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/export.h>
  40 #include <linux/moduleparam.h>
  41 #include <linux/mman.h>
  42 #include <linux/highmem.h>
  43 #include <linux/iommu.h>
  44 #include <linux/intel-iommu.h>
  45 #include <linux/cpufreq.h>
  46 #include <linux/user-return-notifier.h>
  47 #include <linux/srcu.h>
  48 #include <linux/slab.h>
  49 #include <linux/perf_event.h>
  50 #include <linux/uaccess.h>
  51 #include <linux/hash.h>
  52 #include <linux/pci.h>
  53 #include <linux/timekeeper_internal.h>
  54 #include <linux/pvclock_gtod.h>
  55 #include <linux/kvm_irqfd.h>
  56 #include <linux/irqbypass.h>
  57 #include <trace/events/kvm.h>
  58
  59 #include <asm/debugreg.h>
  60 #include <asm/msr.h>
  61 #include <asm/desc.h>
  62 #include <asm/mce.h>
  63 #include <linux/kernel_stat.h>
  64 #include <asm/fpu/internal.h> /* Ugh! */
  65 #include <asm/pvclock.h>
  66 #include <asm/div64.h>
  67 #include <asm/irq_remapping.h>
  68
  69 #define CREATE_TRACE_POINTS
  70 #include "trace.h"
  71
  72 #define MAX_IO_MSRS 256
  73 #define KVM_MAX_MCE_BANKS 32
  74 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
  75 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
  76
  77 #define emul_to_vcpu(ctxt) \
  78         container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
  79
  80 /* EFER defaults:
  81  * - enable syscall per default because its emulated by KVM
  82  * - enable LME and LMA per default on 64 bit KVM
  83  */
  84 #ifdef CONFIG_X86_64
  85 static
  86 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
  87 #else
  88 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
  89 #endif
  90
  91 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  92 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  93
  94 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
  95                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
  96
  97 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  98 static void process_nmi(struct kvm_vcpu *vcpu);
  99 static void enter_smm(struct kvm_vcpu *vcpu);
 100 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
 101
 102 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
 103 EXPORT_SYMBOL_GPL(kvm_x86_ops);
 104
 105 static bool __read_mostly ignore_msrs = 0;
 106 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
 107
 108 unsigned int min_timer_period_us = 500;
 109 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
 110
 111 static bool __read_mostly kvmclock_periodic_sync = true;
 112 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
 113
 114 bool __read_mostly kvm_has_tsc_control;
 115 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
 116 u32  __read_mostly kvm_max_guest_tsc_khz;
 117 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
 118 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
 119 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
 120 u64  __read_mostly kvm_max_tsc_scaling_ratio;
 121 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
 122 u64 __read_mostly kvm_default_tsc_scaling_ratio;
 123 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
 124
 125 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
 126 static u32 __read_mostly tsc_tolerance_ppm = 250;
 127 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
 128
 129 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
 130 unsigned int __read_mostly lapic_timer_advance_ns = 0;
 131 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
 132
 133 static bool __read_mostly vector_hashing = true;
 134 module_param(vector_hashing, bool, S_IRUGO);
 135
 136 static bool __read_mostly backwards_tsc_observed = false;
 137
 138 #define KVM_NR_SHARED_MSRS 16
 139
 140 struct kvm_shared_msrs_global {
 141         int nr;
 142         u32 msrs[KVM_NR_SHARED_MSRS];
 143 };
 144
 145 struct kvm_shared_msrs {
 146         struct user_return_notifier urn;
 147         bool registered;
 148         struct kvm_shared_msr_values {
 149                 u64 host;
 150                 u64 curr;
 151         } values[KVM_NR_SHARED_MSRS];
 152 };
 153
 154 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 155 static struct kvm_shared_msrs __percpu *shared_msrs;
 156
 157 struct kvm_stats_debugfs_item debugfs_entries[] = {
 158         { "pf_fixed", VCPU_STAT(pf_fixed) },
 159         { "pf_guest", VCPU_STAT(pf_guest) },
 160         { "tlb_flush", VCPU_STAT(tlb_flush) },
 161         { "invlpg", VCPU_STAT(invlpg) },
 162         { "exits", VCPU_STAT(exits) },
 163         { "io_exits", VCPU_STAT(io_exits) },
 164         { "mmio_exits", VCPU_STAT(mmio_exits) },
 165         { "signal_exits", VCPU_STAT(signal_exits) },
 166         { "irq_window", VCPU_STAT(irq_window_exits) },
 167         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 168         { "halt_exits", VCPU_STAT(halt_exits) },
 169         { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
 170         { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
 171         { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
 172         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 173         { "hypercalls", VCPU_STAT(hypercalls) },
 174         { "request_irq", VCPU_STAT(request_irq_exits) },
 175         { "irq_exits", VCPU_STAT(irq_exits) },
 176         { "host_state_reload", VCPU_STAT(host_state_reload) },
 177         { "efer_reload", VCPU_STAT(efer_reload) },
 178         { "fpu_reload", VCPU_STAT(fpu_reload) },
 179         { "insn_emulation", VCPU_STAT(insn_emulation) },
 180         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 181         { "irq_injections", VCPU_STAT(irq_injections) },
 182         { "nmi_injections", VCPU_STAT(nmi_injections) },
 183         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 184         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 185         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 186         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 187         { "mmu_flooded", VM_STAT(mmu_flooded) },
 188         { "mmu_recycled", VM_STAT(mmu_recycled) },
 189         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 190         { "mmu_unsync", VM_STAT(mmu_unsync) },
 191         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 192         { "largepages", VM_STAT(lpages) },
 193         { NULL }
 194 };
 195
 196 u64 __read_mostly host_xcr0;
 197
 198 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
 199
 200 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
 201 {
 202         int i;
 203         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
 204                 vcpu->arch.apf.gfns[i] = ~0;
 205 }
 206
 207 static void kvm_on_user_return(struct user_return_notifier *urn)
 208 {
 209         unsigned slot;
 210         struct kvm_shared_msrs *locals
 211                 = container_of(urn, struct kvm_shared_msrs, urn);
 212         struct kvm_shared_msr_values *values;
 213         unsigned long flags;
 214
 215         /*
 216          * Disabling irqs at this point since the following code could be
 217          * interrupted and executed through kvm_arch_hardware_disable()
 218          */
 219         local_irq_save(flags);
 220         if (locals->registered) {
 221                 locals->registered = false;
 222                 user_return_notifier_unregister(urn);
 223         }
 224         local_irq_restore(flags);
 225         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 226                 values = &locals->values[slot];
 227                 if (values->host != values->curr) {
 228                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 229                         values->curr = values->host;
 230                 }
 231         }
 232 }
 233
 234 static void shared_msr_update(unsigned slot, u32 msr)
 235 {
 236         u64 value;
 237         unsigned int cpu = smp_processor_id();
 238         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 239
 240         /* only read, and nobody should modify it at this time,
 241          * so don't need lock */
 242         if (slot >= shared_msrs_global.nr) {
 243                 printk(KERN_ERR "kvm: invalid MSR slot!");
 244                 return;
 245         }
 246         rdmsrl_safe(msr, &value);
 247         smsr->values[slot].host = value;
 248         smsr->values[slot].curr = value;
 249 }
 250
 251 void kvm_define_shared_msr(unsigned slot, u32 msr)
 252 {
 253         BUG_ON(slot >= KVM_NR_SHARED_MSRS);
 254         shared_msrs_global.msrs[slot] = msr;
 255         if (slot >= shared_msrs_global.nr)
 256                 shared_msrs_global.nr = slot + 1;
 257 }
 258 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 259
 260 static void kvm_shared_msr_cpu_online(void)
 261 {
 262         unsigned i;
 263
 264         for (i = 0; i < shared_msrs_global.nr; ++i)
 265                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 266 }
 267
 268 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 269 {
 270         unsigned int cpu = smp_processor_id();
 271         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 272         int err;
 273
 274         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 275                 return 0;
 276         smsr->values[slot].curr = value;
 277         err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
 278         if (err)
 279                 return 1;
 280
 281         if (!smsr->registered) {
 282                 smsr->urn.on_user_return = kvm_on_user_return;
 283                 user_return_notifier_register(&smsr->urn);
 284                 smsr->registered = true;
 285         }
 286         return 0;
 287 }
 288 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 289
 290 static void drop_user_return_notifiers(void)
 291 {
 292         unsigned int cpu = smp_processor_id();
 293         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 294
 295         if (smsr->registered)
 296                 kvm_on_user_return(&smsr->urn);
 297 }
 298
 299 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 300 {
 301         return vcpu->arch.apic_base;
 302 }
 303 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 304
 305 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 306 {
 307         u64 old_state = vcpu->arch.apic_base &
 308                 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
 309         u64 new_state = msr_info->data &
 310                 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
 311         u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
 312                 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
 313
 314         if (!msr_info->host_initiated &&
 315             ((msr_info->data & reserved_bits) != 0 ||
 316              new_state == X2APIC_ENABLE ||
 317              (new_state == MSR_IA32_APICBASE_ENABLE &&
 318               old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
 319              (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
 320               old_state == 0)))
 321                 return 1;
 322
 323         kvm_lapic_set_base(vcpu, msr_info->data);
 324         return 0;
 325 }
 326 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 327
 328 asmlinkage __visible void kvm_spurious_fault(void)
 329 {
 330         /* Fault while not rebooting.  We want the trace. */
 331         BUG();
 332 }
 333 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
 334
 335 #define EXCPT_BENIGN            0
 336 #define EXCPT_CONTRIBUTORY      1
 337 #define EXCPT_PF                2
 338
 339 static int exception_class(int vector)
 340 {
 341         switch (vector) {
 342         case PF_VECTOR:
 343                 return EXCPT_PF;
 344         case DE_VECTOR:
 345         case TS_VECTOR:
 346         case NP_VECTOR:
 347         case SS_VECTOR:
 348         case GP_VECTOR:
 349                 return EXCPT_CONTRIBUTORY;
 350         default:
 351                 break;
 352         }
 353         return EXCPT_BENIGN;
 354 }
 355
 356 #define EXCPT_FAULT             0
 357 #define EXCPT_TRAP              1
 358 #define EXCPT_ABORT             2
 359 #define EXCPT_INTERRUPT         3
 360
 361 static int exception_type(int vector)
 362 {
 363         unsigned int mask;
 364
 365         if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
 366                 return EXCPT_INTERRUPT;
 367
 368         mask = 1 << vector;
 369
 370         /* #DB is trap, as instruction watchpoints are handled elsewhere */
 371         if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
 372                 return EXCPT_TRAP;
 373
 374         if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
 375                 return EXCPT_ABORT;
 376
 377         /* Reserved exceptions will result in fault */
 378         return EXCPT_FAULT;
 379 }
 380
 381 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 382                 unsigned nr, bool has_error, u32 error_code,
 383                 bool reinject)
 384 {
 385         u32 prev_nr;
 386         int class1, class2;
 387
 388         kvm_make_request(KVM_REQ_EVENT, vcpu);
 389
 390         if (!vcpu->arch.exception.pending) {
 391         queue:
 392                 if (has_error && !is_protmode(vcpu))
 393                         has_error = false;
 394                 vcpu->arch.exception.pending = true;
 395                 vcpu->arch.exception.has_error_code = has_error;
 396                 vcpu->arch.exception.nr = nr;
 397                 vcpu->arch.exception.error_code = error_code;
 398                 vcpu->arch.exception.reinject = reinject;
 399                 return;
 400         }
 401
 402         /* to check exception */
 403         prev_nr = vcpu->arch.exception.nr;
 404         if (prev_nr == DF_VECTOR) {
 405                 /* triple fault -> shutdown */
 406                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
 407                 return;
 408         }
 409         class1 = exception_class(prev_nr);
 410         class2 = exception_class(nr);
 411         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 412                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 413                 /* generate double fault per SDM Table 5-5 */
 414                 vcpu->arch.exception.pending = true;
 415                 vcpu->arch.exception.has_error_code = true;
 416                 vcpu->arch.exception.nr = DF_VECTOR;
 417                 vcpu->arch.exception.error_code = 0;
 418         } else
 419                 /* replace previous exception with a new one in a hope
 420                    that instruction re-execution will regenerate lost
 421                    exception */
 422                 goto queue;
 423 }
 424
 425 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 426 {
 427         kvm_multiple_exception(vcpu, nr, false, 0, false);
 428 }
 429 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 430
 431 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 432 {
 433         kvm_multiple_exception(vcpu, nr, false, 0, true);
 434 }
 435 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 436
 437 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
 438 {
 439         if (err)
 440                 kvm_inject_gp(vcpu, 0);
 441         else
 442                 return kvm_skip_emulated_instruction(vcpu);
 443
 444         return 1;
 445 }
 446 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
 447
 448 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 449 {
 450         ++vcpu->stat.pf_guest;
 451         vcpu->arch.cr2 = fault->address;
 452         kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
 453 }
 454 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
 455
 456 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 457 {
 458         if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
 459                 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
 460         else
 461                 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
 462
 463         return fault->nested_page_fault;
 464 }
 465
 466 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 467 {
 468         atomic_inc(&vcpu->arch.nmi_queued);
 469         kvm_make_request(KVM_REQ_NMI, vcpu);
 470 }
 471 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 472
 473 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 474 {
 475         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 476 }
 477 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 478
 479 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 480 {
 481         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 482 }
 483 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 484
 485 /*
 486  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 487  * a #GP and return false.
 488  */
 489 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 490 {
 491         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 492                 return true;
 493         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 494         return false;
 495 }
 496 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 497
 498 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
 499 {
 500         if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 501                 return true;
 502
 503         kvm_queue_exception(vcpu, UD_VECTOR);
 504         return false;
 505 }
 506 EXPORT_SYMBOL_GPL(kvm_require_dr);
 507
 508 /*
 509  * This function will be used to read from the physical memory of the currently
 510  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
 511  * can read from guest physical or from the guest's guest physical memory.
 512  */
 513 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 514                             gfn_t ngfn, void *data, int offset, int len,
 515                             u32 access)
 516 {
 517         struct x86_exception exception;
 518         gfn_t real_gfn;
 519         gpa_t ngpa;
 520
 521         ngpa     = gfn_to_gpa(ngfn);
 522         real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
 523         if (real_gfn == UNMAPPED_GVA)
 524                 return -EFAULT;
 525
 526         real_gfn = gpa_to_gfn(real_gfn);
 527
 528         return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
 529 }
 530 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
 531
 532 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
 533                                void *data, int offset, int len, u32 access)
 534 {
 535         return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
 536                                        data, offset, len, access);
 537 }
 538
 539 /*
 540  * Load the pae pdptrs.  Return true is they are all valid.
 541  */
 542 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
 543 {
 544         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 545         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 546         int i;
 547         int ret;
 548         u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
 549
 550         ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
 551                                       offset * sizeof(u64), sizeof(pdpte),
 552                                       PFERR_USER_MASK|PFERR_WRITE_MASK);
 553         if (ret < 0) {
 554                 ret = 0;
 555                 goto out;
 556         }
 557         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 558                 if ((pdpte[i] & PT_PRESENT_MASK) &&
 559                     (pdpte[i] &
 560                      vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
 561                         ret = 0;
 562                         goto out;
 563                 }
 564         }
 565         ret = 1;
 566
 567         memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
 568         __set_bit(VCPU_EXREG_PDPTR,
 569                   (unsigned long *)&vcpu->arch.regs_avail);
 570         __set_bit(VCPU_EXREG_PDPTR,
 571                   (unsigned long *)&vcpu->arch.regs_dirty);
 572 out:
 573
 574         return ret;
 575 }
 576 EXPORT_SYMBOL_GPL(load_pdptrs);
 577
 578 bool pdptrs_changed(struct kvm_vcpu *vcpu)
 579 {
 580         u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
 581         bool changed = true;
 582         int offset;
 583         gfn_t gfn;
 584         int r;
 585
 586         if (is_long_mode(vcpu) || !is_pae(vcpu))
 587                 return false;
 588
 589         if (!test_bit(VCPU_EXREG_PDPTR,
 590                       (unsigned long *)&vcpu->arch.regs_avail))
 591                 return true;
 592
 593         gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
 594         offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
 595         r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
 596                                        PFERR_USER_MASK | PFERR_WRITE_MASK);
 597         if (r < 0)
 598                 goto out;
 599         changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
 600 out:
 601
 602         return changed;
 603 }
 604 EXPORT_SYMBOL_GPL(pdptrs_changed);
 605
 606 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 607 {
 608         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 609         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
 610
 611         cr0 |= X86_CR0_ET;
 612
 613 #ifdef CONFIG_X86_64
 614         if (cr0 & 0xffffffff00000000UL)
 615                 return 1;
 616 #endif
 617
 618         cr0 &= ~CR0_RESERVED_BITS;
 619
 620         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 621                 return 1;
 622
 623         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 624                 return 1;
 625
 626         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 627 #ifdef CONFIG_X86_64
 628                 if ((vcpu->arch.efer & EFER_LME)) {
 629                         int cs_db, cs_l;
 630
 631                         if (!is_pae(vcpu))
 632                                 return 1;
 633                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 634                         if (cs_l)
 635                                 return 1;
 636                 } else
 637 #endif
 638                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 639                                                  kvm_read_cr3(vcpu)))
 640                         return 1;
 641         }
 642
 643         if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
 644                 return 1;
 645
 646         kvm_x86_ops->set_cr0(vcpu, cr0);
 647
 648         if ((cr0 ^ old_cr0) & X86_CR0_PG) {
 649                 kvm_clear_async_pf_completion_queue(vcpu);
 650                 kvm_async_pf_hash_reset(vcpu);
 651         }
 652
 653         if ((cr0 ^ old_cr0) & update_bits)
 654                 kvm_mmu_reset_context(vcpu);
 655
 656         if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
 657             kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
 658             !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
 659                 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
 660
 661         return 0;
 662 }
 663 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 664
 665 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 666 {
 667         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 668 }
 669 EXPORT_SYMBOL_GPL(kvm_lmsw);
 670
 671 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
 672 {
 673         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
 674                         !vcpu->guest_xcr0_loaded) {
 675                 /* kvm_set_xcr() also depends on this */
 676                 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
 677                 vcpu->guest_xcr0_loaded = 1;
 678         }
 679 }
 680
 681 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
 682 {
 683         if (vcpu->guest_xcr0_loaded) {
 684                 if (vcpu->arch.xcr0 != host_xcr0)
 685                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
 686                 vcpu->guest_xcr0_loaded = 0;
 687         }
 688 }
 689
 690 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 691 {
 692         u64 xcr0 = xcr;
 693         u64 old_xcr0 = vcpu->arch.xcr0;
 694         u64 valid_bits;
 695
 696         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
 697         if (index != XCR_XFEATURE_ENABLED_MASK)
 698                 return 1;
 699         if (!(xcr0 & XFEATURE_MASK_FP))
 700                 return 1;
 701         if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
 702                 return 1;
 703
 704         /*
 705          * Do not allow the guest to set bits that we do not support
 706          * saving.  However, xcr0 bit 0 is always set, even if the
 707          * emulated CPU does not support XSAVE (see fx_init).
 708          */
 709         valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
 710         if (xcr0 & ~valid_bits)
 711                 return 1;
 712
 713         if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
 714             (!(xcr0 & XFEATURE_MASK_BNDCSR)))
 715                 return 1;
 716
 717         if (xcr0 & XFEATURE_MASK_AVX512) {
 718                 if (!(xcr0 & XFEATURE_MASK_YMM))
 719                         return 1;
 720                 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
 721                         return 1;
 722         }
 723         vcpu->arch.xcr0 = xcr0;
 724
 725         if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
 726                 kvm_update_cpuid(vcpu);
 727         return 0;
 728 }
 729
 730 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 731 {
 732         if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
 733             __kvm_set_xcr(vcpu, index, xcr)) {
 734                 kvm_inject_gp(vcpu, 0);
 735                 return 1;
 736         }
 737         return 0;
 738 }
 739 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 740
 741 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 742 {
 743         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 744         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
 745                                    X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
 746
 747         if (cr4 & CR4_RESERVED_BITS)
 748                 return 1;
 749
 750         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
 751                 return 1;
 752
 753         if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
 754                 return 1;
 755
 756         if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
 757                 return 1;
 758
 759         if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
 760                 return 1;
 761
 762         if (!guest_cpuid_has_pku(vcpu) && (cr4 & X86_CR4_PKE))
 763                 return 1;
 764
 765         if (is_long_mode(vcpu)) {
 766                 if (!(cr4 & X86_CR4_PAE))
 767                         return 1;
 768         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 769                    && ((cr4 ^ old_cr4) & pdptr_bits)
 770                    && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 771                                    kvm_read_cr3(vcpu)))
 772                 return 1;
 773
 774         if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
 775                 if (!guest_cpuid_has_pcid(vcpu))
 776                         return 1;
 777
 778                 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
 779                 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
 780                         return 1;
 781         }
 782
 783         if (kvm_x86_ops->set_cr4(vcpu, cr4))
 784                 return 1;
 785
 786         if (((cr4 ^ old_cr4) & pdptr_bits) ||
 787             (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
 788                 kvm_mmu_reset_context(vcpu);
 789
 790         if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
 791                 kvm_update_cpuid(vcpu);
 792
 793         return 0;
 794 }
 795 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 796
 797 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 798 {
 799 #ifdef CONFIG_X86_64
 800         cr3 &= ~CR3_PCID_INVD;
 801 #endif
 802
 803         if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
 804                 kvm_mmu_sync_roots(vcpu);
 805                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 806                 return 0;
 807         }
 808
 809         if (is_long_mode(vcpu)) {
 810                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 811                         return 1;
 812         } else if (is_pae(vcpu) && is_paging(vcpu) &&
 813                    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 814                 return 1;
 815
 816         vcpu->arch.cr3 = cr3;
 817         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
 818         kvm_mmu_new_cr3(vcpu);
 819         return 0;
 820 }
 821 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 822
 823 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 824 {
 825         if (cr8 & CR8_RESERVED_BITS)
 826                 return 1;
 827         if (lapic_in_kernel(vcpu))
 828                 kvm_lapic_set_tpr(vcpu, cr8);
 829         else
 830                 vcpu->arch.cr8 = cr8;
 831         return 0;
 832 }
 833 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 834
 835 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 836 {
 837         if (lapic_in_kernel(vcpu))
 838                 return kvm_lapic_get_cr8(vcpu);
 839         else
 840                 return vcpu->arch.cr8;
 841 }
 842 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 843
 844 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
 845 {
 846         int i;
 847
 848         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 849                 for (i = 0; i < KVM_NR_DB_REGS; i++)
 850                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
 851                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
 852         }
 853 }
 854
 855 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
 856 {
 857         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 858                 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
 859 }
 860
 861 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
 862 {
 863         unsigned long dr7;
 864
 865         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 866                 dr7 = vcpu->arch.guest_debug_dr7;
 867         else
 868                 dr7 = vcpu->arch.dr7;
 869         kvm_x86_ops->set_dr7(vcpu, dr7);
 870         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
 871         if (dr7 & DR7_BP_EN_MASK)
 872                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
 873 }
 874
 875 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
 876 {
 877         u64 fixed = DR6_FIXED_1;
 878
 879         if (!guest_cpuid_has_rtm(vcpu))
 880                 fixed |= DR6_RTM;
 881         return fixed;
 882 }
 883
 884 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 885 {
 886         switch (dr) {
 887         case 0 ... 3:
 888                 vcpu->arch.db[dr] = val;
 889                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 890                         vcpu->arch.eff_db[dr] = val;
 891                 break;
 892         case 4:
 893                 /* fall through */
 894         case 6:
 895                 if (val & 0xffffffff00000000ULL)
 896                         return -1; /* #GP */
 897                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
 898                 kvm_update_dr6(vcpu);
 899                 break;
 900         case 5:
 901                 /* fall through */
 902         default: /* 7 */
 903                 if (val & 0xffffffff00000000ULL)
 904                         return -1; /* #GP */
 905                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 906                 kvm_update_dr7(vcpu);
 907                 break;
 908         }
 909
 910         return 0;
 911 }
 912
 913 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 914 {
 915         if (__kvm_set_dr(vcpu, dr, val)) {
 916                 kvm_inject_gp(vcpu, 0);
 917                 return 1;
 918         }
 919         return 0;
 920 }
 921 EXPORT_SYMBOL_GPL(kvm_set_dr);
 922
 923 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 924 {
 925         switch (dr) {
 926         case 0 ... 3:
 927                 *val = vcpu->arch.db[dr];
 928                 break;
 929         case 4:
 930                 /* fall through */
 931         case 6:
 932                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 933                         *val = vcpu->arch.dr6;
 934                 else
 935                         *val = kvm_x86_ops->get_dr6(vcpu);
 936                 break;
 937         case 5:
 938                 /* fall through */
 939         default: /* 7 */
 940                 *val = vcpu->arch.dr7;
 941                 break;
 942         }
 943         return 0;
 944 }
 945 EXPORT_SYMBOL_GPL(kvm_get_dr);
 946
 947 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
 948 {
 949         u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 950         u64 data;
 951         int err;
 952
 953         err = kvm_pmu_rdpmc(vcpu, ecx, &data);
 954         if (err)
 955                 return err;
 956         kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
 957         kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
 958         return err;
 959 }
 960 EXPORT_SYMBOL_GPL(kvm_rdpmc);
 961
 962 /*
 963  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 964  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 965  *
 966  * This list is modified at module load time to reflect the
 967  * capabilities of the host cpu. This capabilities test skips MSRs that are
 968  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
 969  * may depend on host virtualization features rather than host cpu features.
 970  */
 971
 972 static u32 msrs_to_save[] = {
 973         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 974         MSR_STAR,
 975 #ifdef CONFIG_X86_64
 976         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 977 #endif
 978         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
 979         MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
 980 };
 981
 982 static unsigned num_msrs_to_save;
 983
 984 static u32 emulated_msrs[] = {
 985         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 986         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 987         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 988         HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
 989         HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
 990         HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
 991         HV_X64_MSR_RESET,
 992         HV_X64_MSR_VP_INDEX,
 993         HV_X64_MSR_VP_RUNTIME,
 994         HV_X64_MSR_SCONTROL,
 995         HV_X64_MSR_STIMER0_CONFIG,
 996         HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
 997         MSR_KVM_PV_EOI_EN,
 998
 999         MSR_IA32_TSC_ADJUST,
1000         MSR_IA32_TSCDEADLINE,
1001         MSR_IA32_MISC_ENABLE,
1002         MSR_IA32_MCG_STATUS,
1003         MSR_IA32_MCG_CTL,
1004         MSR_IA32_MCG_EXT_CTL,
1005         MSR_IA32_SMBASE,
1006 };
1007
1008 static unsigned num_emulated_msrs;
1009
1010 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1011 {
1012         if (efer & efer_reserved_bits)
1013                 return false;
1014
1015         if (efer & EFER_FFXSR) {
1016                 struct kvm_cpuid_entry2 *feat;
1017
1018                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1019                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
1020                         return false;
1021         }
1022
1023         if (efer & EFER_SVME) {
1024                 struct kvm_cpuid_entry2 *feat;
1025
1026                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1027                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1028                         return false;
1029         }
1030
1031         return true;
1032 }
1033 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1034
1035 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1036 {
1037         u64 old_efer = vcpu->arch.efer;
1038
1039         if (!kvm_valid_efer(vcpu, efer))
1040                 return 1;
1041
1042         if (is_paging(vcpu)
1043             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1044                 return 1;
1045
1046         efer &= ~EFER_LMA;
1047         efer |= vcpu->arch.efer & EFER_LMA;
1048
1049         kvm_x86_ops->set_efer(vcpu, efer);
1050
1051         /* Update reserved bits */
1052         if ((efer ^ old_efer) & EFER_NX)
1053                 kvm_mmu_reset_context(vcpu);
1054
1055         return 0;
1056 }
1057
1058 void kvm_enable_efer_bits(u64 mask)
1059 {
1060        efer_reserved_bits &= ~mask;
1061 }
1062 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1063
1064 /*
1065  * Writes msr value into into the appropriate "register".
1066  * Returns 0 on success, non-0 otherwise.
1067  * Assumes vcpu_load() was already called.
1068  */
1069 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1070 {
1071         switch (msr->index) {
1072         case MSR_FS_BASE:
1073         case MSR_GS_BASE:
1074         case MSR_KERNEL_GS_BASE:
1075         case MSR_CSTAR:
1076         case MSR_LSTAR:
1077                 if (is_noncanonical_address(msr->data))
1078                         return 1;
1079                 break;
1080         case MSR_IA32_SYSENTER_EIP:
1081         case MSR_IA32_SYSENTER_ESP:
1082                 /*
1083                  * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1084                  * non-canonical address is written on Intel but not on
1085                  * AMD (which ignores the top 32-bits, because it does
1086                  * not implement 64-bit SYSENTER).
1087                  *
1088                  * 64-bit code should hence be able to write a non-canonical
1089                  * value on AMD.  Making the address canonical ensures that
1090                  * vmentry does not fail on Intel after writing a non-canonical
1091                  * value, and that something deterministic happens if the guest
1092                  * invokes 64-bit SYSENTER.
1093                  */
1094                 msr->data = get_canonical(msr->data);
1095         }
1096         return kvm_x86_ops->set_msr(vcpu, msr);
1097 }
1098 EXPORT_SYMBOL_GPL(kvm_set_msr);
1099
1100 /*
1101  * Adapt set_msr() to msr_io()'s calling convention
1102  */
1103 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1104 {
1105         struct msr_data msr;
1106         int r;
1107
1108         msr.index = index;
1109         msr.host_initiated = true;
1110         r = kvm_get_msr(vcpu, &msr);
1111         if (r)
1112                 return r;
1113
1114         *data = msr.data;
1115         return 0;
1116 }
1117
1118 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1119 {
1120         struct msr_data msr;
1121
1122         msr.data = *data;
1123         msr.index = index;
1124         msr.host_initiated = true;
1125         return kvm_set_msr(vcpu, &msr);
1126 }
1127
1128 #ifdef CONFIG_X86_64
1129 struct pvclock_gtod_data {
1130         seqcount_t      seq;
1131
1132         struct { /* extract of a clocksource struct */
1133                 int vclock_mode;
1134                 u64     cycle_last;
1135                 u64     mask;
1136                 u32     mult;
1137                 u32     shift;
1138         } clock;
1139
1140         u64             boot_ns;
1141         u64             nsec_base;
1142 };
1143
1144 static struct pvclock_gtod_data pvclock_gtod_data;
1145
1146 static void update_pvclock_gtod(struct timekeeper *tk)
1147 {
1148         struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1149         u64 boot_ns;
1150
1151         boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1152
1153         write_seqcount_begin(&vdata->seq);
1154
1155         /* copy pvclock gtod data */
1156         vdata->clock.vclock_mode        = tk->tkr_mono.clock->archdata.vclock_mode;
1157         vdata->clock.cycle_last         = tk->tkr_mono.cycle_last;
1158         vdata->clock.mask               = tk->tkr_mono.mask;
1159         vdata->clock.mult               = tk->tkr_mono.mult;
1160         vdata->clock.shift              = tk->tkr_mono.shift;
1161
1162         vdata->boot_ns                  = boot_ns;
1163         vdata->nsec_base                = tk->tkr_mono.xtime_nsec;
1164
1165         write_seqcount_end(&vdata->seq);
1166 }
1167 #endif
1168
1169 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1170 {
1171         /*
1172          * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1173          * vcpu_enter_guest.  This function is only called from
1174          * the physical CPU that is running vcpu.
1175          */
1176         kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1177 }
1178
1179 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1180 {
1181         int version;
1182         int r;
1183         struct pvclock_wall_clock wc;
1184         struct timespec64 boot;
1185
1186         if (!wall_clock)
1187                 return;
1188
1189         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1190         if (r)
1191                 return;
1192
1193         if (version & 1)
1194                 ++version;  /* first time write, random junk */
1195
1196         ++version;
1197
1198         if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1199                 return;
1200
1201         /*
1202          * The guest calculates current wall clock time by adding
1203          * system time (updated by kvm_guest_time_update below) to the
1204          * wall clock specified here.  guest system time equals host
1205          * system time for us, thus we must fill in host boot time here.
1206          */
1207         getboottime64(&boot);
1208
1209         if (kvm->arch.kvmclock_offset) {
1210                 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1211                 boot = timespec64_sub(boot, ts);
1212         }
1213         wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1214         wc.nsec = boot.tv_nsec;
1215         wc.version = version;
1216
1217         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1218
1219         version++;
1220         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1221 }
1222
1223 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1224 {
1225         do_shl32_div32(dividend, divisor);
1226         return dividend;
1227 }
1228
1229 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1230                                s8 *pshift, u32 *pmultiplier)
1231 {
1232         uint64_t scaled64;
1233         int32_t  shift = 0;
1234         uint64_t tps64;
1235         uint32_t tps32;
1236
1237         tps64 = base_hz;
1238         scaled64 = scaled_hz;
1239         while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1240                 tps64 >>= 1;
1241                 shift--;
1242         }
1243
1244         tps32 = (uint32_t)tps64;
1245         while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1246                 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1247                         scaled64 >>= 1;
1248                 else
1249                         tps32 <<= 1;
1250                 shift++;
1251         }
1252
1253         *pshift = shift;
1254         *pmultiplier = div_frac(scaled64, tps32);
1255
1256         pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1257                  __func__, base_hz, scaled_hz, shift, *pmultiplier);
1258 }
1259
1260 #ifdef CONFIG_X86_64
1261 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1262 #endif
1263
1264 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1265 static unsigned long max_tsc_khz;
1266
1267 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1268 {
1269         u64 v = (u64)khz * (1000000 + ppm);
1270         do_div(v, 1000000);
1271         return v;
1272 }
1273
1274 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1275 {
1276         u64 ratio;
1277
1278         /* Guest TSC same frequency as host TSC? */
1279         if (!scale) {
1280                 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1281                 return 0;
1282         }
1283
1284         /* TSC scaling supported? */
1285         if (!kvm_has_tsc_control) {
1286                 if (user_tsc_khz > tsc_khz) {
1287                         vcpu->arch.tsc_catchup = 1;
1288                         vcpu->arch.tsc_always_catchup = 1;
1289                         return 0;
1290                 } else {
1291                         WARN(1, "user requested TSC rate below hardware speed\n");
1292                         return -1;
1293                 }
1294         }
1295
1296         /* TSC scaling required  - calculate ratio */
1297         ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1298                                 user_tsc_khz, tsc_khz);
1299
1300         if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1301                 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1302                           user_tsc_khz);
1303                 return -1;
1304         }
1305
1306         vcpu->arch.tsc_scaling_ratio = ratio;
1307         return 0;
1308 }
1309
1310 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1311 {
1312         u32 thresh_lo, thresh_hi;
1313         int use_scaling = 0;
1314
1315         /* tsc_khz can be zero if TSC calibration fails */
1316         if (user_tsc_khz == 0) {
1317                 /* set tsc_scaling_ratio to a safe value */
1318                 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1319                 return -1;
1320         }
1321
1322         /* Compute a scale to convert nanoseconds in TSC cycles */
1323         kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1324                            &vcpu->arch.virtual_tsc_shift,
1325                            &vcpu->arch.virtual_tsc_mult);
1326         vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1327
1328         /*
1329          * Compute the variation in TSC rate which is acceptable
1330          * within the range of tolerance and decide if the
1331          * rate being applied is within that bounds of the hardware
1332          * rate.  If so, no scaling or compensation need be done.
1333          */
1334         thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1335         thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1336         if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1337                 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1338                 use_scaling = 1;
1339         }
1340         return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1341 }
1342
1343 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1344 {
1345         u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1346                                       vcpu->arch.virtual_tsc_mult,
1347                                       vcpu->arch.virtual_tsc_shift);
1348         tsc += vcpu->arch.this_tsc_write;
1349         return tsc;
1350 }
1351
1352 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1353 {
1354 #ifdef CONFIG_X86_64
1355         bool vcpus_matched;
1356         struct kvm_arch *ka = &vcpu->kvm->arch;
1357         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1358
1359         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1360                          atomic_read(&vcpu->kvm->online_vcpus));
1361
1362         /*
1363          * Once the masterclock is enabled, always perform request in
1364          * order to update it.
1365          *
1366          * In order to enable masterclock, the host clocksource must be TSC
1367          * and the vcpus need to have matched TSCs.  When that happens,
1368          * perform request to enable masterclock.
1369          */
1370         if (ka->use_master_clock ||
1371             (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1372                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1373
1374         trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1375                             atomic_read(&vcpu->kvm->online_vcpus),
1376                             ka->use_master_clock, gtod->clock.vclock_mode);
1377 #endif
1378 }
1379
1380 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1381 {
1382         u64 curr_offset = vcpu->arch.tsc_offset;
1383         vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1384 }
1385
1386 /*
1387  * Multiply tsc by a fixed point number represented by ratio.
1388  *
1389  * The most significant 64-N bits (mult) of ratio represent the
1390  * integral part of the fixed point number; the remaining N bits
1391  * (frac) represent the fractional part, ie. ratio represents a fixed
1392  * point number (mult + frac * 2^(-N)).
1393  *
1394  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1395  */
1396 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1397 {
1398         return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1399 }
1400
1401 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1402 {
1403         u64 _tsc = tsc;
1404         u64 ratio = vcpu->arch.tsc_scaling_ratio;
1405
1406         if (ratio != kvm_default_tsc_scaling_ratio)
1407                 _tsc = __scale_tsc(ratio, tsc);
1408
1409         return _tsc;
1410 }
1411 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1412
1413 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1414 {
1415         u64 tsc;
1416
1417         tsc = kvm_scale_tsc(vcpu, rdtsc());
1418
1419         return target_tsc - tsc;
1420 }
1421
1422 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1423 {
1424         return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1425 }
1426 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1427
1428 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1429 {
1430         kvm_x86_ops->write_tsc_offset(vcpu, offset);
1431         vcpu->arch.tsc_offset = offset;
1432 }
1433
1434 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1435 {
1436         struct kvm *kvm = vcpu->kvm;
1437         u64 offset, ns, elapsed;
1438         unsigned long flags;
1439         s64 usdiff;
1440         bool matched;
1441         bool already_matched;
1442         u64 data = msr->data;
1443
1444         raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1445         offset = kvm_compute_tsc_offset(vcpu, data);
1446         ns = ktime_get_boot_ns();
1447         elapsed = ns - kvm->arch.last_tsc_nsec;
1448
1449         if (vcpu->arch.virtual_tsc_khz) {
1450                 int faulted = 0;
1451
1452                 /* n.b - signed multiplication and division required */
1453                 usdiff = data - kvm->arch.last_tsc_write;
1454 #ifdef CONFIG_X86_64
1455                 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1456 #else
1457                 /* do_div() only does unsigned */
1458                 asm("1: idivl %[divisor]\n"
1459                     "2: xor %%edx, %%edx\n"
1460                     "   movl $0, %[faulted]\n"
1461                     "3:\n"
1462                     ".section .fixup,\"ax\"\n"
1463                     "4: movl $1, %[faulted]\n"
1464                     "   jmp  3b\n"
1465                     ".previous\n"
1466
1467                 _ASM_EXTABLE(1b, 4b)
1468
1469                 : "=A"(usdiff), [faulted] "=r" (faulted)
1470                 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1471
1472 #endif
1473                 do_div(elapsed, 1000);
1474                 usdiff -= elapsed;
1475                 if (usdiff < 0)
1476                         usdiff = -usdiff;
1477
1478                 /* idivl overflow => difference is larger than USEC_PER_SEC */
1479                 if (faulted)
1480                         usdiff = USEC_PER_SEC;
1481         } else
1482                 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1483
1484         /*
1485          * Special case: TSC write with a small delta (1 second) of virtual
1486          * cycle time against real time is interpreted as an attempt to
1487          * synchronize the CPU.
1488          *
1489          * For a reliable TSC, we can match TSC offsets, and for an unstable
1490          * TSC, we add elapsed time in this computation.  We could let the
1491          * compensation code attempt to catch up if we fall behind, but
1492          * it's better to try to match offsets from the beginning.
1493          */
1494         if (usdiff < USEC_PER_SEC &&
1495             vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1496                 if (!check_tsc_unstable()) {
1497                         offset = kvm->arch.cur_tsc_offset;
1498                         pr_debug("kvm: matched tsc offset for %llu\n", data);
1499                 } else {
1500                         u64 delta = nsec_to_cycles(vcpu, elapsed);
1501                         data += delta;
1502                         offset = kvm_compute_tsc_offset(vcpu, data);
1503                         pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1504                 }
1505                 matched = true;
1506                 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1507         } else {
1508                 /*
1509                  * We split periods of matched TSC writes into generations.
1510                  * For each generation, we track the original measured
1511                  * nanosecond time, offset, and write, so if TSCs are in
1512                  * sync, we can match exact offset, and if not, we can match
1513                  * exact software computation in compute_guest_tsc()
1514                  *
1515                  * These values are tracked in kvm->arch.cur_xxx variables.
1516                  */
1517                 kvm->arch.cur_tsc_generation++;
1518                 kvm->arch.cur_tsc_nsec = ns;
1519                 kvm->arch.cur_tsc_write = data;
1520                 kvm->arch.cur_tsc_offset = offset;
1521                 matched = false;
1522                 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1523                          kvm->arch.cur_tsc_generation, data);
1524         }
1525
1526         /*
1527          * We also track th most recent recorded KHZ, write and time to
1528          * allow the matching interval to be extended at each write.
1529          */
1530         kvm->arch.last_tsc_nsec = ns;
1531         kvm->arch.last_tsc_write = data;
1532         kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1533
1534         vcpu->arch.last_guest_tsc = data;
1535
1536         /* Keep track of which generation this VCPU has synchronized to */
1537         vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1538         vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1539         vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1540
1541         if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1542                 update_ia32_tsc_adjust_msr(vcpu, offset);
1543         kvm_vcpu_write_tsc_offset(vcpu, offset);
1544         raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1545
1546         spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1547         if (!matched) {
1548                 kvm->arch.nr_vcpus_matched_tsc = 0;
1549         } else if (!already_matched) {
1550                 kvm->arch.nr_vcpus_matched_tsc++;
1551         }
1552
1553         kvm_track_tsc_matching(vcpu);
1554         spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1555 }
1556
1557 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1558
1559 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1560                                            s64 adjustment)
1561 {
1562         kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1563 }
1564
1565 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1566 {
1567         if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1568                 WARN_ON(adjustment < 0);
1569         adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1570         adjust_tsc_offset_guest(vcpu, adjustment);
1571 }
1572
1573 #ifdef CONFIG_X86_64
1574
1575 static u64 read_tsc(void)
1576 {
1577         u64 ret = (u64)rdtsc_ordered();
1578         u64 last = pvclock_gtod_data.clock.cycle_last;
1579
1580         if (likely(ret >= last))
1581                 return ret;
1582
1583         /*
1584          * GCC likes to generate cmov here, but this branch is extremely
1585          * predictable (it's just a function of time and the likely is
1586          * very likely) and there's a data dependence, so force GCC
1587          * to generate a branch instead.  I don't barrier() because
1588          * we don't actually need a barrier, and if this function
1589          * ever gets inlined it will generate worse code.
1590          */
1591         asm volatile ("");
1592         return last;
1593 }
1594
1595 static inline u64 vgettsc(u64 *cycle_now)
1596 {
1597         long v;
1598         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1599
1600         *cycle_now = read_tsc();
1601
1602         v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1603         return v * gtod->clock.mult;
1604 }
1605
1606 static int do_monotonic_boot(s64 *t, u64 *cycle_now)
1607 {
1608         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1609         unsigned long seq;
1610         int mode;
1611         u64 ns;
1612
1613         do {
1614                 seq = read_seqcount_begin(&gtod->seq);
1615                 mode = gtod->clock.vclock_mode;
1616                 ns = gtod->nsec_base;
1617                 ns += vgettsc(cycle_now);
1618                 ns >>= gtod->clock.shift;
1619                 ns += gtod->boot_ns;
1620         } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1621         *t = ns;
1622
1623         return mode;
1624 }
1625
1626 /* returns true if host is using tsc clocksource */
1627 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
1628 {
1629         /* checked again under seqlock below */
1630         if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1631                 return false;
1632
1633         return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1634 }
1635 #endif
1636
1637 /*
1638  *
1639  * Assuming a stable TSC across physical CPUS, and a stable TSC
1640  * across virtual CPUs, the following condition is possible.
1641  * Each numbered line represents an event visible to both
1642  * CPUs at the next numbered event.
1643  *
1644  * "timespecX" represents host monotonic time. "tscX" represents
1645  * RDTSC value.
1646  *
1647  *              VCPU0 on CPU0           |       VCPU1 on CPU1
1648  *
1649  * 1.  read timespec0,tsc0
1650  * 2.                                   | timespec1 = timespec0 + N
1651  *                                      | tsc1 = tsc0 + M
1652  * 3. transition to guest               | transition to guest
1653  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1654  * 5.                                   | ret1 = timespec1 + (rdtsc - tsc1)
1655  *                                      | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1656  *
1657  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1658  *
1659  *      - ret0 < ret1
1660  *      - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1661  *              ...
1662  *      - 0 < N - M => M < N
1663  *
1664  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1665  * always the case (the difference between two distinct xtime instances
1666  * might be smaller then the difference between corresponding TSC reads,
1667  * when updating guest vcpus pvclock areas).
1668  *
1669  * To avoid that problem, do not allow visibility of distinct
1670  * system_timestamp/tsc_timestamp values simultaneously: use a master
1671  * copy of host monotonic time values. Update that master copy
1672  * in lockstep.
1673  *
1674  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1675  *
1676  */
1677
1678 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1679 {
1680 #ifdef CONFIG_X86_64
1681         struct kvm_arch *ka = &kvm->arch;
1682         int vclock_mode;
1683         bool host_tsc_clocksource, vcpus_matched;
1684
1685         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1686                         atomic_read(&kvm->online_vcpus));
1687
1688         /*
1689          * If the host uses TSC clock, then passthrough TSC as stable
1690          * to the guest.
1691          */
1692         host_tsc_clocksource = kvm_get_time_and_clockread(
1693                                         &ka->master_kernel_ns,
1694                                         &ka->master_cycle_now);
1695
1696         ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1697                                 && !backwards_tsc_observed
1698                                 && !ka->boot_vcpu_runs_old_kvmclock;
1699
1700         if (ka->use_master_clock)
1701                 atomic_set(&kvm_guest_has_master_clock, 1);
1702
1703         vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1704         trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1705                                         vcpus_matched);
1706 #endif
1707 }
1708
1709 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1710 {
1711         kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1712 }
1713
1714 static void kvm_gen_update_masterclock(struct kvm *kvm)
1715 {
1716 #ifdef CONFIG_X86_64
1717         int i;
1718         struct kvm_vcpu *vcpu;
1719         struct kvm_arch *ka = &kvm->arch;
1720
1721         spin_lock(&ka->pvclock_gtod_sync_lock);
1722         kvm_make_mclock_inprogress_request(kvm);
1723         /* no guest entries from this point */
1724         pvclock_update_vm_gtod_copy(kvm);
1725
1726         kvm_for_each_vcpu(i, vcpu, kvm)
1727                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1728
1729         /* guest entries allowed */
1730         kvm_for_each_vcpu(i, vcpu, kvm)
1731                 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1732
1733         spin_unlock(&ka->pvclock_gtod_sync_lock);
1734 #endif
1735 }
1736
1737 static u64 __get_kvmclock_ns(struct kvm *kvm)
1738 {
1739         struct kvm_arch *ka = &kvm->arch;
1740         struct pvclock_vcpu_time_info hv_clock;
1741
1742         spin_lock(&ka->pvclock_gtod_sync_lock);
1743         if (!ka->use_master_clock) {
1744                 spin_unlock(&ka->pvclock_gtod_sync_lock);
1745                 return ktime_get_boot_ns() + ka->kvmclock_offset;
1746         }
1747
1748         hv_clock.tsc_timestamp = ka->master_cycle_now;
1749         hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1750         spin_unlock(&ka->pvclock_gtod_sync_lock);
1751
1752         kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1753                            &hv_clock.tsc_shift,
1754                            &hv_clock.tsc_to_system_mul);
1755         return __pvclock_read_cycles(&hv_clock, rdtsc());
1756 }
1757
1758 u64 get_kvmclock_ns(struct kvm *kvm)
1759 {
1760         unsigned long flags;
1761         s64 ns;
1762
1763         local_irq_save(flags);
1764         ns = __get_kvmclock_ns(kvm);
1765         local_irq_restore(flags);
1766
1767         return ns;
1768 }
1769
1770 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1771 {
1772         struct kvm_vcpu_arch *vcpu = &v->arch;
1773         struct pvclock_vcpu_time_info guest_hv_clock;
1774
1775         if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1776                 &guest_hv_clock, sizeof(guest_hv_clock))))
1777                 return;
1778
1779         /* This VCPU is paused, but it's legal for a guest to read another
1780          * VCPU's kvmclock, so we really have to follow the specification where
1781          * it says that version is odd if data is being modified, and even after
1782          * it is consistent.
1783          *
1784          * Version field updates must be kept separate.  This is because
1785          * kvm_write_guest_cached might use a "rep movs" instruction, and
1786          * writes within a string instruction are weakly ordered.  So there
1787          * are three writes overall.
1788          *
1789          * As a small optimization, only write the version field in the first
1790          * and third write.  The vcpu->pv_time cache is still valid, because the
1791          * version field is the first in the struct.
1792          */
1793         BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1794
1795         vcpu->hv_clock.version = guest_hv_clock.version + 1;
1796         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1797                                 &vcpu->hv_clock,
1798                                 sizeof(vcpu->hv_clock.version));
1799
1800         smp_wmb();
1801
1802         /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1803         vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1804
1805         if (vcpu->pvclock_set_guest_stopped_request) {
1806                 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1807                 vcpu->pvclock_set_guest_stopped_request = false;
1808         }
1809
1810         trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1811
1812         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1813                                 &vcpu->hv_clock,
1814                                 sizeof(vcpu->hv_clock));
1815
1816         smp_wmb();
1817
1818         vcpu->hv_clock.version++;
1819         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1820                                 &vcpu->hv_clock,
1821                                 sizeof(vcpu->hv_clock.version));
1822 }
1823
1824 static int kvm_guest_time_update(struct kvm_vcpu *v)
1825 {
1826         unsigned long flags, tgt_tsc_khz;
1827         struct kvm_vcpu_arch *vcpu = &v->arch;
1828         struct kvm_arch *ka = &v->kvm->arch;
1829         s64 kernel_ns;
1830         u64 tsc_timestamp, host_tsc;
1831         u8 pvclock_flags;
1832         bool use_master_clock;
1833
1834         kernel_ns = 0;
1835         host_tsc = 0;
1836
1837         /*
1838          * If the host uses TSC clock, then passthrough TSC as stable
1839          * to the guest.
1840          */
1841         spin_lock(&ka->pvclock_gtod_sync_lock);
1842         use_master_clock = ka->use_master_clock;
1843         if (use_master_clock) {
1844                 host_tsc = ka->master_cycle_now;
1845                 kernel_ns = ka->master_kernel_ns;
1846         }
1847         spin_unlock(&ka->pvclock_gtod_sync_lock);
1848
1849         /* Keep irq disabled to prevent changes to the clock */
1850         local_irq_save(flags);
1851         tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1852         if (unlikely(tgt_tsc_khz == 0)) {
1853                 local_irq_restore(flags);
1854                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1855                 return 1;
1856         }
1857         if (!use_master_clock) {
1858                 host_tsc = rdtsc();
1859                 kernel_ns = ktime_get_boot_ns();
1860         }
1861
1862         tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1863
1864         /*
1865          * We may have to catch up the TSC to match elapsed wall clock
1866          * time for two reasons, even if kvmclock is used.
1867          *   1) CPU could have been running below the maximum TSC rate
1868          *   2) Broken TSC compensation resets the base at each VCPU
1869          *      entry to avoid unknown leaps of TSC even when running
1870          *      again on the same CPU.  This may cause apparent elapsed
1871          *      time to disappear, and the guest to stand still or run
1872          *      very slowly.
1873          */
1874         if (vcpu->tsc_catchup) {
1875                 u64 tsc = compute_guest_tsc(v, kernel_ns);
1876                 if (tsc > tsc_timestamp) {
1877                         adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1878                         tsc_timestamp = tsc;
1879                 }
1880         }
1881
1882         local_irq_restore(flags);
1883
1884         /* With all the info we got, fill in the values */
1885
1886         if (kvm_has_tsc_control)
1887                 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1888
1889         if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1890                 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1891                                    &vcpu->hv_clock.tsc_shift,
1892                                    &vcpu->hv_clock.tsc_to_system_mul);
1893                 vcpu->hw_tsc_khz = tgt_tsc_khz;
1894         }
1895
1896         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1897         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1898         vcpu->last_guest_tsc = tsc_timestamp;
1899
1900         /* If the host uses TSC clocksource, then it is stable */
1901         pvclock_flags = 0;
1902         if (use_master_clock)
1903                 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1904
1905         vcpu->hv_clock.flags = pvclock_flags;
1906
1907         if (vcpu->pv_time_enabled)
1908                 kvm_setup_pvclock_page(v);
1909         if (v == kvm_get_vcpu(v->kvm, 0))
1910                 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
1911         return 0;
1912 }
1913
1914 /*
1915  * kvmclock updates which are isolated to a given vcpu, such as
1916  * vcpu->cpu migration, should not allow system_timestamp from
1917  * the rest of the vcpus to remain static. Otherwise ntp frequency
1918  * correction applies to one vcpu's system_timestamp but not
1919  * the others.
1920  *
1921  * So in those cases, request a kvmclock update for all vcpus.
1922  * We need to rate-limit these requests though, as they can
1923  * considerably slow guests that have a large number of vcpus.
1924  * The time for a remote vcpu to update its kvmclock is bound
1925  * by the delay we use to rate-limit the updates.
1926  */
1927
1928 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1929
1930 static void kvmclock_update_fn(struct work_struct *work)
1931 {
1932         int i;
1933         struct delayed_work *dwork = to_delayed_work(work);
1934         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1935                                            kvmclock_update_work);
1936         struct kvm *kvm = container_of(ka, struct kvm, arch);
1937         struct kvm_vcpu *vcpu;
1938
1939         kvm_for_each_vcpu(i, vcpu, kvm) {
1940                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1941                 kvm_vcpu_kick(vcpu);
1942         }
1943 }
1944
1945 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1946 {
1947         struct kvm *kvm = v->kvm;
1948
1949         kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1950         schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1951                                         KVMCLOCK_UPDATE_DELAY);
1952 }
1953
1954 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1955
1956 static void kvmclock_sync_fn(struct work_struct *work)
1957 {
1958         struct delayed_work *dwork = to_delayed_work(work);
1959         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1960                                            kvmclock_sync_work);
1961         struct kvm *kvm = container_of(ka, struct kvm, arch);
1962
1963         if (!kvmclock_periodic_sync)
1964                 return;
1965
1966         schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1967         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1968                                         KVMCLOCK_SYNC_PERIOD);
1969 }
1970
1971 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1972 {
1973         u64 mcg_cap = vcpu->arch.mcg_cap;
1974         unsigned bank_num = mcg_cap & 0xff;
1975
1976         switch (msr) {
1977         case MSR_IA32_MCG_STATUS:
1978                 vcpu->arch.mcg_status = data;
1979                 break;
1980         case MSR_IA32_MCG_CTL:
1981                 if (!(mcg_cap & MCG_CTL_P))
1982                         return 1;
1983                 if (data != 0 && data != ~(u64)0)
1984                         return -1;
1985                 vcpu->arch.mcg_ctl = data;
1986                 break;
1987         default:
1988                 if (msr >= MSR_IA32_MC0_CTL &&
1989                     msr < MSR_IA32_MCx_CTL(bank_num)) {
1990                         u32 offset = msr - MSR_IA32_MC0_CTL;
1991                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1992                          * some Linux kernels though clear bit 10 in bank 4 to
1993                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1994                          * this to avoid an uncatched #GP in the guest
1995                          */
1996                         if ((offset & 0x3) == 0 &&
1997                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1998                                 return -1;
1999                         vcpu->arch.mce_banks[offset] = data;
2000                         break;
2001                 }
2002                 return 1;
2003         }
2004         return 0;
2005 }
2006
2007 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2008 {
2009         struct kvm *kvm = vcpu->kvm;
2010         int lm = is_long_mode(vcpu);
2011         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2012                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2013         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2014                 : kvm->arch.xen_hvm_config.blob_size_32;
2015         u32 page_num = data & ~PAGE_MASK;
2016         u64 page_addr = data & PAGE_MASK;
2017         u8 *page;
2018         int r;
2019
2020         r = -E2BIG;
2021         if (page_num >= blob_size)
2022                 goto out;
2023         r = -ENOMEM;
2024         page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2025         if (IS_ERR(page)) {
2026                 r = PTR_ERR(page);
2027                 goto out;
2028         }
2029         if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2030                 goto out_free;
2031         r = 0;
2032 out_free:
2033         kfree(page);
2034 out:
2035         return r;
2036 }
2037
2038 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2039 {
2040         gpa_t gpa = data & ~0x3f;
2041
2042         /* Bits 2:5 are reserved, Should be zero */
2043         if (data & 0x3c)
2044                 return 1;
2045
2046         vcpu->arch.apf.msr_val = data;
2047
2048         if (!(data & KVM_ASYNC_PF_ENABLED)) {
2049                 kvm_clear_async_pf_completion_queue(vcpu);
2050                 kvm_async_pf_hash_reset(vcpu);
2051                 return 0;
2052         }
2053
2054         if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2055                                         sizeof(u32)))
2056                 return 1;
2057
2058         vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2059         kvm_async_pf_wakeup_all(vcpu);
2060         return 0;
2061 }
2062
2063 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2064 {
2065         vcpu->arch.pv_time_enabled = false;
2066 }
2067
2068 static void record_steal_time(struct kvm_vcpu *vcpu)
2069 {
2070         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2071                 return;
2072
2073         if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2074                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2075                 return;
2076
2077         vcpu->arch.st.steal.preempted = 0;
2078
2079         if (vcpu->arch.st.steal.version & 1)
2080                 vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2081
2082         vcpu->arch.st.steal.version += 1;
2083
2084         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2085                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2086
2087         smp_wmb();
2088
2089         vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2090                 vcpu->arch.st.last_steal;
2091         vcpu->arch.st.last_steal = current->sched_info.run_delay;
2092
2093         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2094                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2095
2096         smp_wmb();
2097
2098         vcpu->arch.st.steal.version += 1;
2099
2100         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2101                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2102 }
2103
2104 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2105 {
2106         bool pr = false;
2107         u32 msr = msr_info->index;
2108         u64 data = msr_info->data;
2109
2110         switch (msr) {
2111         case MSR_AMD64_NB_CFG:
2112         case MSR_IA32_UCODE_REV:
2113         case MSR_IA32_UCODE_WRITE:
2114         case MSR_VM_HSAVE_PA:
2115         case MSR_AMD64_PATCH_LOADER:
2116         case MSR_AMD64_BU_CFG2:
2117                 break;
2118
2119         case MSR_EFER:
2120                 return set_efer(vcpu, data);
2121         case MSR_K7_HWCR:
2122                 data &= ~(u64)0x40;     /* ignore flush filter disable */
2123                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
2124                 data &= ~(u64)0x8;      /* ignore TLB cache disable */
2125                 data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2126                 if (data != 0) {
2127                         vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2128                                     data);
2129                         return 1;
2130                 }
2131                 break;
2132         case MSR_FAM10H_MMIO_CONF_BASE:
2133                 if (data != 0) {
2134                         vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2135                                     "0x%llx\n", data);
2136                         return 1;
2137                 }
2138                 break;
2139         case MSR_IA32_DEBUGCTLMSR:
2140                 if (!data) {
2141                         /* We support the non-activated case already */
2142                         break;
2143                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2144                         /* Values other than LBR and BTF are vendor-specific,
2145                            thus reserved and should throw a #GP */
2146                         return 1;
2147                 }
2148                 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2149                             __func__, data);
2150                 break;
2151         case 0x200 ... 0x2ff:
2152                 return kvm_mtrr_set_msr(vcpu, msr, data);
2153         case MSR_IA32_APICBASE:
2154                 return kvm_set_apic_base(vcpu, msr_info);
2155         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2156                 return kvm_x2apic_msr_write(vcpu, msr, data);
2157         case MSR_IA32_TSCDEADLINE:
2158                 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2159                 break;
2160         case MSR_IA32_TSC_ADJUST:
2161                 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2162                         if (!msr_info->host_initiated) {
2163                                 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2164                                 adjust_tsc_offset_guest(vcpu, adj);
2165                         }
2166                         vcpu->arch.ia32_tsc_adjust_msr = data;
2167                 }
2168                 break;
2169         case MSR_IA32_MISC_ENABLE:
2170                 vcpu->arch.ia32_misc_enable_msr = data;
2171                 break;
2172         case MSR_IA32_SMBASE:
2173                 if (!msr_info->host_initiated)
2174                         return 1;
2175                 vcpu->arch.smbase = data;
2176                 break;
2177         case MSR_KVM_WALL_CLOCK_NEW:
2178         case MSR_KVM_WALL_CLOCK:
2179                 vcpu->kvm->arch.wall_clock = data;
2180                 kvm_write_wall_clock(vcpu->kvm, data);
2181                 break;
2182         case MSR_KVM_SYSTEM_TIME_NEW:
2183         case MSR_KVM_SYSTEM_TIME: {
2184                 struct kvm_arch *ka = &vcpu->kvm->arch;
2185
2186                 kvmclock_reset(vcpu);
2187
2188                 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2189                         bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2190
2191                         if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2192                                 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2193                                         &vcpu->requests);
2194
2195                         ka->boot_vcpu_runs_old_kvmclock = tmp;
2196                 }
2197
2198                 vcpu->arch.time = data;
2199                 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2200
2201                 /* we verify if the enable bit is set... */
2202                 if (!(data & 1))
2203                         break;
2204
2205                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2206                      &vcpu->arch.pv_time, data & ~1ULL,
2207                      sizeof(struct pvclock_vcpu_time_info)))
2208                         vcpu->arch.pv_time_enabled = false;
2209                 else
2210                         vcpu->arch.pv_time_enabled = true;
2211
2212                 break;
2213         }
2214         case MSR_KVM_ASYNC_PF_EN:
2215                 if (kvm_pv_enable_async_pf(vcpu, data))
2216                         return 1;
2217                 break;
2218         case MSR_KVM_STEAL_TIME:
2219
2220                 if (unlikely(!sched_info_on()))
2221                         return 1;
2222
2223                 if (data & KVM_STEAL_RESERVED_MASK)
2224                         return 1;
2225
2226                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2227                                                 data & KVM_STEAL_VALID_BITS,
2228                                                 sizeof(struct kvm_steal_time)))
2229                         return 1;
2230
2231                 vcpu->arch.st.msr_val = data;
2232
2233                 if (!(data & KVM_MSR_ENABLED))
2234                         break;
2235
2236                 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2237
2238                 break;
2239         case MSR_KVM_PV_EOI_EN:
2240                 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2241                         return 1;
2242                 break;
2243
2244         case MSR_IA32_MCG_CTL:
2245         case MSR_IA32_MCG_STATUS:
2246         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2247                 return set_msr_mce(vcpu, msr, data);
2248
2249         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2250         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2251                 pr = true; /* fall through */
2252         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2253         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2254                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2255                         return kvm_pmu_set_msr(vcpu, msr_info);
2256
2257                 if (pr || data != 0)
2258                         vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2259                                     "0x%x data 0x%llx\n", msr, data);
2260                 break;
2261         case MSR_K7_CLK_CTL:
2262                 /*
2263                  * Ignore all writes to this no longer documented MSR.
2264                  * Writes are only relevant for old K7 processors,
2265                  * all pre-dating SVM, but a recommended workaround from
2266                  * AMD for these chips. It is possible to specify the
2267                  * affected processor models on the command line, hence
2268                  * the need to ignore the workaround.
2269                  */
2270                 break;
2271         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2272         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2273         case HV_X64_MSR_CRASH_CTL:
2274         case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2275                 return kvm_hv_set_msr_common(vcpu, msr, data,
2276                                              msr_info->host_initiated);
2277         case MSR_IA32_BBL_CR_CTL3:
2278                 /* Drop writes to this legacy MSR -- see rdmsr
2279                  * counterpart for further detail.
2280                  */
2281                 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", msr, data);
2282                 break;
2283         case MSR_AMD64_OSVW_ID_LENGTH:
2284                 if (!guest_cpuid_has_osvw(vcpu))
2285                         return 1;
2286                 vcpu->arch.osvw.length = data;
2287                 break;
2288         case MSR_AMD64_OSVW_STATUS:
2289                 if (!guest_cpuid_has_osvw(vcpu))
2290                         return 1;
2291                 vcpu->arch.osvw.status = data;
2292                 break;
2293         default:
2294                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2295                         return xen_hvm_config(vcpu, data);
2296                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2297                         return kvm_pmu_set_msr(vcpu, msr_info);
2298                 if (!ignore_msrs) {
2299                         vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2300                                     msr, data);
2301                         return 1;
2302                 } else {
2303                         vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2304                                     msr, data);
2305                         break;
2306                 }
2307         }
2308         return 0;
2309 }
2310 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2311
2312
2313 /*
2314  * Reads an msr value (of 'msr_index') into 'pdata'.
2315  * Returns 0 on success, non-0 otherwise.
2316  * Assumes vcpu_load() was already called.
2317  */
2318 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2319 {
2320         return kvm_x86_ops->get_msr(vcpu, msr);
2321 }
2322 EXPORT_SYMBOL_GPL(kvm_get_msr);
2323
2324 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2325 {
2326         u64 data;
2327         u64 mcg_cap = vcpu->arch.mcg_cap;
2328         unsigned bank_num = mcg_cap & 0xff;
2329
2330         switch (msr) {
2331         case MSR_IA32_P5_MC_ADDR:
2332         case MSR_IA32_P5_MC_TYPE:
2333                 data = 0;
2334                 break;
2335         case MSR_IA32_MCG_CAP:
2336                 data = vcpu->arch.mcg_cap;
2337                 break;
2338         case MSR_IA32_MCG_CTL:
2339                 if (!(mcg_cap & MCG_CTL_P))
2340                         return 1;
2341                 data = vcpu->arch.mcg_ctl;
2342                 break;
2343         case MSR_IA32_MCG_STATUS:
2344                 data = vcpu->arch.mcg_status;
2345                 break;
2346         default:
2347                 if (msr >= MSR_IA32_MC0_CTL &&
2348                     msr < MSR_IA32_MCx_CTL(bank_num)) {
2349                         u32 offset = msr - MSR_IA32_MC0_CTL;
2350                         data = vcpu->arch.mce_banks[offset];
2351                         break;
2352                 }
2353                 return 1;
2354         }
2355         *pdata = data;
2356         return 0;
2357 }
2358
2359 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2360 {
2361         switch (msr_info->index) {
2362         case MSR_IA32_PLATFORM_ID:
2363         case MSR_IA32_EBL_CR_POWERON:
2364         case MSR_IA32_DEBUGCTLMSR:
2365         case MSR_IA32_LASTBRANCHFROMIP:
2366         case MSR_IA32_LASTBRANCHTOIP:
2367         case MSR_IA32_LASTINTFROMIP:
2368         case MSR_IA32_LASTINTTOIP:
2369         case MSR_K8_SYSCFG:
2370         case MSR_K8_TSEG_ADDR:
2371         case MSR_K8_TSEG_MASK:
2372         case MSR_K7_HWCR:
2373         case MSR_VM_HSAVE_PA:
2374         case MSR_K8_INT_PENDING_MSG:
2375         case MSR_AMD64_NB_CFG:
2376         case MSR_FAM10H_MMIO_CONF_BASE:
2377         case MSR_AMD64_BU_CFG2:
2378         case MSR_IA32_PERF_CTL:
2379                 msr_info->data = 0;
2380                 break;
2381         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2382         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2383         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2384         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2385                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2386                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2387                 msr_info->data = 0;
2388                 break;
2389         case MSR_IA32_UCODE_REV:
2390                 msr_info->data = 0x100000000ULL;
2391                 break;
2392         case MSR_MTRRcap:
2393         case 0x200 ... 0x2ff:
2394                 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2395         case 0xcd: /* fsb frequency */
2396                 msr_info->data = 3;
2397                 break;
2398                 /*
2399                  * MSR_EBC_FREQUENCY_ID
2400                  * Conservative value valid for even the basic CPU models.
2401                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2402                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2403                  * and 266MHz for model 3, or 4. Set Core Clock
2404                  * Frequency to System Bus Frequency Ratio to 1 (bits
2405                  * 31:24) even though these are only valid for CPU
2406                  * models > 2, however guests may end up dividing or
2407                  * multiplying by zero otherwise.
2408                  */
2409         case MSR_EBC_FREQUENCY_ID:
2410                 msr_info->data = 1 << 24;
2411                 break;
2412         case MSR_IA32_APICBASE:
2413                 msr_info->data = kvm_get_apic_base(vcpu);
2414                 break;
2415         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2416                 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2417                 break;
2418         case MSR_IA32_TSCDEADLINE:
2419                 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2420                 break;
2421         case MSR_IA32_TSC_ADJUST:
2422                 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2423                 break;
2424         case MSR_IA32_MISC_ENABLE:
2425                 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2426                 break;
2427         case MSR_IA32_SMBASE:
2428                 if (!msr_info->host_initiated)
2429                         return 1;
2430                 msr_info->data = vcpu->arch.smbase;
2431                 break;
2432         case MSR_IA32_PERF_STATUS:
2433                 /* TSC increment by tick */
2434                 msr_info->data = 1000ULL;
2435                 /* CPU multiplier */
2436                 msr_info->data |= (((uint64_t)4ULL) << 40);
2437                 break;
2438         case MSR_EFER:
2439                 msr_info->data = vcpu->arch.efer;
2440                 break;
2441         case MSR_KVM_WALL_CLOCK:
2442         case MSR_KVM_WALL_CLOCK_NEW:
2443                 msr_info->data = vcpu->kvm->arch.wall_clock;
2444                 break;
2445         case MSR_KVM_SYSTEM_TIME:
2446         case MSR_KVM_SYSTEM_TIME_NEW:
2447                 msr_info->data = vcpu->arch.time;
2448                 break;
2449         case MSR_KVM_ASYNC_PF_EN:
2450                 msr_info->data = vcpu->arch.apf.msr_val;
2451                 break;
2452         case MSR_KVM_STEAL_TIME:
2453                 msr_info->data = vcpu->arch.st.msr_val;
2454                 break;
2455         case MSR_KVM_PV_EOI_EN:
2456                 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2457                 break;
2458         case MSR_IA32_P5_MC_ADDR:
2459         case MSR_IA32_P5_MC_TYPE:
2460         case MSR_IA32_MCG_CAP:
2461         case MSR_IA32_MCG_CTL:
2462         case MSR_IA32_MCG_STATUS:
2463         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2464                 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2465         case MSR_K7_CLK_CTL:
2466                 /*
2467                  * Provide expected ramp-up count for K7. All other
2468                  * are set to zero, indicating minimum divisors for
2469                  * every field.
2470                  *
2471                  * This prevents guest kernels on AMD host with CPU
2472                  * type 6, model 8 and higher from exploding due to
2473                  * the rdmsr failing.
2474                  */
2475                 msr_info->data = 0x20000000;
2476                 break;
2477         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2478         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2479         case HV_X64_MSR_CRASH_CTL:
2480         case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2481                 return kvm_hv_get_msr_common(vcpu,
2482                                              msr_info->index, &msr_info->data);
2483                 break;
2484         case MSR_IA32_BBL_CR_CTL3:
2485                 /* This legacy MSR exists but isn't fully documented in current
2486                  * silicon.  It is however accessed by winxp in very narrow
2487                  * scenarios where it sets bit #19, itself documented as
2488                  * a "reserved" bit.  Best effort attempt to source coherent
2489                  * read data here should the balance of the register be
2490                  * interpreted by the guest:
2491                  *
2492                  * L2 cache control register 3: 64GB range, 256KB size,
2493                  * enabled, latency 0x1, configured
2494                  */
2495                 msr_info->data = 0xbe702111;
2496                 break;
2497         case MSR_AMD64_OSVW_ID_LENGTH:
2498                 if (!guest_cpuid_has_osvw(vcpu))
2499                         return 1;
2500                 msr_info->data = vcpu->arch.osvw.length;
2501                 break;
2502         case MSR_AMD64_OSVW_STATUS:
2503                 if (!guest_cpuid_has_osvw(vcpu))
2504                         return 1;
2505                 msr_info->data = vcpu->arch.osvw.status;
2506                 break;
2507         default:
2508                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2509                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2510                 if (!ignore_msrs) {
2511                         vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2512                                                msr_info->index);
2513                         return 1;
2514                 } else {
2515                         vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2516                         msr_info->data = 0;
2517                 }
2518                 break;
2519         }
2520         return 0;
2521 }
2522 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2523
2524 /*
2525  * Read or write a bunch of msrs. All parameters are kernel addresses.
2526  *
2527  * @return number of msrs set successfully.
2528  */
2529 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2530                     struct kvm_msr_entry *entries,
2531                     int (*do_msr)(struct kvm_vcpu *vcpu,
2532                                   unsigned index, u64 *data))
2533 {
2534         int i, idx;
2535
2536         idx = srcu_read_lock(&vcpu->kvm->srcu);
2537         for (i = 0; i < msrs->nmsrs; ++i)
2538                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2539                         break;
2540         srcu_read_unlock(&vcpu->kvm->srcu, idx);
2541
2542         return i;
2543 }
2544
2545 /*
2546  * Read or write a bunch of msrs. Parameters are user addresses.
2547  *
2548  * @return number of msrs set successfully.
2549  */
2550 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2551                   int (*do_msr)(struct kvm_vcpu *vcpu,
2552                                 unsigned index, u64 *data),
2553                   int writeback)
2554 {
2555         struct kvm_msrs msrs;
2556         struct kvm_msr_entry *entries;
2557         int r, n;
2558         unsigned size;
2559
2560         r = -EFAULT;
2561         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2562                 goto out;
2563
2564         r = -E2BIG;
2565         if (msrs.nmsrs >= MAX_IO_MSRS)
2566                 goto out;
2567
2568         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2569         entries = memdup_user(user_msrs->entries, size);
2570         if (IS_ERR(entries)) {
2571                 r = PTR_ERR(entries);
2572                 goto out;
2573         }
2574
2575         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2576         if (r < 0)
2577                 goto out_free;
2578
2579         r = -EFAULT;
2580         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2581                 goto out_free;
2582
2583         r = n;
2584
2585 out_free:
2586         kfree(entries);
2587 out:
2588         return r;
2589 }
2590
2591 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2592 {
2593         int r;
2594
2595         switch (ext) {
2596         case KVM_CAP_IRQCHIP:
2597         case KVM_CAP_HLT:
2598         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2599         case KVM_CAP_SET_TSS_ADDR:
2600         case KVM_CAP_EXT_CPUID:
2601         case KVM_CAP_EXT_EMUL_CPUID:
2602         case KVM_CAP_CLOCKSOURCE:
2603         case KVM_CAP_PIT:
2604         case KVM_CAP_NOP_IO_DELAY:
2605         case KVM_CAP_MP_STATE:
2606         case KVM_CAP_SYNC_MMU:
2607         case KVM_CAP_USER_NMI:
2608         case KVM_CAP_REINJECT_CONTROL:
2609         case KVM_CAP_IRQ_INJECT_STATUS:
2610         case KVM_CAP_IOEVENTFD:
2611         case KVM_CAP_IOEVENTFD_NO_LENGTH:
2612         case KVM_CAP_PIT2:
2613         case KVM_CAP_PIT_STATE2:
2614         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2615         case KVM_CAP_XEN_HVM:
2616         case KVM_CAP_VCPU_EVENTS:
2617         case KVM_CAP_HYPERV:
2618         case KVM_CAP_HYPERV_VAPIC:
2619         case KVM_CAP_HYPERV_SPIN:
2620         case KVM_CAP_HYPERV_SYNIC:
2621         case KVM_CAP_PCI_SEGMENT:
2622         case KVM_CAP_DEBUGREGS:
2623         case KVM_CAP_X86_ROBUST_SINGLESTEP:
2624         case KVM_CAP_XSAVE:
2625         case KVM_CAP_ASYNC_PF:
2626         case KVM_CAP_GET_TSC_KHZ:
2627         case KVM_CAP_KVMCLOCK_CTRL:
2628         case KVM_CAP_READONLY_MEM:
2629         case KVM_CAP_HYPERV_TIME:
2630         case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2631         case KVM_CAP_TSC_DEADLINE_TIMER:
2632         case KVM_CAP_ENABLE_CAP_VM:
2633         case KVM_CAP_DISABLE_QUIRKS:
2634         case KVM_CAP_SET_BOOT_CPU_ID:
2635         case KVM_CAP_SPLIT_IRQCHIP:
2636 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2637         case KVM_CAP_ASSIGN_DEV_IRQ:
2638         case KVM_CAP_PCI_2_3:
2639 #endif
2640                 r = 1;
2641                 break;
2642         case KVM_CAP_ADJUST_CLOCK:
2643                 r = KVM_CLOCK_TSC_STABLE;
2644                 break;
2645         case KVM_CAP_X86_SMM:
2646                 /* SMBASE is usually relocated above 1M on modern chipsets,
2647                  * and SMM handlers might indeed rely on 4G segment limits,
2648                  * so do not report SMM to be available if real mode is
2649                  * emulated via vm86 mode.  Still, do not go to great lengths
2650                  * to avoid userspace's usage of the feature, because it is a
2651                  * fringe case that is not enabled except via specific settings
2652                  * of the module parameters.
2653                  */
2654                 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2655                 break;
2656         case KVM_CAP_COALESCED_MMIO:
2657                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2658                 break;
2659         case KVM_CAP_VAPIC:
2660                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2661                 break;
2662         case KVM_CAP_NR_VCPUS:
2663                 r = KVM_SOFT_MAX_VCPUS;
2664                 break;
2665         case KVM_CAP_MAX_VCPUS:
2666                 r = KVM_MAX_VCPUS;
2667                 break;
2668         case KVM_CAP_NR_MEMSLOTS:
2669                 r = KVM_USER_MEM_SLOTS;
2670                 break;
2671         case KVM_CAP_PV_MMU:    /* obsolete */
2672                 r = 0;
2673                 break;
2674 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2675         case KVM_CAP_IOMMU:
2676                 r = iommu_present(&pci_bus_type);
2677                 break;
2678 #endif
2679         case KVM_CAP_MCE:
2680                 r = KVM_MAX_MCE_BANKS;
2681                 break;
2682         case KVM_CAP_XCRS:
2683                 r = boot_cpu_has(X86_FEATURE_XSAVE);
2684                 break;
2685         case KVM_CAP_TSC_CONTROL:
2686                 r = kvm_has_tsc_control;
2687                 break;
2688         case KVM_CAP_X2APIC_API:
2689                 r = KVM_X2APIC_API_VALID_FLAGS;
2690                 break;
2691         default:
2692                 r = 0;
2693                 break;
2694         }
2695         return r;
2696
2697 }
2698
2699 long kvm_arch_dev_ioctl(struct file *filp,
2700                         unsigned int ioctl, unsigned long arg)
2701 {
2702         void __user *argp = (void __user *)arg;
2703         long r;
2704
2705         switch (ioctl) {
2706         case KVM_GET_MSR_INDEX_LIST: {
2707                 struct kvm_msr_list __user *user_msr_list = argp;
2708                 struct kvm_msr_list msr_list;
2709                 unsigned n;
2710
2711                 r = -EFAULT;
2712                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2713                         goto out;
2714                 n = msr_list.nmsrs;
2715                 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2716                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2717                         goto out;
2718                 r = -E2BIG;
2719                 if (n < msr_list.nmsrs)
2720                         goto out;
2721                 r = -EFAULT;
2722                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2723                                  num_msrs_to_save * sizeof(u32)))
2724                         goto out;
2725                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2726                                  &emulated_msrs,
2727                                  num_emulated_msrs * sizeof(u32)))
2728                         goto out;
2729                 r = 0;
2730                 break;
2731         }
2732         case KVM_GET_SUPPORTED_CPUID:
2733         case KVM_GET_EMULATED_CPUID: {
2734                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2735                 struct kvm_cpuid2 cpuid;
2736
2737                 r = -EFAULT;
2738                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2739                         goto out;
2740
2741                 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2742                                             ioctl);
2743                 if (r)
2744                         goto out;
2745
2746                 r = -EFAULT;
2747                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2748                         goto out;
2749                 r = 0;
2750                 break;
2751         }
2752         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2753                 r = -EFAULT;
2754                 if (copy_to_user(argp, &kvm_mce_cap_supported,
2755                                  sizeof(kvm_mce_cap_supported)))
2756                         goto out;
2757                 r = 0;
2758                 break;
2759         }
2760         default:
2761                 r = -EINVAL;
2762         }
2763 out:
2764         return r;
2765 }
2766
2767 static void wbinvd_ipi(void *garbage)
2768 {
2769         wbinvd();
2770 }
2771
2772 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2773 {
2774         return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2775 }
2776
2777 static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu)
2778 {
2779         set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
2780 }
2781
2782 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2783 {
2784         /* Address WBINVD may be executed by guest */
2785         if (need_emulate_wbinvd(vcpu)) {
2786                 if (kvm_x86_ops->has_wbinvd_exit())
2787                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2788                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2789                         smp_call_function_single(vcpu->cpu,
2790                                         wbinvd_ipi, NULL, 1);
2791         }
2792
2793         kvm_x86_ops->vcpu_load(vcpu, cpu);
2794
2795         /* Apply any externally detected TSC adjustments (due to suspend) */
2796         if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2797                 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2798                 vcpu->arch.tsc_offset_adjustment = 0;
2799                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2800         }
2801
2802         if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2803                 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2804                                 rdtsc() - vcpu->arch.last_host_tsc;
2805                 if (tsc_delta < 0)
2806                         mark_tsc_unstable("KVM discovered backwards TSC");
2807
2808                 if (check_tsc_unstable()) {
2809                         u64 offset = kvm_compute_tsc_offset(vcpu,
2810                                                 vcpu->arch.last_guest_tsc);
2811                         kvm_vcpu_write_tsc_offset(vcpu, offset);
2812                         vcpu->arch.tsc_catchup = 1;
2813                 }
2814                 if (kvm_lapic_hv_timer_in_use(vcpu) &&
2815                                 kvm_x86_ops->set_hv_timer(vcpu,
2816                                         kvm_get_lapic_target_expiration_tsc(vcpu)))
2817                         kvm_lapic_switch_to_sw_timer(vcpu);
2818                 /*
2819                  * On a host with synchronized TSC, there is no need to update
2820                  * kvmclock on vcpu->cpu migration
2821                  */
2822                 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2823                         kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2824                 if (vcpu->cpu != cpu)
2825                         kvm_migrate_timers(vcpu);
2826                 vcpu->cpu = cpu;
2827         }
2828
2829         kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2830 }
2831
2832 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2833 {
2834         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2835                 return;
2836
2837         vcpu->arch.st.steal.preempted = 1;
2838
2839         kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
2840                         &vcpu->arch.st.steal.preempted,
2841                         offsetof(struct kvm_steal_time, preempted),
2842                         sizeof(vcpu->arch.st.steal.preempted));
2843 }
2844
2845 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2846 {
2847         int idx;
2848         /*
2849          * Disable page faults because we're in atomic context here.
2850          * kvm_write_guest_offset_cached() would call might_fault()
2851          * that relies on pagefault_disable() to tell if there's a
2852          * bug. NOTE: the write to guest memory may not go through if
2853          * during postcopy live migration or if there's heavy guest
2854          * paging.
2855          */
2856         pagefault_disable();
2857         /*
2858          * kvm_memslots() will be called by
2859          * kvm_write_guest_offset_cached() so take the srcu lock.
2860          */
2861         idx = srcu_read_lock(&vcpu->kvm->srcu);
2862         kvm_steal_time_set_preempted(vcpu);
2863         srcu_read_unlock(&vcpu->kvm->srcu, idx);
2864         pagefault_enable();
2865         kvm_x86_ops->vcpu_put(vcpu);
2866         kvm_put_guest_fpu(vcpu);
2867         vcpu->arch.last_host_tsc = rdtsc();
2868 }
2869
2870 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2871                                     struct kvm_lapic_state *s)
2872 {
2873         if (vcpu->arch.apicv_active)
2874                 kvm_x86_ops->sync_pir_to_irr(vcpu);
2875
2876         return kvm_apic_get_state(vcpu, s);
2877 }
2878
2879 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2880                                     struct kvm_lapic_state *s)
2881 {
2882         int r;
2883
2884         r = kvm_apic_set_state(vcpu, s);
2885         if (r)
2886                 return r;
2887         update_cr8_intercept(vcpu);
2888
2889         return 0;
2890 }
2891
2892 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2893 {
2894         return (!lapic_in_kernel(vcpu) ||
2895                 kvm_apic_accept_pic_intr(vcpu));
2896 }
2897
2898 /*
2899  * if userspace requested an interrupt window, check that the
2900  * interrupt window is open.
2901  *
2902  * No need to exit to userspace if we already have an interrupt queued.
2903  */
2904 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2905 {
2906         return kvm_arch_interrupt_allowed(vcpu) &&
2907                 !kvm_cpu_has_interrupt(vcpu) &&
2908                 !kvm_event_needs_reinjection(vcpu) &&
2909                 kvm_cpu_accept_dm_intr(vcpu);
2910 }
2911
2912 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2913                                     struct kvm_interrupt *irq)
2914 {
2915         if (irq->irq >= KVM_NR_INTERRUPTS)
2916                 return -EINVAL;
2917
2918         if (!irqchip_in_kernel(vcpu->kvm)) {
2919                 kvm_queue_interrupt(vcpu, irq->irq, false);
2920                 kvm_make_request(KVM_REQ_EVENT, vcpu);
2921                 return 0;
2922         }
2923
2924         /*
2925          * With in-kernel LAPIC, we only use this to inject EXTINT, so
2926          * fail for in-kernel 8259.
2927          */
2928         if (pic_in_kernel(vcpu->kvm))
2929                 return -ENXIO;
2930
2931         if (vcpu->arch.pending_external_vector != -1)
2932                 return -EEXIST;
2933
2934         vcpu->arch.pending_external_vector = irq->irq;
2935         kvm_make_request(KVM_REQ_EVENT, vcpu);
2936         return 0;
2937 }
2938
2939 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2940 {
2941         kvm_inject_nmi(vcpu);
2942
2943         return 0;
2944 }
2945
2946 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2947 {
2948         kvm_make_request(KVM_REQ_SMI, vcpu);
2949
2950         return 0;
2951 }
2952
2953 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2954                                            struct kvm_tpr_access_ctl *tac)
2955 {
2956         if (tac->flags)
2957                 return -EINVAL;
2958         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2959         return 0;
2960 }
2961
2962 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2963                                         u64 mcg_cap)
2964 {
2965         int r;
2966         unsigned bank_num = mcg_cap & 0xff, bank;
2967
2968         r = -EINVAL;
2969         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2970                 goto out;
2971         if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
2972                 goto out;
2973         r = 0;
2974         vcpu->arch.mcg_cap = mcg_cap;
2975         /* Init IA32_MCG_CTL to all 1s */
2976         if (mcg_cap & MCG_CTL_P)
2977                 vcpu->arch.mcg_ctl = ~(u64)0;
2978         /* Init IA32_MCi_CTL to all 1s */
2979         for (bank = 0; bank < bank_num; bank++)
2980                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2981
2982         if (kvm_x86_ops->setup_mce)
2983                 kvm_x86_ops->setup_mce(vcpu);
2984 out:
2985         return r;
2986 }
2987
2988 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2989                                       struct kvm_x86_mce *mce)
2990 {
2991         u64 mcg_cap = vcpu->arch.mcg_cap;
2992         unsigned bank_num = mcg_cap & 0xff;
2993         u64 *banks = vcpu->arch.mce_banks;
2994
2995         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2996                 return -EINVAL;
2997         /*
2998          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2999          * reporting is disabled
3000          */
3001         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3002             vcpu->arch.mcg_ctl != ~(u64)0)
3003                 return 0;
3004         banks += 4 * mce->bank;
3005         /*
3006          * if IA32_MCi_CTL is not all 1s, the uncorrected error
3007          * reporting is disabled for the bank
3008          */
3009         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3010                 return 0;
3011         if (mce->status & MCI_STATUS_UC) {
3012                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3013                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3014                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3015                         return 0;
3016                 }
3017                 if (banks[1] & MCI_STATUS_VAL)
3018                         mce->status |= MCI_STATUS_OVER;
3019                 banks[2] = mce->addr;
3020                 banks[3] = mce->misc;
3021                 vcpu->arch.mcg_status = mce->mcg_status;
3022                 banks[1] = mce->status;
3023                 kvm_queue_exception(vcpu, MC_VECTOR);
3024         } else if (!(banks[1] & MCI_STATUS_VAL)
3025                    || !(banks[1] & MCI_STATUS_UC)) {
3026                 if (banks[1] & MCI_STATUS_VAL)
3027                         mce->status |= MCI_STATUS_OVER;
3028                 banks[2] = mce->addr;
3029                 banks[3] = mce->misc;
3030                 banks[1] = mce->status;
3031         } else
3032                 banks[1] |= MCI_STATUS_OVER;
3033         return 0;
3034 }
3035
3036 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3037                                                struct kvm_vcpu_events *events)
3038 {
3039         process_nmi(vcpu);
3040         events->exception.injected =
3041                 vcpu->arch.exception.pending &&
3042                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3043         events->exception.nr = vcpu->arch.exception.nr;
3044         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3045         events->exception.pad = 0;
3046         events->exception.error_code = vcpu->arch.exception.error_code;
3047
3048         events->interrupt.injected =
3049                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3050         events->interrupt.nr = vcpu->arch.interrupt.nr;
3051         events->interrupt.soft = 0;
3052         events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3053
3054         events->nmi.injected = vcpu->arch.nmi_injected;
3055         events->nmi.pending = vcpu->arch.nmi_pending != 0;
3056         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3057         events->nmi.pad = 0;
3058
3059         events->sipi_vector = 0; /* never valid when reporting to user space */
3060
3061         events->smi.smm = is_smm(vcpu);
3062         events->smi.pending = vcpu->arch.smi_pending;
3063         events->smi.smm_inside_nmi =
3064                 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3065         events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3066
3067         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3068                          | KVM_VCPUEVENT_VALID_SHADOW
3069                          | KVM_VCPUEVENT_VALID_SMM);
3070         memset(&events->reserved, 0, sizeof(events->reserved));
3071 }
3072
3073 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3074
3075 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3076                                               struct kvm_vcpu_events *events)
3077 {
3078         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3079                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3080                               | KVM_VCPUEVENT_VALID_SHADOW
3081                               | KVM_VCPUEVENT_VALID_SMM))
3082                 return -EINVAL;
3083
3084         if (events->exception.injected &&
3085             (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3086                 return -EINVAL;
3087
3088         process_nmi(vcpu);
3089         vcpu->arch.exception.pending = events->exception.injected;
3090         vcpu->arch.exception.nr = events->exception.nr;
3091         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3092         vcpu->arch.exception.error_code = events->exception.error_code;
3093
3094         vcpu->arch.interrupt.pending = events->interrupt.injected;
3095         vcpu->arch.interrupt.nr = events->interrupt.nr;
3096         vcpu->arch.interrupt.soft = events->interrupt.soft;
3097         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3098                 kvm_x86_ops->set_interrupt_shadow(vcpu,
3099                                                   events->interrupt.shadow);
3100
3101         vcpu->arch.nmi_injected = events->nmi.injected;
3102         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3103                 vcpu->arch.nmi_pending = events->nmi.pending;
3104         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3105
3106         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3107             lapic_in_kernel(vcpu))
3108                 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3109
3110         if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3111                 u32 hflags = vcpu->arch.hflags;
3112                 if (events->smi.smm)
3113                         hflags |= HF_SMM_MASK;
3114                 else
3115                         hflags &= ~HF_SMM_MASK;
3116                 kvm_set_hflags(vcpu, hflags);
3117
3118                 vcpu->arch.smi_pending = events->smi.pending;
3119                 if (events->smi.smm_inside_nmi)
3120                         vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3121                 else
3122                         vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3123                 if (lapic_in_kernel(vcpu)) {
3124                         if (events->smi.latched_init)
3125                                 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3126                         else
3127                                 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3128                 }
3129         }
3130
3131         kvm_make_request(KVM_REQ_EVENT, vcpu);
3132
3133         return 0;
3134 }
3135
3136 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3137                                              struct kvm_debugregs *dbgregs)
3138 {
3139         unsigned long val;
3140
3141         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3142         kvm_get_dr(vcpu, 6, &val);
3143         dbgregs->dr6 = val;
3144         dbgregs->dr7 = vcpu->arch.dr7;
3145         dbgregs->flags = 0;
3146         memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3147 }
3148
3149 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3150                                             struct kvm_debugregs *dbgregs)
3151 {
3152         if (dbgregs->flags)
3153                 return -EINVAL;
3154
3155         if (dbgregs->dr6 & ~0xffffffffull)
3156                 return -EINVAL;
3157         if (dbgregs->dr7 & ~0xffffffffull)
3158                 return -EINVAL;
3159
3160         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3161         kvm_update_dr0123(vcpu);
3162         vcpu->arch.dr6 = dbgregs->dr6;
3163         kvm_update_dr6(vcpu);
3164         vcpu->arch.dr7 = dbgregs->dr7;
3165         kvm_update_dr7(vcpu);
3166
3167         return 0;
3168 }
3169
3170 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3171
3172 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3173 {
3174         struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3175         u64 xstate_bv = xsave->header.xfeatures;
3176         u64 valid;
3177
3178         /*
3179          * Copy legacy XSAVE area, to avoid complications with CPUID
3180          * leaves 0 and 1 in the loop below.
3181          */
3182         memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3183
3184         /* Set XSTATE_BV */
3185         *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3186
3187         /*
3188          * Copy each region from the possibly compacted offset to the
3189          * non-compacted offset.
3190          */
3191         valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3192         while (valid) {
3193                 u64 feature = valid & -valid;
3194                 int index = fls64(feature) - 1;
3195                 void *src = get_xsave_addr(xsave, feature);
3196
3197                 if (src) {
3198                         u32 size, offset, ecx, edx;
3199                         cpuid_count(XSTATE_CPUID, index,
3200                                     &size, &offset, &ecx, &edx);
3201                         memcpy(dest + offset, src, size);
3202                 }
3203
3204                 valid -= feature;
3205         }
3206 }
3207
3208 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3209 {
3210         struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3211         u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3212         u64 valid;
3213
3214         /*
3215          * Copy legacy XSAVE area, to avoid complications with CPUID
3216          * leaves 0 and 1 in the loop below.
3217          */
3218         memcpy(xsave, src, XSAVE_HDR_OFFSET);
3219
3220         /* Set XSTATE_BV and possibly XCOMP_BV.  */
3221         xsave->header.xfeatures = xstate_bv;
3222         if (boot_cpu_has(X86_FEATURE_XSAVES))
3223                 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3224
3225         /*
3226          * Copy each region from the non-compacted offset to the
3227          * possibly compacted offset.
3228          */
3229         valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3230         while (valid) {
3231                 u64 feature = valid & -valid;
3232                 int index = fls64(feature) - 1;
3233                 void *dest = get_xsave_addr(xsave, feature);
3234
3235                 if (dest) {
3236                         u32 size, offset, ecx, edx;
3237                         cpuid_count(XSTATE_CPUID, index,
3238                                     &size, &offset, &ecx, &edx);
3239                         memcpy(dest, src + offset, size);
3240                 }
3241
3242                 valid -= feature;
3243         }
3244 }
3245
3246 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3247                                          struct kvm_xsave *guest_xsave)
3248 {
3249         if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3250                 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3251                 fill_xsave((u8 *) guest_xsave->region, vcpu);
3252         } else {
3253                 memcpy(guest_xsave->region,
3254                         &vcpu->arch.guest_fpu.state.fxsave,
3255                         sizeof(struct fxregs_state));
3256                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3257                         XFEATURE_MASK_FPSSE;
3258         }
3259 }
3260
3261 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3262                                         struct kvm_xsave *guest_xsave)
3263 {
3264         u64 xstate_bv =
3265                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3266
3267         if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3268                 /*
3269                  * Here we allow setting states that are not present in
3270                  * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3271                  * with old userspace.
3272                  */
3273                 if (xstate_bv & ~kvm_supported_xcr0())
3274                         return -EINVAL;
3275                 load_xsave(vcpu, (u8 *)guest_xsave->region);
3276         } else {
3277                 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3278                         return -EINVAL;
3279                 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3280                         guest_xsave->region, sizeof(struct fxregs_state));
3281         }
3282         return 0;
3283 }
3284
3285 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3286                                         struct kvm_xcrs *guest_xcrs)
3287 {
3288         if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3289                 guest_xcrs->nr_xcrs = 0;
3290                 return;
3291         }
3292
3293         guest_xcrs->nr_xcrs = 1;
3294         guest_xcrs->flags = 0;
3295         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3296         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3297 }
3298
3299 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3300                                        struct kvm_xcrs *guest_xcrs)
3301 {
3302         int i, r = 0;
3303
3304         if (!boot_cpu_has(X86_FEATURE_XSAVE))
3305                 return -EINVAL;
3306
3307         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3308                 return -EINVAL;
3309
3310         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3311                 /* Only support XCR0 currently */
3312                 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3313                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3314                                 guest_xcrs->xcrs[i].value);
3315                         break;
3316                 }
3317         if (r)
3318                 r = -EINVAL;
3319         return r;
3320 }
3321
3322 /*
3323  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3324  * stopped by the hypervisor.  This function will be called from the host only.
3325  * EINVAL is returned when the host attempts to set the flag for a guest that
3326  * does not support pv clocks.
3327  */
3328 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3329 {
3330         if (!vcpu->arch.pv_time_enabled)
3331                 return -EINVAL;
3332         vcpu->arch.pvclock_set_guest_stopped_request = true;
3333         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3334         return 0;
3335 }
3336
3337 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3338                                      struct kvm_enable_cap *cap)
3339 {
3340         if (cap->flags)
3341                 return -EINVAL;
3342
3343         switch (cap->cap) {
3344         case KVM_CAP_HYPERV_SYNIC:
3345                 if (!irqchip_in_kernel(vcpu->kvm))
3346                         return -EINVAL;
3347                 return kvm_hv_activate_synic(vcpu);
3348         default:
3349                 return -EINVAL;
3350         }
3351 }
3352
3353 long kvm_arch_vcpu_ioctl(struct file *filp,
3354                          unsigned int ioctl, unsigned long arg)
3355 {
3356         struct kvm_vcpu *vcpu = filp->private_data;
3357         void __user *argp = (void __user *)arg;
3358         int r;
3359         union {
3360                 struct kvm_lapic_state *lapic;
3361                 struct kvm_xsave *xsave;
3362                 struct kvm_xcrs *xcrs;
3363                 void *buffer;
3364         } u;
3365
3366         u.buffer = NULL;
3367         switch (ioctl) {
3368         case KVM_GET_LAPIC: {
3369                 r = -EINVAL;
3370                 if (!lapic_in_kernel(vcpu))
3371                         goto out;
3372                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3373
3374                 r = -ENOMEM;
3375                 if (!u.lapic)
3376                         goto out;
3377                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3378                 if (r)
3379                         goto out;
3380                 r = -EFAULT;
3381                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3382                         goto out;
3383                 r = 0;
3384                 break;
3385         }
3386         case KVM_SET_LAPIC: {
3387                 r = -EINVAL;
3388                 if (!lapic_in_kernel(vcpu))
3389                         goto out;
3390                 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3391                 if (IS_ERR(u.lapic))
3392                         return PTR_ERR(u.lapic);
3393
3394                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3395                 break;
3396         }
3397         case KVM_INTERRUPT: {
3398                 struct kvm_interrupt irq;
3399
3400                 r = -EFAULT;
3401                 if (copy_from_user(&irq, argp, sizeof irq))
3402                         goto out;
3403                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3404                 break;
3405         }
3406         case KVM_NMI: {
3407                 r = kvm_vcpu_ioctl_nmi(vcpu);
3408                 break;
3409         }
3410         case KVM_SMI: {
3411                 r = kvm_vcpu_ioctl_smi(vcpu);
3412                 break;
3413         }
3414         case KVM_SET_CPUID: {
3415                 struct kvm_cpuid __user *cpuid_arg = argp;
3416                 struct kvm_cpuid cpuid;
3417
3418                 r = -EFAULT;
3419                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3420                         goto out;
3421                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3422                 break;
3423         }
3424         case KVM_SET_CPUID2: {
3425                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3426                 struct kvm_cpuid2 cpuid;
3427
3428                 r = -EFAULT;
3429                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3430                         goto out;
3431                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3432                                               cpuid_arg->entries);
3433                 break;
3434         }
3435         case KVM_GET_CPUID2: {
3436                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3437                 struct kvm_cpuid2 cpuid;
3438
3439                 r = -EFAULT;
3440                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3441                         goto out;
3442                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3443                                               cpuid_arg->entries);
3444                 if (r)
3445                         goto out;
3446                 r = -EFAULT;
3447                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3448                         goto out;
3449                 r = 0;
3450                 break;
3451         }
3452         case KVM_GET_MSRS:
3453                 r = msr_io(vcpu, argp, do_get_msr, 1);
3454                 break;
3455         case KVM_SET_MSRS:
3456                 r = msr_io(vcpu, argp, do_set_msr, 0);
3457                 break;
3458         case KVM_TPR_ACCESS_REPORTING: {
3459                 struct kvm_tpr_access_ctl tac;
3460
3461                 r = -EFAULT;
3462                 if (copy_from_user(&tac, argp, sizeof tac))
3463                         goto out;
3464                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3465                 if (r)
3466                         goto out;
3467                 r = -EFAULT;
3468                 if (copy_to_user(argp, &tac, sizeof tac))
3469                         goto out;
3470                 r = 0;
3471                 break;
3472         };
3473         case KVM_SET_VAPIC_ADDR: {
3474                 struct kvm_vapic_addr va;
3475                 int idx;
3476
3477                 r = -EINVAL;
3478                 if (!lapic_in_kernel(vcpu))
3479                         goto out;
3480                 r = -EFAULT;
3481                 if (copy_from_user(&va, argp, sizeof va))
3482                         goto out;
3483                 idx = srcu_read_lock(&vcpu->kvm->srcu);
3484                 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3485                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3486                 break;
3487         }
3488         case KVM_X86_SETUP_MCE: {
3489                 u64 mcg_cap;
3490
3491                 r = -EFAULT;
3492                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3493                         goto out;
3494                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3495                 break;
3496         }
3497         case KVM_X86_SET_MCE: {
3498                 struct kvm_x86_mce mce;
3499
3500                 r = -EFAULT;
3501                 if (copy_from_user(&mce, argp, sizeof mce))
3502                         goto out;
3503                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3504                 break;
3505         }
3506         case KVM_GET_VCPU_EVENTS: {
3507                 struct kvm_vcpu_events events;
3508
3509                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3510
3511                 r = -EFAULT;
3512                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3513                         break;
3514                 r = 0;
3515                 break;
3516         }
3517         case KVM_SET_VCPU_EVENTS: {
3518                 struct kvm_vcpu_events events;
3519
3520                 r = -EFAULT;
3521                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3522                         break;
3523
3524                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3525                 break;
3526         }
3527         case KVM_GET_DEBUGREGS: {
3528                 struct kvm_debugregs dbgregs;
3529
3530                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3531
3532                 r = -EFAULT;
3533                 if (copy_to_user(argp, &dbgregs,
3534                                  sizeof(struct kvm_debugregs)))
3535                         break;
3536                 r = 0;
3537                 break;
3538         }
3539         case KVM_SET_DEBUGREGS: {
3540                 struct kvm_debugregs dbgregs;
3541
3542                 r = -EFAULT;
3543                 if (copy_from_user(&dbgregs, argp,
3544                                    sizeof(struct kvm_debugregs)))
3545                         break;
3546
3547                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3548                 break;
3549         }
3550         case KVM_GET_XSAVE: {
3551                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3552                 r = -ENOMEM;
3553                 if (!u.xsave)
3554                         break;
3555
3556                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3557
3558                 r = -EFAULT;
3559                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3560                         break;
3561                 r = 0;
3562                 break;
3563         }
3564         case KVM_SET_XSAVE: {
3565                 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3566                 if (IS_ERR(u.xsave))
3567                         return PTR_ERR(u.xsave);
3568
3569                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3570                 break;
3571         }
3572         case KVM_GET_XCRS: {
3573                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3574                 r = -ENOMEM;
3575                 if (!u.xcrs)
3576                         break;
3577
3578                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3579
3580                 r = -EFAULT;
3581                 if (copy_to_user(argp, u.xcrs,
3582                                  sizeof(struct kvm_xcrs)))
3583                         break;
3584                 r = 0;
3585                 break;
3586         }
3587         case KVM_SET_XCRS: {
3588                 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3589                 if (IS_ERR(u.xcrs))
3590                         return PTR_ERR(u.xcrs);
3591
3592                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3593                 break;
3594         }
3595         case KVM_SET_TSC_KHZ: {
3596                 u32 user_tsc_khz;
3597
3598                 r = -EINVAL;
3599                 user_tsc_khz = (u32)arg;
3600
3601                 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3602                         goto out;
3603
3604                 if (user_tsc_khz == 0)
3605                         user_tsc_khz = tsc_khz;
3606
3607                 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3608                         r = 0;
3609
3610                 goto out;
3611         }
3612         case KVM_GET_TSC_KHZ: {
3613                 r = vcpu->arch.virtual_tsc_khz;
3614                 goto out;
3615         }
3616         case KVM_KVMCLOCK_CTRL: {
3617                 r = kvm_set_guest_paused(vcpu);
3618                 goto out;
3619         }
3620         case KVM_ENABLE_CAP: {
3621                 struct kvm_enable_cap cap;
3622
3623                 r = -EFAULT;
3624                 if (copy_from_user(&cap, argp, sizeof(cap)))
3625                         goto out;
3626                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3627                 break;
3628         }
3629         default:
3630                 r = -EINVAL;
3631         }
3632 out:
3633         kfree(u.buffer);
3634         return r;
3635 }
3636
3637 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3638 {
3639         return VM_FAULT_SIGBUS;
3640 }
3641
3642 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3643 {
3644         int ret;
3645
3646         if (addr > (unsigned int)(-3 * PAGE_SIZE))
3647                 return -EINVAL;
3648         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3649         return ret;
3650 }
3651
3652 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3653                                               u64 ident_addr)
3654 {
3655         kvm->arch.ept_identity_map_addr = ident_addr;
3656         return 0;
3657 }
3658
3659 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3660                                           u32 kvm_nr_mmu_pages)
3661 {
3662         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3663                 return -EINVAL;
3664
3665         mutex_lock(&kvm->slots_lock);
3666
3667         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3668         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3669
3670         mutex_unlock(&kvm->slots_lock);
3671         return 0;
3672 }
3673
3674 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3675 {
3676         return kvm->arch.n_max_mmu_pages;
3677 }
3678
3679 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3680 {
3681         int r;
3682
3683         r = 0;
3684         switch (chip->chip_id) {
3685         case KVM_IRQCHIP_PIC_MASTER:
3686                 memcpy(&chip->chip.pic,
3687                         &pic_irqchip(kvm)->pics[0],
3688                         sizeof(struct kvm_pic_state));
3689                 break;
3690         case KVM_IRQCHIP_PIC_SLAVE:
3691                 memcpy(&chip->chip.pic,
3692                         &pic_irqchip(kvm)->pics[1],
3693                         sizeof(struct kvm_pic_state));
3694                 break;
3695         case KVM_IRQCHIP_IOAPIC:
3696                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3697                 break;
3698         default:
3699                 r = -EINVAL;
3700                 break;
3701         }
3702         return r;
3703 }
3704
3705 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3706 {
3707         int r;
3708
3709         r = 0;
3710         switch (chip->chip_id) {
3711         case KVM_IRQCHIP_PIC_MASTER:
3712                 spin_lock(&pic_irqchip(kvm)->lock);
3713                 memcpy(&pic_irqchip(kvm)->pics[0],
3714                         &chip->chip.pic,
3715                         sizeof(struct kvm_pic_state));
3716                 spin_unlock(&pic_irqchip(kvm)->lock);
3717                 break;
3718         case KVM_IRQCHIP_PIC_SLAVE:
3719                 spin_lock(&pic_irqchip(kvm)->lock);
3720                 memcpy(&pic_irqchip(kvm)->pics[1],
3721                         &chip->chip.pic,
3722                         sizeof(struct kvm_pic_state));
3723                 spin_unlock(&pic_irqchip(kvm)->lock);
3724                 break;
3725         case KVM_IRQCHIP_IOAPIC:
3726                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3727                 break;
3728         default:
3729                 r = -EINVAL;
3730                 break;
3731         }
3732         kvm_pic_update_irq(pic_irqchip(kvm));
3733         return r;
3734 }
3735
3736 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3737 {
3738         struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3739
3740         BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3741
3742         mutex_lock(&kps->lock);
3743         memcpy(ps, &kps->channels, sizeof(*ps));
3744         mutex_unlock(&kps->lock);
3745         return 0;
3746 }
3747
3748 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3749 {
3750         int i;
3751         struct kvm_pit *pit = kvm->arch.vpit;
3752
3753         mutex_lock(&pit->pit_state.lock);
3754         memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3755         for (i = 0; i < 3; i++)
3756                 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3757         mutex_unlock(&pit->pit_state.lock);
3758         return 0;
3759 }
3760
3761 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3762 {
3763         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3764         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3765                 sizeof(ps->channels));
3766         ps->flags = kvm->arch.vpit->pit_state.flags;
3767         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3768         memset(&ps->reserved, 0, sizeof(ps->reserved));
3769         return 0;
3770 }
3771
3772 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3773 {
3774         int start = 0;
3775         int i;
3776         u32 prev_legacy, cur_legacy;
3777         struct kvm_pit *pit = kvm->arch.vpit;
3778
3779         mutex_lock(&pit->pit_state.lock);
3780         prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3781         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3782         if (!prev_legacy && cur_legacy)
3783                 start = 1;
3784         memcpy(&pit->pit_state.channels, &ps->channels,
3785                sizeof(pit->pit_state.channels));
3786         pit->pit_state.flags = ps->flags;
3787         for (i = 0; i < 3; i++)
3788                 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3789                                    start && i == 0);
3790         mutex_unlock(&pit->pit_state.lock);
3791         return 0;
3792 }
3793
3794 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3795                                  struct kvm_reinject_control *control)
3796 {
3797         struct kvm_pit *pit = kvm->arch.vpit;
3798
3799         if (!pit)
3800                 return -ENXIO;
3801
3802         /* pit->pit_state.lock was overloaded to prevent userspace from getting
3803          * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3804          * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
3805          */
3806         mutex_lock(&pit->pit_state.lock);
3807         kvm_pit_set_reinject(pit, control->pit_reinject);
3808         mutex_unlock(&pit->pit_state.lock);
3809
3810         return 0;
3811 }
3812
3813 /**
3814  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3815  * @kvm: kvm instance
3816  * @log: slot id and address to which we copy the log
3817  *
3818  * Steps 1-4 below provide general overview of dirty page logging. See
3819  * kvm_get_dirty_log_protect() function description for additional details.
3820  *
3821  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3822  * always flush the TLB (step 4) even if previous step failed  and the dirty
3823  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3824  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3825  * writes will be marked dirty for next log read.
3826  *
3827  *   1. Take a snapshot of the bit and clear it if needed.
3828  *   2. Write protect the corresponding page.
3829  *   3. Copy the snapshot to the userspace.
3830  *   4. Flush TLB's if needed.
3831  */
3832 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3833 {
3834         bool is_dirty = false;
3835         int r;
3836
3837         mutex_lock(&kvm->slots_lock);
3838
3839         /*
3840          * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3841          */
3842         if (kvm_x86_ops->flush_log_dirty)
3843                 kvm_x86_ops->flush_log_dirty(kvm);
3844
3845         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3846
3847         /*
3848          * All the TLBs can be flushed out of mmu lock, see the comments in
3849          * kvm_mmu_slot_remove_write_access().
3850          */
3851         lockdep_assert_held(&kvm->slots_lock);
3852         if (is_dirty)
3853                 kvm_flush_remote_tlbs(kvm);
3854
3855         mutex_unlock(&kvm->slots_lock);
3856         return r;
3857 }
3858
3859 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3860                         bool line_status)
3861 {
3862         if (!irqchip_in_kernel(kvm))
3863                 return -ENXIO;
3864
3865         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3866                                         irq_event->irq, irq_event->level,
3867                                         line_status);
3868         return 0;
3869 }
3870
3871 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3872                                    struct kvm_enable_cap *cap)
3873 {
3874         int r;
3875
3876         if (cap->flags)
3877                 return -EINVAL;
3878
3879         switch (cap->cap) {
3880         case KVM_CAP_DISABLE_QUIRKS:
3881                 kvm->arch.disabled_quirks = cap->args[0];
3882                 r = 0;
3883                 break;
3884         case KVM_CAP_SPLIT_IRQCHIP: {
3885                 mutex_lock(&kvm->lock);
3886                 r = -EINVAL;
3887                 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3888                         goto split_irqchip_unlock;
3889                 r = -EEXIST;
3890                 if (irqchip_in_kernel(kvm))
3891                         goto split_irqchip_unlock;
3892                 if (kvm->created_vcpus)
3893                         goto split_irqchip_unlock;
3894                 r = kvm_setup_empty_irq_routing(kvm);
3895                 if (r)
3896                         goto split_irqchip_unlock;
3897                 /* Pairs with irqchip_in_kernel. */
3898                 smp_wmb();
3899                 kvm->arch.irqchip_split = true;
3900                 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3901                 r = 0;
3902 split_irqchip_unlock:
3903                 mutex_unlock(&kvm->lock);
3904                 break;
3905         }
3906         case KVM_CAP_X2APIC_API:
3907                 r = -EINVAL;
3908                 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3909                         break;
3910
3911                 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
3912                         kvm->arch.x2apic_format = true;
3913                 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
3914                         kvm->arch.x2apic_broadcast_quirk_disabled = true;
3915
3916                 r = 0;
3917                 break;
3918         default:
3919                 r = -EINVAL;
3920                 break;
3921         }
3922         return r;
3923 }
3924
3925 long kvm_arch_vm_ioctl(struct file *filp,
3926                        unsigned int ioctl, unsigned long arg)
3927 {
3928         struct kvm *kvm = filp->private_data;
3929         void __user *argp = (void __user *)arg;
3930         int r = -ENOTTY;
3931         /*
3932          * This union makes it completely explicit to gcc-3.x
3933          * that these two variables' stack usage should be
3934          * combined, not added together.
3935          */
3936         union {
3937                 struct kvm_pit_state ps;
3938                 struct kvm_pit_state2 ps2;
3939                 struct kvm_pit_config pit_config;
3940         } u;
3941
3942         switch (ioctl) {
3943         case KVM_SET_TSS_ADDR:
3944                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3945                 break;
3946         case KVM_SET_IDENTITY_MAP_ADDR: {
3947                 u64 ident_addr;
3948
3949                 r = -EFAULT;
3950                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3951                         goto out;
3952                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3953                 break;
3954         }
3955         case KVM_SET_NR_MMU_PAGES:
3956                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3957                 break;
3958         case KVM_GET_NR_MMU_PAGES:
3959                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3960                 break;
3961         case KVM_CREATE_IRQCHIP: {
3962                 struct kvm_pic *vpic;
3963
3964                 mutex_lock(&kvm->lock);
3965                 r = -EEXIST;
3966                 if (kvm->arch.vpic)
3967                         goto create_irqchip_unlock;
3968                 r = -EINVAL;
3969                 if (kvm->created_vcpus)
3970                         goto create_irqchip_unlock;
3971                 r = -ENOMEM;
3972                 vpic = kvm_create_pic(kvm);
3973                 if (vpic) {
3974                         r = kvm_ioapic_init(kvm);
3975                         if (r) {
3976                                 mutex_lock(&kvm->slots_lock);
3977                                 kvm_destroy_pic(vpic);
3978                                 mutex_unlock(&kvm->slots_lock);
3979                                 goto create_irqchip_unlock;
3980                         }
3981                 } else
3982                         goto create_irqchip_unlock;
3983                 r = kvm_setup_default_irq_routing(kvm);
3984                 if (r) {
3985                         mutex_lock(&kvm->slots_lock);
3986                         mutex_lock(&kvm->irq_lock);
3987                         kvm_ioapic_destroy(kvm);
3988                         kvm_destroy_pic(vpic);
3989                         mutex_unlock(&kvm->irq_lock);
3990                         mutex_unlock(&kvm->slots_lock);
3991                         goto create_irqchip_unlock;
3992                 }
3993                 /* Write kvm->irq_routing before kvm->arch.vpic.  */
3994                 smp_wmb();
3995                 kvm->arch.vpic = vpic;
3996         create_irqchip_unlock:
3997                 mutex_unlock(&kvm->lock);
3998                 break;
3999         }
4000         case KVM_CREATE_PIT:
4001                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4002                 goto create_pit;
4003         case KVM_CREATE_PIT2:
4004                 r = -EFAULT;
4005                 if (copy_from_user(&u.pit_config, argp,
4006                                    sizeof(struct kvm_pit_config)))
4007                         goto out;
4008         create_pit:
4009                 mutex_lock(&kvm->lock);
4010                 r = -EEXIST;
4011                 if (kvm->arch.vpit)
4012                         goto create_pit_unlock;
4013                 r = -ENOMEM;
4014                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4015                 if (kvm->arch.vpit)
4016                         r = 0;
4017         create_pit_unlock:
4018                 mutex_unlock(&kvm->lock);
4019                 break;
4020         case KVM_GET_IRQCHIP: {
4021                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4022                 struct kvm_irqchip *chip;
4023
4024                 chip = memdup_user(argp, sizeof(*chip));
4025                 if (IS_ERR(chip)) {
4026                         r = PTR_ERR(chip);
4027                         goto out;
4028                 }
4029
4030                 r = -ENXIO;
4031                 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
4032                         goto get_irqchip_out;
4033                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4034                 if (r)
4035                         goto get_irqchip_out;
4036                 r = -EFAULT;
4037                 if (copy_to_user(argp, chip, sizeof *chip))
4038                         goto get_irqchip_out;
4039                 r = 0;
4040         get_irqchip_out:
4041                 kfree(chip);
4042                 break;
4043         }
4044         case KVM_SET_IRQCHIP: {
4045                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4046                 struct kvm_irqchip *chip;
4047
4048                 chip = memdup_user(argp, sizeof(*chip));
4049                 if (IS_ERR(chip)) {
4050                         r = PTR_ERR(chip);
4051                         goto out;
4052                 }
4053
4054                 r = -ENXIO;
4055                 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
4056                         goto set_irqchip_out;
4057                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4058                 if (r)
4059                         goto set_irqchip_out;
4060                 r = 0;
4061         set_irqchip_out:
4062                 kfree(chip);
4063                 break;
4064         }
4065         case KVM_GET_PIT: {
4066                 r = -EFAULT;
4067                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4068                         goto out;
4069                 r = -ENXIO;
4070                 if (!kvm->arch.vpit)
4071                         goto out;
4072                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4073                 if (r)
4074                         goto out;
4075                 r = -EFAULT;
4076                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4077                         goto out;
4078                 r = 0;
4079                 break;
4080         }
4081         case KVM_SET_PIT: {
4082                 r = -EFAULT;
4083                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
4084                         goto out;
4085                 r = -ENXIO;
4086                 if (!kvm->arch.vpit)
4087                         goto out;
4088                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4089                 break;
4090         }
4091         case KVM_GET_PIT2: {
4092                 r = -ENXIO;
4093                 if (!kvm->arch.vpit)
4094                         goto out;
4095                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4096                 if (r)
4097                         goto out;
4098                 r = -EFAULT;
4099                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4100                         goto out;
4101                 r = 0;
4102                 break;
4103         }
4104         case KVM_SET_PIT2: {
4105                 r = -EFAULT;
4106                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4107                         goto out;
4108                 r = -ENXIO;
4109                 if (!kvm->arch.vpit)
4110                         goto out;
4111                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4112                 break;
4113         }
4114         case KVM_REINJECT_CONTROL: {
4115                 struct kvm_reinject_control control;
4116                 r =  -EFAULT;
4117                 if (copy_from_user(&control, argp, sizeof(control)))
4118                         goto out;
4119                 r = kvm_vm_ioctl_reinject(kvm, &control);
4120                 break;
4121         }
4122         case KVM_SET_BOOT_CPU_ID:
4123                 r = 0;
4124                 mutex_lock(&kvm->lock);
4125                 if (kvm->created_vcpus)
4126                         r = -EBUSY;
4127                 else
4128                         kvm->arch.bsp_vcpu_id = arg;
4129                 mutex_unlock(&kvm->lock);
4130                 break;
4131         case KVM_XEN_HVM_CONFIG: {
4132                 r = -EFAULT;
4133                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4134                                    sizeof(struct kvm_xen_hvm_config)))
4135                         goto out;
4136                 r = -EINVAL;
4137                 if (kvm->arch.xen_hvm_config.flags)
4138                         goto out;
4139                 r = 0;
4140                 break;
4141         }
4142         case KVM_SET_CLOCK: {
4143                 struct kvm_clock_data user_ns;
4144                 u64 now_ns;
4145
4146                 r = -EFAULT;
4147                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4148                         goto out;
4149
4150                 r = -EINVAL;
4151                 if (user_ns.flags)
4152                         goto out;
4153
4154                 r = 0;
4155                 local_irq_disable();
4156                 now_ns = __get_kvmclock_ns(kvm);
4157                 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4158                 local_irq_enable();
4159                 kvm_gen_update_masterclock(kvm);
4160                 break;
4161         }
4162         case KVM_GET_CLOCK: {
4163                 struct kvm_clock_data user_ns;
4164                 u64 now_ns;
4165
4166                 local_irq_disable();
4167                 now_ns = __get_kvmclock_ns(kvm);
4168                 user_ns.clock = now_ns;
4169                 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4170                 local_irq_enable();
4171                 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4172
4173                 r = -EFAULT;
4174                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4175                         goto out;
4176                 r = 0;
4177                 break;
4178         }
4179         case KVM_ENABLE_CAP: {
4180                 struct kvm_enable_cap cap;
4181
4182                 r = -EFAULT;
4183                 if (copy_from_user(&cap, argp, sizeof(cap)))
4184                         goto out;
4185                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4186                 break;
4187         }
4188         default:
4189                 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4190         }
4191 out:
4192         return r;
4193 }
4194
4195 static void kvm_init_msr_list(void)
4196 {
4197         u32 dummy[2];
4198         unsigned i, j;
4199
4200         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4201                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4202                         continue;
4203
4204                 /*
4205                  * Even MSRs that are valid in the host may not be exposed
4206                  * to the guests in some cases.
4207                  */
4208                 switch (msrs_to_save[i]) {
4209                 case MSR_IA32_BNDCFGS:
4210                         if (!kvm_x86_ops->mpx_supported())
4211                                 continue;
4212                         break;
4213                 case MSR_TSC_AUX:
4214                         if (!kvm_x86_ops->rdtscp_supported())
4215                                 continue;
4216                         break;
4217                 default:
4218                         break;
4219                 }
4220
4221                 if (j < i)
4222                         msrs_to_save[j] = msrs_to_save[i];
4223                 j++;
4224         }
4225         num_msrs_to_save = j;
4226
4227         for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4228                 switch (emulated_msrs[i]) {
4229                 case MSR_IA32_SMBASE:
4230                         if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4231                                 continue;
4232                         break;
4233                 default:
4234                         break;
4235                 }
4236
4237                 if (j < i)
4238                         emulated_msrs[j] = emulated_msrs[i];
4239                 j++;
4240         }
4241         num_emulated_msrs = j;
4242 }
4243
4244 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4245                            const void *v)
4246 {
4247         int handled = 0;
4248         int n;
4249
4250         do {
4251                 n = min(len, 8);
4252                 if (!(lapic_in_kernel(vcpu) &&
4253                       !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4254                     && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4255                         break;
4256                 handled += n;
4257                 addr += n;
4258                 len -= n;
4259                 v += n;
4260         } while (len);
4261
4262         return handled;
4263 }
4264
4265 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4266 {
4267         int handled = 0;
4268         int n;
4269
4270         do {
4271                 n = min(len, 8);
4272                 if (!(lapic_in_kernel(vcpu) &&
4273                       !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4274                                          addr, n, v))
4275                     && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4276                         break;
4277                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4278                 handled += n;
4279                 addr += n;
4280                 len -= n;
4281                 v += n;
4282         } while (len);
4283
4284         return handled;
4285 }
4286
4287 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4288                         struct kvm_segment *var, int seg)
4289 {
4290         kvm_x86_ops->set_segment(vcpu, var, seg);
4291 }
4292
4293 void kvm_get_segment(struct kvm_vcpu *vcpu,
4294                      struct kvm_segment *var, int seg)
4295 {
4296         kvm_x86_ops->get_segment(vcpu, var, seg);
4297 }
4298
4299 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4300                            struct x86_exception *exception)
4301 {
4302         gpa_t t_gpa;
4303
4304         BUG_ON(!mmu_is_nested(vcpu));
4305
4306         /* NPT walks are always user-walks */
4307         access |= PFERR_USER_MASK;
4308         t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4309
4310         return t_gpa;
4311 }
4312
4313 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4314                               struct x86_exception *exception)
4315 {
4316         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4317         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4318 }
4319
4320  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4321                                 struct x86_exception *exception)
4322 {
4323         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4324         access |= PFERR_FETCH_MASK;
4325         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4326 }
4327
4328 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4329                                struct x86_exception *exception)
4330 {
4331         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4332         access |= PFERR_WRITE_MASK;
4333         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4334 }
4335
4336 /* uses this to access any guest's mapped memory without checking CPL */
4337 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4338                                 struct x86_exception *exception)
4339 {
4340         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4341 }
4342
4343 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4344                                       struct kvm_vcpu *vcpu, u32 access,
4345                                       struct x86_exception *exception)
4346 {
4347         void *data = val;
4348         int r = X86EMUL_CONTINUE;
4349
4350         while (bytes) {
4351                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4352                                                             exception);
4353                 unsigned offset = addr & (PAGE_SIZE-1);
4354                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4355                 int ret;
4356
4357                 if (gpa == UNMAPPED_GVA)
4358                         return X86EMUL_PROPAGATE_FAULT;
4359                 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4360                                                offset, toread);
4361                 if (ret < 0) {
4362                         r = X86EMUL_IO_NEEDED;
4363                         goto out;
4364                 }
4365
4366                 bytes -= toread;
4367                 data += toread;
4368                 addr += toread;
4369         }
4370 out:
4371         return r;
4372 }
4373
4374 /* used for instruction fetching */
4375 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4376                                 gva_t addr, void *val, unsigned int bytes,
4377                                 struct x86_exception *exception)
4378 {
4379         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4380         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4381         unsigned offset;
4382         int ret;
4383
4384         /* Inline kvm_read_guest_virt_helper for speed.  */
4385         gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4386                                                     exception);
4387         if (unlikely(gpa == UNMAPPED_GVA))
4388                 return X86EMUL_PROPAGATE_FAULT;
4389
4390         offset = addr & (PAGE_SIZE-1);
4391         if (WARN_ON(offset + bytes > PAGE_SIZE))
4392                 bytes = (unsigned)PAGE_SIZE - offset;
4393         ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4394                                        offset, bytes);
4395         if (unlikely(ret < 0))
4396                 return X86EMUL_IO_NEEDED;
4397
4398         return X86EMUL_CONTINUE;
4399 }
4400
4401 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4402                                gva_t addr, void *val, unsigned int bytes,
4403                                struct x86_exception *exception)
4404 {
4405         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4406         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4407
4408         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4409                                           exception);
4410 }
4411 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4412
4413 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4414                                       gva_t addr, void *val, unsigned int bytes,
4415                                       struct x86_exception *exception)
4416 {
4417         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4418         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4419 }
4420
4421 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4422                 unsigned long addr, void *val, unsigned int bytes)
4423 {
4424         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4425         int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4426
4427         return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4428 }
4429
4430 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4431                                        gva_t addr, void *val,
4432                                        unsigned int bytes,
4433                                        struct x86_exception *exception)
4434 {
4435         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4436         void *data = val;
4437         int r = X86EMUL_CONTINUE;
4438
4439         while (bytes) {
4440                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4441                                                              PFERR_WRITE_MASK,
4442                                                              exception);
4443                 unsigned offset = addr & (PAGE_SIZE-1);
4444                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4445                 int ret;
4446
4447                 if (gpa == UNMAPPED_GVA)
4448                         return X86EMUL_PROPAGATE_FAULT;
4449                 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4450                 if (ret < 0) {
4451                         r = X86EMUL_IO_NEEDED;
4452                         goto out;
4453                 }
4454
4455                 bytes -= towrite;
4456                 data += towrite;
4457                 addr += towrite;
4458         }
4459 out:
4460         return r;
4461 }
4462 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4463
4464 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4465                                 gpa_t *gpa, struct x86_exception *exception,
4466                                 bool write)
4467 {
4468         u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4469                 | (write ? PFERR_WRITE_MASK : 0);
4470
4471         /*
4472          * currently PKRU is only applied to ept enabled guest so
4473          * there is no pkey in EPT page table for L1 guest or EPT
4474          * shadow page table for L2 guest.
4475          */
4476         if (vcpu_match_mmio_gva(vcpu, gva)
4477             && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4478                                  vcpu->arch.access, 0, access)) {
4479                 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4480                                         (gva & (PAGE_SIZE - 1));
4481                 trace_vcpu_match_mmio(gva, *gpa, write, false);
4482                 return 1;
4483         }
4484
4485         *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4486
4487         if (*gpa == UNMAPPED_GVA)
4488                 return -1;
4489
4490         /* For APIC access vmexit */
4491         if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4492                 return 1;
4493
4494         if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4495                 trace_vcpu_match_mmio(gva, *gpa, write, true);
4496                 return 1;
4497         }
4498
4499         return 0;
4500 }
4501
4502 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4503                         const void *val, int bytes)
4504 {
4505         int ret;
4506
4507         ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4508         if (ret < 0)
4509                 return 0;
4510         kvm_page_track_write(vcpu, gpa, val, bytes);
4511         return 1;
4512 }
4513
4514 struct read_write_emulator_ops {
4515         int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4516                                   int bytes);
4517         int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4518                                   void *val, int bytes);
4519         int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4520                                int bytes, void *val);
4521         int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4522                                     void *val, int bytes);
4523         bool write;
4524 };
4525
4526 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4527 {
4528         if (vcpu->mmio_read_completed) {
4529                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4530                                vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4531                 vcpu->mmio_read_completed = 0;
4532                 return 1;
4533         }
4534
4535         return 0;
4536 }
4537
4538 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4539                         void *val, int bytes)
4540 {
4541         return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4542 }
4543
4544 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4545                          void *val, int bytes)
4546 {
4547         return emulator_write_phys(vcpu, gpa, val, bytes);
4548 }
4549
4550 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4551 {
4552         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4553         return vcpu_mmio_write(vcpu, gpa, bytes, val);
4554 }
4555
4556 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4557                           void *val, int bytes)
4558 {
4559         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4560         return X86EMUL_IO_NEEDED;
4561 }
4562
4563 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4564                            void *val, int bytes)
4565 {
4566         struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4567
4568         memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4569         return X86EMUL_CONTINUE;
4570 }
4571
4572 static const struct read_write_emulator_ops read_emultor = {
4573         .read_write_prepare = read_prepare,
4574         .read_write_emulate = read_emulate,
4575         .read_write_mmio = vcpu_mmio_read,
4576         .read_write_exit_mmio = read_exit_mmio,
4577 };
4578
4579 static const struct read_write_emulator_ops write_emultor = {
4580         .read_write_emulate = write_emulate,
4581         .read_write_mmio = write_mmio,
4582         .read_write_exit_mmio = write_exit_mmio,
4583         .write = true,
4584 };
4585
4586 static int emulator_read_write_onepage(unsigned long addr, void *val,
4587                                        unsigned int bytes,
4588                                        struct x86_exception *exception,
4589                                        struct kvm_vcpu *vcpu,
4590                                        const struct read_write_emulator_ops *ops)
4591 {
4592         gpa_t gpa;
4593         int handled, ret;
4594         bool write = ops->write;
4595         struct kvm_mmio_fragment *frag;
4596
4597         ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4598
4599         if (ret < 0)
4600                 return X86EMUL_PROPAGATE_FAULT;
4601
4602         /* For APIC access vmexit */
4603         if (ret)
4604                 goto mmio;
4605
4606         if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4607                 return X86EMUL_CONTINUE;
4608
4609 mmio:
4610         /*
4611          * Is this MMIO handled locally?
4612          */
4613         handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4614         if (handled == bytes)
4615                 return X86EMUL_CONTINUE;
4616
4617         gpa += handled;
4618         bytes -= handled;
4619         val += handled;
4620
4621         WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4622         frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4623         frag->gpa = gpa;
4624         frag->data = val;
4625         frag->len = bytes;
4626         return X86EMUL_CONTINUE;
4627 }
4628
4629 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4630                         unsigned long addr,
4631                         void *val, unsigned int bytes,
4632                         struct x86_exception *exception,
4633                         const struct read_write_emulator_ops *ops)
4634 {
4635         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4636         gpa_t gpa;
4637         int rc;
4638
4639         if (ops->read_write_prepare &&
4640                   ops->read_write_prepare(vcpu, val, bytes))
4641                 return X86EMUL_CONTINUE;
4642
4643         vcpu->mmio_nr_fragments = 0;
4644
4645         /* Crossing a page boundary? */
4646         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4647                 int now;
4648
4649                 now = -addr & ~PAGE_MASK;
4650                 rc = emulator_read_write_onepage(addr, val, now, exception,
4651                                                  vcpu, ops);
4652
4653                 if (rc != X86EMUL_CONTINUE)
4654                         return rc;
4655                 addr += now;
4656                 if (ctxt->mode != X86EMUL_MODE_PROT64)
4657                         addr = (u32)addr;
4658                 val += now;
4659                 bytes -= now;
4660         }
4661
4662         rc = emulator_read_write_onepage(addr, val, bytes, exception,
4663                                          vcpu, ops);
4664         if (rc != X86EMUL_CONTINUE)
4665                 return rc;
4666
4667         if (!vcpu->mmio_nr_fragments)
4668                 return rc;
4669
4670         gpa = vcpu->mmio_fragments[0].gpa;
4671
4672         vcpu->mmio_needed = 1;
4673         vcpu->mmio_cur_fragment = 0;
4674
4675         vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4676         vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4677         vcpu->run->exit_reason = KVM_EXIT_MMIO;
4678         vcpu->run->mmio.phys_addr = gpa;
4679
4680         return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4681 }
4682
4683 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4684                                   unsigned long addr,
4685                                   void *val,
4686                                   unsigned int bytes,
4687                                   struct x86_exception *exception)
4688 {
4689         return emulator_read_write(ctxt, addr, val, bytes,
4690                                    exception, &read_emultor);
4691 }
4692
4693 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4694                             unsigned long addr,
4695                             const void *val,
4696                             unsigned int bytes,
4697                             struct x86_exception *exception)
4698 {
4699         return emulator_read_write(ctxt, addr, (void *)val, bytes,
4700                                    exception, &write_emultor);
4701 }
4702
4703 #define CMPXCHG_TYPE(t, ptr, old, new) \
4704         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4705
4706 #ifdef CONFIG_X86_64
4707 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4708 #else
4709 #  define CMPXCHG64(ptr, old, new) \
4710         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4711 #endif
4712
4713 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4714                                      unsigned long addr,
4715                                      const void *old,
4716                                      const void *new,
4717                                      unsigned int bytes,
4718                                      struct x86_exception *exception)
4719 {
4720         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4721         gpa_t gpa;
4722         struct page *page;
4723         char *kaddr;
4724         bool exchanged;
4725
4726         /* guests cmpxchg8b have to be emulated atomically */
4727         if (bytes > 8 || (bytes & (bytes - 1)))
4728                 goto emul_write;
4729
4730         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4731
4732         if (gpa == UNMAPPED_GVA ||
4733             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4734                 goto emul_write;
4735
4736         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4737                 goto emul_write;
4738
4739         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4740         if (is_error_page(page))
4741                 goto emul_write;
4742
4743         kaddr = kmap_atomic(page);
4744         kaddr += offset_in_page(gpa);
4745         switch (bytes) {
4746         case 1:
4747                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4748                 break;
4749         case 2:
4750                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4751                 break;
4752         case 4:
4753                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4754                 break;
4755         case 8:
4756                 exchanged = CMPXCHG64(kaddr, old, new);
4757                 break;
4758         default:
4759                 BUG();
4760         }
4761         kunmap_atomic(kaddr);
4762         kvm_release_page_dirty(page);
4763
4764         if (!exchanged)
4765                 return X86EMUL_CMPXCHG_FAILED;
4766
4767         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4768         kvm_page_track_write(vcpu, gpa, new, bytes);
4769
4770         return X86EMUL_CONTINUE;
4771
4772 emul_write:
4773         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4774
4775         return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4776 }
4777
4778 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4779 {
4780         /* TODO: String I/O for in kernel device */
4781         int r;
4782
4783         if (vcpu->arch.pio.in)
4784                 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4785                                     vcpu->arch.pio.size, pd);
4786         else
4787                 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4788                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
4789                                      pd);
4790         return r;
4791 }
4792
4793 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4794                                unsigned short port, void *val,
4795                                unsigned int count, bool in)
4796 {
4797         vcpu->arch.pio.port = port;
4798         vcpu->arch.pio.in = in;
4799         vcpu->arch.pio.count  = count;
4800         vcpu->arch.pio.size = size;
4801
4802         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4803                 vcpu->arch.pio.count = 0;
4804                 return 1;
4805         }
4806
4807         vcpu->run->exit_reason = KVM_EXIT_IO;
4808         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4809         vcpu->run->io.size = size;
4810         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4811         vcpu->run->io.count = count;
4812         vcpu->run->io.port = port;
4813
4814         return 0;
4815 }
4816
4817 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4818                                     int size, unsigned short port, void *val,
4819                                     unsigned int count)
4820 {
4821         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4822         int ret;
4823
4824         if (vcpu->arch.pio.count)
4825                 goto data_avail;
4826
4827         ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4828         if (ret) {
4829 data_avail:
4830                 memcpy(val, vcpu->arch.pio_data, size * count);
4831                 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4832                 vcpu->arch.pio.count = 0;
4833                 return 1;
4834         }
4835
4836         return 0;
4837 }
4838
4839 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4840                                      int size, unsigned short port,
4841                                      const void *val, unsigned int count)
4842 {
4843         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4844
4845         memcpy(vcpu->arch.pio_data, val, size * count);
4846         trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4847         return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4848 }
4849
4850 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4851 {
4852         return kvm_x86_ops->get_segment_base(vcpu, seg);
4853 }
4854
4855 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4856 {
4857         kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4858 }
4859
4860 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4861 {
4862         if (!need_emulate_wbinvd(vcpu))
4863                 return X86EMUL_CONTINUE;
4864
4865         if (kvm_x86_ops->has_wbinvd_exit()) {
4866                 int cpu = get_cpu();
4867
4868                 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4869                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4870                                 wbinvd_ipi, NULL, 1);
4871                 put_cpu();
4872                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4873         } else
4874                 wbinvd();
4875         return X86EMUL_CONTINUE;
4876 }
4877
4878 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4879 {
4880         kvm_emulate_wbinvd_noskip(vcpu);
4881         return kvm_skip_emulated_instruction(vcpu);
4882 }
4883 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4884
4885
4886
4887 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4888 {
4889         kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4890 }
4891
4892 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4893                            unsigned long *dest)
4894 {
4895         return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4896 }
4897
4898 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4899                            unsigned long value)
4900 {
4901
4902         return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4903 }
4904
4905 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4906 {
4907         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4908 }
4909
4910 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4911 {
4912         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4913         unsigned long value;
4914
4915         switch (cr) {
4916         case 0:
4917                 value = kvm_read_cr0(vcpu);
4918                 break;
4919         case 2:
4920                 value = vcpu->arch.cr2;
4921                 break;
4922         case 3:
4923                 value = kvm_read_cr3(vcpu);
4924                 break;
4925         case 4:
4926                 value = kvm_read_cr4(vcpu);
4927                 break;
4928         case 8:
4929                 value = kvm_get_cr8(vcpu);
4930                 break;
4931         default:
4932                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4933                 return 0;
4934         }
4935
4936         return value;
4937 }
4938
4939 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4940 {
4941         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4942         int res = 0;
4943
4944         switch (cr) {
4945         case 0:
4946                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4947                 break;
4948         case 2:
4949                 vcpu->arch.cr2 = val;
4950                 break;
4951         case 3:
4952                 res = kvm_set_cr3(vcpu, val);
4953                 break;
4954         case 4:
4955                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4956                 break;
4957         case 8:
4958                 res = kvm_set_cr8(vcpu, val);
4959                 break;
4960         default:
4961                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4962                 res = -1;
4963         }
4964
4965         return res;
4966 }
4967
4968 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4969 {
4970         return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4971 }
4972
4973 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4974 {
4975         kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4976 }
4977
4978 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4979 {
4980         kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4981 }
4982
4983 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4984 {
4985         kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4986 }
4987
4988 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4989 {
4990         kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4991 }
4992
4993 static unsigned long emulator_get_cached_segment_base(
4994         struct x86_emulate_ctxt *ctxt, int seg)
4995 {
4996         return get_segment_base(emul_to_vcpu(ctxt), seg);
4997 }
4998
4999 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5000                                  struct desc_struct *desc, u32 *base3,
5001                                  int seg)
5002 {
5003         struct kvm_segment var;
5004
5005         kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5006         *selector = var.selector;
5007
5008         if (var.unusable) {
5009                 memset(desc, 0, sizeof(*desc));
5010                 return false;
5011         }
5012
5013         if (var.g)
5014                 var.limit >>= 12;
5015         set_desc_limit(desc, var.limit);
5016         set_desc_base(desc, (unsigned long)var.base);
5017 #ifdef CONFIG_X86_64
5018         if (base3)
5019                 *base3 = var.base >> 32;
5020 #endif
5021         desc->type = var.type;
5022         desc->s = var.s;
5023         desc->dpl = var.dpl;
5024         desc->p = var.present;
5025         desc->avl = var.avl;
5026         desc->l = var.l;
5027         desc->d = var.db;
5028         desc->g = var.g;
5029
5030         return true;
5031 }
5032
5033 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5034                                  struct desc_struct *desc, u32 base3,
5035                                  int seg)
5036 {
5037         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5038         struct kvm_segment var;
5039
5040         var.selector = selector;
5041         var.base = get_desc_base(desc);
5042 #ifdef CONFIG_X86_64
5043         var.base |= ((u64)base3) << 32;
5044 #endif
5045         var.limit = get_desc_limit(desc);
5046         if (desc->g)
5047                 var.limit = (var.limit << 12) | 0xfff;
5048         var.type = desc->type;
5049         var.dpl = desc->dpl;
5050         var.db = desc->d;
5051         var.s = desc->s;
5052         var.l = desc->l;
5053         var.g = desc->g;
5054         var.avl = desc->avl;
5055         var.present = desc->p;
5056         var.unusable = !var.present;
5057         var.padding = 0;
5058
5059         kvm_set_segment(vcpu, &var, seg);
5060         return;
5061 }
5062
5063 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5064                             u32 msr_index, u64 *pdata)
5065 {
5066         struct msr_data msr;
5067         int r;
5068
5069         msr.index = msr_index;
5070         msr.host_initiated = false;
5071         r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5072         if (r)
5073                 return r;
5074
5075         *pdata = msr.data;
5076         return 0;
5077 }
5078
5079 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5080                             u32 msr_index, u64 data)
5081 {
5082         struct msr_data msr;
5083
5084         msr.data = data;
5085         msr.index = msr_index;
5086         msr.host_initiated = false;
5087         return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5088 }
5089
5090 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5091 {
5092         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5093
5094         return vcpu->arch.smbase;
5095 }
5096
5097 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5098 {
5099         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5100
5101         vcpu->arch.smbase = smbase;
5102 }
5103
5104 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5105                               u32 pmc)
5106 {
5107         return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5108 }
5109
5110 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5111                              u32 pmc, u64 *pdata)
5112 {
5113         return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5114 }
5115
5116 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5117 {
5118         emul_to_vcpu(ctxt)->arch.halt_request = 1;
5119 }
5120
5121 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5122 {
5123         preempt_disable();
5124         kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5125 }
5126
5127 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5128 {
5129         preempt_enable();
5130 }
5131
5132 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5133                               struct x86_instruction_info *info,
5134                               enum x86_intercept_stage stage)
5135 {
5136         return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5137 }
5138
5139 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5140                                u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5141 {
5142         kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5143 }
5144
5145 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5146 {
5147         return kvm_register_read(emul_to_vcpu(ctxt), reg);
5148 }
5149
5150 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5151 {
5152         kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5153 }
5154
5155 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5156 {
5157         kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5158 }
5159
5160 static const struct x86_emulate_ops emulate_ops = {
5161         .read_gpr            = emulator_read_gpr,
5162         .write_gpr           = emulator_write_gpr,
5163         .read_std            = kvm_read_guest_virt_system,
5164         .write_std           = kvm_write_guest_virt_system,
5165         .read_phys           = kvm_read_guest_phys_system,
5166         .fetch               = kvm_fetch_guest_virt,
5167         .read_emulated       = emulator_read_emulated,
5168         .write_emulated      = emulator_write_emulated,
5169         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
5170         .invlpg              = emulator_invlpg,
5171         .pio_in_emulated     = emulator_pio_in_emulated,
5172         .pio_out_emulated    = emulator_pio_out_emulated,
5173         .get_segment         = emulator_get_segment,
5174         .set_segment         = emulator_set_segment,
5175         .get_cached_segment_base = emulator_get_cached_segment_base,
5176         .get_gdt             = emulator_get_gdt,
5177         .get_idt             = emulator_get_idt,
5178         .set_gdt             = emulator_set_gdt,
5179         .set_idt             = emulator_set_idt,
5180         .get_cr              = emulator_get_cr,
5181         .set_cr              = emulator_set_cr,
5182         .cpl                 = emulator_get_cpl,
5183         .get_dr              = emulator_get_dr,
5184         .set_dr              = emulator_set_dr,
5185         .get_smbase          = emulator_get_smbase,
5186         .set_smbase          = emulator_set_smbase,
5187         .set_msr             = emulator_set_msr,
5188         .get_msr             = emulator_get_msr,
5189         .check_pmc           = emulator_check_pmc,
5190         .read_pmc            = emulator_read_pmc,
5191         .halt                = emulator_halt,
5192         .wbinvd              = emulator_wbinvd,
5193         .fix_hypercall       = emulator_fix_hypercall,
5194         .get_fpu             = emulator_get_fpu,
5195         .put_fpu             = emulator_put_fpu,
5196         .intercept           = emulator_intercept,
5197         .get_cpuid           = emulator_get_cpuid,
5198         .set_nmi_mask        = emulator_set_nmi_mask,
5199 };
5200
5201 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5202 {
5203         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5204         /*
5205          * an sti; sti; sequence only disable interrupts for the first
5206          * instruction. So, if the last instruction, be it emulated or
5207          * not, left the system with the INT_STI flag enabled, it
5208          * means that the last instruction is an sti. We should not
5209          * leave the flag on in this case. The same goes for mov ss
5210          */
5211         if (int_shadow & mask)
5212                 mask = 0;
5213         if (unlikely(int_shadow || mask)) {
5214                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5215                 if (!mask)
5216                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5217         }
5218 }
5219
5220 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5221 {
5222         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5223         if (ctxt->exception.vector == PF_VECTOR)
5224                 return kvm_propagate_fault(vcpu, &ctxt->exception);
5225
5226         if (ctxt->exception.error_code_valid)
5227                 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5228                                       ctxt->exception.error_code);
5229         else
5230                 kvm_queue_exception(vcpu, ctxt->exception.vector);
5231         return false;
5232 }
5233
5234 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5235 {
5236         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5237         int cs_db, cs_l;
5238
5239         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5240
5241         ctxt->eflags = kvm_get_rflags(vcpu);
5242         ctxt->eip = kvm_rip_read(vcpu);
5243         ctxt->mode = (!is_protmode(vcpu))               ? X86EMUL_MODE_REAL :
5244                      (ctxt->eflags & X86_EFLAGS_VM)     ? X86EMUL_MODE_VM86 :
5245                      (cs_l && is_long_mode(vcpu))       ? X86EMUL_MODE_PROT64 :
5246                      cs_db                              ? X86EMUL_MODE_PROT32 :
5247                                                           X86EMUL_MODE_PROT16;
5248         BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5249         BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5250         BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5251         ctxt->emul_flags = vcpu->arch.hflags;
5252
5253         init_decode_cache(ctxt);
5254         vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5255 }
5256
5257 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5258 {
5259         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5260         int ret;
5261
5262         init_emulate_ctxt(vcpu);
5263
5264         ctxt->op_bytes = 2;
5265         ctxt->ad_bytes = 2;
5266         ctxt->_eip = ctxt->eip + inc_eip;
5267         ret = emulate_int_real(ctxt, irq);
5268
5269         if (ret != X86EMUL_CONTINUE)
5270                 return EMULATE_FAIL;
5271
5272         ctxt->eip = ctxt->_eip;
5273         kvm_rip_write(vcpu, ctxt->eip);
5274         kvm_set_rflags(vcpu, ctxt->eflags);
5275
5276         if (irq == NMI_VECTOR)
5277                 vcpu->arch.nmi_pending = 0;
5278         else
5279                 vcpu->arch.interrupt.pending = false;
5280
5281         return EMULATE_DONE;
5282 }
5283 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5284
5285 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5286 {
5287         int r = EMULATE_DONE;
5288
5289         ++vcpu->stat.insn_emulation_fail;
5290         trace_kvm_emulate_insn_failed(vcpu);
5291         if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5292                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5293                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5294                 vcpu->run->internal.ndata = 0;
5295                 r = EMULATE_FAIL;
5296         }
5297         kvm_queue_exception(vcpu, UD_VECTOR);
5298
5299         return r;
5300 }
5301
5302 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5303                                   bool write_fault_to_shadow_pgtable,
5304                                   int emulation_type)
5305 {
5306         gpa_t gpa = cr2;
5307         kvm_pfn_t pfn;
5308
5309         if (emulation_type & EMULTYPE_NO_REEXECUTE)
5310                 return false;
5311
5312         if (!vcpu->arch.mmu.direct_map) {
5313                 /*
5314                  * Write permission should be allowed since only
5315                  * write access need to be emulated.
5316                  */
5317                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5318
5319                 /*
5320                  * If the mapping is invalid in guest, let cpu retry
5321                  * it to generate fault.
5322                  */
5323                 if (gpa == UNMAPPED_GVA)
5324                         return true;
5325         }
5326
5327         /*
5328          * Do not retry the unhandleable instruction if it faults on the
5329          * readonly host memory, otherwise it will goto a infinite loop:
5330          * retry instruction -> write #PF -> emulation fail -> retry
5331          * instruction -> ...
5332          */
5333         pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5334
5335         /*
5336          * If the instruction failed on the error pfn, it can not be fixed,
5337          * report the error to userspace.
5338          */
5339         if (is_error_noslot_pfn(pfn))
5340                 return false;
5341
5342         kvm_release_pfn_clean(pfn);
5343
5344         /* The instructions are well-emulated on direct mmu. */
5345         if (vcpu->arch.mmu.direct_map) {
5346                 unsigned int indirect_shadow_pages;
5347
5348                 spin_lock(&vcpu->kvm->mmu_lock);
5349                 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5350                 spin_unlock(&vcpu->kvm->mmu_lock);
5351
5352                 if (indirect_shadow_pages)
5353                         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5354
5355                 return true;
5356         }
5357
5358         /*
5359          * if emulation was due to access to shadowed page table
5360          * and it failed try to unshadow page and re-enter the
5361          * guest to let CPU execute the instruction.
5362          */
5363         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5364
5365         /*
5366          * If the access faults on its page table, it can not
5367          * be fixed by unprotecting shadow page and it should
5368          * be reported to userspace.
5369          */
5370         return !write_fault_to_shadow_pgtable;
5371 }
5372
5373 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5374                               unsigned long cr2,  int emulation_type)
5375 {
5376         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5377         unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5378
5379         last_retry_eip = vcpu->arch.last_retry_eip;
5380         last_retry_addr = vcpu->arch.last_retry_addr;
5381
5382         /*
5383          * If the emulation is caused by #PF and it is non-page_table
5384          * writing instruction, it means the VM-EXIT is caused by shadow
5385          * page protected, we can zap the shadow page and retry this
5386          * instruction directly.
5387          *
5388          * Note: if the guest uses a non-page-table modifying instruction
5389          * on the PDE that points to the instruction, then we will unmap
5390          * the instruction and go to an infinite loop. So, we cache the
5391          * last retried eip and the last fault address, if we meet the eip
5392          * and the address again, we can break out of the potential infinite
5393          * loop.
5394          */
5395         vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5396
5397         if (!(emulation_type & EMULTYPE_RETRY))
5398                 return false;
5399
5400         if (x86_page_table_writing_insn(ctxt))
5401                 return false;
5402
5403         if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5404                 return false;
5405
5406         vcpu->arch.last_retry_eip = ctxt->eip;
5407         vcpu->arch.last_retry_addr = cr2;
5408
5409         if (!vcpu->arch.mmu.direct_map)
5410                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5411
5412         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5413
5414         return true;
5415 }
5416
5417 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5418 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5419
5420 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5421 {
5422         if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5423                 /* This is a good place to trace that we are exiting SMM.  */
5424                 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5425
5426                 /* Process a latched INIT or SMI, if any.  */
5427                 kvm_make_request(KVM_REQ_EVENT, vcpu);
5428         }
5429
5430         kvm_mmu_reset_context(vcpu);
5431 }
5432
5433 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5434 {
5435         unsigned changed = vcpu->arch.hflags ^ emul_flags;
5436
5437         vcpu->arch.hflags = emul_flags;
5438
5439         if (changed & HF_SMM_MASK)
5440                 kvm_smm_changed(vcpu);
5441 }
5442
5443 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5444                                 unsigned long *db)
5445 {
5446         u32 dr6 = 0;
5447         int i;
5448         u32 enable, rwlen;
5449
5450         enable = dr7;
5451         rwlen = dr7 >> 16;
5452         for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5453                 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5454                         dr6 |= (1 << i);
5455         return dr6;
5456 }
5457
5458 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5459 {
5460         struct kvm_run *kvm_run = vcpu->run;
5461
5462         /*
5463          * rflags is the old, "raw" value of the flags.  The new value has
5464          * not been saved yet.
5465          *
5466          * This is correct even for TF set by the guest, because "the
5467          * processor will not generate this exception after the instruction
5468          * that sets the TF flag".
5469          */
5470         if (unlikely(rflags & X86_EFLAGS_TF)) {
5471                 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5472                         kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5473                                                   DR6_RTM;
5474                         kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5475                         kvm_run->debug.arch.exception = DB_VECTOR;
5476                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
5477                         *r = EMULATE_USER_EXIT;
5478                 } else {
5479                         /*
5480                          * "Certain debug exceptions may clear bit 0-3.  The
5481                          * remaining contents of the DR6 register are never
5482                          * cleared by the processor".
5483                          */
5484                         vcpu->arch.dr6 &= ~15;
5485                         vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5486                         kvm_queue_exception(vcpu, DB_VECTOR);
5487                 }
5488         }
5489 }
5490
5491 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5492 {
5493         unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5494         int r = EMULATE_DONE;
5495
5496         kvm_x86_ops->skip_emulated_instruction(vcpu);
5497         kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5498         return r == EMULATE_DONE;
5499 }
5500 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5501
5502 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5503 {
5504         if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5505             (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5506                 struct kvm_run *kvm_run = vcpu->run;
5507                 unsigned long eip = kvm_get_linear_rip(vcpu);
5508                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5509                                            vcpu->arch.guest_debug_dr7,
5510                                            vcpu->arch.eff_db);
5511
5512                 if (dr6 != 0) {
5513                         kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5514                         kvm_run->debug.arch.pc = eip;
5515                         kvm_run->debug.arch.exception = DB_VECTOR;
5516                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
5517                         *r = EMULATE_USER_EXIT;
5518                         return true;
5519                 }
5520         }
5521
5522         if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5523             !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5524                 unsigned long eip = kvm_get_linear_rip(vcpu);
5525                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5526                                            vcpu->arch.dr7,
5527                                            vcpu->arch.db);
5528
5529                 if (dr6 != 0) {
5530                         vcpu->arch.dr6 &= ~15;
5531                         vcpu->arch.dr6 |= dr6 | DR6_RTM;
5532                         kvm_queue_exception(vcpu, DB_VECTOR);
5533                         *r = EMULATE_DONE;
5534                         return true;
5535                 }
5536         }
5537
5538         return false;
5539 }
5540
5541 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5542                             unsigned long cr2,
5543                             int emulation_type,
5544                             void *insn,
5545                             int insn_len)
5546 {
5547         int r;
5548         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5549         bool writeback = true;
5550         bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5551
5552         /*
5553          * Clear write_fault_to_shadow_pgtable here to ensure it is
5554          * never reused.
5555          */
5556         vcpu->arch.write_fault_to_shadow_pgtable = false;
5557         kvm_clear_exception_queue(vcpu);
5558
5559         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5560                 init_emulate_ctxt(vcpu);
5561
5562                 /*
5563                  * We will reenter on the same instruction since
5564                  * we do not set complete_userspace_io.  This does not
5565                  * handle watchpoints yet, those would be handled in
5566                  * the emulate_ops.
5567                  */
5568                 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5569                         return r;
5570
5571                 ctxt->interruptibility = 0;
5572                 ctxt->have_exception = false;
5573                 ctxt->exception.vector = -1;
5574                 ctxt->perm_ok = false;
5575
5576                 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5577
5578                 r = x86_decode_insn(ctxt, insn, insn_len);
5579
5580                 trace_kvm_emulate_insn_start(vcpu);
5581                 ++vcpu->stat.insn_emulation;
5582                 if (r != EMULATION_OK)  {
5583                         if (emulation_type & EMULTYPE_TRAP_UD)
5584                                 return EMULATE_FAIL;
5585                         if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5586                                                 emulation_type))
5587                                 return EMULATE_DONE;
5588                         if (emulation_type & EMULTYPE_SKIP)
5589                                 return EMULATE_FAIL;
5590                         return handle_emulation_failure(vcpu);
5591                 }
5592         }
5593
5594         if (emulation_type & EMULTYPE_SKIP) {
5595                 kvm_rip_write(vcpu, ctxt->_eip);
5596                 if (ctxt->eflags & X86_EFLAGS_RF)
5597                         kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5598                 return EMULATE_DONE;
5599         }
5600
5601         if (retry_instruction(ctxt, cr2, emulation_type))
5602                 return EMULATE_DONE;
5603
5604         /* this is needed for vmware backdoor interface to work since it
5605            changes registers values  during IO operation */
5606         if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5607                 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5608                 emulator_invalidate_register_cache(ctxt);
5609         }
5610
5611 restart:
5612         r = x86_emulate_insn(ctxt);
5613
5614         if (r == EMULATION_INTERCEPTED)
5615                 return EMULATE_DONE;
5616
5617         if (r == EMULATION_FAILED) {
5618                 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5619                                         emulation_type))
5620                         return EMULATE_DONE;
5621
5622                 return handle_emulation_failure(vcpu);
5623         }
5624
5625         if (ctxt->have_exception) {
5626                 r = EMULATE_DONE;
5627                 if (inject_emulated_exception(vcpu))
5628                         return r;
5629         } else if (vcpu->arch.pio.count) {
5630                 if (!vcpu->arch.pio.in) {
5631                         /* FIXME: return into emulator if single-stepping.  */
5632                         vcpu->arch.pio.count = 0;
5633                 } else {
5634                         writeback = false;
5635                         vcpu->arch.complete_userspace_io = complete_emulated_pio;
5636                 }
5637                 r = EMULATE_USER_EXIT;
5638         } else if (vcpu->mmio_needed) {
5639                 if (!vcpu->mmio_is_write)
5640                         writeback = false;
5641                 r = EMULATE_USER_EXIT;
5642                 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5643         } else if (r == EMULATION_RESTART)
5644                 goto restart;
5645         else
5646                 r = EMULATE_DONE;
5647
5648         if (writeback) {
5649                 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5650                 toggle_interruptibility(vcpu, ctxt->interruptibility);
5651                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5652                 if (vcpu->arch.hflags != ctxt->emul_flags)
5653                         kvm_set_hflags(vcpu, ctxt->emul_flags);
5654                 kvm_rip_write(vcpu, ctxt->eip);
5655                 if (r == EMULATE_DONE)
5656                         kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5657                 if (!ctxt->have_exception ||
5658                     exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5659                         __kvm_set_rflags(vcpu, ctxt->eflags);
5660
5661                 /*
5662                  * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5663                  * do nothing, and it will be requested again as soon as
5664                  * the shadow expires.  But we still need to check here,
5665                  * because POPF has no interrupt shadow.
5666                  */
5667                 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5668                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5669         } else
5670                 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5671
5672         return r;
5673 }
5674 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5675
5676 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5677 {
5678         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5679         int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5680                                             size, port, &val, 1);
5681         /* do not return to emulator after return from userspace */
5682         vcpu->arch.pio.count = 0;
5683         return ret;
5684 }
5685 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5686
5687 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5688 {
5689         unsigned long val;
5690
5691         /* We should only ever be called with arch.pio.count equal to 1 */
5692         BUG_ON(vcpu->arch.pio.count != 1);
5693
5694         /* For size less than 4 we merge, else we zero extend */
5695         val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5696                                         : 0;
5697
5698         /*
5699          * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5700          * the copy and tracing
5701          */
5702         emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5703                                  vcpu->arch.pio.port, &val, 1);
5704         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5705
5706         return 1;
5707 }
5708
5709 int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5710 {
5711         unsigned long val;
5712         int ret;
5713
5714         /* For size less than 4 we merge, else we zero extend */
5715         val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5716
5717         ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5718                                        &val, 1);
5719         if (ret) {
5720                 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5721                 return ret;
5722         }
5723
5724         vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5725
5726         return 0;
5727 }
5728 EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5729
5730 static int kvmclock_cpu_down_prep(unsigned int cpu)
5731 {
5732         __this_cpu_write(cpu_tsc_khz, 0);
5733         return 0;
5734 }
5735
5736 static void tsc_khz_changed(void *data)
5737 {
5738         struct cpufreq_freqs *freq = data;
5739         unsigned long khz = 0;
5740
5741         if (data)
5742                 khz = freq->new;
5743         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5744                 khz = cpufreq_quick_get(raw_smp_processor_id());
5745         if (!khz)
5746                 khz = tsc_khz;
5747         __this_cpu_write(cpu_tsc_khz, khz);
5748 }
5749
5750 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5751                                      void *data)
5752 {
5753         struct cpufreq_freqs *freq = data;
5754         struct kvm *kvm;
5755         struct kvm_vcpu *vcpu;
5756         int i, send_ipi = 0;
5757
5758         /*
5759          * We allow guests to temporarily run on slowing clocks,
5760          * provided we notify them after, or to run on accelerating
5761          * clocks, provided we notify them before.  Thus time never
5762          * goes backwards.
5763          *
5764          * However, we have a problem.  We can't atomically update
5765          * the frequency of a given CPU from this function; it is
5766          * merely a notifier, which can be called from any CPU.
5767          * Changing the TSC frequency at arbitrary points in time
5768          * requires a recomputation of local variables related to
5769          * the TSC for each VCPU.  We must flag these local variables
5770          * to be updated and be sure the update takes place with the
5771          * new frequency before any guests proceed.
5772          *
5773          * Unfortunately, the combination of hotplug CPU and frequency
5774          * change creates an intractable locking scenario; the order
5775          * of when these callouts happen is undefined with respect to
5776          * CPU hotplug, and they can race with each other.  As such,
5777          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5778          * undefined; you can actually have a CPU frequency change take
5779          * place in between the computation of X and the setting of the
5780          * variable.  To protect against this problem, all updates of
5781          * the per_cpu tsc_khz variable are done in an interrupt
5782          * protected IPI, and all callers wishing to update the value
5783          * must wait for a synchronous IPI to complete (which is trivial
5784          * if the caller is on the CPU already).  This establishes the
5785          * necessary total order on variable updates.
5786          *
5787          * Note that because a guest time update may take place
5788          * anytime after the setting of the VCPU's request bit, the
5789          * correct TSC value must be set before the request.  However,
5790          * to ensure the update actually makes it to any guest which
5791          * starts running in hardware virtualization between the set
5792          * and the acquisition of the spinlock, we must also ping the
5793          * CPU after setting the request bit.
5794          *
5795          */
5796
5797         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5798                 return 0;
5799         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5800                 return 0;
5801
5802         smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5803
5804         spin_lock(&kvm_lock);
5805         list_for_each_entry(kvm, &vm_list, vm_list) {
5806                 kvm_for_each_vcpu(i, vcpu, kvm) {
5807                         if (vcpu->cpu != freq->cpu)
5808                                 continue;
5809                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5810                         if (vcpu->cpu != smp_processor_id())
5811                                 send_ipi = 1;
5812                 }
5813         }
5814         spin_unlock(&kvm_lock);
5815
5816         if (freq->old < freq->new && send_ipi) {
5817                 /*
5818                  * We upscale the frequency.  Must make the guest
5819                  * doesn't see old kvmclock values while running with
5820                  * the new frequency, otherwise we risk the guest sees
5821                  * time go backwards.
5822                  *
5823                  * In case we update the frequency for another cpu
5824                  * (which might be in guest context) send an interrupt
5825                  * to kick the cpu out of guest context.  Next time
5826                  * guest context is entered kvmclock will be updated,
5827                  * so the guest will not see stale values.
5828                  */
5829                 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5830         }
5831         return 0;
5832 }
5833
5834 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5835         .notifier_call  = kvmclock_cpufreq_notifier
5836 };
5837
5838 static int kvmclock_cpu_online(unsigned int cpu)
5839 {
5840         tsc_khz_changed(NULL);
5841         return 0;
5842 }
5843
5844 static void kvm_timer_init(void)
5845 {
5846         max_tsc_khz = tsc_khz;
5847
5848         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5849 #ifdef CONFIG_CPU_FREQ
5850                 struct cpufreq_policy policy;
5851                 int cpu;
5852
5853                 memset(&policy, 0, sizeof(policy));
5854                 cpu = get_cpu();
5855                 cpufreq_get_policy(&policy, cpu);
5856                 if (policy.cpuinfo.max_freq)
5857                         max_tsc_khz = policy.cpuinfo.max_freq;
5858                 put_cpu();
5859 #endif
5860                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5861                                           CPUFREQ_TRANSITION_NOTIFIER);
5862         }
5863         pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5864
5865         cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
5866                           kvmclock_cpu_online, kvmclock_cpu_down_prep);
5867 }
5868
5869 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5870
5871 int kvm_is_in_guest(void)
5872 {
5873         return __this_cpu_read(current_vcpu) != NULL;
5874 }
5875
5876 static int kvm_is_user_mode(void)
5877 {
5878         int user_mode = 3;
5879
5880         if (__this_cpu_read(current_vcpu))
5881                 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5882
5883         return user_mode != 0;
5884 }
5885
5886 static unsigned long kvm_get_guest_ip(void)
5887 {
5888         unsigned long ip = 0;
5889
5890         if (__this_cpu_read(current_vcpu))
5891                 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5892
5893         return ip;
5894 }
5895
5896 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5897         .is_in_guest            = kvm_is_in_guest,
5898         .is_user_mode           = kvm_is_user_mode,
5899         .get_guest_ip           = kvm_get_guest_ip,
5900 };
5901
5902 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5903 {
5904         __this_cpu_write(current_vcpu, vcpu);
5905 }
5906 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5907
5908 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5909 {
5910         __this_cpu_write(current_vcpu, NULL);
5911 }
5912 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5913
5914 static void kvm_set_mmio_spte_mask(void)
5915 {
5916         u64 mask;
5917         int maxphyaddr = boot_cpu_data.x86_phys_bits;
5918
5919         /*
5920          * Set the reserved bits and the present bit of an paging-structure
5921          * entry to generate page fault with PFER.RSV = 1.
5922          */
5923          /* Mask the reserved physical address bits. */
5924         mask = rsvd_bits(maxphyaddr, 51);
5925
5926         /* Bit 62 is always reserved for 32bit host. */
5927         mask |= 0x3ull << 62;
5928
5929         /* Set the present bit. */
5930         mask |= 1ull;
5931
5932 #ifdef CONFIG_X86_64
5933         /*
5934          * If reserved bit is not supported, clear the present bit to disable
5935          * mmio page fault.
5936          */
5937         if (maxphyaddr == 52)
5938                 mask &= ~1ull;
5939 #endif
5940
5941         kvm_mmu_set_mmio_spte_mask(mask);
5942 }
5943
5944 #ifdef CONFIG_X86_64
5945 static void pvclock_gtod_update_fn(struct work_struct *work)
5946 {
5947         struct kvm *kvm;
5948
5949         struct kvm_vcpu *vcpu;
5950         int i;
5951
5952         spin_lock(&kvm_lock);
5953         list_for_each_entry(kvm, &vm_list, vm_list)
5954                 kvm_for_each_vcpu(i, vcpu, kvm)
5955                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5956         atomic_set(&kvm_guest_has_master_clock, 0);
5957         spin_unlock(&kvm_lock);
5958 }
5959
5960 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5961
5962 /*
5963  * Notification about pvclock gtod data update.
5964  */
5965 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5966                                void *priv)
5967 {
5968         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5969         struct timekeeper *tk = priv;
5970
5971         update_pvclock_gtod(tk);
5972
5973         /* disable master clock if host does not trust, or does not
5974          * use, TSC clocksource
5975          */
5976         if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5977             atomic_read(&kvm_guest_has_master_clock) != 0)
5978                 queue_work(system_long_wq, &pvclock_gtod_work);
5979
5980         return 0;
5981 }
5982
5983 static struct notifier_block pvclock_gtod_notifier = {
5984         .notifier_call = pvclock_gtod_notify,
5985 };
5986 #endif
5987
5988 int kvm_arch_init(void *opaque)
5989 {
5990         int r;
5991         struct kvm_x86_ops *ops = opaque;
5992
5993         if (kvm_x86_ops) {
5994                 printk(KERN_ERR "kvm: already loaded the other module\n");
5995                 r = -EEXIST;
5996                 goto out;
5997         }
5998
5999         if (!ops->cpu_has_kvm_support()) {
6000                 printk(KERN_ERR "kvm: no hardware support\n");
6001                 r = -EOPNOTSUPP;
6002                 goto out;
6003         }
6004         if (ops->disabled_by_bios()) {
6005                 printk(KERN_ERR "kvm: disabled by bios\n");
6006                 r = -EOPNOTSUPP;
6007                 goto out;
6008         }
6009
6010         r = -ENOMEM;
6011         shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6012         if (!shared_msrs) {
6013                 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6014                 goto out;
6015         }
6016
6017         r = kvm_mmu_module_init();
6018         if (r)
6019                 goto out_free_percpu;
6020
6021         kvm_set_mmio_spte_mask();
6022
6023         kvm_x86_ops = ops;
6024
6025         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6026                         PT_DIRTY_MASK, PT64_NX_MASK, 0,
6027                         PT_PRESENT_MASK);
6028         kvm_timer_init();
6029
6030         perf_register_guest_info_callbacks(&kvm_guest_cbs);
6031
6032         if (boot_cpu_has(X86_FEATURE_XSAVE))
6033                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6034
6035         kvm_lapic_init();
6036 #ifdef CONFIG_X86_64
6037         pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6038 #endif
6039
6040         return 0;
6041
6042 out_free_percpu:
6043         free_percpu(shared_msrs);
6044 out:
6045         return r;
6046 }
6047
6048 void kvm_arch_exit(void)
6049 {
6050         kvm_lapic_exit();
6051         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6052
6053         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6054                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6055                                             CPUFREQ_TRANSITION_NOTIFIER);
6056         cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6057 #ifdef CONFIG_X86_64
6058         pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6059 #endif
6060         kvm_x86_ops = NULL;
6061         kvm_mmu_module_exit();
6062         free_percpu(shared_msrs);
6063 }
6064
6065 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6066 {
6067         ++vcpu->stat.halt_exits;
6068         if (lapic_in_kernel(vcpu)) {
6069                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6070                 return 1;
6071         } else {
6072                 vcpu->run->exit_reason = KVM_EXIT_HLT;
6073                 return 0;
6074         }
6075 }
6076 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6077
6078 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6079 {
6080         int ret = kvm_skip_emulated_instruction(vcpu);
6081         /*
6082          * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6083          * KVM_EXIT_DEBUG here.
6084          */
6085         return kvm_vcpu_halt(vcpu) && ret;
6086 }
6087 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6088
6089 /*
6090  * kvm_pv_kick_cpu_op:  Kick a vcpu.
6091  *
6092  * @apicid - apicid of vcpu to be kicked.
6093  */
6094 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6095 {
6096         struct kvm_lapic_irq lapic_irq;
6097
6098         lapic_irq.shorthand = 0;
6099         lapic_irq.dest_mode = 0;
6100         lapic_irq.dest_id = apicid;
6101         lapic_irq.msi_redir_hint = false;
6102
6103         lapic_irq.delivery_mode = APIC_DM_REMRD;
6104         kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6105 }
6106
6107 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6108 {
6109         vcpu->arch.apicv_active = false;
6110         kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6111 }
6112
6113 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6114 {
6115         unsigned long nr, a0, a1, a2, a3, ret;
6116         int op_64_bit, r;
6117
6118         r = kvm_skip_emulated_instruction(vcpu);
6119
6120         if (kvm_hv_hypercall_enabled(vcpu->kvm))
6121                 return kvm_hv_hypercall(vcpu);
6122
6123         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6124         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6125         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6126         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6127         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6128
6129         trace_kvm_hypercall(nr, a0, a1, a2, a3);
6130
6131         op_64_bit = is_64_bit_mode(vcpu);
6132         if (!op_64_bit) {
6133                 nr &= 0xFFFFFFFF;
6134                 a0 &= 0xFFFFFFFF;
6135                 a1 &= 0xFFFFFFFF;
6136                 a2 &= 0xFFFFFFFF;
6137                 a3 &= 0xFFFFFFFF;
6138         }
6139
6140         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6141                 ret = -KVM_EPERM;
6142                 goto out;
6143         }
6144
6145         switch (nr) {
6146         case KVM_HC_VAPIC_POLL_IRQ:
6147                 ret = 0;
6148                 break;
6149         case KVM_HC_KICK_CPU:
6150                 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6151                 ret = 0;
6152                 break;
6153         default:
6154                 ret = -KVM_ENOSYS;
6155                 break;
6156         }
6157 out:
6158         if (!op_64_bit)
6159                 ret = (u32)ret;
6160         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6161         ++vcpu->stat.hypercalls;
6162         return r;
6163 }
6164 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6165
6166 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6167 {
6168         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6169         char instruction[3];
6170         unsigned long rip = kvm_rip_read(vcpu);
6171
6172         kvm_x86_ops->patch_hypercall(vcpu, instruction);
6173
6174         return emulator_write_emulated(ctxt, rip, instruction, 3,
6175                 &ctxt->exception);
6176 }
6177
6178 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6179 {
6180         return vcpu->run->request_interrupt_window &&
6181                 likely(!pic_in_kernel(vcpu->kvm));
6182 }
6183
6184 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6185 {
6186         struct kvm_run *kvm_run = vcpu->run;
6187
6188         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6189         kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6190         kvm_run->cr8 = kvm_get_cr8(vcpu);
6191         kvm_run->apic_base = kvm_get_apic_base(vcpu);
6192         kvm_run->ready_for_interrupt_injection =
6193                 pic_in_kernel(vcpu->kvm) ||
6194                 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6195 }
6196
6197 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6198 {
6199         int max_irr, tpr;
6200
6201         if (!kvm_x86_ops->update_cr8_intercept)
6202                 return;
6203
6204         if (!lapic_in_kernel(vcpu))
6205                 return;
6206
6207         if (vcpu->arch.apicv_active)
6208                 return;
6209
6210         if (!vcpu->arch.apic->vapic_addr)
6211                 max_irr = kvm_lapic_find_highest_irr(vcpu);
6212         else
6213                 max_irr = -1;
6214
6215         if (max_irr != -1)
6216                 max_irr >>= 4;
6217
6218         tpr = kvm_lapic_get_cr8(vcpu);
6219
6220         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6221 }
6222
6223 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6224 {
6225         int r;
6226
6227         /* try to reinject previous events if any */
6228         if (vcpu->arch.exception.pending) {
6229                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6230                                         vcpu->arch.exception.has_error_code,
6231                                         vcpu->arch.exception.error_code);
6232
6233                 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6234                         __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6235                                              X86_EFLAGS_RF);
6236
6237                 if (vcpu->arch.exception.nr == DB_VECTOR &&
6238                     (vcpu->arch.dr7 & DR7_GD)) {
6239                         vcpu->arch.dr7 &= ~DR7_GD;
6240                         kvm_update_dr7(vcpu);
6241                 }
6242
6243                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6244                                           vcpu->arch.exception.has_error_code,
6245                                           vcpu->arch.exception.error_code,
6246                                           vcpu->arch.exception.reinject);
6247                 return 0;
6248         }
6249
6250         if (vcpu->arch.nmi_injected) {
6251                 kvm_x86_ops->set_nmi(vcpu);
6252                 return 0;
6253         }
6254
6255         if (vcpu->arch.interrupt.pending) {
6256                 kvm_x86_ops->set_irq(vcpu);
6257                 return 0;
6258         }
6259
6260         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6261                 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6262                 if (r != 0)
6263                         return r;
6264         }
6265
6266         /* try to inject new event if pending */
6267         if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6268                 vcpu->arch.smi_pending = false;
6269                 enter_smm(vcpu);
6270         } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6271                 --vcpu->arch.nmi_pending;
6272                 vcpu->arch.nmi_injected = true;
6273                 kvm_x86_ops->set_nmi(vcpu);
6274         } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6275                 /*
6276                  * Because interrupts can be injected asynchronously, we are
6277                  * calling check_nested_events again here to avoid a race condition.
6278                  * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6279                  * proposal and current concerns.  Perhaps we should be setting
6280                  * KVM_REQ_EVENT only on certain events and not unconditionally?
6281                  */
6282                 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6283                         r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6284                         if (r != 0)
6285                                 return r;
6286                 }
6287                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6288                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6289                                             false);
6290                         kvm_x86_ops->set_irq(vcpu);
6291                 }
6292         }
6293
6294         return 0;
6295 }
6296
6297 static void process_nmi(struct kvm_vcpu *vcpu)
6298 {
6299         unsigned limit = 2;
6300
6301         /*
6302          * x86 is limited to one NMI running, and one NMI pending after it.
6303          * If an NMI is already in progress, limit further NMIs to just one.
6304          * Otherwise, allow two (and we'll inject the first one immediately).
6305          */
6306         if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6307                 limit = 1;
6308
6309         vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6310         vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6311         kvm_make_request(KVM_REQ_EVENT, vcpu);
6312 }
6313
6314 #define put_smstate(type, buf, offset, val)                       \
6315         *(type *)((buf) + (offset) - 0x7e00) = val
6316
6317 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6318 {
6319         u32 flags = 0;
6320         flags |= seg->g       << 23;
6321         flags |= seg->db      << 22;
6322         flags |= seg->l       << 21;
6323         flags |= seg->avl     << 20;
6324         flags |= seg->present << 15;
6325         flags |= seg->dpl     << 13;
6326         flags |= seg->s       << 12;
6327         flags |= seg->type    << 8;
6328         return flags;
6329 }
6330
6331 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6332 {
6333         struct kvm_segment seg;
6334         int offset;
6335
6336         kvm_get_segment(vcpu, &seg, n);
6337         put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6338
6339         if (n < 3)
6340                 offset = 0x7f84 + n * 12;
6341         else
6342                 offset = 0x7f2c + (n - 3) * 12;
6343
6344         put_smstate(u32, buf, offset + 8, seg.base);
6345         put_smstate(u32, buf, offset + 4, seg.limit);
6346         put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6347 }
6348
6349 #ifdef CONFIG_X86_64
6350 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6351 {
6352         struct kvm_segment seg;
6353         int offset;
6354         u16 flags;
6355
6356         kvm_get_segment(vcpu, &seg, n);
6357         offset = 0x7e00 + n * 16;
6358
6359         flags = enter_smm_get_segment_flags(&seg) >> 8;
6360         put_smstate(u16, buf, offset, seg.selector);
6361         put_smstate(u16, buf, offset + 2, flags);
6362         put_smstate(u32, buf, offset + 4, seg.limit);
6363         put_smstate(u64, buf, offset + 8, seg.base);
6364 }
6365 #endif
6366
6367 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6368 {
6369         struct desc_ptr dt;
6370         struct kvm_segment seg;
6371         unsigned long val;
6372         int i;
6373
6374         put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6375         put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6376         put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6377         put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6378
6379         for (i = 0; i < 8; i++)
6380                 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6381
6382         kvm_get_dr(vcpu, 6, &val);
6383         put_smstate(u32, buf, 0x7fcc, (u32)val);
6384         kvm_get_dr(vcpu, 7, &val);
6385         put_smstate(u32, buf, 0x7fc8, (u32)val);
6386
6387         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6388         put_smstate(u32, buf, 0x7fc4, seg.selector);
6389         put_smstate(u32, buf, 0x7f64, seg.base);
6390         put_smstate(u32, buf, 0x7f60, seg.limit);
6391         put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6392
6393         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6394         put_smstate(u32, buf, 0x7fc0, seg.selector);
6395         put_smstate(u32, buf, 0x7f80, seg.base);
6396         put_smstate(u32, buf, 0x7f7c, seg.limit);
6397         put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6398
6399         kvm_x86_ops->get_gdt(vcpu, &dt);
6400         put_smstate(u32, buf, 0x7f74, dt.address);
6401         put_smstate(u32, buf, 0x7f70, dt.size);
6402
6403         kvm_x86_ops->get_idt(vcpu, &dt);
6404         put_smstate(u32, buf, 0x7f58, dt.address);
6405         put_smstate(u32, buf, 0x7f54, dt.size);
6406
6407         for (i = 0; i < 6; i++)
6408                 enter_smm_save_seg_32(vcpu, buf, i);
6409
6410         put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6411
6412         /* revision id */
6413         put_smstate(u32, buf, 0x7efc, 0x00020000);
6414         put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6415 }
6416
6417 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6418 {
6419 #ifdef CONFIG_X86_64
6420         struct desc_ptr dt;
6421         struct kvm_segment seg;
6422         unsigned long val;
6423         int i;
6424
6425         for (i = 0; i < 16; i++)
6426                 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6427
6428         put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6429         put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6430
6431         kvm_get_dr(vcpu, 6, &val);
6432         put_smstate(u64, buf, 0x7f68, val);
6433         kvm_get_dr(vcpu, 7, &val);
6434         put_smstate(u64, buf, 0x7f60, val);
6435
6436         put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6437         put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6438         put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6439
6440         put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6441
6442         /* revision id */
6443         put_smstate(u32, buf, 0x7efc, 0x00020064);
6444
6445         put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6446
6447         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6448         put_smstate(u16, buf, 0x7e90, seg.selector);
6449         put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6450         put_smstate(u32, buf, 0x7e94, seg.limit);
6451         put_smstate(u64, buf, 0x7e98, seg.base);
6452
6453         kvm_x86_ops->get_idt(vcpu, &dt);
6454         put_smstate(u32, buf, 0x7e84, dt.size);
6455         put_smstate(u64, buf, 0x7e88, dt.address);
6456
6457         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6458         put_smstate(u16, buf, 0x7e70, seg.selector);
6459         put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6460         put_smstate(u32, buf, 0x7e74, seg.limit);
6461         put_smstate(u64, buf, 0x7e78, seg.base);
6462
6463         kvm_x86_ops->get_gdt(vcpu, &dt);
6464         put_smstate(u32, buf, 0x7e64, dt.size);
6465         put_smstate(u64, buf, 0x7e68, dt.address);
6466
6467         for (i = 0; i < 6; i++)
6468                 enter_smm_save_seg_64(vcpu, buf, i);
6469 #else
6470         WARN_ON_ONCE(1);
6471 #endif
6472 }
6473
6474 static void enter_smm(struct kvm_vcpu *vcpu)
6475 {
6476         struct kvm_segment cs, ds;
6477         struct desc_ptr dt;
6478         char buf[512];
6479         u32 cr0;
6480
6481         trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6482         vcpu->arch.hflags |= HF_SMM_MASK;
6483         memset(buf, 0, 512);
6484         if (guest_cpuid_has_longmode(vcpu))
6485                 enter_smm_save_state_64(vcpu, buf);
6486         else
6487                 enter_smm_save_state_32(vcpu, buf);
6488
6489         kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6490
6491         if (kvm_x86_ops->get_nmi_mask(vcpu))
6492                 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6493         else
6494                 kvm_x86_ops->set_nmi_mask(vcpu, true);
6495
6496         kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6497         kvm_rip_write(vcpu, 0x8000);
6498
6499         cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6500         kvm_x86_ops->set_cr0(vcpu, cr0);
6501         vcpu->arch.cr0 = cr0;
6502
6503         kvm_x86_ops->set_cr4(vcpu, 0);
6504
6505         /* Undocumented: IDT limit is set to zero on entry to SMM.  */
6506         dt.address = dt.size = 0;
6507         kvm_x86_ops->set_idt(vcpu, &dt);
6508
6509         __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6510
6511         cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6512         cs.base = vcpu->arch.smbase;
6513
6514         ds.selector = 0;
6515         ds.base = 0;
6516
6517         cs.limit    = ds.limit = 0xffffffff;
6518         cs.type     = ds.type = 0x3;
6519         cs.dpl      = ds.dpl = 0;
6520         cs.db       = ds.db = 0;
6521         cs.s        = ds.s = 1;
6522         cs.l        = ds.l = 0;
6523         cs.g        = ds.g = 1;
6524         cs.avl      = ds.avl = 0;
6525         cs.present  = ds.present = 1;
6526         cs.unusable = ds.unusable = 0;
6527         cs.padding  = ds.padding = 0;
6528
6529         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6530         kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6531         kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6532         kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6533         kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6534         kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6535
6536         if (guest_cpuid_has_longmode(vcpu))
6537                 kvm_x86_ops->set_efer(vcpu, 0);
6538
6539         kvm_update_cpuid(vcpu);
6540         kvm_mmu_reset_context(vcpu);
6541 }
6542
6543 static void process_smi(struct kvm_vcpu *vcpu)
6544 {
6545         vcpu->arch.smi_pending = true;
6546         kvm_make_request(KVM_REQ_EVENT, vcpu);
6547 }
6548
6549 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6550 {
6551         kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6552 }
6553
6554 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6555 {
6556         u64 eoi_exit_bitmap[4];
6557
6558         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6559                 return;
6560
6561         bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6562
6563         if (irqchip_split(vcpu->kvm))
6564                 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6565         else {
6566                 if (vcpu->arch.apicv_active)
6567                         kvm_x86_ops->sync_pir_to_irr(vcpu);
6568                 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6569         }
6570         bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6571                   vcpu_to_synic(vcpu)->vec_bitmap, 256);
6572         kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6573 }
6574
6575 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6576 {
6577         ++vcpu->stat.tlb_flush;
6578         kvm_x86_ops->tlb_flush(vcpu);
6579 }
6580
6581 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6582 {
6583         struct page *page = NULL;
6584
6585         if (!lapic_in_kernel(vcpu))
6586                 return;
6587
6588         if (!kvm_x86_ops->set_apic_access_page_addr)
6589                 return;
6590
6591         page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6592         if (is_error_page(page))
6593                 return;
6594         kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6595
6596         /*
6597          * Do not pin apic access page in memory, the MMU notifier
6598          * will call us again if it is migrated or swapped out.
6599          */
6600         put_page(page);
6601 }
6602 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6603
6604 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6605                                            unsigned long address)
6606 {
6607         /*
6608          * The physical address of apic access page is stored in the VMCS.
6609          * Update it when it becomes invalid.
6610          */
6611         if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6612                 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6613 }
6614
6615 /*
6616  * Returns 1 to let vcpu_run() continue the guest execution loop without
6617  * exiting to the userspace.  Otherwise, the value will be returned to the
6618  * userspace.
6619  */
6620 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6621 {
6622         int r;
6623         bool req_int_win =
6624                 dm_request_for_irq_injection(vcpu) &&
6625                 kvm_cpu_accept_dm_intr(vcpu);
6626
6627         bool req_immediate_exit = false;
6628
6629         if (vcpu->requests) {
6630                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6631                         kvm_mmu_unload(vcpu);
6632                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6633                         __kvm_migrate_timers(vcpu);
6634                 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6635                         kvm_gen_update_masterclock(vcpu->kvm);
6636                 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6637                         kvm_gen_kvmclock_update(vcpu);
6638                 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6639                         r = kvm_guest_time_update(vcpu);
6640                         if (unlikely(r))
6641                                 goto out;
6642                 }
6643                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6644                         kvm_mmu_sync_roots(vcpu);
6645                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6646                         kvm_vcpu_flush_tlb(vcpu);
6647                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6648                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6649                         r = 0;
6650                         goto out;
6651                 }
6652                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6653                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6654                         r = 0;
6655                         goto out;
6656                 }
6657                 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6658                         vcpu->fpu_active = 0;
6659                         kvm_x86_ops->fpu_deactivate(vcpu);
6660                 }
6661                 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6662                         /* Page is swapped out. Do synthetic halt */
6663                         vcpu->arch.apf.halted = true;
6664                         r = 1;
6665                         goto out;
6666                 }
6667                 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6668                         record_steal_time(vcpu);
6669                 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6670                         process_smi(vcpu);
6671                 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6672                         process_nmi(vcpu);
6673                 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6674                         kvm_pmu_handle_event(vcpu);
6675                 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6676                         kvm_pmu_deliver_pmi(vcpu);
6677                 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6678                         BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6679                         if (test_bit(vcpu->arch.pending_ioapic_eoi,
6680                                      vcpu->arch.ioapic_handled_vectors)) {
6681                                 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6682                                 vcpu->run->eoi.vector =
6683                                                 vcpu->arch.pending_ioapic_eoi;
6684                                 r = 0;
6685                                 goto out;
6686                         }
6687                 }
6688                 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6689                         vcpu_scan_ioapic(vcpu);
6690                 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6691                         kvm_vcpu_reload_apic_access_page(vcpu);
6692                 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6693                         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6694                         vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6695                         r = 0;
6696                         goto out;
6697                 }
6698                 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6699                         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6700                         vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6701                         r = 0;
6702                         goto out;
6703                 }
6704                 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6705                         vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6706                         vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6707                         r = 0;
6708                         goto out;
6709                 }
6710
6711                 /*
6712                  * KVM_REQ_HV_STIMER has to be processed after
6713                  * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6714                  * depend on the guest clock being up-to-date
6715                  */
6716                 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6717                         kvm_hv_process_stimers(vcpu);
6718         }
6719
6720         /*
6721          * KVM_REQ_EVENT is not set when posted interrupts are set by
6722          * VT-d hardware, so we have to update RVI unconditionally.
6723          */
6724         if (kvm_lapic_enabled(vcpu)) {
6725                 /*
6726                  * Update architecture specific hints for APIC
6727                  * virtual interrupt delivery.
6728                  */
6729                 if (vcpu->arch.apicv_active)
6730                         kvm_x86_ops->hwapic_irr_update(vcpu,
6731                                 kvm_lapic_find_highest_irr(vcpu));
6732         }
6733
6734         if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6735                 kvm_apic_accept_events(vcpu);
6736                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6737                         r = 1;
6738                         goto out;
6739                 }
6740
6741                 if (inject_pending_event(vcpu, req_int_win) != 0)
6742                         req_immediate_exit = true;
6743                 else {
6744                         /* Enable NMI/IRQ window open exits if needed.
6745                          *
6746                          * SMIs have two cases: 1) they can be nested, and
6747                          * then there is nothing to do here because RSM will
6748                          * cause a vmexit anyway; 2) or the SMI can be pending
6749                          * because inject_pending_event has completed the
6750                          * injection of an IRQ or NMI from the previous vmexit,
6751                          * and then we request an immediate exit to inject the SMI.
6752                          */
6753                         if (vcpu->arch.smi_pending && !is_smm(vcpu))
6754                                 req_immediate_exit = true;
6755                         if (vcpu->arch.nmi_pending)
6756                                 kvm_x86_ops->enable_nmi_window(vcpu);
6757                         if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6758                                 kvm_x86_ops->enable_irq_window(vcpu);
6759                 }
6760
6761                 if (kvm_lapic_enabled(vcpu)) {
6762                         update_cr8_intercept(vcpu);
6763                         kvm_lapic_sync_to_vapic(vcpu);
6764                 }
6765         }
6766
6767         r = kvm_mmu_reload(vcpu);
6768         if (unlikely(r)) {
6769                 goto cancel_injection;
6770         }
6771
6772         preempt_disable();
6773
6774         kvm_x86_ops->prepare_guest_switch(vcpu);
6775         if (vcpu->fpu_active)
6776                 kvm_load_guest_fpu(vcpu);
6777         vcpu->mode = IN_GUEST_MODE;
6778
6779         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6780
6781         /*
6782          * We should set ->mode before check ->requests,
6783          * Please see the comment in kvm_make_all_cpus_request.
6784          * This also orders the write to mode from any reads
6785          * to the page tables done while the VCPU is running.
6786          * Please see the comment in kvm_flush_remote_tlbs.
6787          */
6788         smp_mb__after_srcu_read_unlock();
6789
6790         local_irq_disable();
6791
6792         if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6793             || need_resched() || signal_pending(current)) {
6794                 vcpu->mode = OUTSIDE_GUEST_MODE;
6795                 smp_wmb();
6796                 local_irq_enable();
6797                 preempt_enable();
6798                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6799                 r = 1;
6800                 goto cancel_injection;
6801         }
6802
6803         kvm_load_guest_xcr0(vcpu);
6804
6805         if (req_immediate_exit) {
6806                 kvm_make_request(KVM_REQ_EVENT, vcpu);
6807                 smp_send_reschedule(vcpu->cpu);
6808         }
6809
6810         trace_kvm_entry(vcpu->vcpu_id);
6811         wait_lapic_expire(vcpu);
6812         guest_enter_irqoff();
6813
6814         if (unlikely(vcpu->arch.switch_db_regs)) {
6815                 set_debugreg(0, 7);
6816                 set_debugreg(vcpu->arch.eff_db[0], 0);
6817                 set_debugreg(vcpu->arch.eff_db[1], 1);
6818                 set_debugreg(vcpu->arch.eff_db[2], 2);
6819                 set_debugreg(vcpu->arch.eff_db[3], 3);
6820                 set_debugreg(vcpu->arch.dr6, 6);
6821                 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6822         }
6823
6824         kvm_x86_ops->run(vcpu);
6825
6826         /*
6827          * Do this here before restoring debug registers on the host.  And
6828          * since we do this before handling the vmexit, a DR access vmexit
6829          * can (a) read the correct value of the debug registers, (b) set
6830          * KVM_DEBUGREG_WONT_EXIT again.
6831          */
6832         if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6833                 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6834                 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6835                 kvm_update_dr0123(vcpu);
6836                 kvm_update_dr6(vcpu);
6837                 kvm_update_dr7(vcpu);
6838                 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6839         }
6840
6841         /*
6842          * If the guest has used debug registers, at least dr7
6843          * will be disabled while returning to the host.
6844          * If we don't have active breakpoints in the host, we don't
6845          * care about the messed up debug address registers. But if
6846          * we have some of them active, restore the old state.
6847          */
6848         if (hw_breakpoint_active())
6849                 hw_breakpoint_restore();
6850
6851         vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6852
6853         vcpu->mode = OUTSIDE_GUEST_MODE;
6854         smp_wmb();
6855
6856         kvm_put_guest_xcr0(vcpu);
6857
6858         kvm_x86_ops->handle_external_intr(vcpu);
6859
6860         ++vcpu->stat.exits;
6861
6862         guest_exit_irqoff();
6863
6864         local_irq_enable();
6865         preempt_enable();
6866
6867         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6868
6869         /*
6870          * Profile KVM exit RIPs:
6871          */
6872         if (unlikely(prof_on == KVM_PROFILING)) {
6873                 unsigned long rip = kvm_rip_read(vcpu);
6874                 profile_hit(KVM_PROFILING, (void *)rip);
6875         }
6876
6877         if (unlikely(vcpu->arch.tsc_always_catchup))
6878                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6879
6880         if (vcpu->arch.apic_attention)
6881                 kvm_lapic_sync_from_vapic(vcpu);
6882
6883         r = kvm_x86_ops->handle_exit(vcpu);
6884         return r;
6885
6886 cancel_injection:
6887         kvm_x86_ops->cancel_injection(vcpu);
6888         if (unlikely(vcpu->arch.apic_attention))
6889                 kvm_lapic_sync_from_vapic(vcpu);
6890 out:
6891         return r;
6892 }
6893
6894 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6895 {
6896         if (!kvm_arch_vcpu_runnable(vcpu) &&
6897             (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6898                 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6899                 kvm_vcpu_block(vcpu);
6900                 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6901
6902                 if (kvm_x86_ops->post_block)
6903                         kvm_x86_ops->post_block(vcpu);
6904
6905                 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6906                         return 1;
6907         }
6908
6909         kvm_apic_accept_events(vcpu);
6910         switch(vcpu->arch.mp_state) {
6911         case KVM_MP_STATE_HALTED:
6912                 vcpu->arch.pv.pv_unhalted = false;
6913                 vcpu->arch.mp_state =
6914                         KVM_MP_STATE_RUNNABLE;
6915         case KVM_MP_STATE_RUNNABLE:
6916                 vcpu->arch.apf.halted = false;
6917                 break;
6918         case KVM_MP_STATE_INIT_RECEIVED:
6919                 break;
6920         default:
6921                 return -EINTR;
6922                 break;
6923         }
6924         return 1;
6925 }
6926
6927 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6928 {
6929         return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6930                 !vcpu->arch.apf.halted);
6931 }
6932
6933 static int vcpu_run(struct kvm_vcpu *vcpu)
6934 {
6935         int r;
6936         struct kvm *kvm = vcpu->kvm;
6937
6938         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6939
6940         for (;;) {
6941                 if (kvm_vcpu_running(vcpu)) {
6942                         r = vcpu_enter_guest(vcpu);
6943                 } else {
6944                         r = vcpu_block(kvm, vcpu);
6945                 }
6946
6947                 if (r <= 0)
6948                         break;
6949
6950                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6951                 if (kvm_cpu_has_pending_timer(vcpu))
6952                         kvm_inject_pending_timer_irqs(vcpu);
6953
6954                 if (dm_request_for_irq_injection(vcpu) &&
6955                         kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
6956                         r = 0;
6957                         vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
6958                         ++vcpu->stat.request_irq_exits;
6959                         break;
6960                 }
6961
6962                 kvm_check_async_pf_completion(vcpu);
6963
6964                 if (signal_pending(current)) {
6965                         r = -EINTR;
6966                         vcpu->run->exit_reason = KVM_EXIT_INTR;
6967                         ++vcpu->stat.signal_exits;
6968                         break;
6969                 }
6970                 if (need_resched()) {
6971                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6972                         cond_resched();
6973                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6974                 }
6975         }
6976
6977         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6978
6979         return r;
6980 }
6981
6982 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6983 {
6984         int r;
6985         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6986         r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6987         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6988         if (r != EMULATE_DONE)
6989                 return 0;
6990         return 1;
6991 }
6992
6993 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6994 {
6995         BUG_ON(!vcpu->arch.pio.count);
6996
6997         return complete_emulated_io(vcpu);
6998 }
6999
7000 /*
7001  * Implements the following, as a state machine:
7002  *
7003  * read:
7004  *   for each fragment
7005  *     for each mmio piece in the fragment
7006  *       write gpa, len
7007  *       exit
7008  *       copy data
7009  *   execute insn
7010  *
7011  * write:
7012  *   for each fragment
7013  *     for each mmio piece in the fragment
7014  *       write gpa, len
7015  *       copy data
7016  *       exit
7017  */
7018 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7019 {
7020         struct kvm_run *run = vcpu->run;
7021         struct kvm_mmio_fragment *frag;
7022         unsigned len;
7023
7024         BUG_ON(!vcpu->mmio_needed);
7025
7026         /* Complete previous fragment */
7027         frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7028         len = min(8u, frag->len);
7029         if (!vcpu->mmio_is_write)
7030                 memcpy(frag->data, run->mmio.data, len);
7031
7032         if (frag->len <= 8) {
7033                 /* Switch to the next fragment. */
7034                 frag++;
7035                 vcpu->mmio_cur_fragment++;
7036         } else {
7037                 /* Go forward to the next mmio piece. */
7038                 frag->data += len;
7039                 frag->gpa += len;
7040                 frag->len -= len;
7041         }
7042
7043         if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7044                 vcpu->mmio_needed = 0;
7045
7046                 /* FIXME: return into emulator if single-stepping.  */
7047                 if (vcpu->mmio_is_write)
7048                         return 1;
7049                 vcpu->mmio_read_completed = 1;
7050                 return complete_emulated_io(vcpu);
7051         }
7052
7053         run->exit_reason = KVM_EXIT_MMIO;
7054         run->mmio.phys_addr = frag->gpa;
7055         if (vcpu->mmio_is_write)
7056                 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7057         run->mmio.len = min(8u, frag->len);
7058         run->mmio.is_write = vcpu->mmio_is_write;
7059         vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7060         return 0;
7061 }
7062
7063
7064 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7065 {
7066         struct fpu *fpu = &current->thread.fpu;
7067         int r;
7068         sigset_t sigsaved;
7069
7070         fpu__activate_curr(fpu);
7071
7072         if (vcpu->sigset_active)
7073                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
7074
7075         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7076                 kvm_vcpu_block(vcpu);
7077                 kvm_apic_accept_events(vcpu);
7078                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
7079                 r = -EAGAIN;
7080                 goto out;
7081         }
7082
7083         /* re-sync apic's tpr */
7084         if (!lapic_in_kernel(vcpu)) {
7085                 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7086                         r = -EINVAL;
7087                         goto out;
7088                 }
7089         }
7090
7091         if (unlikely(vcpu->arch.complete_userspace_io)) {
7092                 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7093                 vcpu->arch.complete_userspace_io = NULL;
7094                 r = cui(vcpu);
7095                 if (r <= 0)
7096                         goto out;
7097         } else
7098                 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7099
7100         r = vcpu_run(vcpu);
7101
7102 out:
7103         post_kvm_run_save(vcpu);
7104         if (vcpu->sigset_active)
7105                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
7106
7107         return r;
7108 }
7109
7110 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7111 {
7112         if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7113                 /*
7114                  * We are here if userspace calls get_regs() in the middle of
7115                  * instruction emulation. Registers state needs to be copied
7116                  * back from emulation context to vcpu. Userspace shouldn't do
7117                  * that usually, but some bad designed PV devices (vmware
7118                  * backdoor interface) need this to work
7119                  */
7120                 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7121                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7122         }
7123         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7124         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7125         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7126         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7127         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7128         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7129         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7130         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7131 #ifdef CONFIG_X86_64
7132         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7133         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7134         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7135         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7136         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7137         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7138         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7139         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7140 #endif
7141
7142         regs->rip = kvm_rip_read(vcpu);
7143         regs->rflags = kvm_get_rflags(vcpu);
7144
7145         return 0;
7146 }
7147
7148 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7149 {
7150         vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7151         vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7152
7153         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7154         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7155         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7156         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7157         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7158         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7159         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7160         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7161 #ifdef CONFIG_X86_64
7162         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7163         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7164         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7165         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7166         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7167         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7168         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7169         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7170 #endif
7171
7172         kvm_rip_write(vcpu, regs->rip);
7173         kvm_set_rflags(vcpu, regs->rflags);
7174
7175         vcpu->arch.exception.pending = false;
7176
7177         kvm_make_request(KVM_REQ_EVENT, vcpu);
7178
7179         return 0;
7180 }
7181
7182 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7183 {
7184         struct kvm_segment cs;
7185
7186         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7187         *db = cs.db;
7188         *l = cs.l;
7189 }
7190 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7191
7192 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7193                                   struct kvm_sregs *sregs)
7194 {
7195         struct desc_ptr dt;
7196
7197         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7198         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7199         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7200         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7201         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7202         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7203
7204         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7205         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7206
7207         kvm_x86_ops->get_idt(vcpu, &dt);
7208         sregs->idt.limit = dt.size;
7209         sregs->idt.base = dt.address;
7210         kvm_x86_ops->get_gdt(vcpu, &dt);
7211         sregs->gdt.limit = dt.size;
7212         sregs->gdt.base = dt.address;
7213
7214         sregs->cr0 = kvm_read_cr0(vcpu);
7215         sregs->cr2 = vcpu->arch.cr2;
7216         sregs->cr3 = kvm_read_cr3(vcpu);
7217         sregs->cr4 = kvm_read_cr4(vcpu);
7218         sregs->cr8 = kvm_get_cr8(vcpu);
7219         sregs->efer = vcpu->arch.efer;
7220         sregs->apic_base = kvm_get_apic_base(vcpu);
7221
7222         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7223
7224         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7225                 set_bit(vcpu->arch.interrupt.nr,
7226                         (unsigned long *)sregs->interrupt_bitmap);
7227
7228         return 0;
7229 }
7230
7231 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7232                                     struct kvm_mp_state *mp_state)
7233 {
7234         kvm_apic_accept_events(vcpu);
7235         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7236                                         vcpu->arch.pv.pv_unhalted)
7237                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7238         else
7239                 mp_state->mp_state = vcpu->arch.mp_state;
7240
7241         return 0;
7242 }
7243
7244 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7245                                     struct kvm_mp_state *mp_state)
7246 {
7247         if (!lapic_in_kernel(vcpu) &&
7248             mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7249                 return -EINVAL;
7250
7251         if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7252                 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7253                 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7254         } else
7255                 vcpu->arch.mp_state = mp_state->mp_state;
7256         kvm_make_request(KVM_REQ_EVENT, vcpu);
7257         return 0;
7258 }
7259
7260 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7261                     int reason, bool has_error_code, u32 error_code)
7262 {
7263         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7264         int ret;
7265
7266         init_emulate_ctxt(vcpu);
7267
7268         ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7269                                    has_error_code, error_code);
7270
7271         if (ret)
7272                 return EMULATE_FAIL;
7273
7274         kvm_rip_write(vcpu, ctxt->eip);
7275         kvm_set_rflags(vcpu, ctxt->eflags);
7276         kvm_make_request(KVM_REQ_EVENT, vcpu);
7277         return EMULATE_DONE;
7278 }
7279 EXPORT_SYMBOL_GPL(kvm_task_switch);
7280
7281 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7282                                   struct kvm_sregs *sregs)
7283 {
7284         struct msr_data apic_base_msr;
7285         int mmu_reset_needed = 0;
7286         int pending_vec, max_bits, idx;
7287         struct desc_ptr dt;
7288
7289         if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7290                 return -EINVAL;
7291
7292         dt.size = sregs->idt.limit;
7293         dt.address = sregs->idt.base;
7294         kvm_x86_ops->set_idt(vcpu, &dt);
7295         dt.size = sregs->gdt.limit;
7296         dt.address = sregs->gdt.base;
7297         kvm_x86_ops->set_gdt(vcpu, &dt);
7298
7299         vcpu->arch.cr2 = sregs->cr2;
7300         mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7301         vcpu->arch.cr3 = sregs->cr3;
7302         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7303
7304         kvm_set_cr8(vcpu, sregs->cr8);
7305
7306         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7307         kvm_x86_ops->set_efer(vcpu, sregs->efer);
7308         apic_base_msr.data = sregs->apic_base;
7309         apic_base_msr.host_initiated = true;
7310         kvm_set_apic_base(vcpu, &apic_base_msr);
7311
7312         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7313         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7314         vcpu->arch.cr0 = sregs->cr0;
7315
7316         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7317         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7318         if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7319                 kvm_update_cpuid(vcpu);
7320
7321         idx = srcu_read_lock(&vcpu->kvm->srcu);
7322         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7323                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7324                 mmu_reset_needed = 1;
7325         }
7326         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7327
7328         if (mmu_reset_needed)
7329                 kvm_mmu_reset_context(vcpu);
7330
7331         max_bits = KVM_NR_INTERRUPTS;
7332         pending_vec = find_first_bit(
7333                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7334         if (pending_vec < max_bits) {
7335                 kvm_queue_interrupt(vcpu, pending_vec, false);
7336                 pr_debug("Set back pending irq %d\n", pending_vec);
7337         }
7338
7339         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7340         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7341         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7342         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7343         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7344         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7345
7346         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7347         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7348
7349         update_cr8_intercept(vcpu);
7350
7351         /* Older userspace won't unhalt the vcpu on reset. */
7352         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7353             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7354             !is_protmode(vcpu))
7355                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7356
7357         kvm_make_request(KVM_REQ_EVENT, vcpu);
7358
7359         return 0;
7360 }
7361
7362 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7363                                         struct kvm_guest_debug *dbg)
7364 {
7365         unsigned long rflags;
7366         int i, r;
7367
7368         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7369                 r = -EBUSY;
7370                 if (vcpu->arch.exception.pending)
7371                         goto out;
7372                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7373                         kvm_queue_exception(vcpu, DB_VECTOR);
7374                 else
7375                         kvm_queue_exception(vcpu, BP_VECTOR);
7376         }
7377
7378         /*
7379          * Read rflags as long as potentially injected trace flags are still
7380          * filtered out.
7381          */
7382         rflags = kvm_get_rflags(vcpu);
7383
7384         vcpu->guest_debug = dbg->control;
7385         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7386                 vcpu->guest_debug = 0;
7387
7388         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7389                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7390                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7391                 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7392         } else {
7393                 for (i = 0; i < KVM_NR_DB_REGS; i++)
7394                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7395         }
7396         kvm_update_dr7(vcpu);
7397
7398         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7399                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7400                         get_segment_base(vcpu, VCPU_SREG_CS);
7401
7402         /*
7403          * Trigger an rflags update that will inject or remove the trace
7404          * flags.
7405          */
7406         kvm_set_rflags(vcpu, rflags);
7407
7408         kvm_x86_ops->update_bp_intercept(vcpu);
7409
7410         r = 0;
7411
7412 out:
7413
7414         return r;
7415 }
7416
7417 /*
7418  * Translate a guest virtual address to a guest physical address.
7419  */
7420 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7421                                     struct kvm_translation *tr)
7422 {
7423         unsigned long vaddr = tr->linear_address;
7424         gpa_t gpa;
7425         int idx;
7426
7427         idx = srcu_read_lock(&vcpu->kvm->srcu);
7428         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7429         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7430         tr->physical_address = gpa;
7431         tr->valid = gpa != UNMAPPED_GVA;
7432         tr->writeable = 1;
7433         tr->usermode = 0;
7434
7435         return 0;
7436 }
7437
7438 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7439 {
7440         struct fxregs_state *fxsave =
7441                         &vcpu->arch.guest_fpu.state.fxsave;
7442
7443         memcpy(fpu->fpr, fxsave->st_space, 128);
7444         fpu->fcw = fxsave->cwd;
7445         fpu->fsw = fxsave->swd;
7446         fpu->ftwx = fxsave->twd;
7447         fpu->last_opcode = fxsave->fop;
7448         fpu->last_ip = fxsave->rip;
7449         fpu->last_dp = fxsave->rdp;
7450         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7451
7452         return 0;
7453 }
7454
7455 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7456 {
7457         struct fxregs_state *fxsave =
7458                         &vcpu->arch.guest_fpu.state.fxsave;
7459
7460         memcpy(fxsave->st_space, fpu->fpr, 128);
7461         fxsave->cwd = fpu->fcw;
7462         fxsave->swd = fpu->fsw;
7463         fxsave->twd = fpu->ftwx;
7464         fxsave->fop = fpu->last_opcode;
7465         fxsave->rip = fpu->last_ip;
7466         fxsave->rdp = fpu->last_dp;
7467         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7468
7469         return 0;
7470 }
7471
7472 static void fx_init(struct kvm_vcpu *vcpu)
7473 {
7474         fpstate_init(&vcpu->arch.guest_fpu.state);
7475         if (boot_cpu_has(X86_FEATURE_XSAVES))
7476                 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7477                         host_xcr0 | XSTATE_COMPACTION_ENABLED;
7478
7479         /*
7480          * Ensure guest xcr0 is valid for loading
7481          */
7482         vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7483
7484         vcpu->arch.cr0 |= X86_CR0_ET;
7485 }
7486
7487 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7488 {
7489         if (vcpu->guest_fpu_loaded)
7490                 return;
7491
7492         /*
7493          * Restore all possible states in the guest,
7494          * and assume host would use all available bits.
7495          * Guest xcr0 would be loaded later.
7496          */
7497         vcpu->guest_fpu_loaded = 1;
7498         __kernel_fpu_begin();
7499         __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7500         trace_kvm_fpu(1);
7501 }
7502
7503 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7504 {
7505         if (!vcpu->guest_fpu_loaded)
7506                 return;
7507
7508         vcpu->guest_fpu_loaded = 0;
7509         copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7510         __kernel_fpu_end();
7511         ++vcpu->stat.fpu_reload;
7512         trace_kvm_fpu(0);
7513 }
7514
7515 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7516 {
7517         void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7518
7519         kvmclock_reset(vcpu);
7520
7521         kvm_x86_ops->vcpu_free(vcpu);
7522         free_cpumask_var(wbinvd_dirty_mask);
7523 }
7524
7525 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7526                                                 unsigned int id)
7527 {
7528         struct kvm_vcpu *vcpu;
7529
7530         if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7531                 printk_once(KERN_WARNING
7532                 "kvm: SMP vm created on host with unstable TSC; "
7533                 "guest TSC will not be reliable\n");
7534
7535         vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7536
7537         return vcpu;
7538 }
7539
7540 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7541 {
7542         int r;
7543
7544         kvm_vcpu_mtrr_init(vcpu);
7545         r = vcpu_load(vcpu);
7546         if (r)
7547                 return r;
7548         kvm_vcpu_reset(vcpu, false);
7549         kvm_mmu_setup(vcpu);
7550         vcpu_put(vcpu);
7551         return r;
7552 }
7553
7554 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7555 {
7556         struct msr_data msr;
7557         struct kvm *kvm = vcpu->kvm;
7558
7559         if (vcpu_load(vcpu))
7560                 return;
7561         msr.data = 0x0;
7562         msr.index = MSR_IA32_TSC;
7563         msr.host_initiated = true;
7564         kvm_write_tsc(vcpu, &msr);
7565         vcpu_put(vcpu);
7566
7567         if (!kvmclock_periodic_sync)
7568                 return;
7569
7570         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7571                                         KVMCLOCK_SYNC_PERIOD);
7572 }
7573
7574 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7575 {
7576         int r;
7577         vcpu->arch.apf.msr_val = 0;
7578
7579         r = vcpu_load(vcpu);
7580         BUG_ON(r);
7581         kvm_mmu_unload(vcpu);
7582         vcpu_put(vcpu);
7583
7584         kvm_x86_ops->vcpu_free(vcpu);
7585 }
7586
7587 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7588 {
7589         vcpu->arch.hflags = 0;
7590
7591         vcpu->arch.smi_pending = 0;
7592         atomic_set(&vcpu->arch.nmi_queued, 0);
7593         vcpu->arch.nmi_pending = 0;
7594         vcpu->arch.nmi_injected = false;
7595         kvm_clear_interrupt_queue(vcpu);
7596         kvm_clear_exception_queue(vcpu);
7597
7598         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7599         kvm_update_dr0123(vcpu);
7600         vcpu->arch.dr6 = DR6_INIT;
7601         kvm_update_dr6(vcpu);
7602         vcpu->arch.dr7 = DR7_FIXED_1;
7603         kvm_update_dr7(vcpu);
7604
7605         vcpu->arch.cr2 = 0;
7606
7607         kvm_make_request(KVM_REQ_EVENT, vcpu);
7608         vcpu->arch.apf.msr_val = 0;
7609         vcpu->arch.st.msr_val = 0;
7610
7611         kvmclock_reset(vcpu);
7612
7613         kvm_clear_async_pf_completion_queue(vcpu);
7614         kvm_async_pf_hash_reset(vcpu);
7615         vcpu->arch.apf.halted = false;
7616
7617         if (!init_event) {
7618                 kvm_pmu_reset(vcpu);
7619                 vcpu->arch.smbase = 0x30000;
7620         }
7621
7622         memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7623         vcpu->arch.regs_avail = ~0;
7624         vcpu->arch.regs_dirty = ~0;
7625
7626         kvm_x86_ops->vcpu_reset(vcpu, init_event);
7627 }
7628
7629 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7630 {
7631         struct kvm_segment cs;
7632
7633         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7634         cs.selector = vector << 8;
7635         cs.base = vector << 12;
7636         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7637         kvm_rip_write(vcpu, 0);
7638 }
7639
7640 int kvm_arch_hardware_enable(void)
7641 {
7642         struct kvm *kvm;
7643         struct kvm_vcpu *vcpu;
7644         int i;
7645         int ret;
7646         u64 local_tsc;
7647         u64 max_tsc = 0;
7648         bool stable, backwards_tsc = false;
7649
7650         kvm_shared_msr_cpu_online();
7651         ret = kvm_x86_ops->hardware_enable();
7652         if (ret != 0)
7653                 return ret;
7654
7655         local_tsc = rdtsc();
7656         stable = !check_tsc_unstable();
7657         list_for_each_entry(kvm, &vm_list, vm_list) {
7658                 kvm_for_each_vcpu(i, vcpu, kvm) {
7659                         if (!stable && vcpu->cpu == smp_processor_id())
7660                                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7661                         if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7662                                 backwards_tsc = true;
7663                                 if (vcpu->arch.last_host_tsc > max_tsc)
7664                                         max_tsc = vcpu->arch.last_host_tsc;
7665                         }
7666                 }
7667         }
7668
7669         /*
7670          * Sometimes, even reliable TSCs go backwards.  This happens on
7671          * platforms that reset TSC during suspend or hibernate actions, but
7672          * maintain synchronization.  We must compensate.  Fortunately, we can
7673          * detect that condition here, which happens early in CPU bringup,
7674          * before any KVM threads can be running.  Unfortunately, we can't
7675          * bring the TSCs fully up to date with real time, as we aren't yet far
7676          * enough into CPU bringup that we know how much real time has actually
7677          * elapsed; our helper function, ktime_get_boot_ns() will be using boot
7678          * variables that haven't been updated yet.
7679          *
7680          * So we simply find the maximum observed TSC above, then record the
7681          * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7682          * the adjustment will be applied.  Note that we accumulate
7683          * adjustments, in case multiple suspend cycles happen before some VCPU
7684          * gets a chance to run again.  In the event that no KVM threads get a
7685          * chance to run, we will miss the entire elapsed period, as we'll have
7686          * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7687          * loose cycle time.  This isn't too big a deal, since the loss will be
7688          * uniform across all VCPUs (not to mention the scenario is extremely
7689          * unlikely). It is possible that a second hibernate recovery happens
7690          * much faster than a first, causing the observed TSC here to be
7691          * smaller; this would require additional padding adjustment, which is
7692          * why we set last_host_tsc to the local tsc observed here.
7693          *
7694          * N.B. - this code below runs only on platforms with reliable TSC,
7695          * as that is the only way backwards_tsc is set above.  Also note
7696          * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7697          * have the same delta_cyc adjustment applied if backwards_tsc
7698          * is detected.  Note further, this adjustment is only done once,
7699          * as we reset last_host_tsc on all VCPUs to stop this from being
7700          * called multiple times (one for each physical CPU bringup).
7701          *
7702          * Platforms with unreliable TSCs don't have to deal with this, they
7703          * will be compensated by the logic in vcpu_load, which sets the TSC to
7704          * catchup mode.  This will catchup all VCPUs to real time, but cannot
7705          * guarantee that they stay in perfect synchronization.
7706          */
7707         if (backwards_tsc) {
7708                 u64 delta_cyc = max_tsc - local_tsc;
7709                 backwards_tsc_observed = true;
7710                 list_for_each_entry(kvm, &vm_list, vm_list) {
7711                         kvm_for_each_vcpu(i, vcpu, kvm) {
7712                                 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7713                                 vcpu->arch.last_host_tsc = local_tsc;
7714                                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7715                         }
7716
7717                         /*
7718                          * We have to disable TSC offset matching.. if you were
7719                          * booting a VM while issuing an S4 host suspend....
7720                          * you may have some problem.  Solving this issue is
7721                          * left as an exercise to the reader.
7722                          */
7723                         kvm->arch.last_tsc_nsec = 0;
7724                         kvm->arch.last_tsc_write = 0;
7725                 }
7726
7727         }
7728         return 0;
7729 }
7730
7731 void kvm_arch_hardware_disable(void)
7732 {
7733         kvm_x86_ops->hardware_disable();
7734         drop_user_return_notifiers();
7735 }
7736
7737 int kvm_arch_hardware_setup(void)
7738 {
7739         int r;
7740
7741         r = kvm_x86_ops->hardware_setup();
7742         if (r != 0)
7743                 return r;
7744
7745         if (kvm_has_tsc_control) {
7746                 /*
7747                  * Make sure the user can only configure tsc_khz values that
7748                  * fit into a signed integer.
7749                  * A min value is not calculated needed because it will always
7750                  * be 1 on all machines.
7751                  */
7752                 u64 max = min(0x7fffffffULL,
7753                               __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7754                 kvm_max_guest_tsc_khz = max;
7755
7756                 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7757         }
7758
7759         kvm_init_msr_list();
7760         return 0;
7761 }
7762
7763 void kvm_arch_hardware_unsetup(void)
7764 {
7765         kvm_x86_ops->hardware_unsetup();
7766 }
7767
7768 void kvm_arch_check_processor_compat(void *rtn)
7769 {
7770         kvm_x86_ops->check_processor_compatibility(rtn);
7771 }
7772
7773 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7774 {
7775         return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7776 }
7777 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7778
7779 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7780 {
7781         return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7782 }
7783
7784 struct static_key kvm_no_apic_vcpu __read_mostly;
7785 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7786
7787 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7788 {
7789         struct page *page;
7790         struct kvm *kvm;
7791         int r;
7792
7793         BUG_ON(vcpu->kvm == NULL);
7794         kvm = vcpu->kvm;
7795
7796         vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7797         vcpu->arch.pv.pv_unhalted = false;
7798         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7799         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7800                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7801         else
7802                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7803
7804         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7805         if (!page) {
7806                 r = -ENOMEM;
7807                 goto fail;
7808         }
7809         vcpu->arch.pio_data = page_address(page);
7810
7811         kvm_set_tsc_khz(vcpu, max_tsc_khz);
7812
7813         r = kvm_mmu_create(vcpu);
7814         if (r < 0)
7815                 goto fail_free_pio_data;
7816
7817         if (irqchip_in_kernel(kvm)) {
7818                 r = kvm_create_lapic(vcpu);
7819                 if (r < 0)
7820                         goto fail_mmu_destroy;
7821         } else
7822                 static_key_slow_inc(&kvm_no_apic_vcpu);
7823
7824         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7825                                        GFP_KERNEL);
7826         if (!vcpu->arch.mce_banks) {
7827                 r = -ENOMEM;
7828                 goto fail_free_lapic;
7829         }
7830         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7831
7832         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7833                 r = -ENOMEM;
7834                 goto fail_free_mce_banks;
7835         }
7836
7837         fx_init(vcpu);
7838
7839         vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7840         vcpu->arch.pv_time_enabled = false;
7841
7842         vcpu->arch.guest_supported_xcr0 = 0;
7843         vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7844
7845         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7846
7847         vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7848
7849         kvm_async_pf_hash_reset(vcpu);
7850         kvm_pmu_init(vcpu);
7851
7852         vcpu->arch.pending_external_vector = -1;
7853
7854         kvm_hv_vcpu_init(vcpu);
7855
7856         return 0;
7857
7858 fail_free_mce_banks:
7859         kfree(vcpu->arch.mce_banks);
7860 fail_free_lapic:
7861         kvm_free_lapic(vcpu);
7862 fail_mmu_destroy:
7863         kvm_mmu_destroy(vcpu);
7864 fail_free_pio_data:
7865         free_page((unsigned long)vcpu->arch.pio_data);
7866 fail:
7867         return r;
7868 }
7869
7870 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7871 {
7872         int idx;
7873
7874         kvm_hv_vcpu_uninit(vcpu);
7875         kvm_pmu_destroy(vcpu);
7876         kfree(vcpu->arch.mce_banks);
7877         kvm_free_lapic(vcpu);
7878         idx = srcu_read_lock(&vcpu->kvm->srcu);
7879         kvm_mmu_destroy(vcpu);
7880         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7881         free_page((unsigned long)vcpu->arch.pio_data);
7882         if (!lapic_in_kernel(vcpu))
7883                 static_key_slow_dec(&kvm_no_apic_vcpu);
7884 }
7885
7886 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7887 {
7888         kvm_x86_ops->sched_in(vcpu, cpu);
7889 }
7890
7891 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7892 {
7893         if (type)
7894                 return -EINVAL;
7895
7896         INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7897         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7898         INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7899         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7900         atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7901
7902         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7903         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7904         /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7905         set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7906                 &kvm->arch.irq_sources_bitmap);
7907
7908         raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7909         mutex_init(&kvm->arch.apic_map_lock);
7910         mutex_init(&kvm->arch.hyperv.hv_lock);
7911         spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7912
7913         kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
7914         pvclock_update_vm_gtod_copy(kvm);
7915
7916         INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7917         INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7918
7919         kvm_page_track_init(kvm);
7920         kvm_mmu_init_vm(kvm);
7921
7922         if (kvm_x86_ops->vm_init)
7923                 return kvm_x86_ops->vm_init(kvm);
7924
7925         return 0;
7926 }
7927
7928 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7929 {
7930         int r;
7931         r = vcpu_load(vcpu);
7932         BUG_ON(r);
7933         kvm_mmu_unload(vcpu);
7934         vcpu_put(vcpu);
7935 }
7936
7937 static void kvm_free_vcpus(struct kvm *kvm)
7938 {
7939         unsigned int i;
7940         struct kvm_vcpu *vcpu;
7941
7942         /*
7943          * Unpin any mmu pages first.
7944          */
7945         kvm_for_each_vcpu(i, vcpu, kvm) {
7946                 kvm_clear_async_pf_completion_queue(vcpu);
7947                 kvm_unload_vcpu_mmu(vcpu);
7948         }
7949         kvm_for_each_vcpu(i, vcpu, kvm)
7950                 kvm_arch_vcpu_free(vcpu);
7951
7952         mutex_lock(&kvm->lock);
7953         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7954                 kvm->vcpus[i] = NULL;
7955
7956         atomic_set(&kvm->online_vcpus, 0);
7957         mutex_unlock(&kvm->lock);
7958 }
7959
7960 void kvm_arch_sync_events(struct kvm *kvm)
7961 {
7962         cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7963         cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7964         kvm_free_all_assigned_devices(kvm);
7965         kvm_free_pit(kvm);
7966 }
7967
7968 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7969 {
7970         int i, r;
7971         unsigned long hva;
7972         struct kvm_memslots *slots = kvm_memslots(kvm);
7973         struct kvm_memory_slot *slot, old;
7974
7975         /* Called with kvm->slots_lock held.  */
7976         if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7977                 return -EINVAL;
7978
7979         slot = id_to_memslot(slots, id);
7980         if (size) {
7981                 if (slot->npages)
7982                         return -EEXIST;
7983
7984                 /*
7985                  * MAP_SHARED to prevent internal slot pages from being moved
7986                  * by fork()/COW.
7987                  */
7988                 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7989                               MAP_SHARED | MAP_ANONYMOUS, 0);
7990                 if (IS_ERR((void *)hva))
7991                         return PTR_ERR((void *)hva);
7992         } else {
7993                 if (!slot->npages)
7994                         return 0;
7995
7996                 hva = 0;
7997         }
7998
7999         old = *slot;
8000         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8001                 struct kvm_userspace_memory_region m;
8002
8003                 m.slot = id | (i << 16);
8004                 m.flags = 0;
8005                 m.guest_phys_addr = gpa;
8006                 m.userspace_addr = hva;
8007                 m.memory_size = size;
8008                 r = __kvm_set_memory_region(kvm, &m);
8009                 if (r < 0)
8010                         return r;
8011         }
8012
8013         if (!size) {
8014                 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8015                 WARN_ON(r < 0);
8016         }
8017
8018         return 0;
8019 }
8020 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8021
8022 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8023 {
8024         int r;
8025
8026         mutex_lock(&kvm->slots_lock);
8027         r = __x86_set_memory_region(kvm, id, gpa, size);
8028         mutex_unlock(&kvm->slots_lock);
8029
8030         return r;
8031 }
8032 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8033
8034 void kvm_arch_destroy_vm(struct kvm *kvm)
8035 {
8036         if (current->mm == kvm->mm) {
8037                 /*
8038                  * Free memory regions allocated on behalf of userspace,
8039                  * unless the the memory map has changed due to process exit
8040                  * or fd copying.
8041                  */
8042                 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8043                 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8044                 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8045         }
8046         if (kvm_x86_ops->vm_destroy)
8047                 kvm_x86_ops->vm_destroy(kvm);
8048         kvm_iommu_unmap_guest(kvm);
8049         kfree(kvm->arch.vpic);
8050         kfree(kvm->arch.vioapic);
8051         kvm_free_vcpus(kvm);
8052         kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8053         kvm_mmu_uninit_vm(kvm);
8054 }
8055
8056 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8057                            struct kvm_memory_slot *dont)
8058 {
8059         int i;
8060
8061         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8062                 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8063                         kvfree(free->arch.rmap[i]);
8064                         free->arch.rmap[i] = NULL;
8065                 }
8066                 if (i == 0)
8067                         continue;
8068
8069                 if (!dont || free->arch.lpage_info[i - 1] !=
8070                              dont->arch.lpage_info[i - 1]) {
8071                         kvfree(free->arch.lpage_info[i - 1]);
8072                         free->arch.lpage_info[i - 1] = NULL;
8073                 }
8074         }
8075
8076         kvm_page_track_free_memslot(free, dont);
8077 }
8078
8079 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8080                             unsigned long npages)
8081 {
8082         int i;
8083
8084         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8085                 struct kvm_lpage_info *linfo;
8086                 unsigned long ugfn;
8087                 int lpages;
8088                 int level = i + 1;
8089
8090                 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8091                                       slot->base_gfn, level) + 1;
8092
8093                 slot->arch.rmap[i] =
8094                         kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
8095                 if (!slot->arch.rmap[i])
8096                         goto out_free;
8097                 if (i == 0)
8098                         continue;
8099
8100                 linfo = kvm_kvzalloc(lpages * sizeof(*linfo));
8101                 if (!linfo)
8102                         goto out_free;
8103
8104                 slot->arch.lpage_info[i - 1] = linfo;
8105
8106                 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8107                         linfo[0].disallow_lpage = 1;
8108                 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8109                         linfo[lpages - 1].disallow_lpage = 1;
8110                 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8111                 /*
8112                  * If the gfn and userspace address are not aligned wrt each
8113                  * other, or if explicitly asked to, disable large page
8114                  * support for this slot
8115                  */
8116                 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8117                     !kvm_largepages_enabled()) {
8118                         unsigned long j;
8119
8120                         for (j = 0; j < lpages; ++j)
8121                                 linfo[j].disallow_lpage = 1;
8122                 }
8123         }
8124
8125         if (kvm_page_track_create_memslot(slot, npages))
8126                 goto out_free;
8127
8128         return 0;
8129
8130 out_free:
8131         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8132                 kvfree(slot->arch.rmap[i]);
8133                 slot->arch.rmap[i] = NULL;
8134                 if (i == 0)
8135                         continue;
8136
8137                 kvfree(slot->arch.lpage_info[i - 1]);
8138                 slot->arch.lpage_info[i - 1] = NULL;
8139         }
8140         return -ENOMEM;
8141 }
8142
8143 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8144 {
8145         /*
8146          * memslots->generation has been incremented.
8147          * mmio generation may have reached its maximum value.
8148          */
8149         kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8150 }
8151
8152 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8153                                 struct kvm_memory_slot *memslot,
8154                                 const struct kvm_userspace_memory_region *mem,
8155                                 enum kvm_mr_change change)
8156 {
8157         return 0;
8158 }
8159
8160 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8161                                      struct kvm_memory_slot *new)
8162 {
8163         /* Still write protect RO slot */
8164         if (new->flags & KVM_MEM_READONLY) {
8165                 kvm_mmu_slot_remove_write_access(kvm, new);
8166                 return;
8167         }
8168
8169         /*
8170          * Call kvm_x86_ops dirty logging hooks when they are valid.
8171          *
8172          * kvm_x86_ops->slot_disable_log_dirty is called when:
8173          *
8174          *  - KVM_MR_CREATE with dirty logging is disabled
8175          *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8176          *
8177          * The reason is, in case of PML, we need to set D-bit for any slots
8178          * with dirty logging disabled in order to eliminate unnecessary GPA
8179          * logging in PML buffer (and potential PML buffer full VMEXT). This
8180          * guarantees leaving PML enabled during guest's lifetime won't have
8181          * any additonal overhead from PML when guest is running with dirty
8182          * logging disabled for memory slots.
8183          *
8184          * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8185          * to dirty logging mode.
8186          *
8187          * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8188          *
8189          * In case of write protect:
8190          *
8191          * Write protect all pages for dirty logging.
8192          *
8193          * All the sptes including the large sptes which point to this
8194          * slot are set to readonly. We can not create any new large
8195          * spte on this slot until the end of the logging.
8196          *
8197          * See the comments in fast_page_fault().
8198          */
8199         if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8200                 if (kvm_x86_ops->slot_enable_log_dirty)
8201                         kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8202                 else
8203                         kvm_mmu_slot_remove_write_access(kvm, new);
8204         } else {
8205                 if (kvm_x86_ops->slot_disable_log_dirty)
8206                         kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8207         }
8208 }
8209
8210 void kvm_arch_commit_memory_region(struct kvm *kvm,
8211                                 const struct kvm_userspace_memory_region *mem,
8212                                 const struct kvm_memory_slot *old,
8213                                 const struct kvm_memory_slot *new,
8214                                 enum kvm_mr_change change)
8215 {
8216         int nr_mmu_pages = 0;
8217
8218         if (!kvm->arch.n_requested_mmu_pages)
8219                 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8220
8221         if (nr_mmu_pages)
8222                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8223
8224         /*
8225          * Dirty logging tracks sptes in 4k granularity, meaning that large
8226          * sptes have to be split.  If live migration is successful, the guest
8227          * in the source machine will be destroyed and large sptes will be
8228          * created in the destination. However, if the guest continues to run
8229          * in the source machine (for example if live migration fails), small
8230          * sptes will remain around and cause bad performance.
8231          *
8232          * Scan sptes if dirty logging has been stopped, dropping those
8233          * which can be collapsed into a single large-page spte.  Later
8234          * page faults will create the large-page sptes.
8235          */
8236         if ((change != KVM_MR_DELETE) &&
8237                 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8238                 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8239                 kvm_mmu_zap_collapsible_sptes(kvm, new);
8240
8241         /*
8242          * Set up write protection and/or dirty logging for the new slot.
8243          *
8244          * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8245          * been zapped so no dirty logging staff is needed for old slot. For
8246          * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8247          * new and it's also covered when dealing with the new slot.
8248          *
8249          * FIXME: const-ify all uses of struct kvm_memory_slot.
8250          */
8251         if (change != KVM_MR_DELETE)
8252                 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8253 }
8254
8255 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8256 {
8257         kvm_mmu_invalidate_zap_all_pages(kvm);
8258 }
8259
8260 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8261                                    struct kvm_memory_slot *slot)
8262 {
8263         kvm_page_track_flush_slot(kvm, slot);
8264 }
8265
8266 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8267 {
8268         if (!list_empty_careful(&vcpu->async_pf.done))
8269                 return true;
8270
8271         if (kvm_apic_has_events(vcpu))
8272                 return true;
8273
8274         if (vcpu->arch.pv.pv_unhalted)
8275                 return true;
8276
8277         if (atomic_read(&vcpu->arch.nmi_queued))
8278                 return true;
8279
8280         if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8281                 return true;
8282
8283         if (kvm_arch_interrupt_allowed(vcpu) &&
8284             kvm_cpu_has_interrupt(vcpu))
8285                 return true;
8286
8287         if (kvm_hv_has_stimer_pending(vcpu))
8288                 return true;
8289
8290         return false;
8291 }
8292
8293 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8294 {
8295         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8296                 kvm_x86_ops->check_nested_events(vcpu, false);
8297
8298         return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8299 }
8300
8301 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8302 {
8303         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8304 }
8305
8306 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8307 {
8308         return kvm_x86_ops->interrupt_allowed(vcpu);
8309 }
8310
8311 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8312 {
8313         if (is_64_bit_mode(vcpu))
8314                 return kvm_rip_read(vcpu);
8315         return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8316                      kvm_rip_read(vcpu));
8317 }
8318 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8319
8320 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8321 {
8322         return kvm_get_linear_rip(vcpu) == linear_rip;
8323 }
8324 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8325
8326 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8327 {
8328         unsigned long rflags;
8329
8330         rflags = kvm_x86_ops->get_rflags(vcpu);
8331         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8332                 rflags &= ~X86_EFLAGS_TF;
8333         return rflags;
8334 }
8335 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8336
8337 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8338 {
8339         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8340             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8341                 rflags |= X86_EFLAGS_TF;
8342         kvm_x86_ops->set_rflags(vcpu, rflags);
8343 }
8344
8345 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8346 {
8347         __kvm_set_rflags(vcpu, rflags);
8348         kvm_make_request(KVM_REQ_EVENT, vcpu);
8349 }
8350 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8351
8352 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8353 {
8354         int r;
8355
8356         if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8357               work->wakeup_all)
8358                 return;
8359
8360         r = kvm_mmu_reload(vcpu);
8361         if (unlikely(r))
8362                 return;
8363
8364         if (!vcpu->arch.mmu.direct_map &&
8365               work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8366                 return;
8367
8368         vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8369 }
8370
8371 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8372 {
8373         return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8374 }
8375
8376 static inline u32 kvm_async_pf_next_probe(u32 key)
8377 {
8378         return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8379 }
8380
8381 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8382 {
8383         u32 key = kvm_async_pf_hash_fn(gfn);
8384
8385         while (vcpu->arch.apf.gfns[key] != ~0)
8386                 key = kvm_async_pf_next_probe(key);
8387
8388         vcpu->arch.apf.gfns[key] = gfn;
8389 }
8390
8391 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8392 {
8393         int i;
8394         u32 key = kvm_async_pf_hash_fn(gfn);
8395
8396         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8397                      (vcpu->arch.apf.gfns[key] != gfn &&
8398                       vcpu->arch.apf.gfns[key] != ~0); i++)
8399                 key = kvm_async_pf_next_probe(key);
8400
8401         return key;
8402 }
8403
8404 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8405 {
8406         return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8407 }
8408
8409 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8410 {
8411         u32 i, j, k;
8412
8413         i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8414         while (true) {
8415                 vcpu->arch.apf.gfns[i] = ~0;
8416                 do {
8417                         j = kvm_async_pf_next_probe(j);
8418                         if (vcpu->arch.apf.gfns[j] == ~0)
8419                                 return;
8420                         k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8421                         /*
8422                          * k lies cyclically in ]i,j]
8423                          * |    i.k.j |
8424                          * |....j i.k.| or  |.k..j i...|
8425                          */
8426                 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8427                 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8428                 i = j;
8429         }
8430 }
8431
8432 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8433 {
8434
8435         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8436                                       sizeof(val));
8437 }
8438
8439 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8440                                      struct kvm_async_pf *work)
8441 {
8442         struct x86_exception fault;
8443
8444         trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8445         kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8446
8447         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8448             (vcpu->arch.apf.send_user_only &&
8449              kvm_x86_ops->get_cpl(vcpu) == 0))
8450                 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8451         else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8452                 fault.vector = PF_VECTOR;
8453                 fault.error_code_valid = true;
8454                 fault.error_code = 0;
8455                 fault.nested_page_fault = false;
8456                 fault.address = work->arch.token;
8457                 kvm_inject_page_fault(vcpu, &fault);
8458         }
8459 }
8460
8461 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8462                                  struct kvm_async_pf *work)
8463 {
8464         struct x86_exception fault;
8465
8466         trace_kvm_async_pf_ready(work->arch.token, work->gva);
8467         if (work->wakeup_all)
8468                 work->arch.token = ~0; /* broadcast wakeup */
8469         else
8470                 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8471
8472         if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8473             !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8474                 fault.vector = PF_VECTOR;
8475                 fault.error_code_valid = true;
8476                 fault.error_code = 0;
8477                 fault.nested_page_fault = false;
8478                 fault.address = work->arch.token;
8479                 kvm_inject_page_fault(vcpu, &fault);
8480         }
8481         vcpu->arch.apf.halted = false;
8482         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8483 }
8484
8485 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8486 {
8487         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8488                 return true;
8489         else
8490                 return !kvm_event_needs_reinjection(vcpu) &&
8491                         kvm_x86_ops->interrupt_allowed(vcpu);
8492 }
8493
8494 void kvm_arch_start_assignment(struct kvm *kvm)
8495 {
8496         atomic_inc(&kvm->arch.assigned_device_count);
8497 }
8498 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8499
8500 void kvm_arch_end_assignment(struct kvm *kvm)
8501 {
8502         atomic_dec(&kvm->arch.assigned_device_count);
8503 }
8504 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8505
8506 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8507 {
8508         return atomic_read(&kvm->arch.assigned_device_count);
8509 }
8510 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8511
8512 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8513 {
8514         atomic_inc(&kvm->arch.noncoherent_dma_count);
8515 }
8516 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8517
8518 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8519 {
8520         atomic_dec(&kvm->arch.noncoherent_dma_count);
8521 }
8522 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8523
8524 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8525 {
8526         return atomic_read(&kvm->arch.noncoherent_dma_count);
8527 }
8528 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8529
8530 bool kvm_arch_has_irq_bypass(void)
8531 {
8532         return kvm_x86_ops->update_pi_irte != NULL;
8533 }
8534
8535 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8536                                       struct irq_bypass_producer *prod)
8537 {
8538         struct kvm_kernel_irqfd *irqfd =
8539                 container_of(cons, struct kvm_kernel_irqfd, consumer);
8540
8541         irqfd->producer = prod;
8542
8543         return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8544                                            prod->irq, irqfd->gsi, 1);
8545 }
8546
8547 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8548                                       struct irq_bypass_producer *prod)
8549 {
8550         int ret;
8551         struct kvm_kernel_irqfd *irqfd =
8552                 container_of(cons, struct kvm_kernel_irqfd, consumer);
8553
8554         WARN_ON(irqfd->producer != prod);
8555         irqfd->producer = NULL;
8556
8557         /*
8558          * When producer of consumer is unregistered, we change back to
8559          * remapped mode, so we can re-use the current implementation
8560          * when the irq is masked/disabled or the consumer side (KVM
8561          * int this case doesn't want to receive the interrupts.
8562         */
8563         ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8564         if (ret)
8565                 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8566                        " fails: %d\n", irqfd->consumer.token, ret);
8567 }
8568
8569 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8570                                    uint32_t guest_irq, bool set)
8571 {
8572         if (!kvm_x86_ops->update_pi_irte)
8573                 return -EINVAL;
8574
8575         return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8576 }
8577
8578 bool kvm_vector_hashing_enabled(void)
8579 {
8580         return vector_hashing;
8581 }
8582 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8583
8584 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8585 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8586 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8587 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8588 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8589 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8590 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8591 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8592 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8593 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8594 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8595 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8596 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8597 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8598 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8599 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8600 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8601 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8602 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);