reason, bool has_error_code, u32 error_code)
7091 {
7092         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7093         int ret;
7094
7095         init_emulate_ctxt(vcpu);
7096
7097         ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7098                                    has_error_code, error_code);
7099
7100         if (ret)
7101                 return EMULATE_FAIL;
7102
7103         kvm_rip_write(vcpu, ctxt->eip);
7104         kvm_set_rflags(vcpu, ctxt->eflags);
7105         kvm_make_request(KVM_REQ_EVENT, vcpu);
7106         return EMULATE_DONE;
7107 }
7108 EXPORT_SYMBOL_GPL(kvm_task_switch);
7109
7110 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7111                                   struct kvm_sregs *sregs)
7112 {
7113         struct msr_data apic_base_msr;
7114         int mmu_reset_needed = 0;
7115         int pending_vec, max_bits, idx;
7116         struct desc_ptr dt;
7117
7118         if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7119                 return -EINVAL;
7120
7121         dt.size = sregs->idt.limit;
7122         dt.address = sregs->idt.base;
7123         kvm_x86_ops->set_idt(vcpu, &dt);
7124         dt.size = sregs->gdt.limit;
7125         dt.address = sregs->gdt.base;
7126         kvm_x86_ops->set_gdt(vcpu, &dt);
7127
7128         vcpu->arch.cr2 = sregs->cr2;
7129         mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7130         vcpu->arch.cr3 = sregs->cr3;
7131         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7132
7133         kvm_set_cr8(vcpu, sregs->cr8);
7134
7135         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7136         kvm_x86_ops->set_efer(vcpu, sregs->efer);
7137         apic_base_msr.data = sregs->apic_base;
7138         apic_base_msr.host_initiated = true;
7139         kvm_set_apic_base(vcpu, &apic_base_msr);
7140
7141         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7142         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7143         vcpu->arch.cr0 = sregs->cr0;
7144
7145         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7146         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7147         if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7148                 kvm_update_cpuid(vcpu);
7149
7150         idx = srcu_read_lock(&vcpu->kvm->srcu);
7151         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7152                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7153                 mmu_reset_needed = 1;
7154         }
7155         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7156
7157         if (mmu_reset_needed)
7158                 kvm_mmu_reset_context(vcpu);
7159
7160         max_bits = KVM_NR_INTERRUPTS;
7161         pending_vec = find_first_bit(
7162                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7163         if (pending_vec < max_bits) {
7164                 kvm_queue_interrupt(vcpu, pending_vec, false);
7165                 pr_debug("Set back pending irq %d\n", pending_vec);
7166         }
7167
7168         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7169         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7170         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7171         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7172         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7173         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7174
7175         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7176         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7177
7178         update_cr8_intercept(vcpu);
7179
7180         /* Older userspace won't unhalt the vcpu on reset. */
7181         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7182             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7183             !is_protmode(vcpu))
7184                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7185
7186         kvm_make_request(KVM_REQ_EVENT, vcpu);
7187
7188         return 0;
7189 }
7190
7191 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7192                                         struct kvm_guest_debug *dbg)
7193 {
7194         unsigned long rflags;
7195         int i, r;
7196
7197         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7198                 r = -EBUSY;
7199                 if (vcpu->arch.exception.pending)
7200                         goto out;
7201                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7202                         kvm_queue_exception(vcpu, DB_VECTOR);
7203                 else
7204                         kvm_queue_exception(vcpu, BP_VECTOR);
7205         }
7206
7207         /*
7208          * Read rflags as long as potentially injected trace flags are still
7209          * filtered out.
7210          */
7211         rflags = kvm_get_rflags(vcpu);
7212
7213         vcpu->guest_debug = dbg->control;
7214         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7215                 vcpu->guest_debug = 0;
7216
7217         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7218                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7219                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7220                 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7221         } else {
7222                 for (i = 0; i < KVM_NR_DB_REGS; i++)
7223                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7224         }
7225         kvm_update_dr7(vcpu);
7226
7227         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7228                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7229                         get_segment_base(vcpu, VCPU_SREG_CS);
7230
7231         /*
7232          * Trigger an rflags update that will inject or remove the trace
7233          * flags.
7234          */
7235         kvm_set_rflags(vcpu, rflags);
7236
7237         kvm_x86_ops->update_bp_intercept(vcpu);
7238
7239         r = 0;
7240
7241 out:
7242
7243         return r;
7244 }
7245
7246 /*
7247  * Translate a guest virtual address to a guest physical address.
7248  */
7249 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7250                                     struct kvm_translation *tr)
7251 {
7252         unsigned long vaddr = tr->linear_address;
7253         gpa_t gpa;
7254         int idx;
7255
7256         idx = srcu_read_lock(&vcpu->kvm->srcu);
7257         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7258         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7259         tr->physical_address = gpa;
7260         tr->valid = gpa != UNMAPPED_GVA;
7261         tr->writeable = 1;
7262         tr->usermode = 0;
7263
7264         return 0;
7265 }
7266
7267 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7268 {
7269         struct fxregs_state *fxsave =
7270                         &vcpu->arch.guest_fpu.state.fxsave;
7271
7272         memcpy(fpu->fpr, fxsave->st_space, 128);
7273         fpu->fcw = fxsave->cwd;
7274         fpu->fsw = fxsave->swd;
7275         fpu->ftwx = fxsave->twd;
7276         fpu->last_opcode = fxsave->fop;
7277         fpu->last_ip = fxsave->rip;
7278         fpu->last_dp = fxsave->rdp;
7279         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7280
7281         return 0;
7282 }
7283
7284 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7285 {
7286         struct fxregs_state *fxsave =
7287                         &vcpu->arch.guest_fpu.state.fxsave;
7288
7289         memcpy(fxsave->st_space, fpu->fpr, 128);
7290         fxsave->cwd = fpu->fcw;
7291         fxsave->swd = fpu->fsw;
7292         fxsave->twd = fpu->ftwx;
7293         fxsave->fop = fpu->last_opcode;
7294         fxsave->rip = fpu->last_ip;
7295         fxsave->rdp = fpu->last_dp;
7296         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7297
7298         return 0;
7299 }
7300
7301 static void fx_init(struct kvm_vcpu *vcpu)
7302 {
7303         fpstate_init(&vcpu->arch.guest_fpu.state);
7304         if (boot_cpu_has(X86_FEATURE_XSAVES))
7305                 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7306                         host_xcr0 | XSTATE_COMPACTION_ENABLED;
7307
7308         /*
7309          * Ensure guest xcr0 is valid for loading
7310          */
7311         vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7312
7313         vcpu->arch.cr0 |= X86_CR0_ET;
7314 }
7315
7316 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7317 {
7318         if (vcpu->guest_fpu_loaded)
7319                 return;
7320
7321         /*
7322          * Restore all possible states in the guest,
7323          * and assume host would use all available bits.
7324          * Guest xcr0 would be loaded later.
7325          */
7326         vcpu->guest_fpu_loaded = 1;
7327         __kernel_fpu_begin();
7328         __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7329         trace_kvm_fpu(1);
7330 }
7331
7332 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7333 {
7334         if (!vcpu->guest_fpu_loaded) {
7335                 vcpu->fpu_counter = 0;
7336                 return;
7337         }
7338
7339         vcpu->guest_fpu_loaded = 0;
7340         copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7341         __kernel_fpu_end();
7342         ++vcpu->stat.fpu_reload;
7343         /*
7344          * If using eager FPU mode, or if the guest is a frequent user
7345          * of the FPU, just leave the FPU active for next time.
7346          * Every 255 times fpu_counter rolls over to 0; a guest that uses
7347          * the FPU in bursts will revert to loading it on demand.
7348          */
7349         if (!use_eager_fpu()) {
7350                 if (++vcpu->fpu_counter < 5)
7351                         kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7352         }
7353         trace_kvm_fpu(0);
7354 }
7355
7356 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7357 {
7358         kvmclock_reset(vcpu);
7359
7360         free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7361         kvm_x86_ops->vcpu_free(vcpu);
7362 }
7363
7364 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7365                                                 unsigned int id)
7366 {
7367         struct kvm_vcpu *vcpu;
7368
7369         if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7370                 printk_once(KERN_WARNING
7371                 "kvm: SMP vm created on host with unstable TSC; "
7372                 "guest TSC will not be reliable\n");
7373
7374         vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7375
7376         return vcpu;
7377 }
7378
7379 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7380 {
7381         int r;
7382
7383         kvm_vcpu_mtrr_init(vcpu);
7384         r = vcpu_load(vcpu);
7385         if (r)
7386                 return r;
7387         kvm_vcpu_reset(vcpu, false);
7388         kvm_mmu_setup(vcpu);
7389         vcpu_put(vcpu);
7390         return r;
7391 }
7392
7393 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7394 {
7395         struct msr_data msr;
7396         struct kvm *kvm = vcpu->kvm;
7397
7398         if (vcpu_load(vcpu))
7399                 return;
7400         msr.data = 0x0;
7401         msr.index = MSR_IA32_TSC;
7402         msr.host_initiated = true;
7403         kvm_write_tsc(vcpu, &msr);
7404         vcpu_put(vcpu);
7405
7406         if (!kvmclock_periodic_sync)
7407                 return;
7408
7409         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7410                                         KVMCLOCK_SYNC_PERIOD);
7411 }
7412
7413 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7414 {
7415         int r;
7416         vcpu->arch.apf.msr_val = 0;
7417
7418         r = vcpu_load(vcpu);
7419         BUG_ON(r);
7420         kvm_mmu_unload(vcpu);
7421         vcpu_put(vcpu);
7422
7423         kvm_x86_ops->vcpu_free(vcpu);
7424 }
7425
7426 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7427 {
7428         vcpu->arch.hflags = 0;
7429
7430         atomic_set(&vcpu->arch.nmi_queued, 0);
7431         vcpu->arch.nmi_pending = 0;
7432         vcpu->arch.nmi_injected = false;
7433         kvm_clear_interrupt_queue(vcpu);
7434         kvm_clear_exception_queue(vcpu);
7435
7436         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7437         kvm_update_dr0123(vcpu);
7438         vcpu->arch.dr6 = DR6_INIT;
7439         kvm_update_dr6(vcpu);
7440         vcpu->arch.dr7 = DR7_FIXED_1;
7441         kvm_update_dr7(vcpu);
7442
7443         vcpu->arch.cr2 = 0;
7444
7445         kvm_make_request(KVM_REQ_EVENT, vcpu);
7446         vcpu->arch.apf.msr_val = 0;
7447         vcpu->arch.st.msr_val = 0;
7448
7449         kvmclock_reset(vcpu);
7450
7451         kvm_clear_async_pf_completion_queue(vcpu);
7452         kvm_async_pf_hash_reset(vcpu);
7453         vcpu->arch.apf.halted = false;
7454
7455         if (!init_event) {
7456                 kvm_pmu_reset(vcpu);
7457                 vcpu->arch.smbase = 0x30000;
7458         }
7459
7460         memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7461         vcpu->arch.regs_avail = ~0;
7462         vcpu->arch.regs_dirty = ~0;
7463
7464         kvm_x86_ops->vcpu_reset(vcpu, init_event);
7465 }
7466
7467 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7468 {
7469         struct kvm_segment cs;
7470
7471         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7472         cs.selector = vector << 8;
7473         cs.base = vector << 12;
7474         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7475         kvm_rip_write(vcpu, 0);
7476 }
7477
7478 int kvm_arch_hardware_enable(void)
7479 {
7480         struct kvm *kvm;
7481         struct kvm_vcpu *vcpu;
7482         int i;
7483         int ret;
7484         u64 local_tsc;
7485         u64 max_tsc = 0;
7486         bool stable, backwards_tsc = false;
7487
7488         kvm_shared_msr_cpu_online();
7489         ret = kvm_x86_ops->hardware_enable();
7490         if (ret != 0)
7491                 return ret;
7492
7493         local_tsc = rdtsc();
7494         stable = !check_tsc_unstable();
7495         list_for_each_entry(kvm, &vm_list, vm_list) {
7496                 kvm_for_each_vcpu(i, vcpu, kvm) {
7497                         if (!stable && vcpu->cpu == smp_processor_id())
7498                                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7499                         if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7500                                 backwards_tsc = true;
7501                                 if (vcpu->arch.last_host_tsc > max_tsc)
7502                                         max_tsc = vcpu->arch.last_host_tsc;
7503                         }
7504                 }
7505         }
7506
7507         /*
7508          * Sometimes, even reliable TSCs go backwards.  This happens on
7509          * platforms that reset TSC during suspend or hibernate actions, but
7510          * maintain synchronization.  We must compensate.  Fortunately, we can
7511          * detect that condition here, which happens early in CPU bringup,
7512          * before any KVM threads can be running.  Unfortunately, we can't
7513          * bring the TSCs fully up to date with real time, as we aren't yet far
7514          * enough into CPU bringup that we know how much real time has actually
7515          * elapsed; our helper function, get_kernel_ns() will be using boot
7516          * variables that haven't been updated yet.
7517          *
7518          * So we simply find the maximum observed TSC above, then record the
7519          * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7520          * the adjustment will be applied.  Note that we accumulate
7521          * adjustments, in case multiple suspend cycles happen before some VCPU
7522          * gets a chance to run again.  In the event that no KVM threads get a
7523          * chance to run, we will miss the entire elapsed period, as we'll have
7524          * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7525          * loose cycle time.  This isn't too big a deal, since the loss will be
7526          * uniform across all VCPUs (not to mention the scenario is extremely
7527          * unlikely). It is possible that a second hibernate recovery happens
7528          * much faster than a first, causing the observed TSC here to be
7529          * smaller; this would require additional padding adjustment, which is
7530          * why we set last_host_tsc to the local tsc observed here.
7531          *
7532          * N.B. - this code below runs only on platforms with reliable TSC,
7533          * as that is the only way backwards_tsc is set above.  Also note
7534          * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7535          * have the same delta_cyc adjustment applied if backwards_tsc
7536          * is detected.  Note further, this adjustment is only done once,
7537          * as we reset last_host_tsc on all VCPUs to stop this from being
7538          * called multiple times (one for each physical CPU bringup).
7539          *
7540          * Platforms with unreliable TSCs don't have to deal with this, they
7541          * will be compensated by the logic in vcpu_load, which sets the TSC to
7542          * catchup mode.  This will catchup all VCPUs to real time, but cannot
7543          * guarantee that they stay in perfect synchronization.
7544          */
7545         if (backwards_tsc) {
7546                 u64 delta_cyc = max_tsc - local_tsc;
7547                 backwards_tsc_observed = true;
7548                 list_for_each_entry(kvm, &vm_list, vm_list) {
7549                         kvm_for_each_vcpu(i, vcpu, kvm) {
7550                                 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7551                                 vcpu->arch.last_host_tsc = local_tsc;
7552                                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7553                         }
7554
7555                         /*
7556                          * We have to disable TSC offset matching.. if you were
7557                          * booting a VM while issuing an S4 host suspend....
7558                          * you may have some problem.  Solving this issue is
7559                          * left as an exercise to the reader.
7560                          */
7561                         kvm->arch.last_tsc_nsec = 0;
7562                         kvm->arch.last_tsc_write = 0;
7563                 }
7564
7565         }
7566         return 0;
7567 }
7568
7569 void kvm_arch_hardware_disable(void)
7570 {
7571         kvm_x86_ops->hardware_disable();
7572         drop_user_return_notifiers();
7573 }
7574
7575 int kvm_arch_hardware_setup(void)
7576 {
7577         int r;
7578
7579         r = kvm_x86_ops->hardware_setup();
7580         if (r != 0)
7581                 return r;
7582
7583         if (kvm_has_tsc_control) {
7584                 /*
7585                  * Make sure the user can only configure tsc_khz values that
7586                  * fit into a signed integer.
7587                  * A min value is not calculated needed because it will always
7588                  * be 1 on all machines.
7589                  */
7590                 u64 max = min(0x7fffffffULL,
7591                               __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7592                 kvm_max_guest_tsc_khz = max;
7593
7594                 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7595         }
7596
7597         kvm_init_msr_list();
7598         return 0;
7599 }
7600
7601 void kvm_arch_hardware_unsetup(void)
7602 {
7603         kvm_x86_ops->hardware_unsetup();
7604 }
7605
7606 void kvm_arch_check_processor_compat(void *rtn)
7607 {
7608         kvm_x86_ops->check_processor_compatibility(rtn);
7609 }
7610
7611 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7612 {
7613         return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7614 }
7615 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7616
7617 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7618 {
7619         return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7620 }
7621
7622 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7623 {
7624         return irqchip_in_kernel(vcpu->kvm) == lapic_in_kernel(vcpu);
7625 }
7626
7627 struct static_key kvm_no_apic_vcpu __read_mostly;
7628 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7629
7630 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7631 {
7632         struct page *page;
7633         struct kvm *kvm;
7634         int r;
7635
7636         BUG_ON(vcpu->kvm == NULL);
7637         kvm = vcpu->kvm;
7638
7639         vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7640         vcpu->arch.pv.pv_unhalted = false;
7641         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7642         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7643                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7644         else
7645                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7646
7647         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7648         if (!page) {
7649                 r = -ENOMEM;
7650                 goto fail;
7651         }
7652         vcpu->arch.pio_data = page_address(page);
7653
7654         kvm_set_tsc_khz(vcpu, max_tsc_khz);
7655
7656         r = kvm_mmu_create(vcpu);
7657         if (r < 0)
7658                 goto fail_free_pio_data;
7659
7660         if (irqchip_in_kernel(kvm)) {
7661                 r = kvm_create_lapic(vcpu);
7662                 if (r < 0)
7663                         goto fail_mmu_destroy;
7664         } else
7665                 static_key_slow_inc(&kvm_no_apic_vcpu);
7666
7667         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7668                                        GFP_KERNEL);
7669         if (!vcpu->arch.mce_banks) {
7670                 r = -ENOMEM;
7671                 goto fail_free_lapic;
7672         }
7673         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7674
7675         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7676                 r = -ENOMEM;
7677                 goto fail_free_mce_banks;
7678         }
7679
7680         fx_init(vcpu);
7681
7682         vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7683         vcpu->arch.pv_time_enabled = false;
7684
7685         vcpu->arch.guest_supported_xcr0 = 0;
7686         vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7687
7688         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7689
7690         vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7691
7692         kvm_async_pf_hash_reset(vcpu);
7693         kvm_pmu_init(vcpu);
7694
7695         vcpu->arch.pending_external_vector = -1;
7696
7697         kvm_hv_vcpu_init(vcpu);
7698
7699         return 0;
7700
7701 fail_free_mce_banks:
7702         kfree(vcpu->arch.mce_banks);
7703 fail_free_lapic:
7704         kvm_free_lapic(vcpu);
7705 fail_mmu_destroy:
7706         kvm_mmu_destroy(vcpu);
7707 fail_free_pio_data:
7708         free_page((unsigned long)vcpu->arch.pio_data);
7709 fail:
7710         return r;
7711 }
7712
7713 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7714 {
7715         int idx;
7716
7717         kvm_hv_vcpu_uninit(vcpu);
7718         kvm_pmu_destroy(vcpu);
7719         kfree(vcpu->arch.mce_banks);
7720         kvm_free_lapic(vcpu);
7721         idx = srcu_read_lock(&vcpu->kvm->srcu);
7722         kvm_mmu_destroy(vcpu);
7723         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7724         free_page((unsigned long)vcpu->arch.pio_data);
7725         if (!lapic_in_kernel(vcpu))
7726                 static_key_slow_dec(&kvm_no_apic_vcpu);
7727 }
7728
7729 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7730 {
7731         kvm_x86_ops->sched_in(vcpu, cpu);
7732 }
7733
7734 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7735 {
7736         if (type)
7737                 return -EINVAL;
7738
7739         INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7740         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7741         INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7742         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7743         atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7744
7745         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7746         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7747         /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7748         set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7749                 &kvm->arch.irq_sources_bitmap);
7750
7751         raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7752         mutex_init(&kvm->arch.apic_map_lock);
7753         spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7754
7755         pvclock_update_vm_gtod_copy(kvm);
7756
7757         INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7758         INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7759
7760         kvm_page_track_init(kvm);
7761         kvm_mmu_init_vm(kvm);
7762
7763         if (kvm_x86_ops->vm_init)
7764                 return kvm_x86_ops->vm_init(kvm);
7765
7766         return 0;
7767 }
7768
7769 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7770 {
7771         int r;
7772         r = vcpu_load(vcpu);
7773         BUG_ON(r);
7774         kvm_mmu_unload(vcpu);
7775         vcpu_put(vcpu);
7776 }
7777
7778 static void kvm_free_vcpus(struct kvm *kvm)
7779 {
7780         unsigned int i;
7781         struct kvm_vcpu *vcpu;
7782
7783         /*
7784          * Unpin any mmu pages first.
7785          */
7786         kvm_for_each_vcpu(i, vcpu, kvm) {
7787                 kvm_clear_async_pf_completion_queue(vcpu);
7788                 kvm_unload_vcpu_mmu(vcpu);
7789         }
7790         kvm_for_each_vcpu(i, vcpu, kvm)
7791                 kvm_arch_vcpu_free(vcpu);
7792
7793         mutex_lock(&kvm->lock);
7794         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7795                 kvm->vcpus[i] = NULL;
7796
7797         atomic_set(&kvm->online_vcpus, 0);
7798         mutex_unlock(&kvm->lock);
7799 }
7800
7801 void kvm_arch_sync_events(struct kvm *kvm)
7802 {
7803         cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7804         cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7805         kvm_free_all_assigned_devices(kvm);
7806         kvm_free_pit(kvm);
7807 }
7808
7809 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7810 {
7811         int i, r;
7812         unsigned long hva;
7813         struct kvm_memslots *slots = kvm_memslots(kvm);
7814         struct kvm_memory_slot *slot, old;
7815
7816         /* Called with kvm->slots_lock held.  */
7817         if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7818                 return -EINVAL;
7819
7820         slot = id_to_memslot(slots, id);
7821         if (size) {
7822                 if (slot->npages)
7823                         return -EEXIST;
7824
7825                 /*
7826                  * MAP_SHARED to prevent internal slot pages from being moved
7827                  * by fork()/COW.
7828                  */
7829                 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7830                               MAP_SHARED | MAP_ANONYMOUS, 0);
7831                 if (IS_ERR((void *)hva))
7832                         return PTR_ERR((void *)hva);
7833         } else {
7834                 if (!slot->npages)
7835                         return 0;
7836
7837                 hva = 0;
7838         }
7839
7840         old = *slot;
7841         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7842                 struct kvm_userspace_memory_region m;
7843
7844                 m.slot = id | (i << 16);
7845                 m.flags = 0;
7846                 m.guest_phys_addr = gpa;
7847                 m.userspace_addr = hva;
7848                 m.memory_size = size;
7849                 r = __kvm_set_memory_region(kvm, &m);
7850                 if (r < 0)
7851                         return r;
7852         }
7853
7854         if (!size) {
7855                 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7856                 WARN_ON(r < 0);
7857         }
7858
7859         return 0;
7860 }
7861 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7862
7863 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7864 {
7865         int r;
7866
7867         mutex_lock(&kvm->slots_lock);
7868         r = __x86_set_memory_region(kvm, id, gpa, size);
7869         mutex_unlock(&kvm->slots_lock);
7870
7871         return r;
7872 }
7873 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7874
7875 void kvm_arch_destroy_vm(struct kvm *kvm)
7876 {
7877         if (current->mm == kvm->mm) {
7878                 /*
7879                  * Free memory regions allocated on behalf of userspace,
7880                  * unless the the memory map has changed due to process exit
7881                  * or fd copying.
7882                  */
7883                 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7884                 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7885                 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
7886         }
7887         if (kvm_x86_ops->vm_destroy)
7888                 kvm_x86_ops->vm_destroy(kvm);
7889         kvm_iommu_unmap_guest(kvm);
7890         kfree(kvm->arch.vpic);
7891         kfree(kvm->arch.vioapic);
7892         kvm_free_vcpus(kvm);
7893         kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7894         kvm_mmu_uninit_vm(kvm);
7895 }
7896
7897 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7898                            struct kvm_memory_slot *dont)
7899 {
7900         int i;
7901
7902         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7903                 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7904                         kvfree(free->arch.rmap[i]);
7905                         free->arch.rmap[i] = NULL;
7906                 }
7907                 if (i == 0)
7908                         continue;
7909
7910                 if (!dont || free->arch.lpage_info[i - 1] !=
7911                              dont->arch.lpage_info[i - 1]) {
7912                         kvfree(free->arch.lpage_info[i - 1]);
7913                         free->arch.lpage_info[i - 1] = NULL;
7914                 }
7915         }
7916
7917         kvm_page_track_free_memslot(free, dont);
7918 }
7919
7920 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7921                             unsigned long npages)
7922 {
7923         int i;
7924
7925         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7926                 struct kvm_lpage_info *linfo;
7927                 unsigned long ugfn;
7928                 int lpages;
7929                 int level = i + 1;
7930
7931                 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7932                                       slot->base_gfn, level) + 1;
7933
7934                 slot->arch.rmap[i] =
7935                         kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7936                 if (!slot->arch.rmap[i])
7937                         goto out_free;
7938                 if (i == 0)
7939                         continue;
7940
7941                 linfo = kvm_kvzalloc(lpages * sizeof(*linfo));
7942                 if (!linfo)
7943                         goto out_free;
7944
7945                 slot->arch.lpage_info[i - 1] = linfo;
7946
7947                 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7948                         linfo[0].disallow_lpage = 1;
7949                 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7950                         linfo[lpages - 1].disallow_lpage = 1;
7951                 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7952                 /*
7953                  * If the gfn and userspace address are not aligned wrt each
7954                  * other, or if explicitly asked to, disable large page
7955                  * support for this slot
7956                  */
7957                 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7958                     !kvm_largepages_enabled()) {
7959                         unsigned long j;
7960
7961                         for (j = 0; j < lpages; ++j)
7962                                 linfo[j].disallow_lpage = 1;
7963                 }
7964         }
7965
7966         if (kvm_page_track_create_memslot(slot, npages))
7967                 goto out_free;
7968
7969         return 0;
7970
7971 out_free:
7972         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7973                 kvfree(slot->arch.rmap[i]);
7974                 slot->arch.rmap[i] = NULL;
7975                 if (i == 0)
7976                         continue;
7977
7978                 kvfree(slot->arch.lpage_info[i - 1]);
7979                 slot->arch.lpage_info[i - 1] = NULL;
7980         }
7981         return -ENOMEM;
7982 }
7983
7984 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7985 {
7986         /*
7987          * memslots->generation has been incremented.
7988          * mmio generation may have reached its maximum value.
7989          */
7990         kvm_mmu_invalidate_mmio_sptes(kvm, slots);
7991 }
7992
7993 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7994                                 struct kvm_memory_slot *memslot,
7995                                 const struct kvm_userspace_memory_region *mem,
7996                                 enum kvm_mr_change change)
7997 {
7998         return 0;
7999 }
8000
8001 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8002                                      struct kvm_memory_slot *new)
8003 {
8004         /* Still write protect RO slot */
8005         if (new->flags & KVM_MEM_READONLY) {
8006                 kvm_mmu_slot_remove_write_access(kvm, new);
8007                 return;
8008         }
8009
8010         /*
8011          * Call kvm_x86_ops dirty logging hooks when they are valid.
8012          *
8013          * kvm_x86_ops->slot_disable_log_dirty is called when:
8014          *
8015          *  - KVM_MR_CREATE with dirty logging is disabled
8016          *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8017          *
8018          * The reason is, in case of PML, we need to set D-bit for any slots
8019          * with dirty logging disabled in order to eliminate unnecessary GPA
8020          * logging in PML buffer (and potential PML buffer full VMEXT). This
8021          * guarantees leaving PML enabled during guest's lifetime won't have
8022          * any additonal overhead from PML when guest is running with dirty
8023          * logging disabled for memory slots.
8024          *
8025          * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8026          * to dirty logging mode.
8027          *
8028          * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8029          *
8030          * In case of write protect:
8031          *
8032          * Write protect all pages for dirty logging.
8033          *
8034          * All the sptes including the large sptes which point to this
8035          * slot are set to readonly. We can not create any new large
8036          * spte on this slot until the end of the logging.
8037          *
8038          * See the comments in fast_page_fault().
8039          */
8040         if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8041                 if (kvm_x86_ops->slot_enable_log_dirty)
8042                         kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8043                 else
8044                         kvm_mmu_slot_remove_write_access(kvm, new);
8045         } else {
8046                 if (kvm_x86_ops->slot_disable_log_dirty)
8047                         kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8048         }
8049 }
8050
8051 void kvm_arch_commit_memory_region(struct kvm *kvm,
8052                                 const struct kvm_userspace_memory_region *mem,
8053                                 const struct kvm_memory_slot *old,
8054                                 const struct kvm_memory_slot *new,
8055                                 enum kvm_mr_change change)
8056 {
8057         int nr_mmu_pages = 0;
8058
8059         if (!kvm->arch.n_requested_mmu_pages)
8060                 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8061
8062         if (nr_mmu_pages)
8063                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8064
8065         /*
8066          * Dirty logging tracks sptes in 4k granularity, meaning that large
8067          * sptes have to be split.  If live migration is successful, the guest
8068          * in the source machine will be destroyed and large sptes will be
8069          * created in the destination. However, if the guest continues to run
8070          * in the source machine (for example if live migration fails), small
8071          * sptes will remain around and cause bad performance.
8072          *
8073          * Scan sptes if dirty logging has been stopped, dropping those
8074          * which can be collapsed into a single large-page spte.  Later
8075          * page faults will create the large-page sptes.
8076          */
8077         if ((change != KVM_MR_DELETE) &&
8078                 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8079                 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8080                 kvm_mmu_zap_collapsible_sptes(kvm, new);
8081
8082         /*
8083          * Set up write protection and/or dirty logging for the new slot.
8084          *
8085          * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8086          * been zapped so no dirty logging staff is needed for old slot. For
8087          * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8088          * new and it's also covered when dealing with the new slot.
8089          *
8090          * FIXME: const-ify all uses of struct kvm_memory_slot.
8091          */
8092         if (change != KVM_MR_DELETE)
8093                 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8094 }
8095
8096 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8097 {
8098         kvm_mmu_invalidate_zap_all_pages(kvm);
8099 }
8100
8101 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8102                                    struct kvm_memory_slot *slot)
8103 {
8104         kvm_mmu_invalidate_zap_all_pages(kvm);
8105 }
8106
8107 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8108 {
8109         if (!list_empty_careful(&vcpu->async_pf.done))
8110                 return true;
8111
8112         if (kvm_apic_has_events(vcpu))
8113                 return true;
8114
8115         if (vcpu->arch.pv.pv_unhalted)
8116                 return true;
8117
8118         if (atomic_read(&vcpu->arch.nmi_queued))
8119                 return true;
8120
8121         if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8122                 return true;
8123
8124         if (kvm_arch_interrupt_allowed(vcpu) &&
8125             kvm_cpu_has_interrupt(vcpu))
8126                 return true;
8127
8128         if (kvm_hv_has_stimer_pending(vcpu))
8129                 return true;
8130
8131         return false;
8132 }
8133
8134 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8135 {
8136         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8137                 kvm_x86_ops->check_nested_events(vcpu, false);
8138
8139         return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8140 }
8141
8142 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8143 {
8144         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8145 }
8146
8147 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8148 {
8149         return kvm_x86_ops->interrupt_allowed(vcpu);
8150 }
8151
8152 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8153 {
8154         if (is_64_bit_mode(vcpu))
8155                 return kvm_rip_read(vcpu);
8156         return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8157                      kvm_rip_read(vcpu));
8158 }
8159 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8160
8161 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8162 {
8163         return kvm_get_linear_rip(vcpu) == linear_rip;
8164 }
8165 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8166
8167 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8168 {
8169         unsigned long rflags;
8170
8171         rflags = kvm_x86_ops->get_rflags(vcpu);
8172         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8173                 rflags &= ~X86_EFLAGS_TF;
8174         return rflags;
8175 }
8176 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8177
8178 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8179 {
8180         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8181             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8182                 rflags |= X86_EFLAGS_TF;
8183         kvm_x86_ops->set_rflags(vcpu, rflags);
8184 }
8185
8186 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8187 {
8188         __kvm_set_rflags(vcpu, rflags);
8189         kvm_make_request(KVM_REQ_EVENT, vcpu);
8190 }
8191 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8192
8193 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8194 {
8195         int r;
8196
8197         if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8198               work->wakeup_all)
8199                 return;
8200
8201         r = kvm_mmu_reload(vcpu);
8202         if (unlikely(r))
8203                 return;
8204
8205         if (!vcpu->arch.mmu.direct_map &&
8206               work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8207                 return;
8208
8209         vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8210 }
8211
8212 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8213 {
8214         return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8215 }
8216
8217 static inline u32 kvm_async_pf_next_probe(u32 key)
8218 {
8219         return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8220 }
8221
8222 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8223 {
8224         u32 key = kvm_async_pf_hash_fn(gfn);
8225
8226         while (vcpu->arch.apf.gfns[key] != ~0)
8227                 key = kvm_async_pf_next_probe(key);
8228
8229         vcpu->arch.apf.gfns[key] = gfn;
8230 }
8231
8232 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8233 {
8234         int i;
8235         u32 key = kvm_async_pf_hash_fn(gfn);
8236
8237         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8238                      (vcpu->arch.apf.gfns[key] != gfn &&
8239                       vcpu->arch.apf.gfns[key] != ~0); i++)
8240                 key = kvm_async_pf_next_probe(key);
8241
8242         return key;
8243 }
8244
8245 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8246 {
8247         return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8248 }
8249
8250 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8251 {
8252         u32 i, j, k;
8253
8254         i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8255         while (true) {
8256                 vcpu->arch.apf.gfns[i] = ~0;
8257                 do {
8258                         j = kvm_async_pf_next_probe(j);
8259                         if (vcpu->arch.apf.gfns[j] == ~0)
8260                                 return;
8261                         k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8262                         /*
8263                          * k lies cyclically in ]i,j]
8264                          * |    i.k.j |
8265                          * |....j i.k.| or  |.k..j i...|
8266                          */
8267                 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8268                 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8269                 i = j;
8270         }
8271 }
8272
8273 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8274 {
8275
8276         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8277                                       sizeof(val));
8278 }
8279
8280 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8281                                      struct kvm_async_pf *work)
8282 {
8283         struct x86_exception fault;
8284
8285         trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8286         kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8287
8288         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8289             (vcpu->arch.apf.send_user_only &&
8290              kvm_x86_ops->get_cpl(vcpu) == 0))
8291                 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8292         else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8293                 fault.vector = PF_VECTOR;
8294                 fault.error_code_valid = true;
8295                 fault.error_code = 0;
8296                 fault.nested_page_fault = false;
8297                 fault.address = work->arch.token;
8298                 kvm_inject_page_fault(vcpu, &fault);
8299         }
8300 }
8301
8302 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8303                                  struct kvm_async_pf *work)
8304 {
8305         struct x86_exception fault;
8306
8307         trace_kvm_async_pf_ready(work->arch.token, work->gva);
8308         if (work->wakeup_all)
8309                 work->arch.token = ~0; /* broadcast wakeup */
8310         else
8311                 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8312
8313         if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8314             !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8315                 fault.vector = PF_VECTOR;
8316                 fault.error_code_valid = true;
8317                 fault.error_code = 0;
8318                 fault.nested_page_fault = false;
8319                 fault.address = work->arch.token;
8320                 kvm_inject_page_fault(vcpu, &fault);
8321         }
8322         vcpu->arch.apf.halted = false;
8323         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8324 }
8325
8326 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8327 {
8328         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8329                 return true;
8330         else
8331                 return !kvm_event_needs_reinjection(vcpu) &&
8332                         kvm_x86_ops->interrupt_allowed(vcpu);
8333 }
8334
8335 void kvm_arch_start_assignment(struct kvm *kvm)
8336 {
8337         atomic_inc(&kvm->arch.assigned_device_count);
8338 }
8339 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8340
8341 void kvm_arch_end_assignment(struct kvm *kvm)
8342 {
8343         atomic_dec(&kvm->arch.assigned_device_count);
8344 }
8345 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8346
8347 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8348 {
8349         return atomic_read(&kvm->arch.assigned_device_count);
8350 }
8351 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8352
8353 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8354 {
8355         atomic_inc(&kvm->arch.noncoherent_dma_count);
8356 }
8357 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8358
8359 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8360 {
8361         atomic_dec(&kvm->arch.noncoherent_dma_count);
8362 }
8363 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8364
8365 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8366 {
8367         return atomic_read(&kvm->arch.noncoherent_dma_count);
8368 }
8369 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8370
8371 bool kvm_arch_has_irq_bypass(void)
8372 {
8373         return kvm_x86_ops->update_pi_irte != NULL;
8374 }
8375
8376 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8377                                       struct irq_bypass_producer *prod)
8378 {
8379         struct kvm_kernel_irqfd *irqfd =
8380                 container_of(cons, struct kvm_kernel_irqfd, consumer);
8381
8382         irqfd->producer = prod;
8383
8384         return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8385                                            prod->irq, irqfd->gsi, 1);
8386 }
8387
8388 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8389                                       struct irq_bypass_producer *prod)
8390 {
8391         int ret;
8392         struct kvm_kernel_irqfd *irqfd =
8393                 container_of(cons, struct kvm_kernel_irqfd, consumer);
8394
8395         WARN_ON(irqfd->producer != prod);
8396         irqfd->producer = NULL;
8397
8398         /*
8399          * When producer of consumer is unregistered, we change back to
8400          * remapped mode, so we can re-use the current implementation
8401          * when the irq is masked/disabed or the consumer side (KVM
8402          * int this case doesn't want to receive the interrupts.
8403         */
8404         ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8405         if (ret)
8406                 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8407                        " fails: %d\n", irqfd->consumer.token, ret);
8408 }
8409
8410 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8411                                    uint32_t guest_irq, bool set)
8412 {
8413         if (!kvm_x86_ops->update_pi_irte)
8414                 return -EINVAL;
8415
8416         return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8417 }
8418
8419 bool kvm_vector_hashing_enabled(void)
8420 {
8421         return vector_hashing;
8422 }
8423 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8424
8425 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8426 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8427 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8428 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8429 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8430 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8431 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8432 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8433 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8434 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8435 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8436 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8437 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8438 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8439 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8440 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8441 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8442 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8443 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);