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
 214         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 215                 values = &locals->values[slot];
 216                 if (values->host != values->curr) {
 217                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 218                         values->curr = values->host;
 219                 }
 220         }
 221         locals->registered = false;
 222         user_return_notifier_unregister(urn);
 223 }
 224
 225 static void shared_msr_update(unsigned slot, u32 msr)
 226 {
 227         u64 value;
 228         unsigned int cpu = smp_processor_id();
 229         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 230
 231         /* only read, and nobody should modify it at this time,
 232          * so don't need lock */
 233         if (slot >= shared_msrs_global.nr) {
 234                 printk(KERN_ERR "kvm: invalid MSR slot!");
 235                 return;
 236         }
 237         rdmsrl_safe(msr, &value);
 238         smsr->values[slot].host = value;
 239         smsr->values[slot].curr = value;
 240 }
 241
 242 void kvm_define_shared_msr(unsigned slot, u32 msr)
 243 {
 244         BUG_ON(slot >= KVM_NR_SHARED_MSRS);
 245         shared_msrs_global.msrs[slot] = msr;
 246         if (slot >= shared_msrs_global.nr)
 247                 shared_msrs_global.nr = slot + 1;
 248 }
 249 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 250
 251 static void kvm_shared_msr_cpu_online(void)
 252 {
 253         unsigned i;
 254
 255         for (i = 0; i < shared_msrs_global.nr; ++i)
 256                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 257 }
 258
 259 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 260 {
 261         unsigned int cpu = smp_processor_id();
 262         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 263         int err;
 264
 265         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 266                 return 0;
 267         smsr->values[slot].curr = value;
 268         err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
 269         if (err)
 270                 return 1;
 271
 272         if (!smsr->registered) {
 273                 smsr->urn.on_user_return = kvm_on_user_return;
 274                 user_return_notifier_register(&smsr->urn);
 275                 smsr->registered = true;
 276         }
 277         return 0;
 278 }
 279 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 280
 281 static void drop_user_return_notifiers(void)
 282 {
 283         unsigned int cpu = smp_processor_id();
 284         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 285
 286         if (smsr->registered)
 287                 kvm_on_user_return(&smsr->urn);
 288 }
 289
 290 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 291 {
 292         return vcpu->arch.apic_base;
 293 }
 294 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 295
 296 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 297 {
 298         u64 old_state = vcpu->arch.apic_base &
 299                 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
 300         u64 new_state = msr_info->data &
 301                 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
 302         u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
 303                 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
 304
 305         if (!msr_info->host_initiated &&
 306             ((msr_info->data & reserved_bits) != 0 ||
 307              new_state == X2APIC_ENABLE ||
 308              (new_state == MSR_IA32_APICBASE_ENABLE &&
 309               old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
 310              (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
 311               old_state == 0)))
 312                 return 1;
 313
 314         kvm_lapic_set_base(vcpu, msr_info->data);
 315         return 0;
 316 }
 317 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 318
 319 asmlinkage __visible void kvm_spurious_fault(void)
 320 {
 321         /* Fault while not rebooting.  We want the trace. */
 322         BUG();
 323 }
 324 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
 325
 326 #define EXCPT_BENIGN            0
 327 #define EXCPT_CONTRIBUTORY      1
 328 #define EXCPT_PF                2
 329
 330 static int exception_class(int vector)
 331 {
 332         switch (vector) {
 333         case PF_VECTOR:
 334                 return EXCPT_PF;
 335         case DE_VECTOR:
 336         case TS_VECTOR:
 337         case NP_VECTOR:
 338         case SS_VECTOR:
 339         case GP_VECTOR:
 340                 return EXCPT_CONTRIBUTORY;
 341         default:
 342                 break;
 343         }
 344         return EXCPT_BENIGN;
 345 }
 346
 347 #define EXCPT_FAULT             0
 348 #define EXCPT_TRAP              1
 349 #define EXCPT_ABORT             2
 350 #define EXCPT_INTERRUPT         3
 351
 352 static int exception_type(int vector)
 353 {
 354         unsigned int mask;
 355
 356         if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
 357                 return EXCPT_INTERRUPT;
 358
 359         mask = 1 << vector;
 360
 361         /* #DB is trap, as instruction watchpoints are handled elsewhere */
 362         if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
 363                 return EXCPT_TRAP;
 364
 365         if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
 366                 return EXCPT_ABORT;
 367
 368         /* Reserved exceptions will result in fault */
 369         return EXCPT_FAULT;
 370 }
 371
 372 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 373                 unsigned nr, bool has_error, u32 error_code,
 374                 bool reinject)
 375 {
 376         u32 prev_nr;
 377         int class1, class2;
 378
 379         kvm_make_request(KVM_REQ_EVENT, vcpu);
 380
 381         if (!vcpu->arch.exception.pending) {
 382         queue:
 383                 if (has_error && !is_protmode(vcpu))
 384                         has_error = false;
 385                 vcpu->arch.exception.pending = true;
 386                 vcpu->arch.exception.has_error_code = has_error;
 387                 vcpu->arch.exception.nr = nr;
 388                 vcpu->arch.exception.error_code = error_code;
 389                 vcpu->arch.exception.reinject = reinject;
 390                 return;
 391         }
 392
 393         /* to check exception */
 394         prev_nr = vcpu->arch.exception.nr;
 395         if (prev_nr == DF_VECTOR) {
 396                 /* triple fault -> shutdown */
 397                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
 398                 return;
 399         }
 400         class1 = exception_class(prev_nr);
 401         class2 = exception_class(nr);
 402         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 403                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 404                 /* generate double fault per SDM Table 5-5 */
 405                 vcpu->arch.exception.pending = true;
 406                 vcpu->arch.exception.has_error_code = true;
 407                 vcpu->arch.exception.nr = DF_VECTOR;
 408                 vcpu->arch.exception.error_code = 0;
 409         } else
 410                 /* replace previous exception with a new one in a hope
 411                    that instruction re-execution will regenerate lost
 412                    exception */
 413                 goto queue;
 414 }
 415
 416 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 417 {
 418         kvm_multiple_exception(vcpu, nr, false, 0, false);
 419 }
 420 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 421
 422 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 423 {
 424         kvm_multiple_exception(vcpu, nr, false, 0, true);
 425 }
 426 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 427
 428 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
 429 {
 430         if (err)
 431                 kvm_inject_gp(vcpu, 0);
 432         else
 433                 kvm_x86_ops->skip_emulated_instruction(vcpu);
 434 }
 435 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
 436
 437 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 438 {
 439         ++vcpu->stat.pf_guest;
 440         vcpu->arch.cr2 = fault->address;
 441         kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
 442 }
 443 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
 444
 445 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 446 {
 447         if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
 448                 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
 449         else
 450                 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
 451
 452         return fault->nested_page_fault;
 453 }
 454
 455 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 456 {
 457         atomic_inc(&vcpu->arch.nmi_queued);
 458         kvm_make_request(KVM_REQ_NMI, vcpu);
 459 }
 460 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 461
 462 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 463 {
 464         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 465 }
 466 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 467
 468 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 469 {
 470         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 471 }
 472 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 473
 474 /*
 475  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 476  * a #GP and return false.
 477  */
 478 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 479 {
 480         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 481                 return true;
 482         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 483         return false;
 484 }
 485 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 486
 487 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
 488 {
 489         if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 490                 return true;
 491
 492         kvm_queue_exception(vcpu, UD_VECTOR);
 493         return false;
 494 }
 495 EXPORT_SYMBOL_GPL(kvm_require_dr);
 496
 497 /*
 498  * This function will be used to read from the physical memory of the currently
 499  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
 500  * can read from guest physical or from the guest's guest physical memory.
 501  */
 502 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 503                             gfn_t ngfn, void *data, int offset, int len,
 504                             u32 access)
 505 {
 506         struct x86_exception exception;
 507         gfn_t real_gfn;
 508         gpa_t ngpa;
 509
 510         ngpa     = gfn_to_gpa(ngfn);
 511         real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
 512         if (real_gfn == UNMAPPED_GVA)
 513                 return -EFAULT;
 514
 515         real_gfn = gpa_to_gfn(real_gfn);
 516
 517         return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
 518 }
 519 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
 520
 521 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
 522                                void *data, int offset, int len, u32 access)
 523 {
 524         return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
 525                                        data, offset, len, access);
 526 }
 527
 528 /*
 529  * Load the pae pdptrs.  Return true is they are all valid.
 530  */
 531 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
 532 {
 533         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 534         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 535         int i;
 536         int ret;
 537         u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
 538
 539         ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
 540                                       offset * sizeof(u64), sizeof(pdpte),
 541                                       PFERR_USER_MASK|PFERR_WRITE_MASK);
 542         if (ret < 0) {
 543                 ret = 0;
 544                 goto out;
 545         }
 546         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 547                 if ((pdpte[i] & PT_PRESENT_MASK) &&
 548                     (pdpte[i] &
 549                      vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
 550                         ret = 0;
 551                         goto out;
 552                 }
 553         }
 554         ret = 1;
 555
 556         memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
 557         __set_bit(VCPU_EXREG_PDPTR,
 558                   (unsigned long *)&vcpu->arch.regs_avail);
 559         __set_bit(VCPU_EXREG_PDPTR,
 560                   (unsigned long *)&vcpu->arch.regs_dirty);
 561 out:
 562
 563         return ret;
 564 }
 565 EXPORT_SYMBOL_GPL(load_pdptrs);
 566
 567 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 568 {
 569         u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
 570         bool changed = true;
 571         int offset;
 572         gfn_t gfn;
 573         int r;
 574
 575         if (is_long_mode(vcpu) || !is_pae(vcpu))
 576                 return false;
 577
 578         if (!test_bit(VCPU_EXREG_PDPTR,
 579                       (unsigned long *)&vcpu->arch.regs_avail))
 580                 return true;
 581
 582         gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
 583         offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
 584         r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
 585                                        PFERR_USER_MASK | PFERR_WRITE_MASK);
 586         if (r < 0)
 587                 goto out;
 588         changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
 589 out:
 590
 591         return changed;
 592 }
 593
 594 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 595 {
 596         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 597         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
 598
 599         cr0 |= X86_CR0_ET;
 600
 601 #ifdef CONFIG_X86_64
 602         if (cr0 & 0xffffffff00000000UL)
 603                 return 1;
 604 #endif
 605
 606         cr0 &= ~CR0_RESERVED_BITS;
 607
 608         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 609                 return 1;
 610
 611         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 612                 return 1;
 613
 614         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 615 #ifdef CONFIG_X86_64
 616                 if ((vcpu->arch.efer & EFER_LME)) {
 617                         int cs_db, cs_l;
 618
 619                         if (!is_pae(vcpu))
 620                                 return 1;
 621                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 622                         if (cs_l)
 623                                 return 1;
 624                 } else
 625 #endif
 626                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 627                                                  kvm_read_cr3(vcpu)))
 628                         return 1;
 629         }
 630
 631         if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
 632                 return 1;
 633
 634         kvm_x86_ops->set_cr0(vcpu, cr0);
 635
 636         if ((cr0 ^ old_cr0) & X86_CR0_PG) {
 637                 kvm_clear_async_pf_completion_queue(vcpu);
 638                 kvm_async_pf_hash_reset(vcpu);
 639         }
 640
 641         if ((cr0 ^ old_cr0) & update_bits)
 642                 kvm_mmu_reset_context(vcpu);
 643
 644         if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
 645             kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
 646             !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
 647                 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
 648
 649         return 0;
 650 }
 651 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 652
 653 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 654 {
 655         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 656 }
 657 EXPORT_SYMBOL_GPL(kvm_lmsw);
 658
 659 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
 660 {
 661         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
 662                         !vcpu->guest_xcr0_loaded) {
 663                 /* kvm_set_xcr() also depends on this */
 664                 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
 665                 vcpu->guest_xcr0_loaded = 1;
 666         }
 667 }
 668
 669 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
 670 {
 671         if (vcpu->guest_xcr0_loaded) {
 672                 if (vcpu->arch.xcr0 != host_xcr0)
 673                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
 674                 vcpu->guest_xcr0_loaded = 0;
 675         }
 676 }
 677
 678 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 679 {
 680         u64 xcr0 = xcr;
 681         u64 old_xcr0 = vcpu->arch.xcr0;
 682         u64 valid_bits;
 683
 684         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
 685         if (index != XCR_XFEATURE_ENABLED_MASK)
 686                 return 1;
 687         if (!(xcr0 & XFEATURE_MASK_FP))
 688                 return 1;
 689         if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
 690                 return 1;
 691
 692         /*
 693          * Do not allow the guest to set bits that we do not support
 694          * saving.  However, xcr0 bit 0 is always set, even if the
 695          * emulated CPU does not support XSAVE (see fx_init).
 696          */
 697         valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
 698         if (xcr0 & ~valid_bits)
 699                 return 1;
 700
 701         if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
 702             (!(xcr0 & XFEATURE_MASK_BNDCSR)))
 703                 return 1;
 704
 705         if (xcr0 & XFEATURE_MASK_AVX512) {
 706                 if (!(xcr0 & XFEATURE_MASK_YMM))
 707                         return 1;
 708                 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
 709                         return 1;
 710         }
 711         vcpu->arch.xcr0 = xcr0;
 712
 713         if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
 714                 kvm_update_cpuid(vcpu);
 715         return 0;
 716 }
 717
 718 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 719 {
 720         if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
 721             __kvm_set_xcr(vcpu, index, xcr)) {
 722                 kvm_inject_gp(vcpu, 0);
 723                 return 1;
 724         }
 725         return 0;
 726 }
 727 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 728
 729 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 730 {
 731         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 732         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
 733                                    X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
 734
 735         if (cr4 & CR4_RESERVED_BITS)
 736                 return 1;
 737
 738         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
 739                 return 1;
 740
 741         if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
 742                 return 1;
 743
 744         if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
 745                 return 1;
 746
 747         if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
 748                 return 1;
 749
 750         if (!guest_cpuid_has_pku(vcpu) && (cr4 & X86_CR4_PKE))
 751                 return 1;
 752
 753         if (is_long_mode(vcpu)) {
 754                 if (!(cr4 & X86_CR4_PAE))
 755                         return 1;
 756         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 757                    && ((cr4 ^ old_cr4) & pdptr_bits)
 758                    && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 759                                    kvm_read_cr3(vcpu)))
 760                 return 1;
 761
 762         if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
 763                 if (!guest_cpuid_has_pcid(vcpu))
 764                         return 1;
 765
 766                 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
 767                 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
 768                         return 1;
 769         }
 770
 771         if (kvm_x86_ops->set_cr4(vcpu, cr4))
 772                 return 1;
 773
 774         if (((cr4 ^ old_cr4) & pdptr_bits) ||
 775             (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
 776                 kvm_mmu_reset_context(vcpu);
 777
 778         if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
 779                 kvm_update_cpuid(vcpu);
 780
 781         return 0;
 782 }
 783 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 784
 785 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 786 {
 787 #ifdef CONFIG_X86_64
 788         cr3 &= ~CR3_PCID_INVD;
 789 #endif
 790
 791         if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
 792                 kvm_mmu_sync_roots(vcpu);
 793                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 794                 return 0;
 795         }
 796
 797         if (is_long_mode(vcpu)) {
 798                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 799                         return 1;
 800         } else if (is_pae(vcpu) && is_paging(vcpu) &&
 801                    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 802                 return 1;
 803
 804         vcpu->arch.cr3 = cr3;
 805         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
 806         kvm_mmu_new_cr3(vcpu);
 807         return 0;
 808 }
 809 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 810
 811 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 812 {
 813         if (cr8 & CR8_RESERVED_BITS)
 814                 return 1;
 815         if (lapic_in_kernel(vcpu))
 816                 kvm_lapic_set_tpr(vcpu, cr8);
 817         else
 818                 vcpu->arch.cr8 = cr8;
 819         return 0;
 820 }
 821 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 822
 823 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 824 {
 825         if (lapic_in_kernel(vcpu))
 826                 return kvm_lapic_get_cr8(vcpu);
 827         else
 828                 return vcpu->arch.cr8;
 829 }
 830 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 831
 832 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
 833 {
 834         int i;
 835
 836         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 837                 for (i = 0; i < KVM_NR_DB_REGS; i++)
 838                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
 839                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
 840         }
 841 }
 842
 843 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
 844 {
 845         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 846                 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
 847 }
 848
 849 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
 850 {
 851         unsigned long dr7;
 852
 853         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 854                 dr7 = vcpu->arch.guest_debug_dr7;
 855         else
 856                 dr7 = vcpu->arch.dr7;
 857         kvm_x86_ops->set_dr7(vcpu, dr7);
 858         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
 859         if (dr7 & DR7_BP_EN_MASK)
 860                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
 861 }
 862
 863 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
 864 {
 865         u64 fixed = DR6_FIXED_1;
 866
 867         if (!guest_cpuid_has_rtm(vcpu))
 868                 fixed |= DR6_RTM;
 869         return fixed;
 870 }
 871
 872 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 873 {
 874         switch (dr) {
 875         case 0 ... 3:
 876                 vcpu->arch.db[dr] = val;
 877                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 878                         vcpu->arch.eff_db[dr] = val;
 879                 break;
 880         case 4:
 881                 /* fall through */
 882         case 6:
 883                 if (val & 0xffffffff00000000ULL)
 884                         return -1; /* #GP */
 885                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
 886                 kvm_update_dr6(vcpu);
 887                 break;
 888         case 5:
 889                 /* fall through */
 890         default: /* 7 */
 891                 if (val & 0xffffffff00000000ULL)
 892                         return -1; /* #GP */
 893                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 894                 kvm_update_dr7(vcpu);
 895                 break;
 896         }
 897
 898         return 0;
 899 }
 900
 901 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 902 {
 903         if (__kvm_set_dr(vcpu, dr, val)) {
 904                 kvm_inject_gp(vcpu, 0);
 905                 return 1;
 906         }
 907         return 0;
 908 }
 909 EXPORT_SYMBOL_GPL(kvm_set_dr);
 910
 911 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 912 {
 913         switch (dr) {
 914         case 0 ... 3:
 915                 *val = vcpu->arch.db[dr];
 916                 break;
 917         case 4:
 918                 /* fall through */
 919         case 6:
 920                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 921                         *val = vcpu->arch.dr6;
 922                 else
 923                         *val = kvm_x86_ops->get_dr6(vcpu);
 924                 break;
 925         case 5:
 926                 /* fall through */
 927         default: /* 7 */
 928                 *val = vcpu->arch.dr7;
 929                 break;
 930         }
 931         return 0;
 932 }
 933 EXPORT_SYMBOL_GPL(kvm_get_dr);
 934
 935 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
 936 {
 937         u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 938         u64 data;
 939         int err;
 940
 941         err = kvm_pmu_rdpmc(vcpu, ecx, &data);
 942         if (err)
 943                 return err;
 944         kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
 945         kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
 946         return err;
 947 }
 948 EXPORT_SYMBOL_GPL(kvm_rdpmc);
 949
 950 /*
 951  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 952  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 953  *
 954  * This list is modified at module load time to reflect the
 955  * capabilities of the host cpu. This capabilities test skips MSRs that are
 956  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
 957  * may depend on host virtualization features rather than host cpu features.
 958  */
 959
 960 static u32 msrs_to_save[] = {
 961         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 962         MSR_STAR,
 963 #ifdef CONFIG_X86_64
 964         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 965 #endif
 966         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
 967         MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
 968 };
 969
 970 static unsigned num_msrs_to_save;
 971
 972 static u32 emulated_msrs[] = {
 973         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 974         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 975         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 976         HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
 977         HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
 978         HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
 979         HV_X64_MSR_RESET,
 980         HV_X64_MSR_VP_INDEX,
 981         HV_X64_MSR_VP_RUNTIME,
 982         HV_X64_MSR_SCONTROL,
 983         HV_X64_MSR_STIMER0_CONFIG,
 984         HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
 985         MSR_KVM_PV_EOI_EN,
 986
 987         MSR_IA32_TSC_ADJUST,
 988         MSR_IA32_TSCDEADLINE,
 989         MSR_IA32_MISC_ENABLE,
 990         MSR_IA32_MCG_STATUS,
 991         MSR_IA32_MCG_CTL,
 992         MSR_IA32_MCG_EXT_CTL,
 993         MSR_IA32_SMBASE,
 994 };
 995
 996 static unsigned num_emulated_msrs;
 997
 998 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
 999 {
1000         if (efer & efer_reserved_bits)
1001                 return false;
1002
1003         if (efer & EFER_FFXSR) {
1004                 struct kvm_cpuid_entry2 *feat;
1005
1006                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1007                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
1008                         return false;
1009         }
1010
1011         if (efer & EFER_SVME) {
1012                 struct kvm_cpuid_entry2 *feat;
1013
1014                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1015                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1016                         return false;
1017         }
1018
1019         return true;
1020 }
1021 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1022
1023 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1024 {
1025         u64 old_efer = vcpu->arch.efer;
1026
1027         if (!kvm_valid_efer(vcpu, efer))
1028                 return 1;
1029
1030         if (is_paging(vcpu)
1031             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1032                 return 1;
1033
1034         efer &= ~EFER_LMA;
1035         efer |= vcpu->arch.efer & EFER_LMA;
1036
1037         kvm_x86_ops->set_efer(vcpu, efer);
1038
1039         /* Update reserved bits */
1040         if ((efer ^ old_efer) & EFER_NX)
1041                 kvm_mmu_reset_context(vcpu);
1042
1043         return 0;
1044 }
1045
1046 void kvm_enable_efer_bits(u64 mask)
1047 {
1048        efer_reserved_bits &= ~mask;
1049 }
1050 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1051
1052 /*
1053  * Writes msr value into into the appropriate "register".
1054  * Returns 0 on success, non-0 otherwise.
1055  * Assumes vcpu_load() was already called.
1056  */
1057 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1058 {
1059         switch (msr->index) {
1060         case MSR_FS_BASE:
1061         case MSR_GS_BASE:
1062         case MSR_KERNEL_GS_BASE:
1063         case MSR_CSTAR:
1064         case MSR_LSTAR:
1065                 if (is_noncanonical_address(msr->data))
1066                         return 1;
1067                 break;
1068         case MSR_IA32_SYSENTER_EIP:
1069         case MSR_IA32_SYSENTER_ESP:
1070                 /*
1071                  * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1072                  * non-canonical address is written on Intel but not on
1073                  * AMD (which ignores the top 32-bits, because it does
1074                  * not implement 64-bit SYSENTER).
1075                  *
1076                  * 64-bit code should hence be able to write a non-canonical
1077                  * value on AMD.  Making the address canonical ensures that
1078                  * vmentry does not fail on Intel after writing a non-canonical
1079                  * value, and that something deterministic happens if the guest
1080                  * invokes 64-bit SYSENTER.
1081                  */
1082                 msr->data = get_canonical(msr->data);
1083         }
1084         return kvm_x86_ops->set_msr(vcpu, msr);
1085 }
1086 EXPORT_SYMBOL_GPL(kvm_set_msr);
1087
1088 /*
1089  * Adapt set_msr() to msr_io()'s calling convention
1090  */
1091 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1092 {
1093         struct msr_data msr;
1094         int r;
1095
1096         msr.index = index;
1097         msr.host_initiated = true;
1098         r = kvm_get_msr(vcpu, &msr);
1099         if (r)
1100                 return r;
1101
1102         *data = msr.data;
1103         return 0;
1104 }
1105
1106 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1107 {
1108         struct msr_data msr;
1109
1110         msr.data = *data;
1111         msr.index = index;
1112         msr.host_initiated = true;
1113         return kvm_set_msr(vcpu, &msr);
1114 }
1115
1116 #ifdef CONFIG_X86_64
1117 struct pvclock_gtod_data {
1118         seqcount_t      seq;
1119
1120         struct { /* extract of a clocksource struct */
1121                 int vclock_mode;
1122                 cycle_t cycle_last;
1123                 cycle_t mask;
1124                 u32     mult;
1125                 u32     shift;
1126         } clock;
1127
1128         u64             boot_ns;
1129         u64             nsec_base;
1130 };
1131
1132 static struct pvclock_gtod_data pvclock_gtod_data;
1133
1134 static void update_pvclock_gtod(struct timekeeper *tk)
1135 {
1136         struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1137         u64 boot_ns;
1138
1139         boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1140
1141         write_seqcount_begin(&vdata->seq);
1142
1143         /* copy pvclock gtod data */
1144         vdata->clock.vclock_mode        = tk->tkr_mono.clock->archdata.vclock_mode;
1145         vdata->clock.cycle_last         = tk->tkr_mono.cycle_last;
1146         vdata->clock.mask               = tk->tkr_mono.mask;
1147         vdata->clock.mult               = tk->tkr_mono.mult;
1148         vdata->clock.shift              = tk->tkr_mono.shift;
1149
1150         vdata->boot_ns                  = boot_ns;
1151         vdata->nsec_base                = tk->tkr_mono.xtime_nsec;
1152
1153         write_seqcount_end(&vdata->seq);
1154 }
1155 #endif
1156
1157 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1158 {
1159         /*
1160          * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1161          * vcpu_enter_guest.  This function is only called from
1162          * the physical CPU that is running vcpu.
1163          */
1164         kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1165 }
1166
1167 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1168 {
1169         int version;
1170         int r;
1171         struct pvclock_wall_clock wc;
1172         struct timespec64 boot;
1173
1174         if (!wall_clock)
1175                 return;
1176
1177         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1178         if (r)
1179                 return;
1180
1181         if (version & 1)
1182                 ++version;  /* first time write, random junk */
1183
1184         ++version;
1185
1186         if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1187                 return;
1188
1189         /*
1190          * The guest calculates current wall clock time by adding
1191          * system time (updated by kvm_guest_time_update below) to the
1192          * wall clock specified here.  guest system time equals host
1193          * system time for us, thus we must fill in host boot time here.
1194          */
1195         getboottime64(&boot);
1196
1197         if (kvm->arch.kvmclock_offset) {
1198                 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1199                 boot = timespec64_sub(boot, ts);
1200         }
1201         wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1202         wc.nsec = boot.tv_nsec;
1203         wc.version = version;
1204
1205         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1206
1207         version++;
1208         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1209 }
1210
1211 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1212 {
1213         do_shl32_div32(dividend, divisor);
1214         return dividend;
1215 }
1216
1217 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1218                                s8 *pshift, u32 *pmultiplier)
1219 {
1220         uint64_t scaled64;
1221         int32_t  shift = 0;
1222         uint64_t tps64;
1223         uint32_t tps32;
1224
1225         tps64 = base_hz;
1226         scaled64 = scaled_hz;
1227         while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1228                 tps64 >>= 1;
1229                 shift--;
1230         }
1231
1232         tps32 = (uint32_t)tps64;
1233         while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1234                 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1235                         scaled64 >>= 1;
1236                 else
1237                         tps32 <<= 1;
1238                 shift++;
1239         }
1240
1241         *pshift = shift;
1242         *pmultiplier = div_frac(scaled64, tps32);
1243
1244         pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1245                  __func__, base_hz, scaled_hz, shift, *pmultiplier);
1246 }
1247
1248 #ifdef CONFIG_X86_64
1249 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1250 #endif
1251
1252 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1253 static unsigned long max_tsc_khz;
1254
1255 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1256 {
1257         u64 v = (u64)khz * (1000000 + ppm);
1258         do_div(v, 1000000);
1259         return v;
1260 }
1261
1262 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1263 {
1264         u64 ratio;
1265
1266         /* Guest TSC same frequency as host TSC? */
1267         if (!scale) {
1268                 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1269                 return 0;
1270         }
1271
1272         /* TSC scaling supported? */
1273         if (!kvm_has_tsc_control) {
1274                 if (user_tsc_khz > tsc_khz) {
1275                         vcpu->arch.tsc_catchup = 1;
1276                         vcpu->arch.tsc_always_catchup = 1;
1277                         return 0;
1278                 } else {
1279                         WARN(1, "user requested TSC rate below hardware speed\n");
1280                         return -1;
1281                 }
1282         }
1283
1284         /* TSC scaling required  - calculate ratio */
1285         ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1286                                 user_tsc_khz, tsc_khz);
1287
1288         if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1289                 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1290                           user_tsc_khz);
1291                 return -1;
1292         }
1293
1294         vcpu->arch.tsc_scaling_ratio = ratio;
1295         return 0;
1296 }
1297
1298 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1299 {
1300         u32 thresh_lo, thresh_hi;
1301         int use_scaling = 0;
1302
1303         /* tsc_khz can be zero if TSC calibration fails */
1304         if (user_tsc_khz == 0) {
1305                 /* set tsc_scaling_ratio to a safe value */
1306                 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1307                 return -1;
1308         }
1309
1310         /* Compute a scale to convert nanoseconds in TSC cycles */
1311         kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1312                            &vcpu->arch.virtual_tsc_shift,
1313                            &vcpu->arch.virtual_tsc_mult);
1314         vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1315
1316         /*
1317          * Compute the variation in TSC rate which is acceptable
1318          * within the range of tolerance and decide if the
1319          * rate being applied is within that bounds of the hardware
1320          * rate.  If so, no scaling or compensation need be done.
1321          */
1322         thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1323         thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1324         if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1325                 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1326                 use_scaling = 1;
1327         }
1328         return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1329 }
1330
1331 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1332 {
1333         u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1334                                       vcpu->arch.virtual_tsc_mult,
1335                                       vcpu->arch.virtual_tsc_shift);
1336         tsc += vcpu->arch.this_tsc_write;
1337         return tsc;
1338 }
1339
1340 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1341 {
1342 #ifdef CONFIG_X86_64
1343         bool vcpus_matched;
1344         struct kvm_arch *ka = &vcpu->kvm->arch;
1345         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1346
1347         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1348                          atomic_read(&vcpu->kvm->online_vcpus));
1349
1350         /*
1351          * Once the masterclock is enabled, always perform request in
1352          * order to update it.
1353          *
1354          * In order to enable masterclock, the host clocksource must be TSC
1355          * and the vcpus need to have matched TSCs.  When that happens,
1356          * perform request to enable masterclock.
1357          */
1358         if (ka->use_master_clock ||
1359             (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1360                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1361
1362         trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1363                             atomic_read(&vcpu->kvm->online_vcpus),
1364                             ka->use_master_clock, gtod->clock.vclock_mode);
1365 #endif
1366 }
1367
1368 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1369 {
1370         u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1371         vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1372 }
1373
1374 /*
1375  * Multiply tsc by a fixed point number represented by ratio.
1376  *
1377  * The most significant 64-N bits (mult) of ratio represent the
1378  * integral part of the fixed point number; the remaining N bits
1379  * (frac) represent the fractional part, ie. ratio represents a fixed
1380  * point number (mult + frac * 2^(-N)).
1381  *
1382  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1383  */
1384 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1385 {
1386         return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1387 }
1388
1389 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1390 {
1391         u64 _tsc = tsc;
1392         u64 ratio = vcpu->arch.tsc_scaling_ratio;
1393
1394         if (ratio != kvm_default_tsc_scaling_ratio)
1395                 _tsc = __scale_tsc(ratio, tsc);
1396
1397         return _tsc;
1398 }
1399 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1400
1401 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1402 {
1403         u64 tsc;
1404
1405         tsc = kvm_scale_tsc(vcpu, rdtsc());
1406
1407         return target_tsc - tsc;
1408 }
1409
1410 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1411 {
1412         return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc));
1413 }
1414 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1415
1416 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1417 {
1418         struct kvm *kvm = vcpu->kvm;
1419         u64 offset, ns, elapsed;
1420         unsigned long flags;
1421         s64 usdiff;
1422         bool matched;
1423         bool already_matched;
1424         u64 data = msr->data;
1425
1426         raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1427         offset = kvm_compute_tsc_offset(vcpu, data);
1428         ns = get_kernel_ns();
1429         elapsed = ns - kvm->arch.last_tsc_nsec;
1430
1431         if (vcpu->arch.virtual_tsc_khz) {
1432                 int faulted = 0;
1433
1434                 /* n.b - signed multiplication and division required */
1435                 usdiff = data - kvm->arch.last_tsc_write;
1436 #ifdef CONFIG_X86_64
1437                 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1438 #else
1439                 /* do_div() only does unsigned */
1440                 asm("1: idivl %[divisor]\n"
1441                     "2: xor %%edx, %%edx\n"
1442                     "   movl $0, %[faulted]\n"
1443                     "3:\n"
1444                     ".section .fixup,\"ax\"\n"
1445                     "4: movl $1, %[faulted]\n"
1446                     "   jmp  3b\n"
1447                     ".previous\n"
1448
1449                 _ASM_EXTABLE(1b, 4b)
1450
1451                 : "=A"(usdiff), [faulted] "=r" (faulted)
1452                 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1453
1454 #endif
1455                 do_div(elapsed, 1000);
1456                 usdiff -= elapsed;
1457                 if (usdiff < 0)
1458                         usdiff = -usdiff;
1459
1460                 /* idivl overflow => difference is larger than USEC_PER_SEC */
1461                 if (faulted)
1462                         usdiff = USEC_PER_SEC;
1463         } else
1464                 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1465
1466         /*
1467          * Special case: TSC write with a small delta (1 second) of virtual
1468          * cycle time against real time is interpreted as an attempt to
1469          * synchronize the CPU.
1470          *
1471          * For a reliable TSC, we can match TSC offsets, and for an unstable
1472          * TSC, we add elapsed time in this computation.  We could let the
1473          * compensation code attempt to catch up if we fall behind, but
1474          * it's better to try to match offsets from the beginning.
1475          */
1476         if (usdiff < USEC_PER_SEC &&
1477             vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1478                 if (!check_tsc_unstable()) {
1479                         offset = kvm->arch.cur_tsc_offset;
1480                         pr_debug("kvm: matched tsc offset for %llu\n", data);
1481                 } else {
1482                         u64 delta = nsec_to_cycles(vcpu, elapsed);
1483                         data += delta;
1484                         offset = kvm_compute_tsc_offset(vcpu, data);
1485                         pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1486                 }
1487                 matched = true;
1488                 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1489         } else {
1490                 /*
1491                  * We split periods of matched TSC writes into generations.
1492                  * For each generation, we track the original measured
1493                  * nanosecond time, offset, and write, so if TSCs are in
1494                  * sync, we can match exact offset, and if not, we can match
1495                  * exact software computation in compute_guest_tsc()
1496                  *
1497                  * These values are tracked in kvm->arch.cur_xxx variables.
1498                  */
1499                 kvm->arch.cur_tsc_generation++;
1500                 kvm->arch.cur_tsc_nsec = ns;
1501                 kvm->arch.cur_tsc_write = data;
1502                 kvm->arch.cur_tsc_offset = offset;
1503                 matched = false;
1504                 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1505                          kvm->arch.cur_tsc_generation, data);
1506         }
1507
1508         /*
1509          * We also track th most recent recorded KHZ, write and time to
1510          * allow the matching interval to be extended at each write.
1511          */
1512         kvm->arch.last_tsc_nsec = ns;
1513         kvm->arch.last_tsc_write = data;
1514         kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1515
1516         vcpu->arch.last_guest_tsc = data;
1517
1518         /* Keep track of which generation this VCPU has synchronized to */
1519         vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1520         vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1521         vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1522
1523         if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1524                 update_ia32_tsc_adjust_msr(vcpu, offset);
1525         kvm_x86_ops->write_tsc_offset(vcpu, offset);
1526         raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1527
1528         spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1529         if (!matched) {
1530                 kvm->arch.nr_vcpus_matched_tsc = 0;
1531         } else if (!already_matched) {
1532                 kvm->arch.nr_vcpus_matched_tsc++;
1533         }
1534
1535         kvm_track_tsc_matching(vcpu);
1536         spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1537 }
1538
1539 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1540
1541 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1542                                            s64 adjustment)
1543 {
1544         kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1545 }
1546
1547 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1548 {
1549         if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1550                 WARN_ON(adjustment < 0);
1551         adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1552         kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1553 }
1554
1555 #ifdef CONFIG_X86_64
1556
1557 static cycle_t read_tsc(void)
1558 {
1559         cycle_t ret = (cycle_t)rdtsc_ordered();
1560         u64 last = pvclock_gtod_data.clock.cycle_last;
1561
1562         if (likely(ret >= last))
1563                 return ret;
1564
1565         /*
1566          * GCC likes to generate cmov here, but this branch is extremely
1567          * predictable (it's just a function of time and the likely is
1568          * very likely) and there's a data dependence, so force GCC
1569          * to generate a branch instead.  I don't barrier() because
1570          * we don't actually need a barrier, and if this function
1571          * ever gets inlined it will generate worse code.
1572          */
1573         asm volatile ("");
1574         return last;
1575 }
1576
1577 static inline u64 vgettsc(cycle_t *cycle_now)
1578 {
1579         long v;
1580         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1581
1582         *cycle_now = read_tsc();
1583
1584         v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1585         return v * gtod->clock.mult;
1586 }
1587
1588 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1589 {
1590         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1591         unsigned long seq;
1592         int mode;
1593         u64 ns;
1594
1595         do {
1596                 seq = read_seqcount_begin(&gtod->seq);
1597                 mode = gtod->clock.vclock_mode;
1598                 ns = gtod->nsec_base;
1599                 ns += vgettsc(cycle_now);
1600                 ns >>= gtod->clock.shift;
1601                 ns += gtod->boot_ns;
1602         } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1603         *t = ns;
1604
1605         return mode;
1606 }
1607
1608 /* returns true if host is using tsc clocksource */
1609 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1610 {
1611         /* checked again under seqlock below */
1612         if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1613                 return false;
1614
1615         return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1616 }
1617 #endif
1618
1619 /*
1620  *
1621  * Assuming a stable TSC across physical CPUS, and a stable TSC
1622  * across virtual CPUs, the following condition is possible.
1623  * Each numbered line represents an event visible to both
1624  * CPUs at the next numbered event.
1625  *
1626  * "timespecX" represents host monotonic time. "tscX" represents
1627  * RDTSC value.
1628  *
1629  *              VCPU0 on CPU0           |       VCPU1 on CPU1
1630  *
1631  * 1.  read timespec0,tsc0
1632  * 2.                                   | timespec1 = timespec0 + N
1633  *                                      | tsc1 = tsc0 + M
1634  * 3. transition to guest               | transition to guest
1635  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1636  * 5.                                   | ret1 = timespec1 + (rdtsc - tsc1)
1637  *                                      | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1638  *
1639  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1640  *
1641  *      - ret0 < ret1
1642  *      - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1643  *              ...
1644  *      - 0 < N - M => M < N
1645  *
1646  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1647  * always the case (the difference between two distinct xtime instances
1648  * might be smaller then the difference between corresponding TSC reads,
1649  * when updating guest vcpus pvclock areas).
1650  *
1651  * To avoid that problem, do not allow visibility of distinct
1652  * system_timestamp/tsc_timestamp values simultaneously: use a master
1653  * copy of host monotonic time values. Update that master copy
1654  * in lockstep.
1655  *
1656  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1657  *
1658  */
1659
1660 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1661 {
1662 #ifdef CONFIG_X86_64
1663         struct kvm_arch *ka = &kvm->arch;
1664         int vclock_mode;
1665         bool host_tsc_clocksource, vcpus_matched;
1666
1667         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1668                         atomic_read(&kvm->online_vcpus));
1669
1670         /*
1671          * If the host uses TSC clock, then passthrough TSC as stable
1672          * to the guest.
1673          */
1674         host_tsc_clocksource = kvm_get_time_and_clockread(
1675                                         &ka->master_kernel_ns,
1676                                         &ka->master_cycle_now);
1677
1678         ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1679                                 && !backwards_tsc_observed
1680                                 && !ka->boot_vcpu_runs_old_kvmclock;
1681
1682         if (ka->use_master_clock)
1683                 atomic_set(&kvm_guest_has_master_clock, 1);
1684
1685         vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1686         trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1687                                         vcpus_matched);
1688 #endif
1689 }
1690
1691 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1692 {
1693         kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1694 }
1695
1696 static void kvm_gen_update_masterclock(struct kvm *kvm)
1697 {
1698 #ifdef CONFIG_X86_64
1699         int i;
1700         struct kvm_vcpu *vcpu;
1701         struct kvm_arch *ka = &kvm->arch;
1702
1703         spin_lock(&ka->pvclock_gtod_sync_lock);
1704         kvm_make_mclock_inprogress_request(kvm);
1705         /* no guest entries from this point */
1706         pvclock_update_vm_gtod_copy(kvm);
1707
1708         kvm_for_each_vcpu(i, vcpu, kvm)
1709                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1710
1711         /* guest entries allowed */
1712         kvm_for_each_vcpu(i, vcpu, kvm)
1713                 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1714
1715         spin_unlock(&ka->pvclock_gtod_sync_lock);
1716 #endif
1717 }
1718
1719 static int kvm_guest_time_update(struct kvm_vcpu *v)
1720 {
1721         unsigned long flags, tgt_tsc_khz;
1722         struct kvm_vcpu_arch *vcpu = &v->arch;
1723         struct kvm_arch *ka = &v->kvm->arch;
1724         s64 kernel_ns;
1725         u64 tsc_timestamp, host_tsc;
1726         struct pvclock_vcpu_time_info guest_hv_clock;
1727         u8 pvclock_flags;
1728         bool use_master_clock;
1729
1730         kernel_ns = 0;
1731         host_tsc = 0;
1732
1733         /*
1734          * If the host uses TSC clock, then passthrough TSC as stable
1735          * to the guest.
1736          */
1737         spin_lock(&ka->pvclock_gtod_sync_lock);
1738         use_master_clock = ka->use_master_clock;
1739         if (use_master_clock) {
1740                 host_tsc = ka->master_cycle_now;
1741                 kernel_ns = ka->master_kernel_ns;
1742         }
1743         spin_unlock(&ka->pvclock_gtod_sync_lock);
1744
1745         /* Keep irq disabled to prevent changes to the clock */
1746         local_irq_save(flags);
1747         tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1748         if (unlikely(tgt_tsc_khz == 0)) {
1749                 local_irq_restore(flags);
1750                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1751                 return 1;
1752         }
1753         if (!use_master_clock) {
1754                 host_tsc = rdtsc();
1755                 kernel_ns = get_kernel_ns();
1756         }
1757
1758         tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1759
1760         /*
1761          * We may have to catch up the TSC to match elapsed wall clock
1762          * time for two reasons, even if kvmclock is used.
1763          *   1) CPU could have been running below the maximum TSC rate
1764          *   2) Broken TSC compensation resets the base at each VCPU
1765          *      entry to avoid unknown leaps of TSC even when running
1766          *      again on the same CPU.  This may cause apparent elapsed
1767          *      time to disappear, and the guest to stand still or run
1768          *      very slowly.
1769          */
1770         if (vcpu->tsc_catchup) {
1771                 u64 tsc = compute_guest_tsc(v, kernel_ns);
1772                 if (tsc > tsc_timestamp) {
1773                         adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1774                         tsc_timestamp = tsc;
1775                 }
1776         }
1777
1778         local_irq_restore(flags);
1779
1780         if (!vcpu->pv_time_enabled)
1781                 return 0;
1782
1783         if (kvm_has_tsc_control)
1784                 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1785
1786         if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1787                 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1788                                    &vcpu->hv_clock.tsc_shift,
1789                                    &vcpu->hv_clock.tsc_to_system_mul);
1790                 vcpu->hw_tsc_khz = tgt_tsc_khz;
1791         }
1792
1793         /* With all the info we got, fill in the values */
1794         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1795         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1796         vcpu->last_guest_tsc = tsc_timestamp;
1797
1798         if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1799                 &guest_hv_clock, sizeof(guest_hv_clock))))
1800                 return 0;
1801
1802         /* This VCPU is paused, but it's legal for a guest to read another
1803          * VCPU's kvmclock, so we really have to follow the specification where
1804          * it says that version is odd if data is being modified, and even after
1805          * it is consistent.
1806          *
1807          * Version field updates must be kept separate.  This is because
1808          * kvm_write_guest_cached might use a "rep movs" instruction, and
1809          * writes within a string instruction are weakly ordered.  So there
1810          * are three writes overall.
1811          *
1812          * As a small optimization, only write the version field in the first
1813          * and third write.  The vcpu->pv_time cache is still valid, because the
1814          * version field is the first in the struct.
1815          */
1816         BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1817
1818         vcpu->hv_clock.version = guest_hv_clock.version + 1;
1819         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1820                                 &vcpu->hv_clock,
1821                                 sizeof(vcpu->hv_clock.version));
1822
1823         smp_wmb();
1824
1825         /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1826         pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1827
1828         if (vcpu->pvclock_set_guest_stopped_request) {
1829                 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1830                 vcpu->pvclock_set_guest_stopped_request = false;
1831         }
1832
1833         /* If the host uses TSC clocksource, then it is stable */
1834         if (use_master_clock)
1835                 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1836
1837         vcpu->hv_clock.flags = pvclock_flags;
1838
1839         trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1840
1841         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1842                                 &vcpu->hv_clock,
1843                                 sizeof(vcpu->hv_clock));
1844
1845         smp_wmb();
1846
1847         vcpu->hv_clock.version++;
1848         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1849                                 &vcpu->hv_clock,
1850                                 sizeof(vcpu->hv_clock.version));
1851         return 0;
1852 }
1853
1854 /*
1855  * kvmclock updates which are isolated to a given vcpu, such as
1856  * vcpu->cpu migration, should not allow system_timestamp from
1857  * the rest of the vcpus to remain static. Otherwise ntp frequency
1858  * correction applies to one vcpu's system_timestamp but not
1859  * the others.
1860  *
1861  * So in those cases, request a kvmclock update for all vcpus.
1862  * We need to rate-limit these requests though, as they can
1863  * considerably slow guests that have a large number of vcpus.
1864  * The time for a remote vcpu to update its kvmclock is bound
1865  * by the delay we use to rate-limit the updates.
1866  */
1867
1868 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1869
1870 static void kvmclock_update_fn(struct work_struct *work)
1871 {
1872         int i;
1873         struct delayed_work *dwork = to_delayed_work(work);
1874         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1875                                            kvmclock_update_work);
1876         struct kvm *kvm = container_of(ka, struct kvm, arch);
1877         struct kvm_vcpu *vcpu;
1878
1879         kvm_for_each_vcpu(i, vcpu, kvm) {
1880                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1881                 kvm_vcpu_kick(vcpu);
1882         }
1883 }
1884
1885 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1886 {
1887         struct kvm *kvm = v->kvm;
1888
1889         kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1890         schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1891                                         KVMCLOCK_UPDATE_DELAY);
1892 }
1893
1894 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1895
1896 static void kvmclock_sync_fn(struct work_struct *work)
1897 {
1898         struct delayed_work *dwork = to_delayed_work(work);
1899         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1900                                            kvmclock_sync_work);
1901         struct kvm *kvm = container_of(ka, struct kvm, arch);
1902
1903         if (!kvmclock_periodic_sync)
1904                 return;
1905
1906         schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1907         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1908                                         KVMCLOCK_SYNC_PERIOD);
1909 }
1910
1911 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1912 {
1913         u64 mcg_cap = vcpu->arch.mcg_cap;
1914         unsigned bank_num = mcg_cap & 0xff;
1915
1916         switch (msr) {
1917         case MSR_IA32_MCG_STATUS:
1918                 vcpu->arch.mcg_status = data;
1919                 break;
1920         case MSR_IA32_MCG_CTL:
1921                 if (!(mcg_cap & MCG_CTL_P))
1922                         return 1;
1923                 if (data != 0 && data != ~(u64)0)
1924                         return -1;
1925                 vcpu->arch.mcg_ctl = data;
1926                 break;
1927         default:
1928                 if (msr >= MSR_IA32_MC0_CTL &&
1929                     msr < MSR_IA32_MCx_CTL(bank_num)) {
1930                         u32 offset = msr - MSR_IA32_MC0_CTL;
1931                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1932                          * some Linux kernels though clear bit 10 in bank 4 to
1933                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1934                          * this to avoid an uncatched #GP in the guest
1935                          */
1936                         if ((offset & 0x3) == 0 &&
1937                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1938                                 return -1;
1939                         vcpu->arch.mce_banks[offset] = data;
1940                         break;
1941                 }
1942                 return 1;
1943         }
1944         return 0;
1945 }
1946
1947 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1948 {
1949         struct kvm *kvm = vcpu->kvm;
1950         int lm = is_long_mode(vcpu);
1951         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1952                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1953         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1954                 : kvm->arch.xen_hvm_config.blob_size_32;
1955         u32 page_num = data & ~PAGE_MASK;
1956         u64 page_addr = data & PAGE_MASK;
1957         u8 *page;
1958         int r;
1959
1960         r = -E2BIG;
1961         if (page_num >= blob_size)
1962                 goto out;
1963         r = -ENOMEM;
1964         page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1965         if (IS_ERR(page)) {
1966                 r = PTR_ERR(page);
1967                 goto out;
1968         }
1969         if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1970                 goto out_free;
1971         r = 0;
1972 out_free:
1973         kfree(page);
1974 out:
1975         return r;
1976 }
1977
1978 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1979 {
1980         gpa_t gpa = data & ~0x3f;
1981
1982         /* Bits 2:5 are reserved, Should be zero */
1983         if (data & 0x3c)
1984                 return 1;
1985
1986         vcpu->arch.apf.msr_val = data;
1987
1988         if (!(data & KVM_ASYNC_PF_ENABLED)) {
1989                 kvm_clear_async_pf_completion_queue(vcpu);
1990                 kvm_async_pf_hash_reset(vcpu);
1991                 return 0;
1992         }
1993
1994         if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1995                                         sizeof(u32)))
1996                 return 1;
1997
1998         vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1999         kvm_async_pf_wakeup_all(vcpu);
2000         return 0;
2001 }
2002
2003 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2004 {
2005         vcpu->arch.pv_time_enabled = false;
2006 }
2007
2008 static void record_steal_time(struct kvm_vcpu *vcpu)
2009 {
2010         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2011                 return;
2012
2013         if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2014                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2015                 return;
2016
2017         if (vcpu->arch.st.steal.version & 1)
2018                 vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2019
2020         vcpu->arch.st.steal.version += 1;
2021
2022         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2023                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2024
2025         smp_wmb();
2026
2027         vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2028                 vcpu->arch.st.last_steal;
2029         vcpu->arch.st.last_steal = current->sched_info.run_delay;
2030
2031         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2032                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2033
2034         smp_wmb();
2035
2036         vcpu->arch.st.steal.version += 1;
2037
2038         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2039                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2040 }
2041
2042 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2043 {
2044         bool pr = false;
2045         u32 msr = msr_info->index;
2046         u64 data = msr_info->data;
2047
2048         switch (msr) {
2049         case MSR_AMD64_NB_CFG:
2050         case MSR_IA32_UCODE_REV:
2051         case MSR_IA32_UCODE_WRITE:
2052         case MSR_VM_HSAVE_PA:
2053         case MSR_AMD64_PATCH_LOADER:
2054         case MSR_AMD64_BU_CFG2:
2055                 break;
2056
2057         case MSR_EFER:
2058                 return set_efer(vcpu, data);
2059         case MSR_K7_HWCR:
2060                 data &= ~(u64)0x40;     /* ignore flush filter disable */
2061                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
2062                 data &= ~(u64)0x8;      /* ignore TLB cache disable */
2063                 data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2064                 if (data != 0) {
2065                         vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2066                                     data);
2067                         return 1;
2068                 }
2069                 break;
2070         case MSR_FAM10H_MMIO_CONF_BASE:
2071                 if (data != 0) {
2072                         vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2073                                     "0x%llx\n", data);
2074                         return 1;
2075                 }
2076                 break;
2077         case MSR_IA32_DEBUGCTLMSR:
2078                 if (!data) {
2079                         /* We support the non-activated case already */
2080                         break;
2081                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2082                         /* Values other than LBR and BTF are vendor-specific,
2083                            thus reserved and should throw a #GP */
2084                         return 1;
2085                 }
2086                 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2087                             __func__, data);
2088                 break;
2089         case 0x200 ... 0x2ff:
2090                 return kvm_mtrr_set_msr(vcpu, msr, data);
2091         case MSR_IA32_APICBASE:
2092                 return kvm_set_apic_base(vcpu, msr_info);
2093         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2094                 return kvm_x2apic_msr_write(vcpu, msr, data);
2095         case MSR_IA32_TSCDEADLINE:
2096                 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2097                 break;
2098         case MSR_IA32_TSC_ADJUST:
2099                 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2100                         if (!msr_info->host_initiated) {
2101                                 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2102                                 adjust_tsc_offset_guest(vcpu, adj);
2103                         }
2104                         vcpu->arch.ia32_tsc_adjust_msr = data;
2105                 }
2106                 break;
2107         case MSR_IA32_MISC_ENABLE:
2108                 vcpu->arch.ia32_misc_enable_msr = data;
2109                 break;
2110         case MSR_IA32_SMBASE:
2111                 if (!msr_info->host_initiated)
2112                         return 1;
2113                 vcpu->arch.smbase = data;
2114                 break;
2115         case MSR_KVM_WALL_CLOCK_NEW:
2116         case MSR_KVM_WALL_CLOCK:
2117                 vcpu->kvm->arch.wall_clock = data;
2118                 kvm_write_wall_clock(vcpu->kvm, data);
2119                 break;
2120         case MSR_KVM_SYSTEM_TIME_NEW:
2121         case MSR_KVM_SYSTEM_TIME: {
2122                 u64 gpa_offset;
2123                 struct kvm_arch *ka = &vcpu->kvm->arch;
2124
2125                 kvmclock_reset(vcpu);
2126
2127                 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2128                         bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2129
2130                         if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2131                                 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2132                                         &vcpu->requests);
2133
2134                         ka->boot_vcpu_runs_old_kvmclock = tmp;
2135                 }
2136
2137                 vcpu->arch.time = data;
2138                 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2139
2140                 /* we verify if the enable bit is set... */
2141                 if (!(data & 1))
2142                         break;
2143
2144                 gpa_offset = data & ~(PAGE_MASK | 1);
2145
2146                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2147                      &vcpu->arch.pv_time, data & ~1ULL,
2148                      sizeof(struct pvclock_vcpu_time_info)))
2149                         vcpu->arch.pv_time_enabled = false;
2150                 else
2151                         vcpu->arch.pv_time_enabled = true;
2152
2153                 break;
2154         }
2155         case MSR_KVM_ASYNC_PF_EN:
2156                 if (kvm_pv_enable_async_pf(vcpu, data))
2157                         return 1;
2158                 break;
2159         case MSR_KVM_STEAL_TIME:
2160
2161                 if (unlikely(!sched_info_on()))
2162                         return 1;
2163
2164                 if (data & KVM_STEAL_RESERVED_MASK)
2165                         return 1;
2166
2167                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2168                                                 data & KVM_STEAL_VALID_BITS,
2169                                                 sizeof(struct kvm_steal_time)))
2170                         return 1;
2171
2172                 vcpu->arch.st.msr_val = data;
2173
2174                 if (!(data & KVM_MSR_ENABLED))
2175                         break;
2176
2177                 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2178
2179                 break;
2180         case MSR_KVM_PV_EOI_EN:
2181                 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2182                         return 1;
2183                 break;
2184
2185         case MSR_IA32_MCG_CTL:
2186         case MSR_IA32_MCG_STATUS:
2187         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2188                 return set_msr_mce(vcpu, msr, data);
2189
2190         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2191         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2192                 pr = true; /* fall through */
2193         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2194         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2195                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2196                         return kvm_pmu_set_msr(vcpu, msr_info);
2197
2198                 if (pr || data != 0)
2199                         vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2200                                     "0x%x data 0x%llx\n", msr, data);
2201                 break;
2202         case MSR_K7_CLK_CTL:
2203                 /*
2204                  * Ignore all writes to this no longer documented MSR.
2205                  * Writes are only relevant for old K7 processors,
2206                  * all pre-dating SVM, but a recommended workaround from
2207                  * AMD for these chips. It is possible to specify the
2208                  * affected processor models on the command line, hence
2209                  * the need to ignore the workaround.
2210                  */
2211                 break;
2212         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2213         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2214         case HV_X64_MSR_CRASH_CTL:
2215         case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2216                 return kvm_hv_set_msr_common(vcpu, msr, data,
2217                                              msr_info->host_initiated);
2218         case MSR_IA32_BBL_CR_CTL3:
2219                 /* Drop writes to this legacy MSR -- see rdmsr
2220                  * counterpart for further detail.
2221                  */
2222                 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2223                 break;
2224         case MSR_AMD64_OSVW_ID_LENGTH:
2225                 if (!guest_cpuid_has_osvw(vcpu))
2226                         return 1;
2227                 vcpu->arch.osvw.length = data;
2228                 break;
2229         case MSR_AMD64_OSVW_STATUS:
2230                 if (!guest_cpuid_has_osvw(vcpu))
2231                         return 1;
2232                 vcpu->arch.osvw.status = data;
2233                 break;
2234         default:
2235                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2236                         return xen_hvm_config(vcpu, data);
2237                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2238                         return kvm_pmu_set_msr(vcpu, msr_info);
2239                 if (!ignore_msrs) {
2240                         vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2241                                     msr, data);
2242                         return 1;
2243                 } else {
2244                         vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2245                                     msr, data);
2246                         break;
2247                 }
2248         }
2249         return 0;
2250 }
2251 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2252
2253
2254 /*
2255  * Reads an msr value (of 'msr_index') into 'pdata'.
2256  * Returns 0 on success, non-0 otherwise.
2257  * Assumes vcpu_load() was already called.
2258  */
2259 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2260 {
2261         return kvm_x86_ops->get_msr(vcpu, msr);
2262 }
2263 EXPORT_SYMBOL_GPL(kvm_get_msr);
2264
2265 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2266 {
2267         u64 data;
2268         u64 mcg_cap = vcpu->arch.mcg_cap;
2269         unsigned bank_num = mcg_cap & 0xff;
2270
2271         switch (msr) {
2272         case MSR_IA32_P5_MC_ADDR:
2273         case MSR_IA32_P5_MC_TYPE:
2274                 data = 0;
2275                 break;
2276         case MSR_IA32_MCG_CAP:
2277                 data = vcpu->arch.mcg_cap;
2278                 break;
2279         case MSR_IA32_MCG_CTL:
2280                 if (!(mcg_cap & MCG_CTL_P))
2281                         return 1;
2282                 data = vcpu->arch.mcg_ctl;
2283                 break;
2284         case MSR_IA32_MCG_STATUS:
2285                 data = vcpu->arch.mcg_status;
2286                 break;
2287         default:
2288                 if (msr >= MSR_IA32_MC0_CTL &&
2289                     msr < MSR_IA32_MCx_CTL(bank_num)) {
2290                         u32 offset = msr - MSR_IA32_MC0_CTL;
2291                         data = vcpu->arch.mce_banks[offset];
2292                         break;
2293                 }
2294                 return 1;
2295         }
2296         *pdata = data;
2297         return 0;
2298 }
2299
2300 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2301 {
2302         switch (msr_info->index) {
2303         case MSR_IA32_PLATFORM_ID:
2304         case MSR_IA32_EBL_CR_POWERON:
2305         case MSR_IA32_DEBUGCTLMSR:
2306         case MSR_IA32_LASTBRANCHFROMIP:
2307         case MSR_IA32_LASTBRANCHTOIP:
2308         case MSR_IA32_LASTINTFROMIP:
2309         case MSR_IA32_LASTINTTOIP:
2310         case MSR_K8_SYSCFG:
2311         case MSR_K8_TSEG_ADDR:
2312         case MSR_K8_TSEG_MASK:
2313         case MSR_K7_HWCR:
2314         case MSR_VM_HSAVE_PA:
2315         case MSR_K8_INT_PENDING_MSG:
2316         case MSR_AMD64_NB_CFG:
2317         case MSR_FAM10H_MMIO_CONF_BASE:
2318         case MSR_AMD64_BU_CFG2:
2319         case MSR_IA32_PERF_CTL:
2320                 msr_info->data = 0;
2321                 break;
2322         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2323         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2324         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2325         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2326                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2327                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2328                 msr_info->data = 0;
2329                 break;
2330         case MSR_IA32_UCODE_REV:
2331                 msr_info->data = 0x100000000ULL;
2332                 break;
2333         case MSR_MTRRcap:
2334         case 0x200 ... 0x2ff:
2335                 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2336         case 0xcd: /* fsb frequency */
2337                 msr_info->data = 3;
2338                 break;
2339                 /*
2340                  * MSR_EBC_FREQUENCY_ID
2341                  * Conservative value valid for even the basic CPU models.
2342                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2343                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2344                  * and 266MHz for model 3, or 4. Set Core Clock
2345                  * Frequency to System Bus Frequency Ratio to 1 (bits
2346                  * 31:24) even though these are only valid for CPU
2347                  * models > 2, however guests may end up dividing or
2348                  * multiplying by zero otherwise.
2349                  */
2350         case MSR_EBC_FREQUENCY_ID:
2351                 msr_info->data = 1 << 24;
2352                 break;
2353         case MSR_IA32_APICBASE:
2354                 msr_info->data = kvm_get_apic_base(vcpu);
2355                 break;
2356         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2357                 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2358                 break;
2359         case MSR_IA32_TSCDEADLINE:
2360                 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2361                 break;
2362         case MSR_IA32_TSC_ADJUST:
2363                 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2364                 break;
2365         case MSR_IA32_MISC_ENABLE:
2366                 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2367                 break;
2368         case MSR_IA32_SMBASE:
2369                 if (!msr_info->host_initiated)
2370                         return 1;
2371                 msr_info->data = vcpu->arch.smbase;
2372                 break;
2373         case MSR_IA32_PERF_STATUS:
2374                 /* TSC increment by tick */
2375                 msr_info->data = 1000ULL;
2376                 /* CPU multiplier */
2377                 msr_info->data |= (((uint64_t)4ULL) << 40);
2378                 break;
2379         case MSR_EFER:
2380                 msr_info->data = vcpu->arch.efer;
2381                 break;
2382         case MSR_KVM_WALL_CLOCK:
2383         case MSR_KVM_WALL_CLOCK_NEW:
2384                 msr_info->data = vcpu->kvm->arch.wall_clock;
2385                 break;
2386         case MSR_KVM_SYSTEM_TIME:
2387         case MSR_KVM_SYSTEM_TIME_NEW:
2388                 msr_info->data = vcpu->arch.time;
2389                 break;
2390         case MSR_KVM_ASYNC_PF_EN:
2391                 msr_info->data = vcpu->arch.apf.msr_val;
2392                 break;
2393         case MSR_KVM_STEAL_TIME:
2394                 msr_info->data = vcpu->arch.st.msr_val;
2395                 break;
2396         case MSR_KVM_PV_EOI_EN:
2397                 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2398                 break;
2399         case MSR_IA32_P5_MC_ADDR:
2400         case MSR_IA32_P5_MC_TYPE:
2401         case MSR_IA32_MCG_CAP:
2402         case MSR_IA32_MCG_CTL:
2403         case MSR_IA32_MCG_STATUS:
2404         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2405                 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2406         case MSR_K7_CLK_CTL:
2407                 /*
2408                  * Provide expected ramp-up count for K7. All other
2409                  * are set to zero, indicating minimum divisors for
2410                  * every field.
2411                  *
2412                  * This prevents guest kernels on AMD host with CPU
2413                  * type 6, model 8 and higher from exploding due to
2414                  * the rdmsr failing.
2415                  */
2416                 msr_info->data = 0x20000000;
2417                 break;
2418         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2419         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2420         case HV_X64_MSR_CRASH_CTL:
2421         case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2422                 return kvm_hv_get_msr_common(vcpu,
2423                                              msr_info->index, &msr_info->data);
2424                 break;
2425         case MSR_IA32_BBL_CR_CTL3:
2426                 /* This legacy MSR exists but isn't fully documented in current
2427                  * silicon.  It is however accessed by winxp in very narrow
2428                  * scenarios where it sets bit #19, itself documented as
2429                  * a "reserved" bit.  Best effort attempt to source coherent
2430                  * read data here should the balance of the register be
2431                  * interpreted by the guest:
2432                  *
2433                  * L2 cache control register 3: 64GB range, 256KB size,
2434                  * enabled, latency 0x1, configured
2435                  */
2436                 msr_info->data = 0xbe702111;
2437                 break;
2438         case MSR_AMD64_OSVW_ID_LENGTH:
2439                 if (!guest_cpuid_has_osvw(vcpu))
2440                         return 1;
2441                 msr_info->data = vcpu->arch.osvw.length;
2442                 break;
2443         case MSR_AMD64_OSVW_STATUS:
2444                 if (!guest_cpuid_has_osvw(vcpu))
2445                         return 1;
2446                 msr_info->data = vcpu->arch.osvw.status;
2447                 break;
2448         default:
2449                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2450                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2451                 if (!ignore_msrs) {
2452                         vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2453                         return 1;
2454                 } else {
2455                         vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2456                         msr_info->data = 0;
2457                 }
2458                 break;
2459         }
2460         return 0;
2461 }
2462 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2463
2464 /*
2465  * Read or write a bunch of msrs. All parameters are kernel addresses.
2466  *
2467  * @return number of msrs set successfully.
2468  */
2469 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2470                     struct kvm_msr_entry *entries,
2471                     int (*do_msr)(struct kvm_vcpu *vcpu,
2472                                   unsigned index, u64 *data))
2473 {
2474         int i, idx;
2475
2476         idx = srcu_read_lock(&vcpu->kvm->srcu);
2477         for (i = 0; i < msrs->nmsrs; ++i)
2478                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2479                         break;
2480         srcu_read_unlock(&vcpu->kvm->srcu, idx);
2481
2482         return i;
2483 }
2484
2485 /*
2486  * Read or write a bunch of msrs. Parameters are user addresses.
2487  *
2488  * @return number of msrs set successfully.
2489  */
2490 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2491                   int (*do_msr)(struct kvm_vcpu *vcpu,
2492                                 unsigned index, u64 *data),
2493                   int writeback)
2494 {
2495         struct kvm_msrs msrs;
2496         struct kvm_msr_entry *entries;
2497         int r, n;
2498         unsigned size;
2499
2500         r = -EFAULT;
2501         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2502                 goto out;
2503
2504         r = -E2BIG;
2505         if (msrs.nmsrs >= MAX_IO_MSRS)
2506                 goto out;
2507
2508         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2509         entries = memdup_user(user_msrs->entries, size);
2510         if (IS_ERR(entries)) {
2511                 r = PTR_ERR(entries);
2512                 goto out;
2513         }
2514
2515         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2516         if (r < 0)
2517                 goto out_free;
2518
2519         r = -EFAULT;
2520         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2521                 goto out_free;
2522
2523         r = n;
2524
2525 out_free:
2526         kfree(entries);
2527 out:
2528         return r;
2529 }
2530
2531 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2532 {
2533         int r;
2534
2535         switch (ext) {
2536         case KVM_CAP_IRQCHIP:
2537         case KVM_CAP_HLT:
2538         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2539         case KVM_CAP_SET_TSS_ADDR:
2540         case KVM_CAP_EXT_CPUID:
2541         case KVM_CAP_EXT_EMUL_CPUID:
2542         case KVM_CAP_CLOCKSOURCE:
2543         case KVM_CAP_PIT:
2544         case KVM_CAP_NOP_IO_DELAY:
2545         case KVM_CAP_MP_STATE:
2546         case KVM_CAP_SYNC_MMU:
2547         case KVM_CAP_USER_NMI:
2548         case KVM_CAP_REINJECT_CONTROL:
2549         case KVM_CAP_IRQ_INJECT_STATUS:
2550         case KVM_CAP_IOEVENTFD:
2551         case KVM_CAP_IOEVENTFD_NO_LENGTH:
2552         case KVM_CAP_PIT2:
2553         case KVM_CAP_PIT_STATE2:
2554         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2555         case KVM_CAP_XEN_HVM:
2556         case KVM_CAP_ADJUST_CLOCK:
2557         case KVM_CAP_VCPU_EVENTS:
2558         case KVM_CAP_HYPERV:
2559         case KVM_CAP_HYPERV_VAPIC:
2560         case KVM_CAP_HYPERV_SPIN:
2561         case KVM_CAP_HYPERV_SYNIC:
2562         case KVM_CAP_PCI_SEGMENT:
2563         case KVM_CAP_DEBUGREGS:
2564         case KVM_CAP_X86_ROBUST_SINGLESTEP:
2565         case KVM_CAP_XSAVE:
2566         case KVM_CAP_ASYNC_PF:
2567         case KVM_CAP_GET_TSC_KHZ:
2568         case KVM_CAP_KVMCLOCK_CTRL:
2569         case KVM_CAP_READONLY_MEM:
2570         case KVM_CAP_HYPERV_TIME:
2571         case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2572         case KVM_CAP_TSC_DEADLINE_TIMER:
2573         case KVM_CAP_ENABLE_CAP_VM:
2574         case KVM_CAP_DISABLE_QUIRKS:
2575         case KVM_CAP_SET_BOOT_CPU_ID:
2576         case KVM_CAP_SPLIT_IRQCHIP:
2577 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2578         case KVM_CAP_ASSIGN_DEV_IRQ:
2579         case KVM_CAP_PCI_2_3:
2580 #endif
2581                 r = 1;
2582                 break;
2583         case KVM_CAP_X86_SMM:
2584                 /* SMBASE is usually relocated above 1M on modern chipsets,
2585                  * and SMM handlers might indeed rely on 4G segment limits,
2586                  * so do not report SMM to be available if real mode is
2587                  * emulated via vm86 mode.  Still, do not go to great lengths
2588                  * to avoid userspace's usage of the feature, because it is a
2589                  * fringe case that is not enabled except via specific settings
2590                  * of the module parameters.
2591                  */
2592                 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2593                 break;
2594         case KVM_CAP_COALESCED_MMIO:
2595                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2596                 break;
2597         case KVM_CAP_VAPIC:
2598                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2599                 break;
2600         case KVM_CAP_NR_VCPUS:
2601                 r = KVM_SOFT_MAX_VCPUS;
2602                 break;
2603         case KVM_CAP_MAX_VCPUS:
2604                 r = KVM_MAX_VCPUS;
2605                 break;
2606         case KVM_CAP_NR_MEMSLOTS:
2607                 r = KVM_USER_MEM_SLOTS;
2608                 break;
2609         case KVM_CAP_PV_MMU:    /* obsolete */
2610                 r = 0;
2611                 break;
2612 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2613         case KVM_CAP_IOMMU:
2614                 r = iommu_present(&pci_bus_type);
2615                 break;
2616 #endif
2617         case KVM_CAP_MCE:
2618                 r = KVM_MAX_MCE_BANKS;
2619                 break;
2620         case KVM_CAP_XCRS:
2621                 r = boot_cpu_has(X86_FEATURE_XSAVE);
2622                 break;
2623         case KVM_CAP_TSC_CONTROL:
2624                 r = kvm_has_tsc_control;
2625                 break;
2626         case KVM_CAP_X2APIC_API:
2627                 r = KVM_X2APIC_API_VALID_FLAGS;
2628                 break;
2629         default:
2630                 r = 0;
2631                 break;
2632         }
2633         return r;
2634
2635 }
2636
2637 long kvm_arch_dev_ioctl(struct file *filp,
2638                         unsigned int ioctl, unsigned long arg)
2639 {
2640         void __user *argp = (void __user *)arg;
2641         long r;
2642
2643         switch (ioctl) {
2644         case KVM_GET_MSR_INDEX_LIST: {
2645                 struct kvm_msr_list __user *user_msr_list = argp;
2646                 struct kvm_msr_list msr_list;
2647                 unsigned n;
2648
2649                 r = -EFAULT;
2650                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2651                         goto out;
2652                 n = msr_list.nmsrs;
2653                 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2654                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2655                         goto out;
2656                 r = -E2BIG;
2657                 if (n < msr_list.nmsrs)
2658                         goto out;
2659                 r = -EFAULT;
2660                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2661                                  num_msrs_to_save * sizeof(u32)))
2662                         goto out;
2663                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2664                                  &emulated_msrs,
2665                                  num_emulated_msrs * sizeof(u32)))
2666                         goto out;
2667                 r = 0;
2668                 break;
2669         }
2670         case KVM_GET_SUPPORTED_CPUID:
2671         case KVM_GET_EMULATED_CPUID: {
2672                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2673                 struct kvm_cpuid2 cpuid;
2674
2675                 r = -EFAULT;
2676                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2677                         goto out;
2678
2679                 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2680                                             ioctl);
2681                 if (r)
2682                         goto out;
2683
2684                 r = -EFAULT;
2685                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2686                         goto out;
2687                 r = 0;
2688                 break;
2689         }
2690         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2691                 r = -EFAULT;
2692                 if (copy_to_user(argp, &kvm_mce_cap_supported,
2693                                  sizeof(kvm_mce_cap_supported)))
2694                         goto out;
2695                 r = 0;
2696                 break;
2697         }
2698         default:
2699                 r = -EINVAL;
2700         }
2701 out:
2702         return r;
2703 }
2704
2705 static void wbinvd_ipi(void *garbage)
2706 {
2707         wbinvd();
2708 }
2709
2710 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2711 {
2712         return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2713 }
2714
2715 static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu)
2716 {
2717         set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
2718 }
2719
2720 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2721 {
2722         /* Address WBINVD may be executed by guest */
2723         if (need_emulate_wbinvd(vcpu)) {
2724                 if (kvm_x86_ops->has_wbinvd_exit())
2725                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2726                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2727                         smp_call_function_single(vcpu->cpu,
2728                                         wbinvd_ipi, NULL, 1);
2729         }
2730
2731         kvm_x86_ops->vcpu_load(vcpu, cpu);
2732
2733         /* Apply any externally detected TSC adjustments (due to suspend) */
2734         if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2735                 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2736                 vcpu->arch.tsc_offset_adjustment = 0;
2737                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2738         }
2739
2740         if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2741                 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2742                                 rdtsc() - vcpu->arch.last_host_tsc;
2743                 if (tsc_delta < 0)
2744                         mark_tsc_unstable("KVM discovered backwards TSC");
2745
2746                 if (check_tsc_unstable()) {
2747                         u64 offset = kvm_compute_tsc_offset(vcpu,
2748                                                 vcpu->arch.last_guest_tsc);
2749                         kvm_x86_ops->write_tsc_offset(vcpu, offset);
2750                         vcpu->arch.tsc_catchup = 1;
2751                 }
2752                 if (kvm_lapic_hv_timer_in_use(vcpu) &&
2753                                 kvm_x86_ops->set_hv_timer(vcpu,
2754                                         kvm_get_lapic_tscdeadline_msr(vcpu)))
2755                         kvm_lapic_switch_to_sw_timer(vcpu);
2756                 /*
2757                  * On a host with synchronized TSC, there is no need to update
2758                  * kvmclock on vcpu->cpu migration
2759                  */
2760                 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2761                         kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2762                 if (vcpu->cpu != cpu)
2763                         kvm_migrate_timers(vcpu);
2764                 vcpu->cpu = cpu;
2765         }
2766
2767         kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2768 }
2769
2770 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2771 {
2772         kvm_x86_ops->vcpu_put(vcpu);
2773         kvm_put_guest_fpu(vcpu);
2774         vcpu->arch.last_host_tsc = rdtsc();
2775 }
2776
2777 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2778                                     struct kvm_lapic_state *s)
2779 {
2780         if (vcpu->arch.apicv_active)
2781                 kvm_x86_ops->sync_pir_to_irr(vcpu);
2782
2783         return kvm_apic_get_state(vcpu, s);
2784 }
2785
2786 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2787                                     struct kvm_lapic_state *s)
2788 {
2789         int r;
2790
2791         r = kvm_apic_set_state(vcpu, s);
2792         if (r)
2793                 return r;
2794         update_cr8_intercept(vcpu);
2795
2796         return 0;
2797 }
2798
2799 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2800 {
2801         return (!lapic_in_kernel(vcpu) ||
2802                 kvm_apic_accept_pic_intr(vcpu));
2803 }
2804
2805 /*
2806  * if userspace requested an interrupt window, check that the
2807  * interrupt window is open.
2808  *
2809  * No need to exit to userspace if we already have an interrupt queued.
2810  */
2811 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2812 {
2813         return kvm_arch_interrupt_allowed(vcpu) &&
2814                 !kvm_cpu_has_interrupt(vcpu) &&
2815                 !kvm_event_needs_reinjection(vcpu) &&
2816                 kvm_cpu_accept_dm_intr(vcpu);
2817 }
2818
2819 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2820                                     struct kvm_interrupt *irq)
2821 {
2822         if (irq->irq >= KVM_NR_INTERRUPTS)
2823                 return -EINVAL;
2824
2825         if (!irqchip_in_kernel(vcpu->kvm)) {
2826                 kvm_queue_interrupt(vcpu, irq->irq, false);
2827                 kvm_make_request(KVM_REQ_EVENT, vcpu);
2828                 return 0;
2829         }
2830
2831         /*
2832          * With in-kernel LAPIC, we only use this to inject EXTINT, so
2833          * fail for in-kernel 8259.
2834          */
2835         if (pic_in_kernel(vcpu->kvm))
2836                 return -ENXIO;
2837
2838         if (vcpu->arch.pending_external_vector != -1)
2839                 return -EEXIST;
2840
2841         vcpu->arch.pending_external_vector = irq->irq;
2842         kvm_make_request(KVM_REQ_EVENT, vcpu);
2843         return 0;
2844 }
2845
2846 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2847 {
2848         kvm_inject_nmi(vcpu);
2849
2850         return 0;
2851 }
2852
2853 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2854 {
2855         kvm_make_request(KVM_REQ_SMI, vcpu);
2856
2857         return 0;
2858 }
2859
2860 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2861                                            struct kvm_tpr_access_ctl *tac)
2862 {
2863         if (tac->flags)
2864                 return -EINVAL;
2865         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2866         return 0;
2867 }
2868
2869 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2870                                         u64 mcg_cap)
2871 {
2872         int r;
2873         unsigned bank_num = mcg_cap & 0xff, bank;
2874
2875         r = -EINVAL;
2876         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2877                 goto out;
2878         if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
2879                 goto out;
2880         r = 0;
2881         vcpu->arch.mcg_cap = mcg_cap;
2882         /* Init IA32_MCG_CTL to all 1s */
2883         if (mcg_cap & MCG_CTL_P)
2884                 vcpu->arch.mcg_ctl = ~(u64)0;
2885         /* Init IA32_MCi_CTL to all 1s */
2886         for (bank = 0; bank < bank_num; bank++)
2887                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2888
2889         if (kvm_x86_ops->setup_mce)
2890                 kvm_x86_ops->setup_mce(vcpu);
2891 out:
2892         return r;
2893 }
2894
2895 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2896                                       struct kvm_x86_mce *mce)
2897 {
2898         u64 mcg_cap = vcpu->arch.mcg_cap;
2899         unsigned bank_num = mcg_cap & 0xff;
2900         u64 *banks = vcpu->arch.mce_banks;
2901
2902         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2903                 return -EINVAL;
2904         /*
2905          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2906          * reporting is disabled
2907          */
2908         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2909             vcpu->arch.mcg_ctl != ~(u64)0)
2910                 return 0;
2911         banks += 4 * mce->bank;
2912         /*
2913          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2914          * reporting is disabled for the bank
2915          */
2916         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2917                 return 0;
2918         if (mce->status & MCI_STATUS_UC) {
2919                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2920                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2921                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2922                         return 0;
2923                 }
2924                 if (banks[1] & MCI_STATUS_VAL)
2925                         mce->status |= MCI_STATUS_OVER;
2926                 banks[2] = mce->addr;
2927                 banks[3] = mce->misc;
2928                 vcpu->arch.mcg_status = mce->mcg_status;
2929                 banks[1] = mce->status;
2930                 kvm_queue_exception(vcpu, MC_VECTOR);
2931         } else if (!(banks[1] & MCI_STATUS_VAL)
2932                    || !(banks[1] & MCI_STATUS_UC)) {
2933                 if (banks[1] & MCI_STATUS_VAL)
2934                         mce->status |= MCI_STATUS_OVER;
2935                 banks[2] = mce->addr;
2936                 banks[3] = mce->misc;
2937                 banks[1] = mce->status;
2938         } else
2939                 banks[1] |= MCI_STATUS_OVER;
2940         return 0;
2941 }
2942
2943 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2944                                                struct kvm_vcpu_events *events)
2945 {
2946         process_nmi(vcpu);
2947         events->exception.injected =
2948                 vcpu->arch.exception.pending &&
2949                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2950         events->exception.nr = vcpu->arch.exception.nr;
2951         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2952         events->exception.pad = 0;
2953         events->exception.error_code = vcpu->arch.exception.error_code;
2954
2955         events->interrupt.injected =
2956                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2957         events->interrupt.nr = vcpu->arch.interrupt.nr;
2958         events->interrupt.soft = 0;
2959         events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2960
2961         events->nmi.injected = vcpu->arch.nmi_injected;
2962         events->nmi.pending = vcpu->arch.nmi_pending != 0;
2963         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2964         events->nmi.pad = 0;
2965
2966         events->sipi_vector = 0; /* never valid when reporting to user space */
2967
2968         events->smi.smm = is_smm(vcpu);
2969         events->smi.pending = vcpu->arch.smi_pending;
2970         events->smi.smm_inside_nmi =
2971                 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2972         events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2973
2974         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2975                          | KVM_VCPUEVENT_VALID_SHADOW
2976                          | KVM_VCPUEVENT_VALID_SMM);
2977         memset(&events->reserved, 0, sizeof(events->reserved));
2978 }
2979
2980 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2981                                               struct kvm_vcpu_events *events)
2982 {
2983         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2984                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2985                               | KVM_VCPUEVENT_VALID_SHADOW
2986                               | KVM_VCPUEVENT_VALID_SMM))
2987                 return -EINVAL;
2988
2989         if (events->exception.injected &&
2990             (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
2991                 return -EINVAL;
2992
2993         process_nmi(vcpu);
2994         vcpu->arch.exception.pending = events->exception.injected;
2995         vcpu->arch.exception.nr = events->exception.nr;
2996         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2997         vcpu->arch.exception.error_code = events->exception.error_code;
2998
2999         vcpu->arch.interrupt.pending = events->interrupt.injected;
3000         vcpu->arch.interrupt.nr = events->interrupt.nr;
3001         vcpu->arch.interrupt.soft = events->interrupt.soft;
3002         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3003                 kvm_x86_ops->set_interrupt_shadow(vcpu,
3004                                                   events->interrupt.shadow);
3005
3006         vcpu->arch.nmi_injected = events->nmi.injected;
3007         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3008                 vcpu->arch.nmi_pending = events->nmi.pending;
3009         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3010
3011         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3012             lapic_in_kernel(vcpu))
3013                 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3014
3015         if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3016                 if (events->smi.smm)
3017                         vcpu->arch.hflags |= HF_SMM_MASK;
3018                 else
3019                         vcpu->arch.hflags &= ~HF_SMM_MASK;
3020                 vcpu->arch.smi_pending = events->smi.pending;
3021                 if (events->smi.smm_inside_nmi)
3022                         vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3023                 else
3024                         vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3025                 if (lapic_in_kernel(vcpu)) {
3026                         if (events->smi.latched_init)
3027                                 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3028                         else
3029                                 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3030                 }
3031         }
3032
3033         kvm_make_request(KVM_REQ_EVENT, vcpu);
3034
3035         return 0;
3036 }
3037
3038 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3039                                              struct kvm_debugregs *dbgregs)
3040 {
3041         unsigned long val;
3042
3043         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3044         kvm_get_dr(vcpu, 6, &val);
3045         dbgregs->dr6 = val;
3046         dbgregs->dr7 = vcpu->arch.dr7;
3047         dbgregs->flags = 0;
3048         memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3049 }
3050
3051 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3052                                             struct kvm_debugregs *dbgregs)
3053 {
3054         if (dbgregs->flags)
3055                 return -EINVAL;
3056
3057         if (dbgregs->dr6 & ~0xffffffffull)
3058                 return -EINVAL;
3059         if (dbgregs->dr7 & ~0xffffffffull)
3060                 return -EINVAL;
3061
3062         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3063         kvm_update_dr0123(vcpu);
3064         vcpu->arch.dr6 = dbgregs->dr6;
3065         kvm_update_dr6(vcpu);
3066         vcpu->arch.dr7 = dbgregs->dr7;
3067         kvm_update_dr7(vcpu);
3068
3069         return 0;
3070 }
3071
3072 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3073
3074 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3075 {
3076         struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3077         u64 xstate_bv = xsave->header.xfeatures;
3078         u64 valid;
3079
3080         /*
3081          * Copy legacy XSAVE area, to avoid complications with CPUID
3082          * leaves 0 and 1 in the loop below.
3083          */
3084         memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3085
3086         /* Set XSTATE_BV */
3087         *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3088
3089         /*
3090          * Copy each region from the possibly compacted offset to the
3091          * non-compacted offset.
3092          */
3093         valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3094         while (valid) {
3095                 u64 feature = valid & -valid;
3096                 int index = fls64(feature) - 1;
3097                 void *src = get_xsave_addr(xsave, feature);
3098
3099                 if (src) {
3100                         u32 size, offset, ecx, edx;
3101                         cpuid_count(XSTATE_CPUID, index,
3102                                     &size, &offset, &ecx, &edx);
3103                         memcpy(dest + offset, src, size);
3104                 }
3105
3106                 valid -= feature;
3107         }
3108 }
3109
3110 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3111 {
3112         struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3113         u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3114         u64 valid;
3115
3116         /*
3117          * Copy legacy XSAVE area, to avoid complications with CPUID
3118          * leaves 0 and 1 in the loop below.
3119          */
3120         memcpy(xsave, src, XSAVE_HDR_OFFSET);
3121
3122         /* Set XSTATE_BV and possibly XCOMP_BV.  */
3123         xsave->header.xfeatures = xstate_bv;
3124         if (boot_cpu_has(X86_FEATURE_XSAVES))
3125                 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3126
3127         /*
3128          * Copy each region from the non-compacted offset to the
3129          * possibly compacted offset.
3130          */
3131         valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3132         while (valid) {
3133                 u64 feature = valid & -valid;
3134                 int index = fls64(feature) - 1;
3135                 void *dest = get_xsave_addr(xsave, feature);
3136
3137                 if (dest) {
3138                         u32 size, offset, ecx, edx;
3139                         cpuid_count(XSTATE_CPUID, index,
3140                                     &size, &offset, &ecx, &edx);
3141                         memcpy(dest, src + offset, size);
3142                 }
3143
3144                 valid -= feature;
3145         }
3146 }
3147
3148 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3149                                          struct kvm_xsave *guest_xsave)
3150 {
3151         if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3152                 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3153                 fill_xsave((u8 *) guest_xsave->region, vcpu);
3154         } else {
3155                 memcpy(guest_xsave->region,
3156                         &vcpu->arch.guest_fpu.state.fxsave,
3157                         sizeof(struct fxregs_state));
3158                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3159                         XFEATURE_MASK_FPSSE;
3160         }
3161 }
3162
3163 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3164                                         struct kvm_xsave *guest_xsave)
3165 {
3166         u64 xstate_bv =
3167                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3168
3169         if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3170                 /*
3171                  * Here we allow setting states that are not present in
3172                  * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3173                  * with old userspace.
3174                  */
3175                 if (xstate_bv & ~kvm_supported_xcr0())
3176                         return -EINVAL;
3177                 load_xsave(vcpu, (u8 *)guest_xsave->region);
3178         } else {
3179                 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3180                         return -EINVAL;
3181                 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3182                         guest_xsave->region, sizeof(struct fxregs_state));
3183         }
3184         return 0;
3185 }
3186
3187 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3188                                         struct kvm_xcrs *guest_xcrs)
3189 {
3190         if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3191                 guest_xcrs->nr_xcrs = 0;
3192                 return;
3193         }
3194
3195         guest_xcrs->nr_xcrs = 1;
3196         guest_xcrs->flags = 0;
3197         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3198         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3199 }
3200
3201 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3202                                        struct kvm_xcrs *guest_xcrs)
3203 {
3204         int i, r = 0;
3205
3206         if (!boot_cpu_has(X86_FEATURE_XSAVE))
3207                 return -EINVAL;
3208
3209         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3210                 return -EINVAL;
3211
3212         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3213                 /* Only support XCR0 currently */
3214                 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3215                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3216                                 guest_xcrs->xcrs[i].value);
3217                         break;
3218                 }
3219         if (r)
3220                 r = -EINVAL;
3221         return r;
3222 }
3223
3224 /*
3225  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3226  * stopped by the hypervisor.  This function will be called from the host only.
3227  * EINVAL is returned when the host attempts to set the flag for a guest that
3228  * does not support pv clocks.
3229  */
3230 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3231 {
3232         if (!vcpu->arch.pv_time_enabled)
3233                 return -EINVAL;
3234         vcpu->arch.pvclock_set_guest_stopped_request = true;
3235         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3236         return 0;
3237 }
3238
3239 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3240                                      struct kvm_enable_cap *cap)
3241 {
3242         if (cap->flags)
3243                 return -EINVAL;
3244
3245         switch (cap->cap) {
3246         case KVM_CAP_HYPERV_SYNIC:
3247                 return kvm_hv_activate_synic(vcpu);
3248         default:
3249                 return -EINVAL;
3250         }
3251 }
3252
3253 long kvm_arch_vcpu_ioctl(struct file *filp,
3254                          unsigned int ioctl, unsigned long arg)
3255 {
3256         struct kvm_vcpu *vcpu = filp->private_data;
3257         void __user *argp = (void __user *)arg;
3258         int r;
3259         union {
3260                 struct kvm_lapic_state *lapic;
3261                 struct kvm_xsave *xsave;
3262                 struct kvm_xcrs *xcrs;
3263                 void *buffer;
3264         } u;
3265
3266         u.buffer = NULL;
3267         switch (ioctl) {
3268         case KVM_GET_LAPIC: {
3269                 r = -EINVAL;
3270                 if (!lapic_in_kernel(vcpu))
3271                         goto out;
3272                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3273
3274                 r = -ENOMEM;
3275                 if (!u.lapic)
3276                         goto out;
3277                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3278                 if (r)
3279                         goto out;
3280                 r = -EFAULT;
3281                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3282                         goto out;
3283                 r = 0;
3284                 break;
3285         }
3286         case KVM_SET_LAPIC: {
3287                 r = -EINVAL;
3288                 if (!lapic_in_kernel(vcpu))
3289                         goto out;
3290                 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3291                 if (IS_ERR(u.lapic))
3292                         return PTR_ERR(u.lapic);
3293
3294                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3295                 break;
3296         }
3297         case KVM_INTERRUPT: {
3298                 struct kvm_interrupt irq;
3299
3300                 r = -EFAULT;
3301                 if (copy_from_user(&irq, argp, sizeof irq))
3302                         goto out;
3303                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3304                 break;
3305         }
3306         case KVM_NMI: {
3307                 r = kvm_vcpu_ioctl_nmi(vcpu);
3308                 break;
3309         }
3310         case KVM_SMI: {
3311                 r = kvm_vcpu_ioctl_smi(vcpu);
3312                 break;
3313         }
3314         case KVM_SET_CPUID: {
3315                 struct kvm_cpuid __user *cpuid_arg = argp;
3316                 struct kvm_cpuid cpuid;
3317
3318                 r = -EFAULT;
3319                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3320                         goto out;
3321                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3322                 break;
3323         }
3324         case KVM_SET_CPUID2: {
3325                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3326                 struct kvm_cpuid2 cpuid;
3327
3328                 r = -EFAULT;
3329                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3330                         goto out;
3331                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3332                                               cpuid_arg->entries);
3333                 break;
3334         }
3335         case KVM_GET_CPUID2: {
3336                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3337                 struct kvm_cpuid2 cpuid;
3338
3339                 r = -EFAULT;
3340                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3341                         goto out;
3342                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3343                                               cpuid_arg->entries);
3344                 if (r)
3345                         goto out;
3346                 r = -EFAULT;
3347                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3348                         goto out;
3349                 r = 0;
3350                 break;
3351         }
3352         case KVM_GET_MSRS:
3353                 r = msr_io(vcpu, argp, do_get_msr, 1);
3354                 break;
3355         case KVM_SET_MSRS:
3356                 r = msr_io(vcpu, argp, do_set_msr, 0);
3357                 break;
3358         case KVM_TPR_ACCESS_REPORTING: {
3359                 struct kvm_tpr_access_ctl tac;
3360
3361                 r = -EFAULT;
3362                 if (copy_from_user(&tac, argp, sizeof tac))
3363                         goto out;
3364                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3365                 if (r)
3366                         goto out;
3367                 r = -EFAULT;
3368                 if (copy_to_user(argp, &tac, sizeof tac))
3369                         goto out;
3370                 r = 0;
3371                 break;
3372         };
3373         case KVM_SET_VAPIC_ADDR: {
3374                 struct kvm_vapic_addr va;
3375
3376                 r = -EINVAL;
3377                 if (!lapic_in_kernel(vcpu))
3378                         goto out;
3379                 r = -EFAULT;
3380                 if (copy_from_user(&va, argp, sizeof va))
3381                         goto out;
3382                 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3383                 break;
3384         }
3385         case KVM_X86_SETUP_MCE: {
3386                 u64 mcg_cap;
3387
3388                 r = -EFAULT;
3389                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3390                         goto out;
3391                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3392                 break;
3393         }
3394         case KVM_X86_SET_MCE: {
3395                 struct kvm_x86_mce mce;
3396
3397                 r = -EFAULT;
3398                 if (copy_from_user(&mce, argp, sizeof mce))
3399                         goto out;
3400                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3401                 break;
3402         }
3403         case KVM_GET_VCPU_EVENTS: {
3404                 struct kvm_vcpu_events events;
3405
3406                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3407
3408                 r = -EFAULT;
3409                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3410                         break;
3411                 r = 0;
3412                 break;
3413         }
3414         case KVM_SET_VCPU_EVENTS: {
3415                 struct kvm_vcpu_events events;
3416
3417                 r = -EFAULT;
3418                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3419                         break;
3420
3421                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3422                 break;
3423         }
3424         case KVM_GET_DEBUGREGS: {
3425                 struct kvm_debugregs dbgregs;
3426
3427                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3428
3429                 r = -EFAULT;
3430                 if (copy_to_user(argp, &dbgregs,
3431                                  sizeof(struct kvm_debugregs)))
3432                         break;
3433                 r = 0;
3434                 break;
3435         }
3436         case KVM_SET_DEBUGREGS: {
3437                 struct kvm_debugregs dbgregs;
3438
3439                 r = -EFAULT;
3440                 if (copy_from_user(&dbgregs, argp,
3441                                    sizeof(struct kvm_debugregs)))
3442                         break;
3443
3444                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3445                 break;
3446         }
3447         case KVM_GET_XSAVE: {
3448                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3449                 r = -ENOMEM;
3450                 if (!u.xsave)
3451                         break;
3452
3453                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3454
3455                 r = -EFAULT;
3456                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3457                         break;
3458                 r = 0;
3459                 break;
3460         }
3461         case KVM_SET_XSAVE: {
3462                 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3463                 if (IS_ERR(u.xsave))
3464                         return PTR_ERR(u.xsave);
3465
3466                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3467                 break;
3468         }
3469         case KVM_GET_XCRS: {
3470                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3471                 r = -ENOMEM;
3472                 if (!u.xcrs)
3473                         break;
3474
3475                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3476
3477                 r = -EFAULT;
3478                 if (copy_to_user(argp, u.xcrs,
3479                                  sizeof(struct kvm_xcrs)))
3480                         break;
3481                 r = 0;
3482                 break;
3483         }
3484         case KVM_SET_XCRS: {
3485                 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3486                 if (IS_ERR(u.xcrs))
3487                         return PTR_ERR(u.xcrs);
3488
3489                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3490                 break;
3491         }
3492         case KVM_SET_TSC_KHZ: {
3493                 u32 user_tsc_khz;
3494
3495                 r = -EINVAL;
3496                 user_tsc_khz = (u32)arg;
3497
3498                 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3499                         goto out;
3500
3501                 if (user_tsc_khz == 0)
3502                         user_tsc_khz = tsc_khz;
3503
3504                 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3505                         r = 0;
3506
3507                 goto out;
3508         }
3509         case KVM_GET_TSC_KHZ: {
3510                 r = vcpu->arch.virtual_tsc_khz;
3511                 goto out;
3512         }
3513         case KVM_KVMCLOCK_CTRL: {
3514                 r = kvm_set_guest_paused(vcpu);
3515                 goto out;
3516         }
3517         case KVM_ENABLE_CAP: {
3518                 struct kvm_enable_cap cap;
3519
3520                 r = -EFAULT;
3521                 if (copy_from_user(&cap, argp, sizeof(cap)))
3522                         goto out;
3523                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3524                 break;
3525         }
3526         default:
3527                 r = -EINVAL;
3528         }
3529 out:
3530         kfree(u.buffer);
3531         return r;
3532 }
3533
3534 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3535 {
3536         return VM_FAULT_SIGBUS;
3537 }
3538
3539 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3540 {
3541         int ret;
3542
3543         if (addr > (unsigned int)(-3 * PAGE_SIZE))
3544                 return -EINVAL;
3545         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3546         return ret;
3547 }
3548
3549 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3550                                               u64 ident_addr)
3551 {
3552         kvm->arch.ept_identity_map_addr = ident_addr;
3553         return 0;
3554 }
3555
3556 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3557                                           u32 kvm_nr_mmu_pages)
3558 {
3559         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3560                 return -EINVAL;
3561
3562         mutex_lock(&kvm->slots_lock);
3563
3564         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3565         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3566
3567         mutex_unlock(&kvm->slots_lock);
3568         return 0;
3569 }
3570
3571 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3572 {
3573         return kvm->arch.n_max_mmu_pages;
3574 }
3575
3576 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3577 {
3578         int r;
3579
3580         r = 0;
3581         switch (chip->chip_id) {
3582         case KVM_IRQCHIP_PIC_MASTER:
3583                 memcpy(&chip->chip.pic,
3584                         &pic_irqchip(kvm)->pics[0],
3585                         sizeof(struct kvm_pic_state));
3586                 break;
3587         case KVM_IRQCHIP_PIC_SLAVE:
3588                 memcpy(&chip->chip.pic,
3589                         &pic_irqchip(kvm)->pics[1],
3590                         sizeof(struct kvm_pic_state));
3591                 break;
3592         case KVM_IRQCHIP_IOAPIC:
3593                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3594                 break;
3595         default:
3596                 r = -EINVAL;
3597                 break;
3598         }
3599         return r;
3600 }
3601
3602 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3603 {
3604         int r;
3605
3606         r = 0;
3607         switch (chip->chip_id) {
3608         case KVM_IRQCHIP_PIC_MASTER:
3609                 spin_lock(&pic_irqchip(kvm)->lock);
3610                 memcpy(&pic_irqchip(kvm)->pics[0],
3611                         &chip->chip.pic,
3612                         sizeof(struct kvm_pic_state));
3613                 spin_unlock(&pic_irqchip(kvm)->lock);
3614                 break;
3615         case KVM_IRQCHIP_PIC_SLAVE:
3616                 spin_lock(&pic_irqchip(kvm)->lock);
3617                 memcpy(&pic_irqchip(kvm)->pics[1],
3618                         &chip->chip.pic,
3619                         sizeof(struct kvm_pic_state));
3620                 spin_unlock(&pic_irqchip(kvm)->lock);
3621                 break;
3622         case KVM_IRQCHIP_IOAPIC:
3623                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3624                 break;
3625         default:
3626                 r = -EINVAL;
3627                 break;
3628         }
3629         kvm_pic_update_irq(pic_irqchip(kvm));
3630         return r;
3631 }
3632
3633 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3634 {
3635         struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3636
3637         BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3638
3639         mutex_lock(&kps->lock);
3640         memcpy(ps, &kps->channels, sizeof(*ps));
3641         mutex_unlock(&kps->lock);
3642         return 0;
3643 }
3644
3645 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3646 {
3647         int i;
3648         struct kvm_pit *pit = kvm->arch.vpit;
3649
3650         mutex_lock(&pit->pit_state.lock);
3651         memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3652         for (i = 0; i < 3; i++)
3653                 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3654         mutex_unlock(&pit->pit_state.lock);
3655         return 0;
3656 }
3657
3658 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3659 {
3660         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3661         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3662                 sizeof(ps->channels));
3663         ps->flags = kvm->arch.vpit->pit_state.flags;
3664         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3665         memset(&ps->reserved, 0, sizeof(ps->reserved));
3666         return 0;
3667 }
3668
3669 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3670 {
3671         int start = 0;
3672         int i;
3673         u32 prev_legacy, cur_legacy;
3674         struct kvm_pit *pit = kvm->arch.vpit;
3675
3676         mutex_lock(&pit->pit_state.lock);
3677         prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3678         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3679         if (!prev_legacy && cur_legacy)
3680                 start = 1;
3681         memcpy(&pit->pit_state.channels, &ps->channels,
3682                sizeof(pit->pit_state.channels));
3683         pit->pit_state.flags = ps->flags;
3684         for (i = 0; i < 3; i++)
3685                 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3686                                    start && i == 0);
3687         mutex_unlock(&pit->pit_state.lock);
3688         return 0;
3689 }
3690
3691 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3692                                  struct kvm_reinject_control *control)
3693 {
3694         struct kvm_pit *pit = kvm->arch.vpit;
3695
3696         if (!pit)
3697                 return -ENXIO;
3698
3699         /* pit->pit_state.lock was overloaded to prevent userspace from getting
3700          * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3701          * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
3702          */
3703         mutex_lock(&pit->pit_state.lock);
3704         kvm_pit_set_reinject(pit, control->pit_reinject);
3705         mutex_unlock(&pit->pit_state.lock);
3706
3707         return 0;
3708 }
3709
3710 /**
3711  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3712  * @kvm: kvm instance
3713  * @log: slot id and address to which we copy the log
3714  *
3715  * Steps 1-4 below provide general overview of dirty page logging. See
3716  * kvm_get_dirty_log_protect() function description for additional details.
3717  *
3718  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3719  * always flush the TLB (step 4) even if previous step failed  and the dirty
3720  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3721  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3722  * writes will be marked dirty for next log read.
3723  *
3724  *   1. Take a snapshot of the bit and clear it if needed.
3725  *   2. Write protect the corresponding page.
3726  *   3. Copy the snapshot to the userspace.
3727  *   4. Flush TLB's if needed.
3728  */
3729 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3730 {
3731         bool is_dirty = false;
3732         int r;
3733
3734         mutex_lock(&kvm->slots_lock);
3735
3736         /*
3737          * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3738          */
3739         if (kvm_x86_ops->flush_log_dirty)
3740                 kvm_x86_ops->flush_log_dirty(kvm);
3741
3742         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3743
3744         /*
3745          * All the TLBs can be flushed out of mmu lock, see the comments in
3746          * kvm_mmu_slot_remove_write_access().
3747          */
3748         lockdep_assert_held(&kvm->slots_lock);
3749         if (is_dirty)
3750                 kvm_flush_remote_tlbs(kvm);
3751
3752         mutex_unlock(&kvm->slots_lock);
3753         return r;
3754 }
3755
3756 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3757                         bool line_status)
3758 {
3759         if (!irqchip_in_kernel(kvm))
3760                 return -ENXIO;
3761
3762         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3763                                         irq_event->irq, irq_event->level,
3764                                         line_status);
3765         return 0;
3766 }
3767
3768 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3769                                    struct kvm_enable_cap *cap)
3770 {
3771         int r;
3772
3773         if (cap->flags)
3774                 return -EINVAL;
3775
3776         switch (cap->cap) {
3777         case KVM_CAP_DISABLE_QUIRKS:
3778                 kvm->arch.disabled_quirks = cap->args[0];
3779                 r = 0;
3780                 break;
3781         case KVM_CAP_SPLIT_IRQCHIP: {
3782                 mutex_lock(&kvm->lock);
3783                 r = -EINVAL;
3784                 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3785                         goto split_irqchip_unlock;
3786                 r = -EEXIST;
3787                 if (irqchip_in_kernel(kvm))
3788                         goto split_irqchip_unlock;
3789                 if (kvm->created_vcpus)
3790                         goto split_irqchip_unlock;
3791                 r = kvm_setup_empty_irq_routing(kvm);
3792                 if (r)
3793                         goto split_irqchip_unlock;
3794                 /* Pairs with irqchip_in_kernel. */
3795                 smp_wmb();
3796                 kvm->arch.irqchip_split = true;
3797                 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3798                 r = 0;
3799 split_irqchip_unlock:
3800                 mutex_unlock(&kvm->lock);
3801                 break;
3802         }
3803         case KVM_CAP_X2APIC_API:
3804                 r = -EINVAL;
3805                 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3806                         break;
3807
3808                 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
3809                         kvm->arch.x2apic_format = true;
3810                 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
3811                         kvm->arch.x2apic_broadcast_quirk_disabled = true;
3812
3813                 r = 0;
3814                 break;
3815         default:
3816                 r = -EINVAL;
3817                 break;
3818         }
3819         return r;
3820 }
3821
3822 long kvm_arch_vm_ioctl(struct file *filp,
3823                        unsigned int ioctl, unsigned long arg)
3824 {
3825         struct kvm *kvm = filp->private_data;
3826         void __user *argp = (void __user *)arg;
3827         int r = -ENOTTY;
3828         /*
3829          * This union makes it completely explicit to gcc-3.x
3830          * that these two variables' stack usage should be
3831          * combined, not added together.
3832          */
3833         union {
3834                 struct kvm_pit_state ps;
3835                 struct kvm_pit_state2 ps2;
3836                 struct kvm_pit_config pit_config;
3837         } u;
3838
3839         switch (ioctl) {
3840         case KVM_SET_TSS_ADDR:
3841                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3842                 break;
3843         case KVM_SET_IDENTITY_MAP_ADDR: {
3844                 u64 ident_addr;
3845
3846                 r = -EFAULT;
3847                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3848                         goto out;
3849                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3850                 break;
3851         }
3852         case KVM_SET_NR_MMU_PAGES:
3853                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3854                 break;
3855         case KVM_GET_NR_MMU_PAGES:
3856                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3857                 break;
3858         case KVM_CREATE_IRQCHIP: {
3859                 struct kvm_pic *vpic;
3860
3861                 mutex_lock(&kvm->lock);
3862                 r = -EEXIST;
3863                 if (kvm->arch.vpic)
3864                         goto create_irqchip_unlock;
3865                 r = -EINVAL;
3866                 if (kvm->created_vcpus)
3867                         goto create_irqchip_unlock;
3868                 r = -ENOMEM;
3869                 vpic = kvm_create_pic(kvm);
3870                 if (vpic) {
3871                         r = kvm_ioapic_init(kvm);
3872                         if (r) {
3873                                 mutex_lock(&kvm->slots_lock);
3874                                 kvm_destroy_pic(vpic);
3875                                 mutex_unlock(&kvm->slots_lock);
3876                                 goto create_irqchip_unlock;
3877                         }
3878                 } else
3879                         goto create_irqchip_unlock;
3880                 r = kvm_setup_default_irq_routing(kvm);
3881                 if (r) {
3882                         mutex_lock(&kvm->slots_lock);
3883                         mutex_lock(&kvm->irq_lock);
3884                         kvm_ioapic_destroy(kvm);
3885                         kvm_destroy_pic(vpic);
3886                         mutex_unlock(&kvm->irq_lock);
3887                         mutex_unlock(&kvm->slots_lock);
3888                         goto create_irqchip_unlock;
3889                 }
3890                 /* Write kvm->irq_routing before kvm->arch.vpic.  */
3891                 smp_wmb();
3892                 kvm->arch.vpic = vpic;
3893         create_irqchip_unlock:
3894                 mutex_unlock(&kvm->lock);
3895                 break;
3896         }
3897         case KVM_CREATE_PIT:
3898                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3899                 goto create_pit;
3900         case KVM_CREATE_PIT2:
3901                 r = -EFAULT;
3902                 if (copy_from_user(&u.pit_config, argp,
3903                                    sizeof(struct kvm_pit_config)))
3904                         goto out;
3905         create_pit:
3906                 mutex_lock(&kvm->lock);
3907                 r = -EEXIST;
3908                 if (kvm->arch.vpit)
3909                         goto create_pit_unlock;
3910                 r = -ENOMEM;
3911                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3912                 if (kvm->arch.vpit)
3913                         r = 0;
3914         create_pit_unlock:
3915                 mutex_unlock(&kvm->lock);
3916                 break;
3917         case KVM_GET_IRQCHIP: {
3918                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3919                 struct kvm_irqchip *chip;
3920
3921                 chip = memdup_user(argp, sizeof(*chip));
3922                 if (IS_ERR(chip)) {
3923                         r = PTR_ERR(chip);
3924                         goto out;
3925                 }
3926
3927                 r = -ENXIO;
3928                 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3929                         goto get_irqchip_out;
3930                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3931                 if (r)
3932                         goto get_irqchip_out;
3933                 r = -EFAULT;
3934                 if (copy_to_user(argp, chip, sizeof *chip))
3935                         goto get_irqchip_out;
3936                 r = 0;
3937         get_irqchip_out:
3938                 kfree(chip);
3939                 break;
3940         }
3941         case KVM_SET_IRQCHIP: {
3942                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3943                 struct kvm_irqchip *chip;
3944
3945                 chip = memdup_user(argp, sizeof(*chip));
3946                 if (IS_ERR(chip)) {
3947                         r = PTR_ERR(chip);
3948                         goto out;
3949                 }
3950
3951                 r = -ENXIO;
3952                 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3953                         goto set_irqchip_out;
3954                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3955                 if (r)
3956                         goto set_irqchip_out;
3957                 r = 0;
3958         set_irqchip_out:
3959                 kfree(chip);
3960                 break;
3961         }
3962         case KVM_GET_PIT: {
3963                 r = -EFAULT;
3964                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3965                         goto out;
3966                 r = -ENXIO;
3967                 if (!kvm->arch.vpit)
3968                         goto out;
3969                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3970                 if (r)
3971                         goto out;
3972                 r = -EFAULT;
3973                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3974                         goto out;
3975                 r = 0;
3976                 break;
3977         }
3978         case KVM_SET_PIT: {
3979                 r = -EFAULT;
3980                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3981                         goto out;
3982                 r = -ENXIO;
3983                 if (!kvm->arch.vpit)
3984                         goto out;
3985                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3986                 break;
3987         }
3988         case KVM_GET_PIT2: {
3989                 r = -ENXIO;
3990                 if (!kvm->arch.vpit)
3991                         goto out;
3992                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3993                 if (r)
3994                         goto out;
3995                 r = -EFAULT;
3996                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3997                         goto out;
3998                 r = 0;
3999                 break;
4000         }
4001         case KVM_SET_PIT2: {
4002                 r = -EFAULT;
4003                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4004                         goto out;
4005                 r = -ENXIO;
4006                 if (!kvm->arch.vpit)
4007                         goto out;
4008                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4009                 break;
4010         }
4011         case KVM_REINJECT_CONTROL: {
4012                 struct kvm_reinject_control control;
4013                 r =  -EFAULT;
4014                 if (copy_from_user(&control, argp, sizeof(control)))
4015                         goto out;
4016                 r = kvm_vm_ioctl_reinject(kvm, &control);
4017                 break;
4018         }
4019         case KVM_SET_BOOT_CPU_ID:
4020                 r = 0;
4021                 mutex_lock(&kvm->lock);
4022                 if (kvm->created_vcpus)
4023                         r = -EBUSY;
4024                 else
4025                         kvm->arch.bsp_vcpu_id = arg;
4026                 mutex_unlock(&kvm->lock);
4027                 break;
4028         case KVM_XEN_HVM_CONFIG: {
4029                 r = -EFAULT;
4030                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4031                                    sizeof(struct kvm_xen_hvm_config)))
4032                         goto out;
4033                 r = -EINVAL;
4034                 if (kvm->arch.xen_hvm_config.flags)
4035                         goto out;
4036                 r = 0;
4037                 break;
4038         }
4039         case KVM_SET_CLOCK: {
4040                 struct kvm_clock_data user_ns;
4041                 u64 now_ns;
4042                 s64 delta;
4043
4044                 r = -EFAULT;
4045                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4046                         goto out;
4047
4048                 r = -EINVAL;
4049                 if (user_ns.flags)
4050                         goto out;
4051
4052                 r = 0;
4053                 local_irq_disable();
4054                 now_ns = get_kernel_ns();
4055                 delta = user_ns.clock - now_ns;
4056                 local_irq_enable();
4057                 kvm->arch.kvmclock_offset = delta;
4058                 kvm_gen_update_masterclock(kvm);
4059                 break;
4060         }
4061         case KVM_GET_CLOCK: {
4062                 struct kvm_clock_data user_ns;
4063                 u64 now_ns;
4064
4065                 local_irq_disable();
4066                 now_ns = get_kernel_ns();
4067                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4068                 local_irq_enable();
4069                 user_ns.flags = 0;
4070                 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4071
4072                 r = -EFAULT;
4073                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4074                         goto out;
4075                 r = 0;
4076                 break;
4077         }
4078         case KVM_ENABLE_CAP: {
4079                 struct kvm_enable_cap cap;
4080
4081                 r = -EFAULT;
4082                 if (copy_from_user(&cap, argp, sizeof(cap)))
4083                         goto out;
4084                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4085                 break;
4086         }
4087         default:
4088                 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4089         }
4090 out:
4091         return r;
4092 }
4093
4094 static void kvm_init_msr_list(void)
4095 {
4096         u32 dummy[2];
4097         unsigned i, j;
4098
4099         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4100                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4101                         continue;
4102
4103                 /*
4104                  * Even MSRs that are valid in the host may not be exposed
4105                  * to the guests in some cases.
4106                  */
4107                 switch (msrs_to_save[i]) {
4108                 case MSR_IA32_BNDCFGS:
4109                         if (!kvm_x86_ops->mpx_supported())
4110                                 continue;
4111                         break;
4112                 case MSR_TSC_AUX:
4113                         if (!kvm_x86_ops->rdtscp_supported())
4114                                 continue;
4115                         break;
4116                 default:
4117                         break;
4118                 }
4119
4120                 if (j < i)
4121                         msrs_to_save[j] = msrs_to_save[i];
4122                 j++;
4123         }
4124         num_msrs_to_save = j;
4125
4126         for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4127                 switch (emulated_msrs[i]) {
4128                 case MSR_IA32_SMBASE:
4129                         if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4130                                 continue;
4131                         break;
4132                 default:
4133                         break;
4134                 }
4135
4136                 if (j < i)
4137                         emulated_msrs[j] = emulated_msrs[i];
4138                 j++;
4139         }
4140         num_emulated_msrs = j;
4141 }
4142
4143 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4144                            const void *v)
4145 {
4146         int handled = 0;
4147         int n;
4148
4149         do {
4150                 n = min(len, 8);
4151                 if (!(lapic_in_kernel(vcpu) &&
4152                       !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4153                     && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4154                         break;
4155                 handled += n;
4156                 addr += n;
4157                 len -= n;
4158                 v += n;
4159         } while (len);
4160
4161         return handled;
4162 }
4163
4164 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4165 {
4166         int handled = 0;
4167         int n;
4168
4169         do {
4170                 n = min(len, 8);
4171                 if (!(lapic_in_kernel(vcpu) &&
4172                       !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4173                                          addr, n, v))
4174                     && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4175                         break;
4176                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4177                 handled += n;
4178                 addr += n;
4179                 len -= n;
4180                 v += n;
4181         } while (len);
4182
4183         return handled;
4184 }
4185
4186 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4187                         struct kvm_segment *var, int seg)
4188 {
4189         kvm_x86_ops->set_segment(vcpu, var, seg);
4190 }
4191
4192 void kvm_get_segment(struct kvm_vcpu *vcpu,
4193                      struct kvm_segment *var, int seg)
4194 {
4195         kvm_x86_ops->get_segment(vcpu, var, seg);
4196 }
4197
4198 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4199                            struct x86_exception *exception)
4200 {
4201         gpa_t t_gpa;
4202
4203         BUG_ON(!mmu_is_nested(vcpu));
4204
4205         /* NPT walks are always user-walks */
4206         access |= PFERR_USER_MASK;
4207         t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4208
4209         return t_gpa;
4210 }
4211
4212 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4213                               struct x86_exception *exception)
4214 {
4215         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4216         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4217 }
4218
4219  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4220                                 struct x86_exception *exception)
4221 {
4222         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4223         access |= PFERR_FETCH_MASK;
4224         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4225 }
4226
4227 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4228                                struct x86_exception *exception)
4229 {
4230         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4231         access |= PFERR_WRITE_MASK;
4232         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4233 }
4234
4235 /* uses this to access any guest's mapped memory without checking CPL */
4236 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4237                                 struct x86_exception *exception)
4238 {
4239         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4240 }
4241
4242 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4243                                       struct kvm_vcpu *vcpu, u32 access,
4244                                       struct x86_exception *exception)
4245 {
4246         void *data = val;
4247         int r = X86EMUL_CONTINUE;
4248
4249         while (bytes) {
4250                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4251                                                             exception);
4252                 unsigned offset = addr & (PAGE_SIZE-1);
4253                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4254                 int ret;
4255
4256                 if (gpa == UNMAPPED_GVA)
4257                         return X86EMUL_PROPAGATE_FAULT;
4258                 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4259                                                offset, toread);
4260                 if (ret < 0) {
4261                         r = X86EMUL_IO_NEEDED;
4262                         goto out;
4263                 }
4264
4265                 bytes -= toread;
4266                 data += toread;
4267                 addr += toread;
4268         }
4269 out:
4270         return r;
4271 }
4272
4273 /* used for instruction fetching */
4274 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4275                                 gva_t addr, void *val, unsigned int bytes,
4276                                 struct x86_exception *exception)
4277 {
4278         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4279         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4280         unsigned offset;
4281         int ret;
4282
4283         /* Inline kvm_read_guest_virt_helper for speed.  */
4284         gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4285                                                     exception);
4286         if (unlikely(gpa == UNMAPPED_GVA))
4287                 return X86EMUL_PROPAGATE_FAULT;
4288
4289         offset = addr & (PAGE_SIZE-1);
4290         if (WARN_ON(offset + bytes > PAGE_SIZE))
4291                 bytes = (unsigned)PAGE_SIZE - offset;
4292         ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4293                                        offset, bytes);
4294         if (unlikely(ret < 0))
4295                 return X86EMUL_IO_NEEDED;
4296
4297         return X86EMUL_CONTINUE;
4298 }
4299
4300 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4301                                gva_t addr, void *val, unsigned int bytes,
4302                                struct x86_exception *exception)
4303 {
4304         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4305         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4306
4307         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4308                                           exception);
4309 }
4310 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4311
4312 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4313                                       gva_t addr, void *val, unsigned int bytes,
4314                                       struct x86_exception *exception)
4315 {
4316         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4317         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4318 }
4319
4320 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4321                 unsigned long addr, void *val, unsigned int bytes)
4322 {
4323         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4324         int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4325
4326         return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4327 }
4328
4329 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4330                                        gva_t addr, void *val,
4331                                        unsigned int bytes,
4332                                        struct x86_exception *exception)
4333 {
4334         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4335         void *data = val;
4336         int r = X86EMUL_CONTINUE;
4337
4338         while (bytes) {
4339                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4340                                                              PFERR_WRITE_MASK,
4341                                                              exception);
4342                 unsigned offset = addr & (PAGE_SIZE-1);
4343                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4344                 int ret;
4345
4346                 if (gpa == UNMAPPED_GVA)
4347                         return X86EMUL_PROPAGATE_FAULT;
4348                 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4349                 if (ret < 0) {
4350                         r = X86EMUL_IO_NEEDED;
4351                         goto out;
4352                 }
4353
4354                 bytes -= towrite;
4355                 data += towrite;
4356                 addr += towrite;
4357         }
4358 out:
4359         return r;
4360 }
4361 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4362
4363 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4364                                 gpa_t *gpa, struct x86_exception *exception,
4365                                 bool write)
4366 {
4367         u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4368                 | (write ? PFERR_WRITE_MASK : 0);
4369
4370         /*
4371          * currently PKRU is only applied to ept enabled guest so
4372          * there is no pkey in EPT page table for L1 guest or EPT
4373          * shadow page table for L2 guest.
4374          */
4375         if (vcpu_match_mmio_gva(vcpu, gva)
4376             && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4377                                  vcpu->arch.access, 0, access)) {
4378                 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4379                                         (gva & (PAGE_SIZE - 1));
4380                 trace_vcpu_match_mmio(gva, *gpa, write, false);
4381                 return 1;
4382         }
4383
4384         *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4385
4386         if (*gpa == UNMAPPED_GVA)
4387                 return -1;
4388
4389         /* For APIC access vmexit */
4390         if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4391                 return 1;
4392
4393         if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4394                 trace_vcpu_match_mmio(gva, *gpa, write, true);
4395                 return 1;
4396         }
4397
4398         return 0;
4399 }
4400
4401 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4402                         const void *val, int bytes)
4403 {
4404         int ret;
4405
4406         ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4407         if (ret < 0)
4408                 return 0;
4409         kvm_page_track_write(vcpu, gpa, val, bytes);
4410         return 1;
4411 }
4412
4413 struct read_write_emulator_ops {
4414         int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4415                                   int bytes);
4416         int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4417                                   void *val, int bytes);
4418         int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4419                                int bytes, void *val);
4420         int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4421                                     void *val, int bytes);
4422         bool write;
4423 };
4424
4425 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4426 {
4427         if (vcpu->mmio_read_completed) {
4428                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4429                                vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4430                 vcpu->mmio_read_completed = 0;
4431                 return 1;
4432         }
4433
4434         return 0;
4435 }
4436
4437 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4438                         void *val, int bytes)
4439 {
4440         return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4441 }
4442
4443 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4444                          void *val, int bytes)
4445 {
4446         return emulator_write_phys(vcpu, gpa, val, bytes);
4447 }
4448
4449 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4450 {
4451         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4452         return vcpu_mmio_write(vcpu, gpa, bytes, val);
4453 }
4454
4455 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4456                           void *val, int bytes)
4457 {
4458         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4459         return X86EMUL_IO_NEEDED;
4460 }
4461
4462 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4463                            void *val, int bytes)
4464 {
4465         struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4466
4467         memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4468         return X86EMUL_CONTINUE;
4469 }
4470
4471 static const struct read_write_emulator_ops read_emultor = {
4472         .read_write_prepare = read_prepare,
4473         .read_write_emulate = read_emulate,
4474         .read_write_mmio = vcpu_mmio_read,
4475         .read_write_exit_mmio = read_exit_mmio,
4476 };
4477
4478 static const struct read_write_emulator_ops write_emultor = {
4479         .read_write_emulate = write_emulate,
4480         .read_write_mmio = write_mmio,
4481         .read_write_exit_mmio = write_exit_mmio,
4482         .write = true,
4483 };
4484
4485 static int emulator_read_write_onepage(unsigned long addr, void *val,
4486                                        unsigned int bytes,
4487                                        struct x86_exception *exception,
4488                                        struct kvm_vcpu *vcpu,
4489                                        const struct read_write_emulator_ops *ops)
4490 {
4491         gpa_t gpa;
4492         int handled, ret;
4493         bool write = ops->write;
4494         struct kvm_mmio_fragment *frag;
4495
4496         ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4497
4498         if (ret < 0)
4499                 return X86EMUL_PROPAGATE_FAULT;
4500
4501         /* For APIC access vmexit */
4502         if (ret)
4503                 goto mmio;
4504
4505         if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4506                 return X86EMUL_CONTINUE;
4507
4508 mmio:
4509         /*
4510          * Is this MMIO handled locally?
4511          */
4512         handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4513         if (handled == bytes)
4514                 return X86EMUL_CONTINUE;
4515
4516         gpa += handled;
4517         bytes -= handled;
4518         val += handled;
4519
4520         WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4521         frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4522         frag->gpa = gpa;
4523         frag->data = val;
4524         frag->len = bytes;
4525         return X86EMUL_CONTINUE;
4526 }
4527
4528 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4529                         unsigned long addr,
4530                         void *val, unsigned int bytes,
4531                         struct x86_exception *exception,
4532                         const struct read_write_emulator_ops *ops)
4533 {
4534         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4535         gpa_t gpa;
4536         int rc;
4537
4538         if (ops->read_write_prepare &&
4539                   ops->read_write_prepare(vcpu, val, bytes))
4540                 return X86EMUL_CONTINUE;
4541
4542         vcpu->mmio_nr_fragments = 0;
4543
4544         /* Crossing a page boundary? */
4545         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4546                 int now;
4547
4548                 now = -addr & ~PAGE_MASK;
4549                 rc = emulator_read_write_onepage(addr, val, now, exception,
4550                                                  vcpu, ops);
4551
4552                 if (rc != X86EMUL_CONTINUE)
4553                         return rc;
4554                 addr += now;
4555                 if (ctxt->mode != X86EMUL_MODE_PROT64)
4556                         addr = (u32)addr;
4557                 val += now;
4558                 bytes -= now;
4559         }
4560
4561         rc = emulator_read_write_onepage(addr, val, bytes, exception,
4562                                          vcpu, ops);
4563         if (rc != X86EMUL_CONTINUE)
4564                 return rc;
4565
4566         if (!vcpu->mmio_nr_fragments)
4567                 return rc;
4568
4569         gpa = vcpu->mmio_fragments[0].gpa;
4570
4571         vcpu->mmio_needed = 1;
4572         vcpu->mmio_cur_fragment = 0;
4573
4574         vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4575         vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4576         vcpu->run->exit_reason = KVM_EXIT_MMIO;
4577         vcpu->run->mmio.phys_addr = gpa;
4578
4579         return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4580 }
4581
4582 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4583                                   unsigned long addr,
4584                                   void *val,
4585                                   unsigned int bytes,
4586                                   struct x86_exception *exception)
4587 {
4588         return emulator_read_write(ctxt, addr, val, bytes,
4589                                    exception, &read_emultor);
4590 }
4591
4592 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4593                             unsigned long addr,
4594                             const void *val,
4595                             unsigned int bytes,
4596                             struct x86_exception *exception)
4597 {
4598         return emulator_read_write(ctxt, addr, (void *)val, bytes,
4599                                    exception, &write_emultor);
4600 }
4601
4602 #define CMPXCHG_TYPE(t, ptr, old, new) \
4603         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4604
4605 #ifdef CONFIG_X86_64
4606 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4607 #else
4608 #  define CMPXCHG64(ptr, old, new) \
4609         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4610 #endif
4611
4612 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4613                                      unsigned long addr,
4614                                      const void *old,
4615                                      const void *new,
4616                                      unsigned int bytes,
4617                                      struct x86_exception *exception)
4618 {
4619         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4620         gpa_t gpa;
4621         struct page *page;
4622         char *kaddr;
4623         bool exchanged;
4624
4625         /* guests cmpxchg8b have to be emulated atomically */
4626         if (bytes > 8 || (bytes & (bytes - 1)))
4627                 goto emul_write;
4628
4629         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4630
4631         if (gpa == UNMAPPED_GVA ||
4632             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4633                 goto emul_write;
4634
4635         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4636                 goto emul_write;
4637
4638         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4639         if (is_error_page(page))
4640                 goto emul_write;
4641
4642         kaddr = kmap_atomic(page);
4643         kaddr += offset_in_page(gpa);
4644         switch (bytes) {
4645         case 1:
4646                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4647                 break;
4648         case 2:
4649                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4650                 break;
4651         case 4:
4652                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4653                 break;
4654         case 8:
4655                 exchanged = CMPXCHG64(kaddr, old, new);
4656                 break;
4657         default:
4658                 BUG();
4659         }
4660         kunmap_atomic(kaddr);
4661         kvm_release_page_dirty(page);
4662
4663         if (!exchanged)
4664                 return X86EMUL_CMPXCHG_FAILED;
4665
4666         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4667         kvm_page_track_write(vcpu, gpa, new, bytes);
4668
4669         return X86EMUL_CONTINUE;
4670
4671 emul_write:
4672         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4673
4674         return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4675 }
4676
4677 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4678 {
4679         /* TODO: String I/O for in kernel device */
4680         int r;
4681
4682         if (vcpu->arch.pio.in)
4683                 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4684                                     vcpu->arch.pio.size, pd);
4685         else
4686                 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4687                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
4688                                      pd);
4689         return r;
4690 }
4691
4692 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4693                                unsigned short port, void *val,
4694                                unsigned int count, bool in)
4695 {
4696         vcpu->arch.pio.port = port;
4697         vcpu->arch.pio.in = in;
4698         vcpu->arch.pio.count  = count;
4699         vcpu->arch.pio.size = size;
4700
4701         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4702                 vcpu->arch.pio.count = 0;
4703                 return 1;
4704         }
4705
4706         vcpu->run->exit_reason = KVM_EXIT_IO;
4707         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4708         vcpu->run->io.size = size;
4709         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4710         vcpu->run->io.count = count;
4711         vcpu->run->io.port = port;
4712
4713         return 0;
4714 }
4715
4716 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4717                                     int size, unsigned short port, void *val,
4718                                     unsigned int count)
4719 {
4720         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4721         int ret;
4722
4723         if (vcpu->arch.pio.count)
4724                 goto data_avail;
4725
4726         ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4727         if (ret) {
4728 data_avail:
4729                 memcpy(val, vcpu->arch.pio_data, size * count);
4730                 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4731                 vcpu->arch.pio.count = 0;
4732                 return 1;
4733         }
4734
4735         return 0;
4736 }
4737
4738 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4739                                      int size, unsigned short port,
4740                                      const void *val, unsigned int count)
4741 {
4742         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4743
4744         memcpy(vcpu->arch.pio_data, val, size * count);
4745         trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4746         return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4747 }
4748
4749 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4750 {
4751         return kvm_x86_ops->get_segment_base(vcpu, seg);
4752 }
4753
4754 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4755 {
4756         kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4757 }
4758
4759 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4760 {
4761         if (!need_emulate_wbinvd(vcpu))
4762                 return X86EMUL_CONTINUE;
4763
4764         if (kvm_x86_ops->has_wbinvd_exit()) {
4765                 int cpu = get_cpu();
4766
4767                 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4768                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4769                                 wbinvd_ipi, NULL, 1);
4770                 put_cpu();
4771                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4772         } else
4773                 wbinvd();
4774         return X86EMUL_CONTINUE;
4775 }
4776
4777 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4778 {
4779         kvm_x86_ops->skip_emulated_instruction(vcpu);
4780         return kvm_emulate_wbinvd_noskip(vcpu);
4781 }
4782 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4783
4784
4785
4786 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4787 {
4788         kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4789 }
4790
4791 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4792                            unsigned long *dest)
4793 {
4794         return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4795 }
4796
4797 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4798                            unsigned long value)
4799 {
4800
4801         return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4802 }
4803
4804 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4805 {
4806         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4807 }
4808
4809 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4810 {
4811         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4812         unsigned long value;
4813
4814         switch (cr) {
4815         case 0:
4816                 value = kvm_read_cr0(vcpu);
4817                 break;
4818         case 2:
4819                 value = vcpu->arch.cr2;
4820                 break;
4821         case 3:
4822                 value = kvm_read_cr3(vcpu);
4823                 break;
4824         case 4:
4825                 value = kvm_read_cr4(vcpu);
4826                 break;
4827         case 8:
4828                 value = kvm_get_cr8(vcpu);
4829                 break;
4830         default:
4831                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4832                 return 0;
4833         }
4834
4835         return value;
4836 }
4837
4838 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4839 {
4840         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4841         int res = 0;
4842
4843         switch (cr) {
4844         case 0:
4845                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4846                 break;
4847         case 2:
4848                 vcpu->arch.cr2 = val;
4849                 break;
4850         case 3:
4851                 res = kvm_set_cr3(vcpu, val);
4852                 break;
4853         case 4:
4854                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4855                 break;
4856         case 8:
4857                 res = kvm_set_cr8(vcpu, val);
4858                 break;
4859         default:
4860                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4861                 res = -1;
4862         }
4863
4864         return res;
4865 }
4866
4867 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4868 {
4869         return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4870 }
4871
4872 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4873 {
4874         kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4875 }
4876
4877 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4878 {
4879         kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4880 }
4881
4882 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4883 {
4884         kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4885 }
4886
4887 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4888 {
4889         kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4890 }
4891
4892 static unsigned long emulator_get_cached_segment_base(
4893         struct x86_emulate_ctxt *ctxt, int seg)
4894 {
4895         return get_segment_base(emul_to_vcpu(ctxt), seg);
4896 }
4897
4898 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4899                                  struct desc_struct *desc, u32 *base3,
4900                                  int seg)
4901 {
4902         struct kvm_segment var;
4903
4904         kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4905         *selector = var.selector;
4906
4907         if (var.unusable) {
4908                 memset(desc, 0, sizeof(*desc));
4909                 return false;
4910         }
4911
4912         if (var.g)
4913                 var.limit >>= 12;
4914         set_desc_limit(desc, var.limit);
4915         set_desc_base(desc, (unsigned long)var.base);
4916 #ifdef CONFIG_X86_64
4917         if (base3)
4918                 *base3 = var.base >> 32;
4919 #endif
4920         desc->type = var.type;
4921         desc->s = var.s;
4922         desc->dpl = var.dpl;
4923         desc->p = var.present;
4924         desc->avl = var.avl;
4925         desc->l = var.l;
4926         desc->d = var.db;
4927         desc->g = var.g;
4928
4929         return true;
4930 }
4931
4932 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4933                                  struct desc_struct *desc, u32 base3,
4934                                  int seg)
4935 {
4936         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4937         struct kvm_segment var;
4938
4939         var.selector = selector;
4940         var.base = get_desc_base(desc);
4941 #ifdef CONFIG_X86_64
4942         var.base |= ((u64)base3) << 32;
4943 #endif
4944         var.limit = get_desc_limit(desc);
4945         if (desc->g)
4946                 var.limit = (var.limit << 12) | 0xfff;
4947         var.type = desc->type;
4948         var.dpl = desc->dpl;
4949         var.db = desc->d;
4950         var.s = desc->s;
4951         var.l = desc->l;
4952         var.g = desc->g;
4953         var.avl = desc->avl;
4954         var.present = desc->p;
4955         var.unusable = !var.present;
4956         var.padding = 0;
4957
4958         kvm_set_segment(vcpu, &var, seg);
4959         return;
4960 }
4961
4962 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4963                             u32 msr_index, u64 *pdata)
4964 {
4965         struct msr_data msr;
4966         int r;
4967
4968         msr.index = msr_index;
4969         msr.host_initiated = false;
4970         r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4971         if (r)
4972                 return r;
4973
4974         *pdata = msr.data;
4975         return 0;
4976 }
4977
4978 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4979                             u32 msr_index, u64 data)
4980 {
4981         struct msr_data msr;
4982
4983         msr.data = data;
4984         msr.index = msr_index;
4985         msr.host_initiated = false;
4986         return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4987 }
4988
4989 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4990 {
4991         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4992
4993         return vcpu->arch.smbase;
4994 }
4995
4996 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4997 {
4998         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4999
5000         vcpu->arch.smbase = smbase;
5001 }
5002
5003 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5004                               u32 pmc)
5005 {
5006         return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5007 }
5008
5009 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5010                              u32 pmc, u64 *pdata)
5011 {
5012         return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5013 }
5014
5015 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5016 {
5017         emul_to_vcpu(ctxt)->arch.halt_request = 1;
5018 }
5019
5020 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5021 {
5022         preempt_disable();
5023         kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5024         /*
5025          * CR0.TS may reference the host fpu state, not the guest fpu state,
5026          * so it may be clear at this point.
5027          */
5028         clts();
5029 }
5030
5031 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5032 {
5033         preempt_enable();
5034 }
5035
5036 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5037                               struct x86_instruction_info *info,
5038                               enum x86_intercept_stage stage)
5039 {
5040         return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5041 }
5042
5043 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5044                                u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5045 {
5046         kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5047 }
5048
5049 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5050 {
5051         return kvm_register_read(emul_to_vcpu(ctxt), reg);
5052 }
5053
5054 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5055 {
5056         kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5057 }
5058
5059 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5060 {
5061         kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5062 }
5063
5064 static const struct x86_emulate_ops emulate_ops = {
5065         .read_gpr            = emulator_read_gpr,
5066         .write_gpr           = emulator_write_gpr,
5067         .read_std            = kvm_read_guest_virt_system,
5068         .write_std           = kvm_write_guest_virt_system,
5069         .read_phys           = kvm_read_guest_phys_system,
5070         .fetch               = kvm_fetch_guest_virt,
5071         .read_emulated       = emulator_read_emulated,
5072         .write_emulated      = emulator_write_emulated,
5073         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
5074         .invlpg              = emulator_invlpg,
5075         .pio_in_emulated     = emulator_pio_in_emulated,
5076         .pio_out_emulated    = emulator_pio_out_emulated,
5077         .get_segment         = emulator_get_segment,
5078         .set_segment         = emulator_set_segment,
5079         .get_cached_segment_base = emulator_get_cached_segment_base,
5080         .get_gdt             = emulator_get_gdt,
5081         .get_idt             = emulator_get_idt,
5082         .set_gdt             = emulator_set_gdt,
5083         .set_idt             = emulator_set_idt,
5084         .get_cr              = emulator_get_cr,
5085         .set_cr              = emulator_set_cr,
5086         .cpl                 = emulator_get_cpl,
5087         .get_dr              = emulator_get_dr,
5088         .set_dr              = emulator_set_dr,
5089         .get_smbase          = emulator_get_smbase,
5090         .set_smbase          = emulator_set_smbase,
5091         .set_msr             = emulator_set_msr,
5092         .get_msr             = emulator_get_msr,
5093         .check_pmc           = emulator_check_pmc,
5094         .read_pmc            = emulator_read_pmc,
5095         .halt                = emulator_halt,
5096         .wbinvd              = emulator_wbinvd,
5097         .fix_hypercall       = emulator_fix_hypercall,
5098         .get_fpu             = emulator_get_fpu,
5099         .put_fpu             = emulator_put_fpu,
5100         .intercept           = emulator_intercept,
5101         .get_cpuid           = emulator_get_cpuid,
5102         .set_nmi_mask        = emulator_set_nmi_mask,
5103 };
5104
5105 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5106 {
5107         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5108         /*
5109          * an sti; sti; sequence only disable interrupts for the first
5110          * instruction. So, if the last instruction, be it emulated or
5111          * not, left the system with the INT_STI flag enabled, it
5112          * means that the last instruction is an sti. We should not
5113          * leave the flag on in this case. The same goes for mov ss
5114          */
5115         if (int_shadow & mask)
5116                 mask = 0;
5117         if (unlikely(int_shadow || mask)) {
5118                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5119                 if (!mask)
5120                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5121         }
5122 }
5123
5124 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5125 {
5126         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5127         if (ctxt->exception.vector == PF_VECTOR)
5128                 return kvm_propagate_fault(vcpu, &ctxt->exception);
5129
5130         if (ctxt->exception.error_code_valid)
5131                 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5132                                       ctxt->exception.error_code);
5133         else
5134                 kvm_queue_exception(vcpu, ctxt->exception.vector);
5135         return false;
5136 }
5137
5138 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5139 {
5140         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5141         int cs_db, cs_l;
5142
5143         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5144
5145         ctxt->eflags = kvm_get_rflags(vcpu);
5146         ctxt->eip = kvm_rip_read(vcpu);
5147         ctxt->mode = (!is_protmode(vcpu))               ? X86EMUL_MODE_REAL :
5148                      (ctxt->eflags & X86_EFLAGS_VM)     ? X86EMUL_MODE_VM86 :
5149                      (cs_l && is_long_mode(vcpu))       ? X86EMUL_MODE_PROT64 :
5150                      cs_db                              ? X86EMUL_MODE_PROT32 :
5151                                                           X86EMUL_MODE_PROT16;
5152         BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5153         BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5154         BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5155         ctxt->emul_flags = vcpu->arch.hflags;
5156
5157         init_decode_cache(ctxt);
5158         vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5159 }
5160
5161 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5162 {
5163         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5164         int ret;
5165
5166         init_emulate_ctxt(vcpu);
5167
5168         ctxt->op_bytes = 2;
5169         ctxt->ad_bytes = 2;
5170         ctxt->_eip = ctxt->eip + inc_eip;
5171         ret = emulate_int_real(ctxt, irq);
5172
5173         if (ret != X86EMUL_CONTINUE)
5174                 return EMULATE_FAIL;
5175
5176         ctxt->eip = ctxt->_eip;
5177         kvm_rip_write(vcpu, ctxt->eip);
5178         kvm_set_rflags(vcpu, ctxt->eflags);
5179
5180         if (irq == NMI_VECTOR)
5181                 vcpu->arch.nmi_pending = 0;
5182         else
5183                 vcpu->arch.interrupt.pending = false;
5184
5185         return EMULATE_DONE;
5186 }
5187 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5188
5189 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5190 {
5191         int r = EMULATE_DONE;
5192
5193         ++vcpu->stat.insn_emulation_fail;
5194         trace_kvm_emulate_insn_failed(vcpu);
5195         if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5196                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5197                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5198                 vcpu->run->internal.ndata = 0;
5199                 r = EMULATE_FAIL;
5200         }
5201         kvm_queue_exception(vcpu, UD_VECTOR);
5202
5203         return r;
5204 }
5205
5206 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5207                                   bool write_fault_to_shadow_pgtable,
5208                                   int emulation_type)
5209 {
5210         gpa_t gpa = cr2;
5211         kvm_pfn_t pfn;
5212
5213         if (emulation_type & EMULTYPE_NO_REEXECUTE)
5214                 return false;
5215
5216         if (!vcpu->arch.mmu.direct_map) {
5217                 /*
5218                  * Write permission should be allowed since only
5219                  * write access need to be emulated.
5220                  */
5221                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5222
5223                 /*
5224                  * If the mapping is invalid in guest, let cpu retry
5225                  * it to generate fault.
5226                  */
5227                 if (gpa == UNMAPPED_GVA)
5228                         return true;
5229         }
5230
5231         /*
5232          * Do not retry the unhandleable instruction if it faults on the
5233          * readonly host memory, otherwise it will goto a infinite loop:
5234          * retry instruction -> write #PF -> emulation fail -> retry
5235          * instruction -> ...
5236          */
5237         pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5238
5239         /*
5240          * If the instruction failed on the error pfn, it can not be fixed,
5241          * report the error to userspace.
5242          */
5243         if (is_error_noslot_pfn(pfn))
5244                 return false;
5245
5246         kvm_release_pfn_clean(pfn);
5247
5248         /* The instructions are well-emulated on direct mmu. */
5249         if (vcpu->arch.mmu.direct_map) {
5250                 unsigned int indirect_shadow_pages;
5251
5252                 spin_lock(&vcpu->kvm->mmu_lock);
5253                 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5254                 spin_unlock(&vcpu->kvm->mmu_lock);
5255
5256                 if (indirect_shadow_pages)
5257                         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5258
5259                 return true;
5260         }
5261
5262         /*
5263          * if emulation was due to access to shadowed page table
5264          * and it failed try to unshadow page and re-enter the
5265          * guest to let CPU execute the instruction.
5266          */
5267         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5268
5269         /*
5270          * If the access faults on its page table, it can not
5271          * be fixed by unprotecting shadow page and it should
5272          * be reported to userspace.
5273          */
5274         return !write_fault_to_shadow_pgtable;
5275 }
5276
5277 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5278                               unsigned long cr2,  int emulation_type)
5279 {
5280         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5281         unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5282
5283         last_retry_eip = vcpu->arch.last_retry_eip;
5284         last_retry_addr = vcpu->arch.last_retry_addr;
5285
5286         /*
5287          * If the emulation is caused by #PF and it is non-page_table
5288          * writing instruction, it means the VM-EXIT is caused by shadow
5289          * page protected, we can zap the shadow page and retry this
5290          * instruction directly.
5291          *
5292          * Note: if the guest uses a non-page-table modifying instruction
5293          * on the PDE that points to the instruction, then we will unmap
5294          * the instruction and go to an infinite loop. So, we cache the
5295          * last retried eip and the last fault address, if we meet the eip
5296          * and the address again, we can break out of the potential infinite
5297          * loop.
5298          */
5299         vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5300
5301         if (!(emulation_type & EMULTYPE_RETRY))
5302                 return false;
5303
5304         if (x86_page_table_writing_insn(ctxt))
5305                 return false;
5306
5307         if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5308                 return false;
5309
5310         vcpu->arch.last_retry_eip = ctxt->eip;
5311         vcpu->arch.last_retry_addr = cr2;
5312
5313         if (!vcpu->arch.mmu.direct_map)
5314                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5315
5316         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5317
5318         return true;
5319 }
5320
5321 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5322 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5323
5324 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5325 {
5326         if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5327                 /* This is a good place to trace that we are exiting SMM.  */
5328                 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5329
5330                 /* Process a latched INIT or SMI, if any.  */
5331                 kvm_make_request(KVM_REQ_EVENT, vcpu);
5332         }
5333
5334         kvm_mmu_reset_context(vcpu);
5335 }
5336
5337 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5338 {
5339         unsigned changed = vcpu->arch.hflags ^ emul_flags;
5340
5341         vcpu->arch.hflags = emul_flags;
5342
5343         if (changed & HF_SMM_MASK)
5344                 kvm_smm_changed(vcpu);
5345 }
5346
5347 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5348                                 unsigned long *db)
5349 {
5350         u32 dr6 = 0;
5351         int i;
5352         u32 enable, rwlen;
5353
5354         enable = dr7;
5355         rwlen = dr7 >> 16;
5356         for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5357                 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5358                         dr6 |= (1 << i);
5359         return dr6;
5360 }
5361
5362 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5363 {
5364         struct kvm_run *kvm_run = vcpu->run;
5365
5366         /*
5367          * rflags is the old, "raw" value of the flags.  The new value has
5368          * not been saved yet.
5369          *
5370          * This is correct even for TF set by the guest, because "the
5371          * processor will not generate this exception after the instruction
5372          * that sets the TF flag".
5373          */
5374         if (unlikely(rflags & X86_EFLAGS_TF)) {
5375                 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5376                         kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5377                                                   DR6_RTM;
5378                         kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5379                         kvm_run->debug.arch.exception = DB_VECTOR;
5380                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
5381                         *r = EMULATE_USER_EXIT;
5382                 } else {
5383                         vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5384                         /*
5385                          * "Certain debug exceptions may clear bit 0-3.  The
5386                          * remaining contents of the DR6 register are never
5387                          * cleared by the processor".
5388                          */
5389                         vcpu->arch.dr6 &= ~15;
5390                         vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5391                         kvm_queue_exception(vcpu, DB_VECTOR);
5392                 }
5393         }
5394 }
5395
5396 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5397 {
5398         if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5399             (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5400                 struct kvm_run *kvm_run = vcpu->run;
5401                 unsigned long eip = kvm_get_linear_rip(vcpu);
5402                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5403                                            vcpu->arch.guest_debug_dr7,
5404                                            vcpu->arch.eff_db);
5405
5406                 if (dr6 != 0) {
5407                         kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5408                         kvm_run->debug.arch.pc = eip;
5409                         kvm_run->debug.arch.exception = DB_VECTOR;
5410                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
5411                         *r = EMULATE_USER_EXIT;
5412                         return true;
5413                 }
5414         }
5415
5416         if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5417             !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5418                 unsigned long eip = kvm_get_linear_rip(vcpu);
5419                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5420                                            vcpu->arch.dr7,
5421                                            vcpu->arch.db);
5422
5423                 if (dr6 != 0) {
5424                         vcpu->arch.dr6 &= ~15;
5425                         vcpu->arch.dr6 |= dr6 | DR6_RTM;
5426                         kvm_queue_exception(vcpu, DB_VECTOR);
5427                         *r = EMULATE_DONE;
5428                         return true;
5429                 }
5430         }
5431
5432         return false;
5433 }
5434
5435 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5436                             unsigned long cr2,
5437                             int emulation_type,
5438                             void *insn,
5439                             int insn_len)
5440 {
5441         int r;
5442         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5443         bool writeback = true;
5444         bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5445
5446         /*
5447          * Clear write_fault_to_shadow_pgtable here to ensure it is
5448          * never reused.
5449          */
5450         vcpu->arch.write_fault_to_shadow_pgtable = false;
5451         kvm_clear_exception_queue(vcpu);
5452
5453         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5454                 init_emulate_ctxt(vcpu);
5455
5456                 /*
5457                  * We will reenter on the same instruction since
5458                  * we do not set complete_userspace_io.  This does not
5459                  * handle watchpoints yet, those would be handled in
5460                  * the emulate_ops.
5461                  */
5462                 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5463                         return r;
5464
5465                 ctxt->interruptibility = 0;
5466                 ctxt->have_exception = false;
5467                 ctxt->exception.vector = -1;
5468                 ctxt->perm_ok = false;
5469
5470                 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5471
5472                 r = x86_decode_insn(ctxt, insn, insn_len);
5473
5474                 trace_kvm_emulate_insn_start(vcpu);
5475                 ++vcpu->stat.insn_emulation;
5476                 if (r != EMULATION_OK)  {
5477                         if (emulation_type & EMULTYPE_TRAP_UD)
5478                                 return EMULATE_FAIL;
5479                         if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5480                                                 emulation_type))
5481                                 return EMULATE_DONE;
5482                         if (emulation_type & EMULTYPE_SKIP)
5483                                 return EMULATE_FAIL;
5484                         return handle_emulation_failure(vcpu);
5485                 }
5486         }
5487
5488         if (emulation_type & EMULTYPE_SKIP) {
5489                 kvm_rip_write(vcpu, ctxt->_eip);
5490                 if (ctxt->eflags & X86_EFLAGS_RF)
5491                         kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5492                 return EMULATE_DONE;
5493         }
5494
5495         if (retry_instruction(ctxt, cr2, emulation_type))
5496                 return EMULATE_DONE;
5497
5498         /* this is needed for vmware backdoor interface to work since it
5499            changes registers values  during IO operation */
5500         if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5501                 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5502                 emulator_invalidate_register_cache(ctxt);
5503         }
5504
5505 restart:
5506         r = x86_emulate_insn(ctxt);
5507
5508         if (r == EMULATION_INTERCEPTED)
5509                 return EMULATE_DONE;
5510
5511         if (r == EMULATION_FAILED) {
5512                 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5513                                         emulation_type))
5514                         return EMULATE_DONE;
5515
5516                 return handle_emulation_failure(vcpu);
5517         }
5518
5519         if (ctxt->have_exception) {
5520                 r = EMULATE_DONE;
5521                 if (inject_emulated_exception(vcpu))
5522                         return r;
5523         } else if (vcpu->arch.pio.count) {
5524                 if (!vcpu->arch.pio.in) {
5525                         /* FIXME: return into emulator if single-stepping.  */
5526                         vcpu->arch.pio.count = 0;
5527                 } else {
5528                         writeback = false;
5529                         vcpu->arch.complete_userspace_io = complete_emulated_pio;
5530                 }
5531                 r = EMULATE_USER_EXIT;
5532         } else if (vcpu->mmio_needed) {
5533                 if (!vcpu->mmio_is_write)
5534                         writeback = false;
5535                 r = EMULATE_USER_EXIT;
5536                 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5537         } else if (r == EMULATION_RESTART)
5538                 goto restart;
5539         else
5540                 r = EMULATE_DONE;
5541
5542         if (writeback) {
5543                 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5544                 toggle_interruptibility(vcpu, ctxt->interruptibility);
5545                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5546                 if (vcpu->arch.hflags != ctxt->emul_flags)
5547                         kvm_set_hflags(vcpu, ctxt->emul_flags);
5548                 kvm_rip_write(vcpu, ctxt->eip);
5549                 if (r == EMULATE_DONE)
5550                         kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5551                 if (!ctxt->have_exception ||
5552                     exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5553                         __kvm_set_rflags(vcpu, ctxt->eflags);
5554
5555                 /*
5556                  * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5557                  * do nothing, and it will be requested again as soon as
5558                  * the shadow expires.  But we still need to check here,
5559                  * because POPF has no interrupt shadow.
5560                  */
5561                 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5562                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5563         } else
5564                 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5565
5566         return r;
5567 }
5568 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5569
5570 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5571 {
5572         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5573         int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5574                                             size, port, &val, 1);
5575         /* do not return to emulator after return from userspace */
5576         vcpu->arch.pio.count = 0;
5577         return ret;
5578 }
5579 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5580
5581 static int kvmclock_cpu_down_prep(unsigned int cpu)
5582 {
5583         __this_cpu_write(cpu_tsc_khz, 0);
5584         return 0;
5585 }
5586
5587 static void tsc_khz_changed(void *data)
5588 {
5589         struct cpufreq_freqs *freq = data;
5590         unsigned long khz = 0;
5591
5592         if (data)
5593                 khz = freq->new;
5594         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5595                 khz = cpufreq_quick_get(raw_smp_processor_id());
5596         if (!khz)
5597                 khz = tsc_khz;
5598         __this_cpu_write(cpu_tsc_khz, khz);
5599 }
5600
5601 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5602                                      void *data)
5603 {
5604         struct cpufreq_freqs *freq = data;
5605         struct kvm *kvm;
5606         struct kvm_vcpu *vcpu;
5607         int i, send_ipi = 0;
5608
5609         /*
5610          * We allow guests to temporarily run on slowing clocks,
5611          * provided we notify them after, or to run on accelerating
5612          * clocks, provided we notify them before.  Thus time never
5613          * goes backwards.
5614          *
5615          * However, we have a problem.  We can't atomically update
5616          * the frequency of a given CPU from this function; it is
5617          * merely a notifier, which can be called from any CPU.
5618          * Changing the TSC frequency at arbitrary points in time
5619          * requires a recomputation of local variables related to
5620          * the TSC for each VCPU.  We must flag these local variables
5621          * to be updated and be sure the update takes place with the
5622          * new frequency before any guests proceed.
5623          *
5624          * Unfortunately, the combination of hotplug CPU and frequency
5625          * change creates an intractable locking scenario; the order
5626          * of when these callouts happen is undefined with respect to
5627          * CPU hotplug, and they can race with each other.  As such,
5628          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5629          * undefined; you can actually have a CPU frequency change take
5630          * place in between the computation of X and the setting of the
5631          * variable.  To protect against this problem, all updates of
5632          * the per_cpu tsc_khz variable are done in an interrupt
5633          * protected IPI, and all callers wishing to update the value
5634          * must wait for a synchronous IPI to complete (which is trivial
5635          * if the caller is on the CPU already).  This establishes the
5636          * necessary total order on variable updates.
5637          *
5638          * Note that because a guest time update may take place
5639          * anytime after the setting of the VCPU's request bit, the
5640          * correct TSC value must be set before the request.  However,
5641          * to ensure the update actually makes it to any guest which
5642          * starts running in hardware virtualization between the set
5643          * and the acquisition of the spinlock, we must also ping the
5644          * CPU after setting the request bit.
5645          *
5646          */
5647
5648         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5649                 return 0;
5650         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5651                 return 0;
5652
5653         smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5654
5655         spin_lock(&kvm_lock);
5656         list_for_each_entry(kvm, &vm_list, vm_list) {
5657                 kvm_for_each_vcpu(i, vcpu, kvm) {
5658                         if (vcpu->cpu != freq->cpu)
5659                                 continue;
5660                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5661                         if (vcpu->cpu != smp_processor_id())
5662                                 send_ipi = 1;
5663                 }
5664         }
5665         spin_unlock(&kvm_lock);
5666
5667         if (freq->old < freq->new && send_ipi) {
5668                 /*
5669                  * We upscale the frequency.  Must make the guest
5670                  * doesn't see old kvmclock values while running with
5671                  * the new frequency, otherwise we risk the guest sees
5672                  * time go backwards.
5673                  *
5674                  * In case we update the frequency for another cpu
5675                  * (which might be in guest context) send an interrupt
5676                  * to kick the cpu out of guest context.  Next time
5677                  * guest context is entered kvmclock will be updated,
5678                  * so the guest will not see stale values.
5679                  */
5680                 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5681         }
5682         return 0;
5683 }
5684
5685 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5686         .notifier_call  = kvmclock_cpufreq_notifier
5687 };
5688
5689 static int kvmclock_cpu_online(unsigned int cpu)
5690 {
5691         tsc_khz_changed(NULL);
5692         return 0;
5693 }
5694
5695 static void kvm_timer_init(void)
5696 {
5697         int cpu;
5698
5699         max_tsc_khz = tsc_khz;
5700
5701         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5702 #ifdef CONFIG_CPU_FREQ
5703                 struct cpufreq_policy policy;
5704                 memset(&policy, 0, sizeof(policy));
5705                 cpu = get_cpu();
5706                 cpufreq_get_policy(&policy, cpu);
5707                 if (policy.cpuinfo.max_freq)
5708                         max_tsc_khz = policy.cpuinfo.max_freq;
5709                 put_cpu();
5710 #endif
5711                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5712                                           CPUFREQ_TRANSITION_NOTIFIER);
5713         }
5714         pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5715
5716         cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "AP_X86_KVM_CLK_ONLINE",
5717                           kvmclock_cpu_online, kvmclock_cpu_down_prep);
5718 }
5719
5720 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5721
5722 int kvm_is_in_guest(void)
5723 {
5724         return __this_cpu_read(current_vcpu) != NULL;
5725 }
5726
5727 static int kvm_is_user_mode(void)
5728 {
5729         int user_mode = 3;
5730
5731         if (__this_cpu_read(current_vcpu))
5732                 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5733
5734         return user_mode != 0;
5735 }
5736
5737 static unsigned long kvm_get_guest_ip(void)
5738 {
5739         unsigned long ip = 0;
5740
5741         if (__this_cpu_read(current_vcpu))
5742                 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5743
5744         return ip;
5745 }
5746
5747 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5748         .is_in_guest            = kvm_is_in_guest,
5749         .is_user_mode           = kvm_is_user_mode,
5750         .get_guest_ip           = kvm_get_guest_ip,
5751 };
5752
5753 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5754 {
5755         __this_cpu_write(current_vcpu, vcpu);
5756 }
5757 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5758
5759 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5760 {
5761         __this_cpu_write(current_vcpu, NULL);
5762 }
5763 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5764
5765 static void kvm_set_mmio_spte_mask(void)
5766 {
5767         u64 mask;
5768         int maxphyaddr = boot_cpu_data.x86_phys_bits;
5769
5770         /*
5771          * Set the reserved bits and the present bit of an paging-structure
5772          * entry to generate page fault with PFER.RSV = 1.
5773          */
5774          /* Mask the reserved physical address bits. */
5775         mask = rsvd_bits(maxphyaddr, 51);
5776
5777         /* Bit 62 is always reserved for 32bit host. */
5778         mask |= 0x3ull << 62;
5779
5780         /* Set the present bit. */
5781         mask |= 1ull;
5782
5783 #ifdef CONFIG_X86_64
5784         /*
5785          * If reserved bit is not supported, clear the present bit to disable
5786          * mmio page fault.
5787          */
5788         if (maxphyaddr == 52)
5789                 mask &= ~1ull;
5790 #endif
5791
5792         kvm_mmu_set_mmio_spte_mask(mask);
5793 }
5794
5795 #ifdef CONFIG_X86_64
5796 static void pvclock_gtod_update_fn(struct work_struct *work)
5797 {
5798         struct kvm *kvm;
5799
5800         struct kvm_vcpu *vcpu;
5801         int i;
5802
5803         spin_lock(&kvm_lock);
5804         list_for_each_entry(kvm, &vm_list, vm_list)
5805                 kvm_for_each_vcpu(i, vcpu, kvm)
5806                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5807         atomic_set(&kvm_guest_has_master_clock, 0);
5808         spin_unlock(&kvm_lock);
5809 }
5810
5811 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5812
5813 /*
5814  * Notification about pvclock gtod data update.
5815  */
5816 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5817                                void *priv)
5818 {
5819         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5820         struct timekeeper *tk = priv;
5821
5822         update_pvclock_gtod(tk);
5823
5824         /* disable master clock if host does not trust, or does not
5825          * use, TSC clocksource
5826          */
5827         if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5828             atomic_read(&kvm_guest_has_master_clock) != 0)
5829                 queue_work(system_long_wq, &pvclock_gtod_work);
5830
5831         return 0;
5832 }
5833
5834 static struct notifier_block pvclock_gtod_notifier = {
5835         .notifier_call = pvclock_gtod_notify,
5836 };
5837 #endif
5838
5839 int kvm_arch_init(void *opaque)
5840 {
5841         int r;
5842         struct kvm_x86_ops *ops = opaque;
5843
5844         if (kvm_x86_ops) {
5845                 printk(KERN_ERR "kvm: already loaded the other module\n");
5846                 r = -EEXIST;
5847                 goto out;
5848         }
5849
5850         if (!ops->cpu_has_kvm_support()) {
5851                 printk(KERN_ERR "kvm: no hardware support\n");
5852                 r = -EOPNOTSUPP;
5853                 goto out;
5854         }
5855         if (ops->disabled_by_bios()) {
5856                 printk(KERN_ERR "kvm: disabled by bios\n");
5857                 r = -EOPNOTSUPP;
5858                 goto out;
5859         }
5860
5861         r = -ENOMEM;
5862         shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5863         if (!shared_msrs) {
5864                 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5865                 goto out;
5866         }
5867
5868         r = kvm_mmu_module_init();
5869         if (r)
5870                 goto out_free_percpu;
5871
5872         kvm_set_mmio_spte_mask();
5873
5874         kvm_x86_ops = ops;
5875
5876         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5877                         PT_DIRTY_MASK, PT64_NX_MASK, 0,
5878                         PT_PRESENT_MASK);
5879         kvm_timer_init();
5880
5881         perf_register_guest_info_callbacks(&kvm_guest_cbs);
5882
5883         if (boot_cpu_has(X86_FEATURE_XSAVE))
5884                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5885
5886         kvm_lapic_init();
5887 #ifdef CONFIG_X86_64
5888         pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5889 #endif
5890
5891         return 0;
5892
5893 out_free_percpu:
5894         free_percpu(shared_msrs);
5895 out:
5896         return r;
5897 }
5898
5899 void kvm_arch_exit(void)
5900 {
5901         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5902
5903         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5904                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5905                                             CPUFREQ_TRANSITION_NOTIFIER);
5906         cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
5907 #ifdef CONFIG_X86_64
5908         pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5909 #endif
5910         kvm_x86_ops = NULL;
5911         kvm_mmu_module_exit();
5912         free_percpu(shared_msrs);
5913 }
5914
5915 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5916 {
5917         ++vcpu->stat.halt_exits;
5918         if (lapic_in_kernel(vcpu)) {
5919                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5920                 return 1;
5921         } else {
5922                 vcpu->run->exit_reason = KVM_EXIT_HLT;
5923                 return 0;
5924         }
5925 }
5926 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5927
5928 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5929 {
5930         kvm_x86_ops->skip_emulated_instruction(vcpu);
5931         return kvm_vcpu_halt(vcpu);
5932 }
5933 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5934
5935 /*
5936  * kvm_pv_kick_cpu_op:  Kick a vcpu.
5937  *
5938  * @apicid - apicid of vcpu to be kicked.
5939  */
5940 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5941 {
5942         struct kvm_lapic_irq lapic_irq;
5943
5944         lapic_irq.shorthand = 0;
5945         lapic_irq.dest_mode = 0;
5946         lapic_irq.dest_id = apicid;
5947         lapic_irq.msi_redir_hint = false;
5948
5949         lapic_irq.delivery_mode = APIC_DM_REMRD;
5950         kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5951 }
5952
5953 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
5954 {
5955         vcpu->arch.apicv_active = false;
5956         kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
5957 }
5958
5959 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5960 {
5961         unsigned long nr, a0, a1, a2, a3, ret;
5962         int op_64_bit, r = 1;
5963
5964         kvm_x86_ops->skip_emulated_instruction(vcpu);
5965
5966         if (kvm_hv_hypercall_enabled(vcpu->kvm))
5967                 return kvm_hv_hypercall(vcpu);
5968
5969         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5970         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5971         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5972         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5973         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5974
5975         trace_kvm_hypercall(nr, a0, a1, a2, a3);
5976
5977         op_64_bit = is_64_bit_mode(vcpu);
5978         if (!op_64_bit) {
5979                 nr &= 0xFFFFFFFF;
5980                 a0 &= 0xFFFFFFFF;
5981                 a1 &= 0xFFFFFFFF;
5982                 a2 &= 0xFFFFFFFF;
5983                 a3 &= 0xFFFFFFFF;
5984         }
5985
5986         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5987                 ret = -KVM_EPERM;
5988                 goto out;
5989         }
5990
5991         switch (nr) {
5992         case KVM_HC_VAPIC_POLL_IRQ:
5993                 ret = 0;
5994                 break;
5995         case KVM_HC_KICK_CPU:
5996                 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5997                 ret = 0;
5998                 break;
5999         default:
6000                 ret = -KVM_ENOSYS;
6001                 break;
6002         }
6003 out:
6004         if (!op_64_bit)
6005                 ret = (u32)ret;
6006         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6007         ++vcpu->stat.hypercalls;
6008         return r;
6009 }
6010 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6011
6012 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6013 {
6014         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6015         char instruction[3];
6016         unsigned long rip = kvm_rip_read(vcpu);
6017
6018         kvm_x86_ops->patch_hypercall(vcpu, instruction);
6019
6020         return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
6021 }
6022
6023 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6024 {
6025         return vcpu->run->request_interrupt_window &&
6026                 likely(!pic_in_kernel(vcpu->kvm));
6027 }
6028
6029 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6030 {
6031         struct kvm_run *kvm_run = vcpu->run;
6032
6033         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6034         kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6035         kvm_run->cr8 = kvm_get_cr8(vcpu);
6036         kvm_run->apic_base = kvm_get_apic_base(vcpu);
6037         kvm_run->ready_for_interrupt_injection =
6038                 pic_in_kernel(vcpu->kvm) ||
6039                 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6040 }
6041
6042 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6043 {
6044         int max_irr, tpr;
6045
6046         if (!kvm_x86_ops->update_cr8_intercept)
6047                 return;
6048
6049         if (!lapic_in_kernel(vcpu))
6050                 return;
6051
6052         if (vcpu->arch.apicv_active)
6053                 return;
6054
6055         if (!vcpu->arch.apic->vapic_addr)
6056                 max_irr = kvm_lapic_find_highest_irr(vcpu);
6057         else
6058                 max_irr = -1;
6059
6060         if (max_irr != -1)
6061                 max_irr >>= 4;
6062
6063         tpr = kvm_lapic_get_cr8(vcpu);
6064
6065         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6066 }
6067
6068 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6069 {
6070         int r;
6071
6072         /* try to reinject previous events if any */
6073         if (vcpu->arch.exception.pending) {
6074                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6075                                         vcpu->arch.exception.has_error_code,
6076                                         vcpu->arch.exception.error_code);
6077
6078                 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6079                         __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6080                                              X86_EFLAGS_RF);
6081
6082                 if (vcpu->arch.exception.nr == DB_VECTOR &&
6083                     (vcpu->arch.dr7 & DR7_GD)) {
6084                         vcpu->arch.dr7 &= ~DR7_GD;
6085                         kvm_update_dr7(vcpu);
6086                 }
6087
6088                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6089                                           vcpu->arch.exception.has_error_code,
6090                                           vcpu->arch.exception.error_code,
6091                                           vcpu->arch.exception.reinject);
6092                 return 0;
6093         }
6094
6095         if (vcpu->arch.nmi_injected) {
6096                 kvm_x86_ops->set_nmi(vcpu);
6097                 return 0;
6098         }
6099
6100         if (vcpu->arch.interrupt.pending) {
6101                 kvm_x86_ops->set_irq(vcpu);
6102                 return 0;
6103         }
6104
6105         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6106                 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6107                 if (r != 0)
6108                         return r;
6109         }
6110
6111         /* try to inject new event if pending */
6112         if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6113                 vcpu->arch.smi_pending = false;
6114                 enter_smm(vcpu);
6115         } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6116                 --vcpu->arch.nmi_pending;
6117                 vcpu->arch.nmi_injected = true;
6118                 kvm_x86_ops->set_nmi(vcpu);
6119         } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6120                 /*
6121                  * Because interrupts can be injected asynchronously, we are
6122                  * calling check_nested_events again here to avoid a race condition.
6123                  * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6124                  * proposal and current concerns.  Perhaps we should be setting
6125                  * KVM_REQ_EVENT only on certain events and not unconditionally?
6126                  */
6127                 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6128                         r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6129                         if (r != 0)
6130                                 return r;
6131                 }
6132                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6133                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6134                                             false);
6135                         kvm_x86_ops->set_irq(vcpu);
6136                 }
6137         }
6138
6139         return 0;
6140 }
6141
6142 static void process_nmi(struct kvm_vcpu *vcpu)
6143 {
6144         unsigned limit = 2;
6145
6146         /*
6147          * x86 is limited to one NMI running, and one NMI pending after it.
6148          * If an NMI is already in progress, limit further NMIs to just one.
6149          * Otherwise, allow two (and we'll inject the first one immediately).
6150          */
6151         if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6152                 limit = 1;
6153
6154         vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6155         vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6156         kvm_make_request(KVM_REQ_EVENT, vcpu);
6157 }
6158
6159 #define put_smstate(type, buf, offset, val)                       \
6160         *(type *)((buf) + (offset) - 0x7e00) = val
6161
6162 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6163 {
6164         u32 flags = 0;
6165         flags |= seg->g       << 23;
6166         flags |= seg->db      << 22;
6167         flags |= seg->l       << 21;
6168         flags |= seg->avl     << 20;
6169         flags |= seg->present << 15;
6170         flags |= seg->dpl     << 13;
6171         flags |= seg->s       << 12;
6172         flags |= seg->type    << 8;
6173         return flags;
6174 }
6175
6176 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6177 {
6178         struct kvm_segment seg;
6179         int offset;
6180
6181         kvm_get_segment(vcpu, &seg, n);
6182         put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6183
6184         if (n < 3)
6185                 offset = 0x7f84 + n * 12;
6186         else
6187                 offset = 0x7f2c + (n - 3) * 12;
6188
6189         put_smstate(u32, buf, offset + 8, seg.base);
6190         put_smstate(u32, buf, offset + 4, seg.limit);
6191         put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6192 }
6193
6194 #ifdef CONFIG_X86_64
6195 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6196 {
6197         struct kvm_segment seg;
6198         int offset;
6199         u16 flags;
6200
6201         kvm_get_segment(vcpu, &seg, n);
6202         offset = 0x7e00 + n * 16;
6203
6204         flags = enter_smm_get_segment_flags(&seg) >> 8;
6205         put_smstate(u16, buf, offset, seg.selector);
6206         put_smstate(u16, buf, offset + 2, flags);
6207         put_smstate(u32, buf, offset + 4, seg.limit);
6208         put_smstate(u64, buf, offset + 8, seg.base);
6209 }
6210 #endif
6211
6212 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6213 {
6214         struct desc_ptr dt;
6215         struct kvm_segment seg;
6216         unsigned long val;
6217         int i;
6218
6219         put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6220         put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6221         put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6222         put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6223
6224         for (i = 0; i < 8; i++)
6225                 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6226
6227         kvm_get_dr(vcpu, 6, &val);
6228         put_smstate(u32, buf, 0x7fcc, (u32)val);
6229         kvm_get_dr(vcpu, 7, &val);
6230         put_smstate(u32, buf, 0x7fc8, (u32)val);
6231
6232         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6233         put_smstate(u32, buf, 0x7fc4, seg.selector);
6234         put_smstate(u32, buf, 0x7f64, seg.base);
6235         put_smstate(u32, buf, 0x7f60, seg.limit);
6236         put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6237
6238         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6239         put_smstate(u32, buf, 0x7fc0, seg.selector);
6240         put_smstate(u32, buf, 0x7f80, seg.base);
6241         put_smstate(u32, buf, 0x7f7c, seg.limit);
6242         put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6243
6244         kvm_x86_ops->get_gdt(vcpu, &dt);
6245         put_smstate(u32, buf, 0x7f74, dt.address);
6246         put_smstate(u32, buf, 0x7f70, dt.size);
6247
6248         kvm_x86_ops->get_idt(vcpu, &dt);
6249         put_smstate(u32, buf, 0x7f58, dt.address);
6250         put_smstate(u32, buf, 0x7f54, dt.size);
6251
6252         for (i = 0; i < 6; i++)
6253                 enter_smm_save_seg_32(vcpu, buf, i);
6254
6255         put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6256
6257         /* revision id */
6258         put_smstate(u32, buf, 0x7efc, 0x00020000);
6259         put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6260 }
6261
6262 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6263 {
6264 #ifdef CONFIG_X86_64
6265         struct desc_ptr dt;
6266         struct kvm_segment seg;
6267         unsigned long val;
6268         int i;
6269
6270         for (i = 0; i < 16; i++)
6271                 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6272
6273         put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6274         put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6275
6276         kvm_get_dr(vcpu, 6, &val);
6277         put_smstate(u64, buf, 0x7f68, val);
6278         kvm_get_dr(vcpu, 7, &val);
6279         put_smstate(u64, buf, 0x7f60, val);
6280
6281         put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6282         put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6283         put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6284
6285         put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6286
6287         /* revision id */
6288         put_smstate(u32, buf, 0x7efc, 0x00020064);
6289
6290         put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6291
6292         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6293         put_smstate(u16, buf, 0x7e90, seg.selector);
6294         put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6295         put_smstate(u32, buf, 0x7e94, seg.limit);
6296         put_smstate(u64, buf, 0x7e98, seg.base);
6297
6298         kvm_x86_ops->get_idt(vcpu, &dt);
6299         put_smstate(u32, buf, 0x7e84, dt.size);
6300         put_smstate(u64, buf, 0x7e88, dt.address);
6301
6302         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6303         put_smstate(u16, buf, 0x7e70, seg.selector);
6304         put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6305         put_smstate(u32, buf, 0x7e74, seg.limit);
6306         put_smstate(u64, buf, 0x7e78, seg.base);
6307
6308         kvm_x86_ops->get_gdt(vcpu, &dt);
6309         put_smstate(u32, buf, 0x7e64, dt.size);
6310         put_smstate(u64, buf, 0x7e68, dt.address);
6311
6312         for (i = 0; i < 6; i++)
6313                 enter_smm_save_seg_64(vcpu, buf, i);
6314 #else
6315         WARN_ON_ONCE(1);
6316 #endif
6317 }
6318
6319 static void enter_smm(struct kvm_vcpu *vcpu)
6320 {
6321         struct kvm_segment cs, ds;
6322         struct desc_ptr dt;
6323         char buf[512];
6324         u32 cr0;
6325
6326         trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6327         vcpu->arch.hflags |= HF_SMM_MASK;
6328         memset(buf, 0, 512);
6329         if (guest_cpuid_has_longmode(vcpu))
6330                 enter_smm_save_state_64(vcpu, buf);
6331         else
6332                 enter_smm_save_state_32(vcpu, buf);
6333
6334         kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6335
6336         if (kvm_x86_ops->get_nmi_mask(vcpu))
6337                 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6338         else
6339                 kvm_x86_ops->set_nmi_mask(vcpu, true);
6340
6341         kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6342         kvm_rip_write(vcpu, 0x8000);
6343
6344         cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6345         kvm_x86_ops->set_cr0(vcpu, cr0);
6346         vcpu->arch.cr0 = cr0;
6347
6348         kvm_x86_ops->set_cr4(vcpu, 0);
6349
6350         /* Undocumented: IDT limit is set to zero on entry to SMM.  */
6351         dt.address = dt.size = 0;
6352         kvm_x86_ops->set_idt(vcpu, &dt);
6353
6354         __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6355
6356         cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6357         cs.base = vcpu->arch.smbase;
6358
6359         ds.selector = 0;
6360         ds.base = 0;
6361
6362         cs.limit    = ds.limit = 0xffffffff;
6363         cs.type     = ds.type = 0x3;
6364         cs.dpl      = ds.dpl = 0;
6365         cs.db       = ds.db = 0;
6366         cs.s        = ds.s = 1;
6367         cs.l        = ds.l = 0;
6368         cs.g        = ds.g = 1;
6369         cs.avl      = ds.avl = 0;
6370         cs.present  = ds.present = 1;
6371         cs.unusable = ds.unusable = 0;
6372         cs.padding  = ds.padding = 0;
6373
6374         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6375         kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6376         kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6377         kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6378         kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6379         kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6380
6381         if (guest_cpuid_has_longmode(vcpu))
6382                 kvm_x86_ops->set_efer(vcpu, 0);
6383
6384         kvm_update_cpuid(vcpu);
6385         kvm_mmu_reset_context(vcpu);
6386 }
6387
6388 static void process_smi(struct kvm_vcpu *vcpu)
6389 {
6390         vcpu->arch.smi_pending = true;
6391         kvm_make_request(KVM_REQ_EVENT, vcpu);
6392 }
6393
6394 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6395 {
6396         kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6397 }
6398
6399 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6400 {
6401         u64 eoi_exit_bitmap[4];
6402
6403         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6404                 return;
6405
6406         bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6407
6408         if (irqchip_split(vcpu->kvm))
6409                 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6410         else {
6411                 if (vcpu->arch.apicv_active)
6412                         kvm_x86_ops->sync_pir_to_irr(vcpu);
6413                 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6414         }
6415         bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6416                   vcpu_to_synic(vcpu)->vec_bitmap, 256);
6417         kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6418 }
6419
6420 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6421 {
6422         ++vcpu->stat.tlb_flush;
6423         kvm_x86_ops->tlb_flush(vcpu);
6424 }
6425
6426 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6427 {
6428         struct page *page = NULL;
6429
6430         if (!lapic_in_kernel(vcpu))
6431                 return;
6432
6433         if (!kvm_x86_ops->set_apic_access_page_addr)
6434                 return;
6435
6436         page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6437         if (is_error_page(page))
6438                 return;
6439         kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6440
6441         /*
6442          * Do not pin apic access page in memory, the MMU notifier
6443          * will call us again if it is migrated or swapped out.
6444          */
6445         put_page(page);
6446 }
6447 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6448
6449 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6450                                            unsigned long address)
6451 {
6452         /*
6453          * The physical address of apic access page is stored in the VMCS.
6454          * Update it when it becomes invalid.
6455          */
6456         if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6457                 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6458 }
6459
6460 /*
6461  * Returns 1 to let vcpu_run() continue the guest execution loop without
6462  * exiting to the userspace.  Otherwise, the value will be returned to the
6463  * userspace.
6464  */
6465 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6466 {
6467         int r;
6468         bool req_int_win =
6469                 dm_request_for_irq_injection(vcpu) &&
6470                 kvm_cpu_accept_dm_intr(vcpu);
6471
6472         bool req_immediate_exit = false;
6473
6474         if (vcpu->requests) {
6475                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6476                         kvm_mmu_unload(vcpu);
6477                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6478                         __kvm_migrate_timers(vcpu);
6479                 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6480                         kvm_gen_update_masterclock(vcpu->kvm);
6481                 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6482                         kvm_gen_kvmclock_update(vcpu);
6483                 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6484                         r = kvm_guest_time_update(vcpu);
6485                         if (unlikely(r))
6486                                 goto out;
6487                 }
6488                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6489                         kvm_mmu_sync_roots(vcpu);
6490                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6491                         kvm_vcpu_flush_tlb(vcpu);
6492                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6493                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6494                         r = 0;
6495                         goto out;
6496                 }
6497                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6498                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6499                         r = 0;
6500                         goto out;
6501                 }
6502                 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6503                         vcpu->fpu_active = 0;
6504                         kvm_x86_ops->fpu_deactivate(vcpu);
6505                 }
6506                 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6507                         /* Page is swapped out. Do synthetic halt */
6508                         vcpu->arch.apf.halted = true;
6509                         r = 1;
6510                         goto out;
6511                 }
6512                 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6513                         record_steal_time(vcpu);
6514                 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6515                         process_smi(vcpu);
6516                 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6517                         process_nmi(vcpu);
6518                 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6519                         kvm_pmu_handle_event(vcpu);
6520                 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6521                         kvm_pmu_deliver_pmi(vcpu);
6522                 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6523                         BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6524                         if (test_bit(vcpu->arch.pending_ioapic_eoi,
6525                                      vcpu->arch.ioapic_handled_vectors)) {
6526                                 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6527                                 vcpu->run->eoi.vector =
6528                                                 vcpu->arch.pending_ioapic_eoi;
6529                                 r = 0;
6530                                 goto out;
6531                         }
6532                 }
6533                 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6534                         vcpu_scan_ioapic(vcpu);
6535                 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6536                         kvm_vcpu_reload_apic_access_page(vcpu);
6537                 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6538                         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6539                         vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6540                         r = 0;
6541                         goto out;
6542                 }
6543                 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6544                         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6545                         vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6546                         r = 0;
6547                         goto out;
6548                 }
6549                 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6550                         vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6551                         vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6552                         r = 0;
6553                         goto out;
6554                 }
6555
6556                 /*
6557                  * KVM_REQ_HV_STIMER has to be processed after
6558                  * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6559                  * depend on the guest clock being up-to-date
6560                  */
6561                 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6562                         kvm_hv_process_stimers(vcpu);
6563         }
6564
6565         /*
6566          * KVM_REQ_EVENT is not set when posted interrupts are set by
6567          * VT-d hardware, so we have to update RVI unconditionally.
6568          */
6569         if (kvm_lapic_enabled(vcpu)) {
6570                 /*
6571                  * Update architecture specific hints for APIC
6572                  * virtual interrupt delivery.
6573                  */
6574                 if (vcpu->arch.apicv_active)
6575                         kvm_x86_ops->hwapic_irr_update(vcpu,
6576                                 kvm_lapic_find_highest_irr(vcpu));
6577         }
6578
6579         if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6580                 kvm_apic_accept_events(vcpu);
6581                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6582                         r = 1;
6583                         goto out;
6584                 }
6585
6586                 if (inject_pending_event(vcpu, req_int_win) != 0)
6587                         req_immediate_exit = true;
6588                 else {
6589                         /* Enable NMI/IRQ window open exits if needed.
6590                          *
6591                          * SMIs have two cases: 1) they can be nested, and
6592                          * then there is nothing to do here because RSM will
6593                          * cause a vmexit anyway; 2) or the SMI can be pending
6594                          * because inject_pending_event has completed the
6595                          * injection of an IRQ or NMI from the previous vmexit,
6596                          * and then we request an immediate exit to inject the SMI.
6597                          */
6598                         if (vcpu->arch.smi_pending && !is_smm(vcpu))
6599                                 req_immediate_exit = true;
6600                         if (vcpu->arch.nmi_pending)
6601                                 kvm_x86_ops->enable_nmi_window(vcpu);
6602                         if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6603                                 kvm_x86_ops->enable_irq_window(vcpu);
6604                 }
6605
6606                 if (kvm_lapic_enabled(vcpu)) {
6607                         update_cr8_intercept(vcpu);
6608                         kvm_lapic_sync_to_vapic(vcpu);
6609                 }
6610         }
6611
6612         r = kvm_mmu_reload(vcpu);
6613         if (unlikely(r)) {
6614                 goto cancel_injection;
6615         }
6616
6617         preempt_disable();
6618
6619         kvm_x86_ops->prepare_guest_switch(vcpu);
6620         if (vcpu->fpu_active)
6621                 kvm_load_guest_fpu(vcpu);
6622         vcpu->mode = IN_GUEST_MODE;
6623
6624         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6625
6626         /*
6627          * We should set ->mode before check ->requests,
6628          * Please see the comment in kvm_make_all_cpus_request.
6629          * This also orders the write to mode from any reads
6630          * to the page tables done while the VCPU is running.
6631          * Please see the comment in kvm_flush_remote_tlbs.
6632          */
6633         smp_mb__after_srcu_read_unlock();
6634
6635         local_irq_disable();
6636
6637         if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6638             || need_resched() || signal_pending(current)) {
6639                 vcpu->mode = OUTSIDE_GUEST_MODE;
6640                 smp_wmb();
6641                 local_irq_enable();
6642                 preempt_enable();
6643                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6644                 r = 1;
6645                 goto cancel_injection;
6646         }
6647
6648         kvm_load_guest_xcr0(vcpu);
6649
6650         if (req_immediate_exit) {
6651                 kvm_make_request(KVM_REQ_EVENT, vcpu);
6652                 smp_send_reschedule(vcpu->cpu);
6653         }
6654
6655         trace_kvm_entry(vcpu->vcpu_id);
6656         wait_lapic_expire(vcpu);
6657         guest_enter_irqoff();
6658
6659         if (unlikely(vcpu->arch.switch_db_regs)) {
6660                 set_debugreg(0, 7);
6661                 set_debugreg(vcpu->arch.eff_db[0], 0);
6662                 set_debugreg(vcpu->arch.eff_db[1], 1);
6663                 set_debugreg(vcpu->arch.eff_db[2], 2);
6664                 set_debugreg(vcpu->arch.eff_db[3], 3);
6665                 set_debugreg(vcpu->arch.dr6, 6);
6666                 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6667         }
6668
6669         kvm_x86_ops->run(vcpu);
6670
6671         /*
6672          * Do this here before restoring debug registers on the host.  And
6673          * since we do this before handling the vmexit, a DR access vmexit
6674          * can (a) read the correct value of the debug registers, (b) set
6675          * KVM_DEBUGREG_WONT_EXIT again.
6676          */
6677         if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6678                 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6679                 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6680                 kvm_update_dr0123(vcpu);
6681                 kvm_update_dr6(vcpu);
6682                 kvm_update_dr7(vcpu);
6683                 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6684         }
6685
6686         /*
6687          * If the guest has used debug registers, at least dr7
6688          * will be disabled while returning to the host.
6689          * If we don't have active breakpoints in the host, we don't
6690          * care about the messed up debug address registers. But if
6691          * we have some of them active, restore the old state.
6692          */
6693         if (hw_breakpoint_active())
6694                 hw_breakpoint_restore();
6695
6696         vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6697
6698         vcpu->mode = OUTSIDE_GUEST_MODE;
6699         smp_wmb();
6700
6701         kvm_put_guest_xcr0(vcpu);
6702
6703         /* Interrupt is enabled by handle_external_intr() */
6704         kvm_x86_ops->handle_external_intr(vcpu);
6705
6706         ++vcpu->stat.exits;
6707
6708         guest_exit_irqoff();
6709
6710         local_irq_enable();
6711         preempt_enable();
6712
6713         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6714
6715         /*
6716          * Profile KVM exit RIPs:
6717          */
6718         if (unlikely(prof_on == KVM_PROFILING)) {
6719                 unsigned long rip = kvm_rip_read(vcpu);
6720                 profile_hit(KVM_PROFILING, (void *)rip);
6721         }
6722
6723         if (unlikely(vcpu->arch.tsc_always_catchup))
6724                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6725
6726         if (vcpu->arch.apic_attention)
6727                 kvm_lapic_sync_from_vapic(vcpu);
6728
6729         r = kvm_x86_ops->handle_exit(vcpu);
6730         return r;
6731
6732 cancel_injection:
6733         kvm_x86_ops->cancel_injection(vcpu);
6734         if (unlikely(vcpu->arch.apic_attention))
6735                 kvm_lapic_sync_from_vapic(vcpu);
6736 out:
6737         return r;
6738 }
6739
6740 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6741 {
6742         if (!kvm_arch_vcpu_runnable(vcpu) &&
6743             (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6744                 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6745                 kvm_vcpu_block(vcpu);
6746                 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6747
6748                 if (kvm_x86_ops->post_block)
6749                         kvm_x86_ops->post_block(vcpu);
6750
6751                 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6752                         return 1;
6753         }
6754
6755         kvm_apic_accept_events(vcpu);
6756         switch(vcpu->arch.mp_state) {
6757         case KVM_MP_STATE_HALTED:
6758                 vcpu->arch.pv.pv_unhalted = false;
6759                 vcpu->arch.mp_state =
6760                         KVM_MP_STATE_RUNNABLE;
6761         case KVM_MP_STATE_RUNNABLE:
6762                 vcpu->arch.apf.halted = false;
6763                 break;
6764         case KVM_MP_STATE_INIT_RECEIVED:
6765                 break;
6766         default:
6767                 return -EINTR;
6768                 break;
6769         }
6770         return 1;
6771 }
6772
6773 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6774 {
6775         return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6776                 !vcpu->arch.apf.halted);
6777 }
6778
6779 static int vcpu_run(struct kvm_vcpu *vcpu)
6780 {
6781         int r;
6782         struct kvm *kvm = vcpu->kvm;
6783
6784         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6785
6786         for (;;) {
6787                 if (kvm_vcpu_running(vcpu)) {
6788                         r = vcpu_enter_guest(vcpu);
6789                 } else {
6790                         r = vcpu_block(kvm, vcpu);
6791                 }
6792
6793                 if (r <= 0)
6794                         break;
6795
6796                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6797                 if (kvm_cpu_has_pending_timer(vcpu))
6798                         kvm_inject_pending_timer_irqs(vcpu);
6799
6800                 if (dm_request_for_irq_injection(vcpu) &&
6801                         kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
6802                         r = 0;
6803                         vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
6804                         ++vcpu->stat.request_irq_exits;
6805                         break;
6806                 }
6807
6808                 kvm_check_async_pf_completion(vcpu);
6809
6810                 if (signal_pending(current)) {
6811                         r = -EINTR;
6812                         vcpu->run->exit_reason = KVM_EXIT_INTR;
6813                         ++vcpu->stat.signal_exits;
6814                         break;
6815                 }
6816                 if (need_resched()) {
6817                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6818                         cond_resched();
6819                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6820                 }
6821         }
6822
6823         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6824
6825         return r;
6826 }
6827
6828 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6829 {
6830         int r;
6831         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6832         r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6833         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6834         if (r != EMULATE_DONE)
6835                 return 0;
6836         return 1;
6837 }
6838
6839 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6840 {
6841         BUG_ON(!vcpu->arch.pio.count);
6842
6843         return complete_emulated_io(vcpu);
6844 }
6845
6846 /*
6847  * Implements the following, as a state machine:
6848  *
6849  * read:
6850  *   for each fragment
6851  *     for each mmio piece in the fragment
6852  *       write gpa, len
6853  *       exit
6854  *       copy data
6855  *   execute insn
6856  *
6857  * write:
6858  *   for each fragment
6859  *     for each mmio piece in the fragment
6860  *       write gpa, len
6861  *       copy data
6862  *       exit
6863  */
6864 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6865 {
6866         struct kvm_run *run = vcpu->run;
6867         struct kvm_mmio_fragment *frag;
6868         unsigned len;
6869
6870         BUG_ON(!vcpu->mmio_needed);
6871
6872         /* Complete previous fragment */
6873         frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6874         len = min(8u, frag->len);
6875         if (!vcpu->mmio_is_write)
6876                 memcpy(frag->data, run->mmio.data, len);
6877
6878         if (frag->len <= 8) {
6879                 /* Switch to the next fragment. */
6880                 frag++;
6881                 vcpu->mmio_cur_fragment++;
6882         } else {
6883                 /* Go forward to the next mmio piece. */
6884                 frag->data += len;
6885                 frag->gpa += len;
6886                 frag->len -= len;
6887         }
6888
6889         if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6890                 vcpu->mmio_needed = 0;
6891
6892                 /* FIXME: return into emulator if single-stepping.  */
6893                 if (vcpu->mmio_is_write)
6894                         return 1;
6895                 vcpu->mmio_read_completed = 1;
6896                 return complete_emulated_io(vcpu);
6897         }
6898
6899         run->exit_reason = KVM_EXIT_MMIO;
6900         run->mmio.phys_addr = frag->gpa;
6901         if (vcpu->mmio_is_write)
6902                 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6903         run->mmio.len = min(8u, frag->len);
6904         run->mmio.is_write = vcpu->mmio_is_write;
6905         vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6906         return 0;
6907 }
6908
6909
6910 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6911 {
6912         struct fpu *fpu = &current->thread.fpu;
6913         int r;
6914         sigset_t sigsaved;
6915
6916         fpu__activate_curr(fpu);
6917
6918         if (vcpu->sigset_active)
6919                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6920
6921         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6922                 kvm_vcpu_block(vcpu);
6923                 kvm_apic_accept_events(vcpu);
6924                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6925                 r = -EAGAIN;
6926                 goto out;
6927         }
6928
6929         /* re-sync apic's tpr */
6930         if (!lapic_in_kernel(vcpu)) {
6931                 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6932                         r = -EINVAL;
6933                         goto out;
6934                 }
6935         }
6936
6937         if (unlikely(vcpu->arch.complete_userspace_io)) {
6938                 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6939                 vcpu->arch.complete_userspace_io = NULL;
6940                 r = cui(vcpu);
6941                 if (r <= 0)
6942                         goto out;
6943         } else
6944                 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6945
6946         r = vcpu_run(vcpu);
6947
6948 out:
6949         post_kvm_run_save(vcpu);
6950         if (vcpu->sigset_active)
6951                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6952
6953         return r;
6954 }
6955
6956 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6957 {
6958         if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6959                 /*
6960                  * We are here if userspace calls get_regs() in the middle of
6961                  * instruction emulation. Registers state needs to be copied
6962                  * back from emulation context to vcpu. Userspace shouldn't do
6963                  * that usually, but some bad designed PV devices (vmware
6964                  * backdoor interface) need this to work
6965                  */
6966                 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6967                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6968         }
6969         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6970         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6971         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6972         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6973         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6974         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6975         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6976         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6977 #ifdef CONFIG_X86_64
6978         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6979         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6980         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6981         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6982         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6983         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6984         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6985         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6986 #endif
6987
6988         regs->rip = kvm_rip_read(vcpu);
6989         regs->rflags = kvm_get_rflags(vcpu);
6990
6991         return 0;
6992 }
6993
6994 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6995 {
6996         vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6997         vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6998
6999         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7000         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7001         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7002         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7003         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7004         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7005         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7006         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7007 #ifdef CONFIG_X86_64
7008         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7009         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7010         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7011         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7012         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7013         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7014         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7015         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7016 #endif
7017
7018         kvm_rip_write(vcpu, regs->rip);
7019         kvm_set_rflags(vcpu, regs->rflags);
7020
7021         vcpu->arch.exception.pending = false;
7022
7023         kvm_make_request(KVM_REQ_EVENT, vcpu);
7024
7025         return 0;
7026 }
7027
7028 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7029 {
7030         struct kvm_segment cs;
7031
7032         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7033         *db = cs.db;
7034         *l = cs.l;
7035 }
7036 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7037
7038 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7039                                   struct kvm_sregs *sregs)
7040 {
7041         struct desc_ptr dt;
7042
7043         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7044         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7045         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7046         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7047         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7048         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7049
7050         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7051         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7052
7053         kvm_x86_ops->get_idt(vcpu, &dt);
7054         sregs->idt.limit = dt.size;
7055         sregs->idt.base = dt.address;
7056         kvm_x86_ops->get_gdt(vcpu, &dt);
7057         sregs->gdt.limit = dt.size;
7058         sregs->gdt.base = dt.address;
7059
7060         sregs->cr0 = kvm_read_cr0(vcpu);
7061         sregs->cr2 = vcpu->arch.cr2;
7062         sregs->cr3 = kvm_read_cr3(vcpu);
7063         sregs->cr4 = kvm_read_cr4(vcpu);
7064         sregs->cr8 = kvm_get_cr8(vcpu);
7065         sregs->efer = vcpu->arch.efer;
7066         sregs->apic_base = kvm_get_apic_base(vcpu);
7067
7068         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7069
7070         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7071                 set_bit(vcpu->arch.interrupt.nr,
7072                         (unsigned long *)sregs->interrupt_bitmap);
7073
7074         return 0;
7075 }
7076
7077 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7078                                     struct kvm_mp_state *mp_state)
7079 {
7080         kvm_apic_accept_events(vcpu);
7081         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7082                                         vcpu->arch.pv.pv_unhalted)
7083                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7084         else
7085                 mp_state->mp_state = vcpu->arch.mp_state;
7086
7087         return 0;
7088 }
7089
7090 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7091                                     struct kvm_mp_state *mp_state)
7092 {
7093         if (!lapic_in_kernel(vcpu) &&
7094             mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7095                 return -EINVAL;
7096
7097         if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7098                 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7099                 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7100         } else
7101                 vcpu->arch.mp_state = mp_state->mp_state;
7102         kvm_make_request(KVM_REQ_EVENT, vcpu);
7103         return 0;
7104 }
7105
7106 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7107                     int reason, bool has_error_code, u32 error_code)
7108 {
7109         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7110         int ret;
7111
7112         init_emulate_ctxt(vcpu);
7113
7114         ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7115                                    has_error_code, error_code);
7116
7117         if (ret)
7118                 return EMULATE_FAIL;
7119
7120         kvm_rip_write(vcpu, ctxt->eip);
7121         kvm_set_rflags(vcpu, ctxt->eflags);
7122         kvm_make_request(KVM_REQ_EVENT, vcpu);
7123         return EMULATE_DONE;
7124 }
7125 EXPORT_SYMBOL_GPL(kvm_task_switch);
7126
7127 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7128                                   struct kvm_sregs *sregs)
7129 {
7130         struct msr_data apic_base_msr;
7131         int mmu_reset_needed = 0;
7132         int pending_vec, max_bits, idx;
7133         struct desc_ptr dt;
7134
7135         if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7136                 return -EINVAL;
7137
7138         dt.size = sregs->idt.limit;
7139         dt.address = sregs->idt.base;
7140         kvm_x86_ops->set_idt(vcpu, &dt);
7141         dt.size = sregs->gdt.limit;
7142         dt.address = sregs->gdt.base;
7143         kvm_x86_ops->set_gdt(vcpu, &dt);
7144
7145         vcpu->arch.cr2 = sregs->cr2;
7146         mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7147         vcpu->arch.cr3 = sregs->cr3;
7148         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7149
7150         kvm_set_cr8(vcpu, sregs->cr8);
7151
7152         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7153         kvm_x86_ops->set_efer(vcpu, sregs->efer);
7154         apic_base_msr.data = sregs->apic_base;
7155         apic_base_msr.host_initiated = true;
7156         kvm_set_apic_base(vcpu, &apic_base_msr);
7157
7158         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7159         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7160         vcpu->arch.cr0 = sregs->cr0;
7161
7162         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7163         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7164         if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7165                 kvm_update_cpuid(vcpu);
7166
7167         idx = srcu_read_lock(&vcpu->kvm->srcu);
7168         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7169                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7170                 mmu_reset_needed = 1;
7171         }
7172         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7173
7174         if (mmu_reset_needed)
7175                 kvm_mmu_reset_context(vcpu);
7176
7177         max_bits = KVM_NR_INTERRUPTS;
7178         pending_vec = find_first_bit(
7179                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7180         if (pending_vec < max_bits) {
7181                 kvm_queue_interrupt(vcpu, pending_vec, false);
7182                 pr_debug("Set back pending irq %d\n", pending_vec);
7183         }
7184
7185         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7186         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7187         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7188         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7189         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7190         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7191
7192         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7193         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7194
7195         update_cr8_intercept(vcpu);
7196
7197         /* Older userspace won't unhalt the vcpu on reset. */
7198         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7199             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7200             !is_protmode(vcpu))
7201                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7202
7203         kvm_make_request(KVM_REQ_EVENT, vcpu);
7204
7205         return 0;
7206 }
7207
7208 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7209                                         struct kvm_guest_debug *dbg)
7210 {
7211         unsigned long rflags;
7212         int i, r;
7213
7214         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7215                 r = -EBUSY;
7216                 if (vcpu->arch.exception.pending)
7217                         goto out;
7218                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7219                         kvm_queue_exception(vcpu, DB_VECTOR);
7220                 else
7221                         kvm_queue_exception(vcpu, BP_VECTOR);
7222         }
7223
7224         /*
7225          * Read rflags as long as potentially injected trace flags are still
7226          * filtered out.
7227          */
7228         rflags = kvm_get_rflags(vcpu);
7229
7230         vcpu->guest_debug = dbg->control;
7231         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7232                 vcpu->guest_debug = 0;
7233
7234         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7235                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7236                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7237                 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7238         } else {
7239                 for (i = 0; i < KVM_NR_DB_REGS; i++)
7240                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7241         }
7242         kvm_update_dr7(vcpu);
7243
7244         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7245                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7246                         get_segment_base(vcpu, VCPU_SREG_CS);
7247
7248         /*
7249          * Trigger an rflags update that will inject or remove the trace
7250          * flags.
7251          */
7252         kvm_set_rflags(vcpu, rflags);
7253
7254         kvm_x86_ops->update_bp_intercept(vcpu);
7255
7256         r = 0;
7257
7258 out:
7259
7260         return r;
7261 }
7262
7263 /*
7264  * Translate a guest virtual address to a guest physical address.
7265  */
7266 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7267                                     struct kvm_translation *tr)
7268 {
7269         unsigned long vaddr = tr->linear_address;
7270         gpa_t gpa;
7271         int idx;
7272
7273         idx = srcu_read_lock(&vcpu->kvm->srcu);
7274         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7275         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7276         tr->physical_address = gpa;
7277         tr->valid = gpa != UNMAPPED_GVA;
7278         tr->writeable = 1;
7279         tr->usermode = 0;
7280
7281         return 0;
7282 }
7283
7284 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7285 {
7286         struct fxregs_state *fxsave =
7287                         &vcpu->arch.guest_fpu.state.fxsave;
7288
7289         memcpy(fpu->fpr, fxsave->st_space, 128);
7290         fpu->fcw = fxsave->cwd;
7291         fpu->fsw = fxsave->swd;
7292         fpu->ftwx = fxsave->twd;
7293         fpu->last_opcode = fxsave->fop;
7294         fpu->last_ip = fxsave->rip;
7295         fpu->last_dp = fxsave->rdp;
7296         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7297
7298         return 0;
7299 }
7300
7301 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7302 {
7303         struct fxregs_state *fxsave =
7304                         &vcpu->arch.guest_fpu.state.fxsave;
7305
7306         memcpy(fxsave->st_space, fpu->fpr, 128);
7307         fxsave->cwd = fpu->fcw;
7308         fxsave->swd = fpu->fsw;
7309         fxsave->twd = fpu->ftwx;
7310         fxsave->fop = fpu->last_opcode;
7311         fxsave->rip = fpu->last_ip;
7312         fxsave->rdp = fpu->last_dp;
7313         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7314
7315         return 0;
7316 }
7317
7318 static void fx_init(struct kvm_vcpu *vcpu)
7319 {
7320         fpstate_init(&vcpu->arch.guest_fpu.state);
7321         if (boot_cpu_has(X86_FEATURE_XSAVES))
7322                 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7323                         host_xcr0 | XSTATE_COMPACTION_ENABLED;
7324
7325         /*
7326          * Ensure guest xcr0 is valid for loading
7327          */
7328         vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7329
7330         vcpu->arch.cr0 |= X86_CR0_ET;
7331 }
7332
7333 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7334 {
7335         if (vcpu->guest_fpu_loaded)
7336                 return;
7337
7338         /*
7339          * Restore all possible states in the guest,
7340          * and assume host would use all available bits.
7341          * Guest xcr0 would be loaded later.
7342          */
7343         vcpu->guest_fpu_loaded = 1;
7344         __kernel_fpu_begin();
7345         __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7346         trace_kvm_fpu(1);
7347 }
7348
7349 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7350 {
7351         if (!vcpu->guest_fpu_loaded) {
7352                 vcpu->fpu_counter = 0;
7353                 return;
7354         }
7355
7356         vcpu->guest_fpu_loaded = 0;
7357         copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7358         __kernel_fpu_end();
7359         ++vcpu->stat.fpu_reload;
7360         /*
7361          * If using eager FPU mode, or if the guest is a frequent user
7362          * of the FPU, just leave the FPU active for next time.
7363          * Every 255 times fpu_counter rolls over to 0; a guest that uses
7364          * the FPU in bursts will revert to loading it on demand.
7365          */
7366         if (!use_eager_fpu()) {
7367                 if (++vcpu->fpu_counter < 5)
7368                         kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7369         }
7370         trace_kvm_fpu(0);
7371 }
7372
7373 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7374 {
7375         kvmclock_reset(vcpu);
7376
7377         free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7378         kvm_x86_ops->vcpu_free(vcpu);
7379 }
7380
7381 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7382                                                 unsigned int id)
7383 {
7384         struct kvm_vcpu *vcpu;
7385
7386         if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7387                 printk_once(KERN_WARNING
7388                 "kvm: SMP vm created on host with unstable TSC; "
7389                 "guest TSC will not be reliable\n");
7390
7391         vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7392
7393         return vcpu;
7394 }
7395
7396 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7397 {
7398         int r;
7399
7400         kvm_vcpu_mtrr_init(vcpu);
7401         r = vcpu_load(vcpu);
7402         if (r)
7403                 return r;
7404         kvm_vcpu_reset(vcpu, false);
7405         kvm_mmu_setup(vcpu);
7406         vcpu_put(vcpu);
7407         return r;
7408 }
7409
7410 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7411 {
7412         struct msr_data msr;
7413         struct kvm *kvm = vcpu->kvm;
7414
7415         if (vcpu_load(vcpu))
7416                 return;
7417         msr.data = 0x0;
7418         msr.index = MSR_IA32_TSC;
7419         msr.host_initiated = true;
7420         kvm_write_tsc(vcpu, &msr);
7421         vcpu_put(vcpu);
7422
7423         if (!kvmclock_periodic_sync)
7424                 return;
7425
7426         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7427                                         KVMCLOCK_SYNC_PERIOD);
7428 }
7429
7430 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7431 {
7432         int r;
7433         vcpu->arch.apf.msr_val = 0;
7434
7435         r = vcpu_load(vcpu);
7436         BUG_ON(r);
7437         kvm_mmu_unload(vcpu);
7438         vcpu_put(vcpu);
7439
7440         kvm_x86_ops->vcpu_free(vcpu);
7441 }
7442
7443 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7444 {
7445         vcpu->arch.hflags = 0;
7446
7447         vcpu->arch.smi_pending = 0;
7448         atomic_set(&vcpu->arch.nmi_queued, 0);
7449         vcpu->arch.nmi_pending = 0;
7450         vcpu->arch.nmi_injected = false;
7451         kvm_clear_interrupt_queue(vcpu);
7452         kvm_clear_exception_queue(vcpu);
7453
7454         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7455         kvm_update_dr0123(vcpu);
7456         vcpu->arch.dr6 = DR6_INIT;
7457         kvm_update_dr6(vcpu);
7458         vcpu->arch.dr7 = DR7_FIXED_1;
7459         kvm_update_dr7(vcpu);
7460
7461         vcpu->arch.cr2 = 0;
7462
7463         kvm_make_request(KVM_REQ_EVENT, vcpu);
7464         vcpu->arch.apf.msr_val = 0;
7465         vcpu->arch.st.msr_val = 0;
7466
7467         kvmclock_reset(vcpu);
7468
7469         kvm_clear_async_pf_completion_queue(vcpu);
7470         kvm_async_pf_hash_reset(vcpu);
7471         vcpu->arch.apf.halted = false;
7472
7473         if (!init_event) {
7474                 kvm_pmu_reset(vcpu);
7475                 vcpu->arch.smbase = 0x30000;
7476         }
7477
7478         memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7479         vcpu->arch.regs_avail = ~0;
7480         vcpu->arch.regs_dirty = ~0;
7481
7482         kvm_x86_ops->vcpu_reset(vcpu, init_event);
7483 }
7484
7485 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7486 {
7487         struct kvm_segment cs;
7488
7489         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7490         cs.selector = vector << 8;
7491         cs.base = vector << 12;
7492         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7493         kvm_rip_write(vcpu, 0);
7494 }
7495
7496 int kvm_arch_hardware_enable(void)
7497 {
7498         struct kvm *kvm;
7499         struct kvm_vcpu *vcpu;
7500         int i;
7501         int ret;
7502         u64 local_tsc;
7503         u64 max_tsc = 0;
7504         bool stable, backwards_tsc = false;
7505
7506         kvm_shared_msr_cpu_online();
7507         ret = kvm_x86_ops->hardware_enable();
7508         if (ret != 0)
7509                 return ret;
7510
7511         local_tsc = rdtsc();
7512         stable = !check_tsc_unstable();
7513         list_for_each_entry(kvm, &vm_list, vm_list) {
7514                 kvm_for_each_vcpu(i, vcpu, kvm) {
7515                         if (!stable && vcpu->cpu == smp_processor_id())
7516                                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7517                         if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7518                                 backwards_tsc = true;
7519                                 if (vcpu->arch.last_host_tsc > max_tsc)
7520                                         max_tsc = vcpu->arch.last_host_tsc;
7521                         }
7522                 }
7523         }
7524
7525         /*
7526          * Sometimes, even reliable TSCs go backwards.  This happens on
7527          * platforms that reset TSC during suspend or hibernate actions, but
7528          * maintain synchronization.  We must compensate.  Fortunately, we can
7529          * detect that condition here, which happens early in CPU bringup,
7530          * before any KVM threads can be running.  Unfortunately, we can't
7531          * bring the TSCs fully up to date with real time, as we aren't yet far
7532          * enough into CPU bringup that we know how much real time has actually
7533          * elapsed; our helper function, get_kernel_ns() will be using boot
7534          * variables that haven't been updated yet.
7535          *
7536          * So we simply find the maximum observed TSC above, then record the
7537          * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7538          * the adjustment will be applied.  Note that we accumulate
7539          * adjustments, in case multiple suspend cycles happen before some VCPU
7540          * gets a chance to run again.  In the event that no KVM threads get a
7541          * chance to run, we will miss the entire elapsed period, as we'll have
7542          * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7543          * loose cycle time.  This isn't too big a deal, since the loss will be
7544          * uniform across all VCPUs (not to mention the scenario is extremely
7545          * unlikely). It is possible that a second hibernate recovery happens
7546          * much faster than a first, causing the observed TSC here to be
7547          * smaller; this would require additional padding adjustment, which is
7548          * why we set last_host_tsc to the local tsc observed here.
7549          *
7550          * N.B. - this code below runs only on platforms with reliable TSC,
7551          * as that is the only way backwards_tsc is set above.  Also note
7552          * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7553          * have the same delta_cyc adjustment applied if backwards_tsc
7554          * is detected.  Note further, this adjustment is only done once,
7555          * as we reset last_host_tsc on all VCPUs to stop this from being
7556          * called multiple times (one for each physical CPU bringup).
7557          *
7558          * Platforms with unreliable TSCs don't have to deal with this, they
7559          * will be compensated by the logic in vcpu_load, which sets the TSC to
7560          * catchup mode.  This will catchup all VCPUs to real time, but cannot
7561          * guarantee that they stay in perfect synchronization.
7562          */
7563         if (backwards_tsc) {
7564                 u64 delta_cyc = max_tsc - local_tsc;
7565                 backwards_tsc_observed = true;
7566                 list_for_each_entry(kvm, &vm_list, vm_list) {
7567                         kvm_for_each_vcpu(i, vcpu, kvm) {
7568                                 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7569                                 vcpu->arch.last_host_tsc = local_tsc;
7570                                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7571                         }
7572
7573                         /*
7574                          * We have to disable TSC offset matching.. if you were
7575                          * booting a VM while issuing an S4 host suspend....
7576                          * you may have some problem.  Solving this issue is
7577                          * left as an exercise to the reader.
7578                          */
7579                         kvm->arch.last_tsc_nsec = 0;
7580                         kvm->arch.last_tsc_write = 0;
7581                 }
7582
7583         }
7584         return 0;
7585 }
7586
7587 void kvm_arch_hardware_disable(void)
7588 {
7589         kvm_x86_ops->hardware_disable();
7590         drop_user_return_notifiers();
7591 }
7592
7593 int kvm_arch_hardware_setup(void)
7594 {
7595         int r;
7596
7597         r = kvm_x86_ops->hardware_setup();
7598         if (r != 0)
7599                 return r;
7600
7601         if (kvm_has_tsc_control) {
7602                 /*
7603                  * Make sure the user can only configure tsc_khz values that
7604                  * fit into a signed integer.
7605                  * A min value is not calculated needed because it will always
7606                  * be 1 on all machines.
7607                  */
7608                 u64 max = min(0x7fffffffULL,
7609                               __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7610                 kvm_max_guest_tsc_khz = max;
7611
7612                 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7613         }
7614
7615         kvm_init_msr_list();
7616         return 0;
7617 }
7618
7619 void kvm_arch_hardware_unsetup(void)
7620 {
7621         kvm_x86_ops->hardware_unsetup();
7622 }
7623
7624 void kvm_arch_check_processor_compat(void *rtn)
7625 {
7626         kvm_x86_ops->check_processor_compatibility(rtn);
7627 }
7628
7629 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7630 {
7631         return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7632 }
7633 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7634
7635 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7636 {
7637         return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7638 }
7639
7640 struct static_key kvm_no_apic_vcpu __read_mostly;
7641 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7642
7643 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7644 {
7645         struct page *page;
7646         struct kvm *kvm;
7647         int r;
7648
7649         BUG_ON(vcpu->kvm == NULL);
7650         kvm = vcpu->kvm;
7651
7652         vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7653         vcpu->arch.pv.pv_unhalted = false;
7654         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7655         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7656                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7657         else
7658                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7659
7660         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7661         if (!page) {
7662                 r = -ENOMEM;
7663                 goto fail;
7664         }
7665         vcpu->arch.pio_data = page_address(page);
7666
7667         kvm_set_tsc_khz(vcpu, max_tsc_khz);
7668
7669         r = kvm_mmu_create(vcpu);
7670         if (r < 0)
7671                 goto fail_free_pio_data;
7672
7673         if (irqchip_in_kernel(kvm)) {
7674                 r = kvm_create_lapic(vcpu);
7675                 if (r < 0)
7676                         goto fail_mmu_destroy;
7677         } else
7678                 static_key_slow_inc(&kvm_no_apic_vcpu);
7679
7680         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7681                                        GFP_KERNEL);
7682         if (!vcpu->arch.mce_banks) {
7683                 r = -ENOMEM;
7684                 goto fail_free_lapic;
7685         }
7686         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7687
7688         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7689                 r = -ENOMEM;
7690                 goto fail_free_mce_banks;
7691         }
7692
7693         fx_init(vcpu);
7694
7695         vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7696         vcpu->arch.pv_time_enabled = false;
7697
7698         vcpu->arch.guest_supported_xcr0 = 0;
7699         vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7700
7701         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7702
7703         vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7704
7705         kvm_async_pf_hash_reset(vcpu);
7706         kvm_pmu_init(vcpu);
7707
7708         vcpu->arch.pending_external_vector = -1;
7709
7710         kvm_hv_vcpu_init(vcpu);
7711
7712         return 0;
7713
7714 fail_free_mce_banks:
7715         kfree(vcpu->arch.mce_banks);
7716 fail_free_lapic:
7717         kvm_free_lapic(vcpu);
7718 fail_mmu_destroy:
7719         kvm_mmu_destroy(vcpu);
7720 fail_free_pio_data:
7721         free_page((unsigned long)vcpu->arch.pio_data);
7722 fail:
7723         return r;
7724 }
7725
7726 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7727 {
7728         int idx;
7729
7730         kvm_hv_vcpu_uninit(vcpu);
7731         kvm_pmu_destroy(vcpu);
7732         kfree(vcpu->arch.mce_banks);
7733         kvm_free_lapic(vcpu);
7734         idx = srcu_read_lock(&vcpu->kvm->srcu);
7735         kvm_mmu_destroy(vcpu);
7736         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7737         free_page((unsigned long)vcpu->arch.pio_data);
7738         if (!lapic_in_kernel(vcpu))
7739                 static_key_slow_dec(&kvm_no_apic_vcpu);
7740 }
7741
7742 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7743 {
7744         kvm_x86_ops->sched_in(vcpu, cpu);
7745 }
7746
7747 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7748 {
7749         if (type)
7750                 return -EINVAL;
7751
7752         INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7753         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7754         INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7755         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7756         atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7757
7758         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7759         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7760         /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7761         set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7762                 &kvm->arch.irq_sources_bitmap);
7763
7764         raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7765         mutex_init(&kvm->arch.apic_map_lock);
7766         spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7767
7768         pvclock_update_vm_gtod_copy(kvm);
7769
7770         INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7771         INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7772
7773         kvm_page_track_init(kvm);
7774         kvm_mmu_init_vm(kvm);
7775
7776         if (kvm_x86_ops->vm_init)
7777                 return kvm_x86_ops->vm_init(kvm);
7778
7779         return 0;
7780 }
7781
7782 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7783 {
7784         int r;
7785         r = vcpu_load(vcpu);
7786         BUG_ON(r);
7787         kvm_mmu_unload(vcpu);
7788         vcpu_put(vcpu);
7789 }
7790
7791 static void kvm_free_vcpus(struct kvm *kvm)
7792 {
7793         unsigned int i;
7794         struct kvm_vcpu *vcpu;
7795
7796         /*
7797          * Unpin any mmu pages first.
7798          */
7799         kvm_for_each_vcpu(i, vcpu, kvm) {
7800                 kvm_clear_async_pf_completion_queue(vcpu);
7801                 kvm_unload_vcpu_mmu(vcpu);
7802         }
7803         kvm_for_each_vcpu(i, vcpu, kvm)
7804                 kvm_arch_vcpu_free(vcpu);
7805
7806         mutex_lock(&kvm->lock);
7807         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7808                 kvm->vcpus[i] = NULL;
7809
7810         atomic_set(&kvm->online_vcpus, 0);
7811         mutex_unlock(&kvm->lock);
7812 }
7813
7814 void kvm_arch_sync_events(struct kvm *kvm)
7815 {
7816         cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7817         cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7818         kvm_free_all_assigned_devices(kvm);
7819         kvm_free_pit(kvm);
7820 }
7821
7822 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7823 {
7824         int i, r;
7825         unsigned long hva;
7826         struct kvm_memslots *slots = kvm_memslots(kvm);
7827         struct kvm_memory_slot *slot, old;
7828
7829         /* Called with kvm->slots_lock held.  */
7830         if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7831                 return -EINVAL;
7832
7833         slot = id_to_memslot(slots, id);
7834         if (size) {
7835                 if (slot->npages)
7836                         return -EEXIST;
7837
7838                 /*
7839                  * MAP_SHARED to prevent internal slot pages from being moved
7840                  * by fork()/COW.
7841                  */
7842                 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7843                               MAP_SHARED | MAP_ANONYMOUS, 0);
7844                 if (IS_ERR((void *)hva))
7845                         return PTR_ERR((void *)hva);
7846         } else {
7847                 if (!slot->npages)
7848                         return 0;
7849
7850                 hva = 0;
7851         }
7852
7853         old = *slot;
7854         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7855                 struct kvm_userspace_memory_region m;
7856
7857                 m.slot = id | (i << 16);
7858                 m.flags = 0;
7859                 m.guest_phys_addr = gpa;
7860                 m.userspace_addr = hva;
7861                 m.memory_size = size;
7862                 r = __kvm_set_memory_region(kvm, &m);
7863                 if (r < 0)
7864                         return r;
7865         }
7866
7867         if (!size) {
7868                 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7869                 WARN_ON(r < 0);
7870         }
7871
7872         return 0;
7873 }
7874 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7875
7876 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7877 {
7878         int r;
7879
7880         mutex_lock(&kvm->slots_lock);
7881         r = __x86_set_memory_region(kvm, id, gpa, size);
7882         mutex_unlock(&kvm->slots_lock);
7883
7884         return r;
7885 }
7886 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7887
7888 void kvm_arch_destroy_vm(struct kvm *kvm)
7889 {
7890         if (current->mm == kvm->mm) {
7891                 /*
7892                  * Free memory regions allocated on behalf of userspace,
7893                  * unless the the memory map has changed due to process exit
7894                  * or fd copying.
7895                  */
7896                 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7897                 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7898                 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
7899         }
7900         if (kvm_x86_ops->vm_destroy)
7901                 kvm_x86_ops->vm_destroy(kvm);
7902         kvm_iommu_unmap_guest(kvm);
7903         kfree(kvm->arch.vpic);
7904         kfree(kvm->arch.vioapic);
7905         kvm_free_vcpus(kvm);
7906         kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7907         kvm_mmu_uninit_vm(kvm);
7908 }
7909
7910 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7911                            struct kvm_memory_slot *dont)
7912 {
7913         int i;
7914
7915         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7916                 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7917                         kvfree(free->arch.rmap[i]);
7918                         free->arch.rmap[i] = NULL;
7919                 }
7920                 if (i == 0)
7921                         continue;
7922
7923                 if (!dont || free->arch.lpage_info[i - 1] !=
7924                              dont->arch.lpage_info[i - 1]) {
7925                         kvfree(free->arch.lpage_info[i - 1]);
7926                         free->arch.lpage_info[i - 1] = NULL;
7927                 }
7928         }
7929
7930         kvm_page_track_free_memslot(free, dont);
7931 }
7932
7933 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7934                             unsigned long npages)
7935 {
7936         int i;
7937
7938         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7939                 struct kvm_lpage_info *linfo;
7940                 unsigned long ugfn;
7941                 int lpages;
7942                 int level = i + 1;
7943
7944                 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7945                                       slot->base_gfn, level) + 1;
7946
7947                 slot->arch.rmap[i] =
7948                         kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7949                 if (!slot->arch.rmap[i])
7950                         goto out_free;
7951                 if (i == 0)
7952                         continue;
7953
7954                 linfo = kvm_kvzalloc(lpages * sizeof(*linfo));
7955                 if (!linfo)
7956                         goto out_free;
7957
7958                 slot->arch.lpage_info[i - 1] = linfo;
7959
7960                 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7961                         linfo[0].disallow_lpage = 1;
7962                 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7963                         linfo[lpages - 1].disallow_lpage = 1;
7964                 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7965                 /*
7966                  * If the gfn and userspace address are not aligned wrt each
7967                  * other, or if explicitly asked to, disable large page
7968                  * support for this slot
7969                  */
7970                 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7971                     !kvm_largepages_enabled()) {
7972                         unsigned long j;
7973
7974                         for (j = 0; j < lpages; ++j)
7975                                 linfo[j].disallow_lpage = 1;
7976                 }
7977         }
7978
7979         if (kvm_page_track_create_memslot(slot, npages))
7980                 goto out_free;
7981
7982         return 0;
7983
7984 out_free:
7985         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7986                 kvfree(slot->arch.rmap[i]);
7987                 slot->arch.rmap[i] = NULL;
7988                 if (i == 0)
7989                         continue;
7990
7991                 kvfree(slot->arch.lpage_info[i - 1]);
7992                 slot->arch.lpage_info[i - 1] = NULL;
7993         }
7994         return -ENOMEM;
7995 }
7996
7997 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7998 {
7999         /*
8000          * memslots->generation has been incremented.
8001          * mmio generation may have reached its maximum value.
8002          */
8003         kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8004 }
8005
8006 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8007                                 struct kvm_memory_slot *memslot,
8008                                 const struct kvm_userspace_memory_region *mem,
8009                                 enum kvm_mr_change change)
8010 {
8011         return 0;
8012 }
8013
8014 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8015                                      struct kvm_memory_slot *new)
8016 {
8017         /* Still write protect RO slot */
8018         if (new->flags & KVM_MEM_READONLY) {
8019                 kvm_mmu_slot_remove_write_access(kvm, new);
8020                 return;
8021         }
8022
8023         /*
8024          * Call kvm_x86_ops dirty logging hooks when they are valid.
8025          *
8026          * kvm_x86_ops->slot_disable_log_dirty is called when:
8027          *
8028          *  - KVM_MR_CREATE with dirty logging is disabled
8029          *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8030          *
8031          * The reason is, in case of PML, we need to set D-bit for any slots
8032          * with dirty logging disabled in order to eliminate unnecessary GPA
8033          * logging in PML buffer (and potential PML buffer full VMEXT). This
8034          * guarantees leaving PML enabled during guest's lifetime won't have
8035          * any additonal overhead from PML when guest is running with dirty
8036          * logging disabled for memory slots.
8037          *
8038          * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8039          * to dirty logging mode.
8040          *
8041          * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8042          *
8043          * In case of write protect:
8044          *
8045          * Write protect all pages for dirty logging.
8046          *
8047          * All the sptes including the large sptes which point to this
8048          * slot are set to readonly. We can not create any new large
8049          * spte on this slot until the end of the logging.
8050          *
8051          * See the comments in fast_page_fault().
8052          */
8053         if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8054                 if (kvm_x86_ops->slot_enable_log_dirty)
8055                         kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8056                 else
8057                         kvm_mmu_slot_remove_write_access(kvm, new);
8058         } else {
8059                 if (kvm_x86_ops->slot_disable_log_dirty)
8060                         kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8061         }
8062 }
8063
8064 void kvm_arch_commit_memory_region(struct kvm *kvm,
8065                                 const struct kvm_userspace_memory_region *mem,
8066                                 const struct kvm_memory_slot *old,
8067                                 const struct kvm_memory_slot *new,
8068                                 enum kvm_mr_change change)
8069 {
8070         int nr_mmu_pages = 0;
8071
8072         if (!kvm->arch.n_requested_mmu_pages)
8073                 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8074
8075         if (nr_mmu_pages)
8076                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8077
8078         /*
8079          * Dirty logging tracks sptes in 4k granularity, meaning that large
8080          * sptes have to be split.  If live migration is successful, the guest
8081          * in the source machine will be destroyed and large sptes will be
8082          * created in the destination. However, if the guest continues to run
8083          * in the source machine (for example if live migration fails), small
8084          * sptes will remain around and cause bad performance.
8085          *
8086          * Scan sptes if dirty logging has been stopped, dropping those
8087          * which can be collapsed into a single large-page spte.  Later
8088          * page faults will create the large-page sptes.
8089          */
8090         if ((change != KVM_MR_DELETE) &&
8091                 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8092                 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8093                 kvm_mmu_zap_collapsible_sptes(kvm, new);
8094
8095         /*
8096          * Set up write protection and/or dirty logging for the new slot.
8097          *
8098          * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8099          * been zapped so no dirty logging staff is needed for old slot. For
8100          * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8101          * new and it's also covered when dealing with the new slot.
8102          *
8103          * FIXME: const-ify all uses of struct kvm_memory_slot.
8104          */
8105         if (change != KVM_MR_DELETE)
8106                 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8107 }
8108
8109 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8110 {
8111         kvm_mmu_invalidate_zap_all_pages(kvm);
8112 }
8113
8114 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8115                                    struct kvm_memory_slot *slot)
8116 {
8117         kvm_mmu_invalidate_zap_all_pages(kvm);
8118 }
8119
8120 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8121 {
8122         if (!list_empty_careful(&vcpu->async_pf.done))
8123                 return true;
8124
8125         if (kvm_apic_has_events(vcpu))
8126                 return true;
8127
8128         if (vcpu->arch.pv.pv_unhalted)
8129                 return true;
8130
8131         if (atomic_read(&vcpu->arch.nmi_queued))
8132                 return true;
8133
8134         if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8135                 return true;
8136
8137         if (kvm_arch_interrupt_allowed(vcpu) &&
8138             kvm_cpu_has_interrupt(vcpu))
8139                 return true;
8140
8141         if (kvm_hv_has_stimer_pending(vcpu))
8142                 return true;
8143
8144         return false;
8145 }
8146
8147 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8148 {
8149         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8150                 kvm_x86_ops->check_nested_events(vcpu, false);
8151
8152         return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8153 }
8154
8155 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8156 {
8157         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8158 }
8159
8160 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8161 {
8162         return kvm_x86_ops->interrupt_allowed(vcpu);
8163 }
8164
8165 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8166 {
8167         if (is_64_bit_mode(vcpu))
8168                 return kvm_rip_read(vcpu);
8169         return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8170                      kvm_rip_read(vcpu));
8171 }
8172 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8173
8174 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8175 {
8176         return kvm_get_linear_rip(vcpu) == linear_rip;
8177 }
8178 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8179
8180 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8181 {
8182         unsigned long rflags;
8183
8184         rflags = kvm_x86_ops->get_rflags(vcpu);
8185         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8186                 rflags &= ~X86_EFLAGS_TF;
8187         return rflags;
8188 }
8189 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8190
8191 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8192 {
8193         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8194             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8195                 rflags |= X86_EFLAGS_TF;
8196         kvm_x86_ops->set_rflags(vcpu, rflags);
8197 }
8198
8199 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8200 {
8201         __kvm_set_rflags(vcpu, rflags);
8202         kvm_make_request(KVM_REQ_EVENT, vcpu);
8203 }
8204 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8205
8206 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8207 {
8208         int r;
8209
8210         if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8211               work->wakeup_all)
8212                 return;
8213
8214         r = kvm_mmu_reload(vcpu);
8215         if (unlikely(r))
8216                 return;
8217
8218         if (!vcpu->arch.mmu.direct_map &&
8219               work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8220                 return;
8221
8222         vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8223 }
8224
8225 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8226 {
8227         return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8228 }
8229
8230 static inline u32 kvm_async_pf_next_probe(u32 key)
8231 {
8232         return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8233 }
8234
8235 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8236 {
8237         u32 key = kvm_async_pf_hash_fn(gfn);
8238
8239         while (vcpu->arch.apf.gfns[key] != ~0)
8240                 key = kvm_async_pf_next_probe(key);
8241
8242         vcpu->arch.apf.gfns[key] = gfn;
8243 }
8244
8245 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8246 {
8247         int i;
8248         u32 key = kvm_async_pf_hash_fn(gfn);
8249
8250         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8251                      (vcpu->arch.apf.gfns[key] != gfn &&
8252                       vcpu->arch.apf.gfns[key] != ~0); i++)
8253                 key = kvm_async_pf_next_probe(key);
8254
8255         return key;
8256 }
8257
8258 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8259 {
8260         return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8261 }
8262
8263 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8264 {
8265         u32 i, j, k;
8266
8267         i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8268         while (true) {
8269                 vcpu->arch.apf.gfns[i] = ~0;
8270                 do {
8271                         j = kvm_async_pf_next_probe(j);
8272                         if (vcpu->arch.apf.gfns[j] == ~0)
8273                                 return;
8274                         k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8275                         /*
8276                          * k lies cyclically in ]i,j]
8277                          * |    i.k.j |
8278                          * |....j i.k.| or  |.k..j i...|
8279                          */
8280                 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8281                 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8282                 i = j;
8283         }
8284 }
8285
8286 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8287 {
8288
8289         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8290                                       sizeof(val));
8291 }
8292
8293 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8294                                      struct kvm_async_pf *work)
8295 {
8296         struct x86_exception fault;
8297
8298         trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8299         kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8300
8301         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8302             (vcpu->arch.apf.send_user_only &&
8303              kvm_x86_ops->get_cpl(vcpu) == 0))
8304                 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8305         else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8306                 fault.vector = PF_VECTOR;
8307                 fault.error_code_valid = true;
8308                 fault.error_code = 0;
8309                 fault.nested_page_fault = false;
8310                 fault.address = work->arch.token;
8311                 kvm_inject_page_fault(vcpu, &fault);
8312         }
8313 }
8314
8315 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8316                                  struct kvm_async_pf *work)
8317 {
8318         struct x86_exception fault;
8319
8320         trace_kvm_async_pf_ready(work->arch.token, work->gva);
8321         if (work->wakeup_all)
8322                 work->arch.token = ~0; /* broadcast wakeup */
8323         else
8324                 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8325
8326         if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8327             !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8328                 fault.vector = PF_VECTOR;
8329                 fault.error_code_valid = true;
8330                 fault.error_code = 0;
8331                 fault.nested_page_fault = false;
8332                 fault.address = work->arch.token;
8333                 kvm_inject_page_fault(vcpu, &fault);
8334         }
8335         vcpu->arch.apf.halted = false;
8336         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8337 }
8338
8339 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8340 {
8341         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8342                 return true;
8343         else
8344                 return !kvm_event_needs_reinjection(vcpu) &&
8345                         kvm_x86_ops->interrupt_allowed(vcpu);
8346 }
8347
8348 void kvm_arch_start_assignment(struct kvm *kvm)
8349 {
8350         atomic_inc(&kvm->arch.assigned_device_count);
8351 }
8352 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8353
8354 void kvm_arch_end_assignment(struct kvm *kvm)
8355 {
8356         atomic_dec(&kvm->arch.assigned_device_count);
8357 }
8358 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8359
8360 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8361 {
8362         return atomic_read(&kvm->arch.assigned_device_count);
8363 }
8364 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8365
8366 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8367 {
8368         atomic_inc(&kvm->arch.noncoherent_dma_count);
8369 }
8370 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8371
8372 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8373 {
8374         atomic_dec(&kvm->arch.noncoherent_dma_count);
8375 }
8376 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8377
8378 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8379 {
8380         return atomic_read(&kvm->arch.noncoherent_dma_count);
8381 }
8382 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8383
8384 bool kvm_arch_has_irq_bypass(void)
8385 {
8386         return kvm_x86_ops->update_pi_irte != NULL;
8387 }
8388
8389 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8390                                       struct irq_bypass_producer *prod)
8391 {
8392         struct kvm_kernel_irqfd *irqfd =
8393                 container_of(cons, struct kvm_kernel_irqfd, consumer);
8394
8395         irqfd->producer = prod;
8396
8397         return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8398                                            prod->irq, irqfd->gsi, 1);
8399 }
8400
8401 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8402                                       struct irq_bypass_producer *prod)
8403 {
8404         int ret;
8405         struct kvm_kernel_irqfd *irqfd =
8406                 container_of(cons, struct kvm_kernel_irqfd, consumer);
8407
8408         WARN_ON(irqfd->producer != prod);
8409         irqfd->producer = NULL;
8410
8411         /*
8412          * When producer of consumer is unregistered, we change back to
8413          * remapped mode, so we can re-use the current implementation
8414          * when the irq is masked/disabled or the consumer side (KVM
8415          * int this case doesn't want to receive the interrupts.
8416         */
8417         ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8418         if (ret)
8419                 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8420                        " fails: %d\n", irqfd->consumer.token, ret);
8421 }
8422
8423 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8424                                    uint32_t guest_irq, bool set)
8425 {
8426         if (!kvm_x86_ops->update_pi_irte)
8427                 return -EINVAL;
8428
8429         return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8430 }
8431
8432 bool kvm_vector_hashing_enabled(void)
8433 {
8434         return vector_hashing;
8435 }
8436 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8437
8438 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8439 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8440 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8441 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8442 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8443 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8444 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8445 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8446 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8447 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8448 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8449 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8450 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8451 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8452 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8453 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8454 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8455 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8456 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);