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
  33 #include <linux/clocksource.h>
  34 #include <linux/interrupt.h>
  35 #include <linux/kvm.h>
  36 #include <linux/fs.h>
  37 #include <linux/vmalloc.h>
  38 #include <linux/module.h>
  39 #include <linux/mman.h>
  40 #include <linux/highmem.h>
  41 #include <linux/iommu.h>
  42 #include <linux/intel-iommu.h>
  43 #include <linux/cpufreq.h>
  44 #include <linux/user-return-notifier.h>
  45 #include <linux/srcu.h>
  46 #include <linux/slab.h>
  47 #include <linux/perf_event.h>
  48 #include <linux/uaccess.h>
  49 #include <linux/hash.h>
  50 #include <linux/pci.h>
  51 #include <linux/timekeeper_internal.h>
  52 #include <linux/pvclock_gtod.h>
  53 #include <trace/events/kvm.h>
  54
  55 #define CREATE_TRACE_POINTS
  56 #include "trace.h"
  57
  58 #include <asm/debugreg.h>
  59 #include <asm/msr.h>
  60 #include <asm/desc.h>
  61 #include <asm/mce.h>
  62 #include <linux/kernel_stat.h>
  63 #include <asm/fpu/internal.h> /* Ugh! */
  64 #include <asm/pvclock.h>
  65 #include <asm/div64.h>
  66
  67 #define MAX_IO_MSRS 256
  68 #define KVM_MAX_MCE_BANKS 32
  69 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
  70
  71 #define emul_to_vcpu(ctxt) \
  72         container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
  73
  74 /* EFER defaults:
  75  * - enable syscall per default because its emulated by KVM
  76  * - enable LME and LMA per default on 64 bit KVM
  77  */
  78 #ifdef CONFIG_X86_64
  79 static
  80 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
  81 #else
  82 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
  83 #endif
  84
  85 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  86 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  87
  88 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  89 static void process_nmi(struct kvm_vcpu *vcpu);
  90 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
  91
  92 struct kvm_x86_ops *kvm_x86_ops;
  93 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  94
  95 static bool ignore_msrs = 0;
  96 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
  97
  98 unsigned int min_timer_period_us = 500;
  99 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
 100
 101 static bool __read_mostly kvmclock_periodic_sync = true;
 102 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
 103
 104 bool kvm_has_tsc_control;
 105 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
 106 u32  kvm_max_guest_tsc_khz;
 107 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
 108
 109 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
 110 static u32 tsc_tolerance_ppm = 250;
 111 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
 112
 113 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
 114 unsigned int lapic_timer_advance_ns = 0;
 115 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
 116
 117 static bool backwards_tsc_observed = false;
 118
 119 #define KVM_NR_SHARED_MSRS 16
 120
 121 struct kvm_shared_msrs_global {
 122         int nr;
 123         u32 msrs[KVM_NR_SHARED_MSRS];
 124 };
 125
 126 struct kvm_shared_msrs {
 127         struct user_return_notifier urn;
 128         bool registered;
 129         struct kvm_shared_msr_values {
 130                 u64 host;
 131                 u64 curr;
 132         } values[KVM_NR_SHARED_MSRS];
 133 };
 134
 135 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 136 static struct kvm_shared_msrs __percpu *shared_msrs;
 137
 138 struct kvm_stats_debugfs_item debugfs_entries[] = {
 139         { "pf_fixed", VCPU_STAT(pf_fixed) },
 140         { "pf_guest", VCPU_STAT(pf_guest) },
 141         { "tlb_flush", VCPU_STAT(tlb_flush) },
 142         { "invlpg", VCPU_STAT(invlpg) },
 143         { "exits", VCPU_STAT(exits) },
 144         { "io_exits", VCPU_STAT(io_exits) },
 145         { "mmio_exits", VCPU_STAT(mmio_exits) },
 146         { "signal_exits", VCPU_STAT(signal_exits) },
 147         { "irq_window", VCPU_STAT(irq_window_exits) },
 148         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 149         { "halt_exits", VCPU_STAT(halt_exits) },
 150         { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
 151         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 152         { "hypercalls", VCPU_STAT(hypercalls) },
 153         { "request_irq", VCPU_STAT(request_irq_exits) },
 154         { "irq_exits", VCPU_STAT(irq_exits) },
 155         { "host_state_reload", VCPU_STAT(host_state_reload) },
 156         { "efer_reload", VCPU_STAT(efer_reload) },
 157         { "fpu_reload", VCPU_STAT(fpu_reload) },
 158         { "insn_emulation", VCPU_STAT(insn_emulation) },
 159         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 160         { "irq_injections", VCPU_STAT(irq_injections) },
 161         { "nmi_injections", VCPU_STAT(nmi_injections) },
 162         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 163         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 164         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 165         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 166         { "mmu_flooded", VM_STAT(mmu_flooded) },
 167         { "mmu_recycled", VM_STAT(mmu_recycled) },
 168         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 169         { "mmu_unsync", VM_STAT(mmu_unsync) },
 170         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 171         { "largepages", VM_STAT(lpages) },
 172         { NULL }
 173 };
 174
 175 u64 __read_mostly host_xcr0;
 176
 177 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
 178
 179 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
 180 {
 181         int i;
 182         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
 183                 vcpu->arch.apf.gfns[i] = ~0;
 184 }
 185
 186 static void kvm_on_user_return(struct user_return_notifier *urn)
 187 {
 188         unsigned slot;
 189         struct kvm_shared_msrs *locals
 190                 = container_of(urn, struct kvm_shared_msrs, urn);
 191         struct kvm_shared_msr_values *values;
 192
 193         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 194                 values = &locals->values[slot];
 195                 if (values->host != values->curr) {
 196                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 197                         values->curr = values->host;
 198                 }
 199         }
 200         locals->registered = false;
 201         user_return_notifier_unregister(urn);
 202 }
 203
 204 static void shared_msr_update(unsigned slot, u32 msr)
 205 {
 206         u64 value;
 207         unsigned int cpu = smp_processor_id();
 208         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 209
 210         /* only read, and nobody should modify it at this time,
 211          * so don't need lock */
 212         if (slot >= shared_msrs_global.nr) {
 213                 printk(KERN_ERR "kvm: invalid MSR slot!");
 214                 return;
 215         }
 216         rdmsrl_safe(msr, &value);
 217         smsr->values[slot].host = value;
 218         smsr->values[slot].curr = value;
 219 }
 220
 221 void kvm_define_shared_msr(unsigned slot, u32 msr)
 222 {
 223         BUG_ON(slot >= KVM_NR_SHARED_MSRS);
 224         if (slot >= shared_msrs_global.nr)
 225                 shared_msrs_global.nr = slot + 1;
 226         shared_msrs_global.msrs[slot] = msr;
 227         /* we need ensured the shared_msr_global have been updated */
 228         smp_wmb();
 229 }
 230 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 231
 232 static void kvm_shared_msr_cpu_online(void)
 233 {
 234         unsigned i;
 235
 236         for (i = 0; i < shared_msrs_global.nr; ++i)
 237                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 238 }
 239
 240 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 241 {
 242         unsigned int cpu = smp_processor_id();
 243         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 244         int err;
 245
 246         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 247                 return 0;
 248         smsr->values[slot].curr = value;
 249         err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
 250         if (err)
 251                 return 1;
 252
 253         if (!smsr->registered) {
 254                 smsr->urn.on_user_return = kvm_on_user_return;
 255                 user_return_notifier_register(&smsr->urn);
 256                 smsr->registered = true;
 257         }
 258         return 0;
 259 }
 260 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 261
 262 static void drop_user_return_notifiers(void)
 263 {
 264         unsigned int cpu = smp_processor_id();
 265         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 266
 267         if (smsr->registered)
 268                 kvm_on_user_return(&smsr->urn);
 269 }
 270
 271 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 272 {
 273         return vcpu->arch.apic_base;
 274 }
 275 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 276
 277 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 278 {
 279         u64 old_state = vcpu->arch.apic_base &
 280                 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
 281         u64 new_state = msr_info->data &
 282                 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
 283         u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
 284                 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
 285
 286         if (!msr_info->host_initiated &&
 287             ((msr_info->data & reserved_bits) != 0 ||
 288              new_state == X2APIC_ENABLE ||
 289              (new_state == MSR_IA32_APICBASE_ENABLE &&
 290               old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
 291              (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
 292               old_state == 0)))
 293                 return 1;
 294
 295         kvm_lapic_set_base(vcpu, msr_info->data);
 296         return 0;
 297 }
 298 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 299
 300 asmlinkage __visible void kvm_spurious_fault(void)
 301 {
 302         /* Fault while not rebooting.  We want the trace. */
 303         BUG();
 304 }
 305 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
 306
 307 #define EXCPT_BENIGN            0
 308 #define EXCPT_CONTRIBUTORY      1
 309 #define EXCPT_PF                2
 310
 311 static int exception_class(int vector)
 312 {
 313         switch (vector) {
 314         case PF_VECTOR:
 315                 return EXCPT_PF;
 316         case DE_VECTOR:
 317         case TS_VECTOR:
 318         case NP_VECTOR:
 319         case SS_VECTOR:
 320         case GP_VECTOR:
 321                 return EXCPT_CONTRIBUTORY;
 322         default:
 323                 break;
 324         }
 325         return EXCPT_BENIGN;
 326 }
 327
 328 #define EXCPT_FAULT             0
 329 #define EXCPT_TRAP              1
 330 #define EXCPT_ABORT             2
 331 #define EXCPT_INTERRUPT         3
 332
 333 static int exception_type(int vector)
 334 {
 335         unsigned int mask;
 336
 337         if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
 338                 return EXCPT_INTERRUPT;
 339
 340         mask = 1 << vector;
 341
 342         /* #DB is trap, as instruction watchpoints are handled elsewhere */
 343         if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
 344                 return EXCPT_TRAP;
 345
 346         if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
 347                 return EXCPT_ABORT;
 348
 349         /* Reserved exceptions will result in fault */
 350         return EXCPT_FAULT;
 351 }
 352
 353 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 354                 unsigned nr, bool has_error, u32 error_code,
 355                 bool reinject)
 356 {
 357         u32 prev_nr;
 358         int class1, class2;
 359
 360         kvm_make_request(KVM_REQ_EVENT, vcpu);
 361
 362         if (!vcpu->arch.exception.pending) {
 363         queue:
 364                 if (has_error && !is_protmode(vcpu))
 365                         has_error = false;
 366                 vcpu->arch.exception.pending = true;
 367                 vcpu->arch.exception.has_error_code = has_error;
 368                 vcpu->arch.exception.nr = nr;
 369                 vcpu->arch.exception.error_code = error_code;
 370                 vcpu->arch.exception.reinject = reinject;
 371                 return;
 372         }
 373
 374         /* to check exception */
 375         prev_nr = vcpu->arch.exception.nr;
 376         if (prev_nr == DF_VECTOR) {
 377                 /* triple fault -> shutdown */
 378                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
 379                 return;
 380         }
 381         class1 = exception_class(prev_nr);
 382         class2 = exception_class(nr);
 383         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 384                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 385                 /* generate double fault per SDM Table 5-5 */
 386                 vcpu->arch.exception.pending = true;
 387                 vcpu->arch.exception.has_error_code = true;
 388                 vcpu->arch.exception.nr = DF_VECTOR;
 389                 vcpu->arch.exception.error_code = 0;
 390         } else
 391                 /* replace previous exception with a new one in a hope
 392                    that instruction re-execution will regenerate lost
 393                    exception */
 394                 goto queue;
 395 }
 396
 397 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 398 {
 399         kvm_multiple_exception(vcpu, nr, false, 0, false);
 400 }
 401 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 402
 403 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 404 {
 405         kvm_multiple_exception(vcpu, nr, false, 0, true);
 406 }
 407 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 408
 409 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
 410 {
 411         if (err)
 412                 kvm_inject_gp(vcpu, 0);
 413         else
 414                 kvm_x86_ops->skip_emulated_instruction(vcpu);
 415 }
 416 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
 417
 418 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 419 {
 420         ++vcpu->stat.pf_guest;
 421         vcpu->arch.cr2 = fault->address;
 422         kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
 423 }
 424 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
 425
 426 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 427 {
 428         if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
 429                 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
 430         else
 431                 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
 432
 433         return fault->nested_page_fault;
 434 }
 435
 436 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 437 {
 438         atomic_inc(&vcpu->arch.nmi_queued);
 439         kvm_make_request(KVM_REQ_NMI, vcpu);
 440 }
 441 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 442
 443 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 444 {
 445         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 446 }
 447 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 448
 449 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 450 {
 451         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 452 }
 453 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 454
 455 /*
 456  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 457  * a #GP and return false.
 458  */
 459 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 460 {
 461         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 462                 return true;
 463         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 464         return false;
 465 }
 466 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 467
 468 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
 469 {
 470         if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 471                 return true;
 472
 473         kvm_queue_exception(vcpu, UD_VECTOR);
 474         return false;
 475 }
 476 EXPORT_SYMBOL_GPL(kvm_require_dr);
 477
 478 /*
 479  * This function will be used to read from the physical memory of the currently
 480  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
 481  * can read from guest physical or from the guest's guest physical memory.
 482  */
 483 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 484                             gfn_t ngfn, void *data, int offset, int len,
 485                             u32 access)
 486 {
 487         struct x86_exception exception;
 488         gfn_t real_gfn;
 489         gpa_t ngpa;
 490
 491         ngpa     = gfn_to_gpa(ngfn);
 492         real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
 493         if (real_gfn == UNMAPPED_GVA)
 494                 return -EFAULT;
 495
 496         real_gfn = gpa_to_gfn(real_gfn);
 497
 498         return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
 499 }
 500 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
 501
 502 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
 503                                void *data, int offset, int len, u32 access)
 504 {
 505         return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
 506                                        data, offset, len, access);
 507 }
 508
 509 /*
 510  * Load the pae pdptrs.  Return true is they are all valid.
 511  */
 512 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
 513 {
 514         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 515         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 516         int i;
 517         int ret;
 518         u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
 519
 520         ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
 521                                       offset * sizeof(u64), sizeof(pdpte),
 522                                       PFERR_USER_MASK|PFERR_WRITE_MASK);
 523         if (ret < 0) {
 524                 ret = 0;
 525                 goto out;
 526         }
 527         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 528                 if (is_present_gpte(pdpte[i]) &&
 529                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 530                         ret = 0;
 531                         goto out;
 532                 }
 533         }
 534         ret = 1;
 535
 536         memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
 537         __set_bit(VCPU_EXREG_PDPTR,
 538                   (unsigned long *)&vcpu->arch.regs_avail);
 539         __set_bit(VCPU_EXREG_PDPTR,
 540                   (unsigned long *)&vcpu->arch.regs_dirty);
 541 out:
 542
 543         return ret;
 544 }
 545 EXPORT_SYMBOL_GPL(load_pdptrs);
 546
 547 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 548 {
 549         u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
 550         bool changed = true;
 551         int offset;
 552         gfn_t gfn;
 553         int r;
 554
 555         if (is_long_mode(vcpu) || !is_pae(vcpu))
 556                 return false;
 557
 558         if (!test_bit(VCPU_EXREG_PDPTR,
 559                       (unsigned long *)&vcpu->arch.regs_avail))
 560                 return true;
 561
 562         gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
 563         offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
 564         r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
 565                                        PFERR_USER_MASK | PFERR_WRITE_MASK);
 566         if (r < 0)
 567                 goto out;
 568         changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
 569 out:
 570
 571         return changed;
 572 }
 573
 574 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 575 {
 576         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 577         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
 578
 579         cr0 |= X86_CR0_ET;
 580
 581 #ifdef CONFIG_X86_64
 582         if (cr0 & 0xffffffff00000000UL)
 583                 return 1;
 584 #endif
 585
 586         cr0 &= ~CR0_RESERVED_BITS;
 587
 588         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 589                 return 1;
 590
 591         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 592                 return 1;
 593
 594         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 595 #ifdef CONFIG_X86_64
 596                 if ((vcpu->arch.efer & EFER_LME)) {
 597                         int cs_db, cs_l;
 598
 599                         if (!is_pae(vcpu))
 600                                 return 1;
 601                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 602                         if (cs_l)
 603                                 return 1;
 604                 } else
 605 #endif
 606                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 607                                                  kvm_read_cr3(vcpu)))
 608                         return 1;
 609         }
 610
 611         if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
 612                 return 1;
 613
 614         kvm_x86_ops->set_cr0(vcpu, cr0);
 615
 616         if ((cr0 ^ old_cr0) & X86_CR0_PG) {
 617                 kvm_clear_async_pf_completion_queue(vcpu);
 618                 kvm_async_pf_hash_reset(vcpu);
 619         }
 620
 621         if ((cr0 ^ old_cr0) & update_bits)
 622                 kvm_mmu_reset_context(vcpu);
 623
 624         if ((cr0 ^ old_cr0) & X86_CR0_CD)
 625                 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
 626
 627         return 0;
 628 }
 629 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 630
 631 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 632 {
 633         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 634 }
 635 EXPORT_SYMBOL_GPL(kvm_lmsw);
 636
 637 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
 638 {
 639         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
 640                         !vcpu->guest_xcr0_loaded) {
 641                 /* kvm_set_xcr() also depends on this */
 642                 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
 643                 vcpu->guest_xcr0_loaded = 1;
 644         }
 645 }
 646
 647 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
 648 {
 649         if (vcpu->guest_xcr0_loaded) {
 650                 if (vcpu->arch.xcr0 != host_xcr0)
 651                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
 652                 vcpu->guest_xcr0_loaded = 0;
 653         }
 654 }
 655
 656 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 657 {
 658         u64 xcr0 = xcr;
 659         u64 old_xcr0 = vcpu->arch.xcr0;
 660         u64 valid_bits;
 661
 662         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
 663         if (index != XCR_XFEATURE_ENABLED_MASK)
 664                 return 1;
 665         if (!(xcr0 & XSTATE_FP))
 666                 return 1;
 667         if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
 668                 return 1;
 669
 670         /*
 671          * Do not allow the guest to set bits that we do not support
 672          * saving.  However, xcr0 bit 0 is always set, even if the
 673          * emulated CPU does not support XSAVE (see fx_init).
 674          */
 675         valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
 676         if (xcr0 & ~valid_bits)
 677                 return 1;
 678
 679         if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
 680                 return 1;
 681
 682         if (xcr0 & XSTATE_AVX512) {
 683                 if (!(xcr0 & XSTATE_YMM))
 684                         return 1;
 685                 if ((xcr0 & XSTATE_AVX512) != XSTATE_AVX512)
 686                         return 1;
 687         }
 688         kvm_put_guest_xcr0(vcpu);
 689         vcpu->arch.xcr0 = xcr0;
 690
 691         if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
 692                 kvm_update_cpuid(vcpu);
 693         return 0;
 694 }
 695
 696 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 697 {
 698         if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
 699             __kvm_set_xcr(vcpu, index, xcr)) {
 700                 kvm_inject_gp(vcpu, 0);
 701                 return 1;
 702         }
 703         return 0;
 704 }
 705 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 706
 707 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 708 {
 709         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 710         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
 711                                    X86_CR4_SMEP | X86_CR4_SMAP;
 712
 713         if (cr4 & CR4_RESERVED_BITS)
 714                 return 1;
 715
 716         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
 717                 return 1;
 718
 719         if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
 720                 return 1;
 721
 722         if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
 723                 return 1;
 724
 725         if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
 726                 return 1;
 727
 728         if (is_long_mode(vcpu)) {
 729                 if (!(cr4 & X86_CR4_PAE))
 730                         return 1;
 731         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 732                    && ((cr4 ^ old_cr4) & pdptr_bits)
 733                    && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 734                                    kvm_read_cr3(vcpu)))
 735                 return 1;
 736
 737         if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
 738                 if (!guest_cpuid_has_pcid(vcpu))
 739                         return 1;
 740
 741                 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
 742                 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
 743                         return 1;
 744         }
 745
 746         if (kvm_x86_ops->set_cr4(vcpu, cr4))
 747                 return 1;
 748
 749         if (((cr4 ^ old_cr4) & pdptr_bits) ||
 750             (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
 751                 kvm_mmu_reset_context(vcpu);
 752
 753         if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
 754                 kvm_update_cpuid(vcpu);
 755
 756         return 0;
 757 }
 758 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 759
 760 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 761 {
 762 #ifdef CONFIG_X86_64
 763         cr3 &= ~CR3_PCID_INVD;
 764 #endif
 765
 766         if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
 767                 kvm_mmu_sync_roots(vcpu);
 768                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 769                 return 0;
 770         }
 771
 772         if (is_long_mode(vcpu)) {
 773                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 774                         return 1;
 775         } else if (is_pae(vcpu) && is_paging(vcpu) &&
 776                    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 777                 return 1;
 778
 779         vcpu->arch.cr3 = cr3;
 780         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
 781         kvm_mmu_new_cr3(vcpu);
 782         return 0;
 783 }
 784 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 785
 786 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 787 {
 788         if (cr8 & CR8_RESERVED_BITS)
 789                 return 1;
 790         if (irqchip_in_kernel(vcpu->kvm))
 791                 kvm_lapic_set_tpr(vcpu, cr8);
 792         else
 793                 vcpu->arch.cr8 = cr8;
 794         return 0;
 795 }
 796 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 797
 798 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 799 {
 800         if (irqchip_in_kernel(vcpu->kvm))
 801                 return kvm_lapic_get_cr8(vcpu);
 802         else
 803                 return vcpu->arch.cr8;
 804 }
 805 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 806
 807 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
 808 {
 809         int i;
 810
 811         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 812                 for (i = 0; i < KVM_NR_DB_REGS; i++)
 813                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
 814                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
 815         }
 816 }
 817
 818 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
 819 {
 820         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 821                 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
 822 }
 823
 824 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
 825 {
 826         unsigned long dr7;
 827
 828         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 829                 dr7 = vcpu->arch.guest_debug_dr7;
 830         else
 831                 dr7 = vcpu->arch.dr7;
 832         kvm_x86_ops->set_dr7(vcpu, dr7);
 833         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
 834         if (dr7 & DR7_BP_EN_MASK)
 835                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
 836 }
 837
 838 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
 839 {
 840         u64 fixed = DR6_FIXED_1;
 841
 842         if (!guest_cpuid_has_rtm(vcpu))
 843                 fixed |= DR6_RTM;
 844         return fixed;
 845 }
 846
 847 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 848 {
 849         switch (dr) {
 850         case 0 ... 3:
 851                 vcpu->arch.db[dr] = val;
 852                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 853                         vcpu->arch.eff_db[dr] = val;
 854                 break;
 855         case 4:
 856                 /* fall through */
 857         case 6:
 858                 if (val & 0xffffffff00000000ULL)
 859                         return -1; /* #GP */
 860                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
 861                 kvm_update_dr6(vcpu);
 862                 break;
 863         case 5:
 864                 /* fall through */
 865         default: /* 7 */
 866                 if (val & 0xffffffff00000000ULL)
 867                         return -1; /* #GP */
 868                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 869                 kvm_update_dr7(vcpu);
 870                 break;
 871         }
 872
 873         return 0;
 874 }
 875
 876 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 877 {
 878         if (__kvm_set_dr(vcpu, dr, val)) {
 879                 kvm_inject_gp(vcpu, 0);
 880                 return 1;
 881         }
 882         return 0;
 883 }
 884 EXPORT_SYMBOL_GPL(kvm_set_dr);
 885
 886 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 887 {
 888         switch (dr) {
 889         case 0 ... 3:
 890                 *val = vcpu->arch.db[dr];
 891                 break;
 892         case 4:
 893                 /* fall through */
 894         case 6:
 895                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 896                         *val = vcpu->arch.dr6;
 897                 else
 898                         *val = kvm_x86_ops->get_dr6(vcpu);
 899                 break;
 900         case 5:
 901                 /* fall through */
 902         default: /* 7 */
 903                 *val = vcpu->arch.dr7;
 904                 break;
 905         }
 906         return 0;
 907 }
 908 EXPORT_SYMBOL_GPL(kvm_get_dr);
 909
 910 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
 911 {
 912         u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 913         u64 data;
 914         int err;
 915
 916         err = kvm_pmu_rdpmc(vcpu, ecx, &data);
 917         if (err)
 918                 return err;
 919         kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
 920         kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
 921         return err;
 922 }
 923 EXPORT_SYMBOL_GPL(kvm_rdpmc);
 924
 925 /*
 926  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 927  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 928  *
 929  * This list is modified at module load time to reflect the
 930  * capabilities of the host cpu. This capabilities test skips MSRs that are
 931  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
 932  * may depend on host virtualization features rather than host cpu features.
 933  */
 934
 935 static u32 msrs_to_save[] = {
 936         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 937         MSR_STAR,
 938 #ifdef CONFIG_X86_64
 939         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 940 #endif
 941         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
 942         MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
 943 };
 944
 945 static unsigned num_msrs_to_save;
 946
 947 static u32 emulated_msrs[] = {
 948         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 949         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 950         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 951         HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
 952         HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
 953         MSR_KVM_PV_EOI_EN,
 954
 955         MSR_IA32_TSC_ADJUST,
 956         MSR_IA32_TSCDEADLINE,
 957         MSR_IA32_MISC_ENABLE,
 958         MSR_IA32_MCG_STATUS,
 959         MSR_IA32_MCG_CTL,
 960         MSR_IA32_SMBASE,
 961 };
 962
 963 static unsigned num_emulated_msrs;
 964
 965 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
 966 {
 967         if (efer & efer_reserved_bits)
 968                 return false;
 969
 970         if (efer & EFER_FFXSR) {
 971                 struct kvm_cpuid_entry2 *feat;
 972
 973                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 974                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
 975                         return false;
 976         }
 977
 978         if (efer & EFER_SVME) {
 979                 struct kvm_cpuid_entry2 *feat;
 980
 981                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 982                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
 983                         return false;
 984         }
 985
 986         return true;
 987 }
 988 EXPORT_SYMBOL_GPL(kvm_valid_efer);
 989
 990 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
 991 {
 992         u64 old_efer = vcpu->arch.efer;
 993
 994         if (!kvm_valid_efer(vcpu, efer))
 995                 return 1;
 996
 997         if (is_paging(vcpu)
 998             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
 999                 return 1;
1000
1001         efer &= ~EFER_LMA;
1002         efer |= vcpu->arch.efer & EFER_LMA;
1003
1004         kvm_x86_ops->set_efer(vcpu, efer);
1005
1006         /* Update reserved bits */
1007         if ((efer ^ old_efer) & EFER_NX)
1008                 kvm_mmu_reset_context(vcpu);
1009
1010         return 0;
1011 }
1012
1013 void kvm_enable_efer_bits(u64 mask)
1014 {
1015        efer_reserved_bits &= ~mask;
1016 }
1017 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1018
1019 /*
1020  * Writes msr value into into the appropriate "register".
1021  * Returns 0 on success, non-0 otherwise.
1022  * Assumes vcpu_load() was already called.
1023  */
1024 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1025 {
1026         switch (msr->index) {
1027         case MSR_FS_BASE:
1028         case MSR_GS_BASE:
1029         case MSR_KERNEL_GS_BASE:
1030         case MSR_CSTAR:
1031         case MSR_LSTAR:
1032                 if (is_noncanonical_address(msr->data))
1033                         return 1;
1034                 break;
1035         case MSR_IA32_SYSENTER_EIP:
1036         case MSR_IA32_SYSENTER_ESP:
1037                 /*
1038                  * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1039                  * non-canonical address is written on Intel but not on
1040                  * AMD (which ignores the top 32-bits, because it does
1041                  * not implement 64-bit SYSENTER).
1042                  *
1043                  * 64-bit code should hence be able to write a non-canonical
1044                  * value on AMD.  Making the address canonical ensures that
1045                  * vmentry does not fail on Intel after writing a non-canonical
1046                  * value, and that something deterministic happens if the guest
1047                  * invokes 64-bit SYSENTER.
1048                  */
1049                 msr->data = get_canonical(msr->data);
1050         }
1051         return kvm_x86_ops->set_msr(vcpu, msr);
1052 }
1053 EXPORT_SYMBOL_GPL(kvm_set_msr);
1054
1055 /*
1056  * Adapt set_msr() to msr_io()'s calling convention
1057  */
1058 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1059 {
1060         struct msr_data msr;
1061         int r;
1062
1063         msr.index = index;
1064         msr.host_initiated = true;
1065         r = kvm_get_msr(vcpu, &msr);
1066         if (r)
1067                 return r;
1068
1069         *data = msr.data;
1070         return 0;
1071 }
1072
1073 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1074 {
1075         struct msr_data msr;
1076
1077         msr.data = *data;
1078         msr.index = index;
1079         msr.host_initiated = true;
1080         return kvm_set_msr(vcpu, &msr);
1081 }
1082
1083 #ifdef CONFIG_X86_64
1084 struct pvclock_gtod_data {
1085         seqcount_t      seq;
1086
1087         struct { /* extract of a clocksource struct */
1088                 int vclock_mode;
1089                 cycle_t cycle_last;
1090                 cycle_t mask;
1091                 u32     mult;
1092                 u32     shift;
1093         } clock;
1094
1095         u64             boot_ns;
1096         u64             nsec_base;
1097 };
1098
1099 static struct pvclock_gtod_data pvclock_gtod_data;
1100
1101 static void update_pvclock_gtod(struct timekeeper *tk)
1102 {
1103         struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1104         u64 boot_ns;
1105
1106         boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1107
1108         write_seqcount_begin(&vdata->seq);
1109
1110         /* copy pvclock gtod data */
1111         vdata->clock.vclock_mode        = tk->tkr_mono.clock->archdata.vclock_mode;
1112         vdata->clock.cycle_last         = tk->tkr_mono.cycle_last;
1113         vdata->clock.mask               = tk->tkr_mono.mask;
1114         vdata->clock.mult               = tk->tkr_mono.mult;
1115         vdata->clock.shift              = tk->tkr_mono.shift;
1116
1117         vdata->boot_ns                  = boot_ns;
1118         vdata->nsec_base                = tk->tkr_mono.xtime_nsec;
1119
1120         write_seqcount_end(&vdata->seq);
1121 }
1122 #endif
1123
1124 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1125 {
1126         /*
1127          * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1128          * vcpu_enter_guest.  This function is only called from
1129          * the physical CPU that is running vcpu.
1130          */
1131         kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1132 }
1133
1134 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1135 {
1136         int version;
1137         int r;
1138         struct pvclock_wall_clock wc;
1139         struct timespec boot;
1140
1141         if (!wall_clock)
1142                 return;
1143
1144         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1145         if (r)
1146                 return;
1147
1148         if (version & 1)
1149                 ++version;  /* first time write, random junk */
1150
1151         ++version;
1152
1153         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1154
1155         /*
1156          * The guest calculates current wall clock time by adding
1157          * system time (updated by kvm_guest_time_update below) to the
1158          * wall clock specified here.  guest system time equals host
1159          * system time for us, thus we must fill in host boot time here.
1160          */
1161         getboottime(&boot);
1162
1163         if (kvm->arch.kvmclock_offset) {
1164                 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1165                 boot = timespec_sub(boot, ts);
1166         }
1167         wc.sec = boot.tv_sec;
1168         wc.nsec = boot.tv_nsec;
1169         wc.version = version;
1170
1171         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1172
1173         version++;
1174         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1175 }
1176
1177 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1178 {
1179         uint32_t quotient, remainder;
1180
1181         /* Don't try to replace with do_div(), this one calculates
1182          * "(dividend << 32) / divisor" */
1183         __asm__ ( "divl %4"
1184                   : "=a" (quotient), "=d" (remainder)
1185                   : "0" (0), "1" (dividend), "r" (divisor) );
1186         return quotient;
1187 }
1188
1189 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1190                                s8 *pshift, u32 *pmultiplier)
1191 {
1192         uint64_t scaled64;
1193         int32_t  shift = 0;
1194         uint64_t tps64;
1195         uint32_t tps32;
1196
1197         tps64 = base_khz * 1000LL;
1198         scaled64 = scaled_khz * 1000LL;
1199         while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1200                 tps64 >>= 1;
1201                 shift--;
1202         }
1203
1204         tps32 = (uint32_t)tps64;
1205         while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1206                 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1207                         scaled64 >>= 1;
1208                 else
1209                         tps32 <<= 1;
1210                 shift++;
1211         }
1212
1213         *pshift = shift;
1214         *pmultiplier = div_frac(scaled64, tps32);
1215
1216         pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1217                  __func__, base_khz, scaled_khz, shift, *pmultiplier);
1218 }
1219
1220 static inline u64 get_kernel_ns(void)
1221 {
1222         return ktime_get_boot_ns();
1223 }
1224
1225 #ifdef CONFIG_X86_64
1226 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1227 #endif
1228
1229 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1230 static unsigned long max_tsc_khz;
1231
1232 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1233 {
1234         return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1235                                    vcpu->arch.virtual_tsc_shift);
1236 }
1237
1238 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1239 {
1240         u64 v = (u64)khz * (1000000 + ppm);
1241         do_div(v, 1000000);
1242         return v;
1243 }
1244
1245 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1246 {
1247         u32 thresh_lo, thresh_hi;
1248         int use_scaling = 0;
1249
1250         /* tsc_khz can be zero if TSC calibration fails */
1251         if (this_tsc_khz == 0)
1252                 return;
1253
1254         /* Compute a scale to convert nanoseconds in TSC cycles */
1255         kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1256                            &vcpu->arch.virtual_tsc_shift,
1257                            &vcpu->arch.virtual_tsc_mult);
1258         vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1259
1260         /*
1261          * Compute the variation in TSC rate which is acceptable
1262          * within the range of tolerance and decide if the
1263          * rate being applied is within that bounds of the hardware
1264          * rate.  If so, no scaling or compensation need be done.
1265          */
1266         thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1267         thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1268         if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1269                 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1270                 use_scaling = 1;
1271         }
1272         kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1273 }
1274
1275 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1276 {
1277         u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1278                                       vcpu->arch.virtual_tsc_mult,
1279                                       vcpu->arch.virtual_tsc_shift);
1280         tsc += vcpu->arch.this_tsc_write;
1281         return tsc;
1282 }
1283
1284 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1285 {
1286 #ifdef CONFIG_X86_64
1287         bool vcpus_matched;
1288         struct kvm_arch *ka = &vcpu->kvm->arch;
1289         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1290
1291         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1292                          atomic_read(&vcpu->kvm->online_vcpus));
1293
1294         /*
1295          * Once the masterclock is enabled, always perform request in
1296          * order to update it.
1297          *
1298          * In order to enable masterclock, the host clocksource must be TSC
1299          * and the vcpus need to have matched TSCs.  When that happens,
1300          * perform request to enable masterclock.
1301          */
1302         if (ka->use_master_clock ||
1303             (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1304                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1305
1306         trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1307                             atomic_read(&vcpu->kvm->online_vcpus),
1308                             ka->use_master_clock, gtod->clock.vclock_mode);
1309 #endif
1310 }
1311
1312 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1313 {
1314         u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1315         vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1316 }
1317
1318 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1319 {
1320         struct kvm *kvm = vcpu->kvm;
1321         u64 offset, ns, elapsed;
1322         unsigned long flags;
1323         s64 usdiff;
1324         bool matched;
1325         bool already_matched;
1326         u64 data = msr->data;
1327
1328         raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1329         offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1330         ns = get_kernel_ns();
1331         elapsed = ns - kvm->arch.last_tsc_nsec;
1332
1333         if (vcpu->arch.virtual_tsc_khz) {
1334                 int faulted = 0;
1335
1336                 /* n.b - signed multiplication and division required */
1337                 usdiff = data - kvm->arch.last_tsc_write;
1338 #ifdef CONFIG_X86_64
1339                 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1340 #else
1341                 /* do_div() only does unsigned */
1342                 asm("1: idivl %[divisor]\n"
1343                     "2: xor %%edx, %%edx\n"
1344                     "   movl $0, %[faulted]\n"
1345                     "3:\n"
1346                     ".section .fixup,\"ax\"\n"
1347                     "4: movl $1, %[faulted]\n"
1348                     "   jmp  3b\n"
1349                     ".previous\n"
1350
1351                 _ASM_EXTABLE(1b, 4b)
1352
1353                 : "=A"(usdiff), [faulted] "=r" (faulted)
1354                 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1355
1356 #endif
1357                 do_div(elapsed, 1000);
1358                 usdiff -= elapsed;
1359                 if (usdiff < 0)
1360                         usdiff = -usdiff;
1361
1362                 /* idivl overflow => difference is larger than USEC_PER_SEC */
1363                 if (faulted)
1364                         usdiff = USEC_PER_SEC;
1365         } else
1366                 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1367
1368         /*
1369          * Special case: TSC write with a small delta (1 second) of virtual
1370          * cycle time against real time is interpreted as an attempt to
1371          * synchronize the CPU.
1372          *
1373          * For a reliable TSC, we can match TSC offsets, and for an unstable
1374          * TSC, we add elapsed time in this computation.  We could let the
1375          * compensation code attempt to catch up if we fall behind, but
1376          * it's better to try to match offsets from the beginning.
1377          */
1378         if (usdiff < USEC_PER_SEC &&
1379             vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1380                 if (!check_tsc_unstable()) {
1381                         offset = kvm->arch.cur_tsc_offset;
1382                         pr_debug("kvm: matched tsc offset for %llu\n", data);
1383                 } else {
1384                         u64 delta = nsec_to_cycles(vcpu, elapsed);
1385                         data += delta;
1386                         offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1387                         pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1388                 }
1389                 matched = true;
1390                 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1391         } else {
1392                 /*
1393                  * We split periods of matched TSC writes into generations.
1394                  * For each generation, we track the original measured
1395                  * nanosecond time, offset, and write, so if TSCs are in
1396                  * sync, we can match exact offset, and if not, we can match
1397                  * exact software computation in compute_guest_tsc()
1398                  *
1399                  * These values are tracked in kvm->arch.cur_xxx variables.
1400                  */
1401                 kvm->arch.cur_tsc_generation++;
1402                 kvm->arch.cur_tsc_nsec = ns;
1403                 kvm->arch.cur_tsc_write = data;
1404                 kvm->arch.cur_tsc_offset = offset;
1405                 matched = false;
1406                 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1407                          kvm->arch.cur_tsc_generation, data);
1408         }
1409
1410         /*
1411          * We also track th most recent recorded KHZ, write and time to
1412          * allow the matching interval to be extended at each write.
1413          */
1414         kvm->arch.last_tsc_nsec = ns;
1415         kvm->arch.last_tsc_write = data;
1416         kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1417
1418         vcpu->arch.last_guest_tsc = data;
1419
1420         /* Keep track of which generation this VCPU has synchronized to */
1421         vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1422         vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1423         vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1424
1425         if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1426                 update_ia32_tsc_adjust_msr(vcpu, offset);
1427         kvm_x86_ops->write_tsc_offset(vcpu, offset);
1428         raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1429
1430         spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1431         if (!matched) {
1432                 kvm->arch.nr_vcpus_matched_tsc = 0;
1433         } else if (!already_matched) {
1434                 kvm->arch.nr_vcpus_matched_tsc++;
1435         }
1436
1437         kvm_track_tsc_matching(vcpu);
1438         spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1439 }
1440
1441 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1442
1443 #ifdef CONFIG_X86_64
1444
1445 static cycle_t read_tsc(void)
1446 {
1447         cycle_t ret;
1448         u64 last;
1449
1450         /*
1451          * Empirically, a fence (of type that depends on the CPU)
1452          * before rdtsc is enough to ensure that rdtsc is ordered
1453          * with respect to loads.  The various CPU manuals are unclear
1454          * as to whether rdtsc can be reordered with later loads,
1455          * but no one has ever seen it happen.
1456          */
1457         rdtsc_barrier();
1458         ret = (cycle_t)vget_cycles();
1459
1460         last = pvclock_gtod_data.clock.cycle_last;
1461
1462         if (likely(ret >= last))
1463                 return ret;
1464
1465         /*
1466          * GCC likes to generate cmov here, but this branch is extremely
1467          * predictable (it's just a funciton of time and the likely is
1468          * very likely) and there's a data dependence, so force GCC
1469          * to generate a branch instead.  I don't barrier() because
1470          * we don't actually need a barrier, and if this function
1471          * ever gets inlined it will generate worse code.
1472          */
1473         asm volatile ("");
1474         return last;
1475 }
1476
1477 static inline u64 vgettsc(cycle_t *cycle_now)
1478 {
1479         long v;
1480         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1481
1482         *cycle_now = read_tsc();
1483
1484         v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1485         return v * gtod->clock.mult;
1486 }
1487
1488 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1489 {
1490         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1491         unsigned long seq;
1492         int mode;
1493         u64 ns;
1494
1495         do {
1496                 seq = read_seqcount_begin(&gtod->seq);
1497                 mode = gtod->clock.vclock_mode;
1498                 ns = gtod->nsec_base;
1499                 ns += vgettsc(cycle_now);
1500                 ns >>= gtod->clock.shift;
1501                 ns += gtod->boot_ns;
1502         } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1503         *t = ns;
1504
1505         return mode;
1506 }
1507
1508 /* returns true if host is using tsc clocksource */
1509 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1510 {
1511         /* checked again under seqlock below */
1512         if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1513                 return false;
1514
1515         return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1516 }
1517 #endif
1518
1519 /*
1520  *
1521  * Assuming a stable TSC across physical CPUS, and a stable TSC
1522  * across virtual CPUs, the following condition is possible.
1523  * Each numbered line represents an event visible to both
1524  * CPUs at the next numbered event.
1525  *
1526  * "timespecX" represents host monotonic time. "tscX" represents
1527  * RDTSC value.
1528  *
1529  *              VCPU0 on CPU0           |       VCPU1 on CPU1
1530  *
1531  * 1.  read timespec0,tsc0
1532  * 2.                                   | timespec1 = timespec0 + N
1533  *                                      | tsc1 = tsc0 + M
1534  * 3. transition to guest               | transition to guest
1535  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1536  * 5.                                   | ret1 = timespec1 + (rdtsc - tsc1)
1537  *                                      | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1538  *
1539  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1540  *
1541  *      - ret0 < ret1
1542  *      - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1543  *              ...
1544  *      - 0 < N - M => M < N
1545  *
1546  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1547  * always the case (the difference between two distinct xtime instances
1548  * might be smaller then the difference between corresponding TSC reads,
1549  * when updating guest vcpus pvclock areas).
1550  *
1551  * To avoid that problem, do not allow visibility of distinct
1552  * system_timestamp/tsc_timestamp values simultaneously: use a master
1553  * copy of host monotonic time values. Update that master copy
1554  * in lockstep.
1555  *
1556  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1557  *
1558  */
1559
1560 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1561 {
1562 #ifdef CONFIG_X86_64
1563         struct kvm_arch *ka = &kvm->arch;
1564         int vclock_mode;
1565         bool host_tsc_clocksource, vcpus_matched;
1566
1567         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1568                         atomic_read(&kvm->online_vcpus));
1569
1570         /*
1571          * If the host uses TSC clock, then passthrough TSC as stable
1572          * to the guest.
1573          */
1574         host_tsc_clocksource = kvm_get_time_and_clockread(
1575                                         &ka->master_kernel_ns,
1576                                         &ka->master_cycle_now);
1577
1578         ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1579                                 && !backwards_tsc_observed
1580                                 && !ka->boot_vcpu_runs_old_kvmclock;
1581
1582         if (ka->use_master_clock)
1583                 atomic_set(&kvm_guest_has_master_clock, 1);
1584
1585         vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1586         trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1587                                         vcpus_matched);
1588 #endif
1589 }
1590
1591 static void kvm_gen_update_masterclock(struct kvm *kvm)
1592 {
1593 #ifdef CONFIG_X86_64
1594         int i;
1595         struct kvm_vcpu *vcpu;
1596         struct kvm_arch *ka = &kvm->arch;
1597
1598         spin_lock(&ka->pvclock_gtod_sync_lock);
1599         kvm_make_mclock_inprogress_request(kvm);
1600         /* no guest entries from this point */
1601         pvclock_update_vm_gtod_copy(kvm);
1602
1603         kvm_for_each_vcpu(i, vcpu, kvm)
1604                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1605
1606         /* guest entries allowed */
1607         kvm_for_each_vcpu(i, vcpu, kvm)
1608                 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1609
1610         spin_unlock(&ka->pvclock_gtod_sync_lock);
1611 #endif
1612 }
1613
1614 static int kvm_guest_time_update(struct kvm_vcpu *v)
1615 {
1616         unsigned long flags, this_tsc_khz;
1617         struct kvm_vcpu_arch *vcpu = &v->arch;
1618         struct kvm_arch *ka = &v->kvm->arch;
1619         s64 kernel_ns;
1620         u64 tsc_timestamp, host_tsc;
1621         struct pvclock_vcpu_time_info guest_hv_clock;
1622         u8 pvclock_flags;
1623         bool use_master_clock;
1624
1625         kernel_ns = 0;
1626         host_tsc = 0;
1627
1628         /*
1629          * If the host uses TSC clock, then passthrough TSC as stable
1630          * to the guest.
1631          */
1632         spin_lock(&ka->pvclock_gtod_sync_lock);
1633         use_master_clock = ka->use_master_clock;
1634         if (use_master_clock) {
1635                 host_tsc = ka->master_cycle_now;
1636                 kernel_ns = ka->master_kernel_ns;
1637         }
1638         spin_unlock(&ka->pvclock_gtod_sync_lock);
1639
1640         /* Keep irq disabled to prevent changes to the clock */
1641         local_irq_save(flags);
1642         this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1643         if (unlikely(this_tsc_khz == 0)) {
1644                 local_irq_restore(flags);
1645                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1646                 return 1;
1647         }
1648         if (!use_master_clock) {
1649                 host_tsc = native_read_tsc();
1650                 kernel_ns = get_kernel_ns();
1651         }
1652
1653         tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1654
1655         /*
1656          * We may have to catch up the TSC to match elapsed wall clock
1657          * time for two reasons, even if kvmclock is used.
1658          *   1) CPU could have been running below the maximum TSC rate
1659          *   2) Broken TSC compensation resets the base at each VCPU
1660          *      entry to avoid unknown leaps of TSC even when running
1661          *      again on the same CPU.  This may cause apparent elapsed
1662          *      time to disappear, and the guest to stand still or run
1663          *      very slowly.
1664          */
1665         if (vcpu->tsc_catchup) {
1666                 u64 tsc = compute_guest_tsc(v, kernel_ns);
1667                 if (tsc > tsc_timestamp) {
1668                         adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1669                         tsc_timestamp = tsc;
1670                 }
1671         }
1672
1673         local_irq_restore(flags);
1674
1675         if (!vcpu->pv_time_enabled)
1676                 return 0;
1677
1678         if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1679                 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1680                                    &vcpu->hv_clock.tsc_shift,
1681                                    &vcpu->hv_clock.tsc_to_system_mul);
1682                 vcpu->hw_tsc_khz = this_tsc_khz;
1683         }
1684
1685         /* With all the info we got, fill in the values */
1686         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1687         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1688         vcpu->last_guest_tsc = tsc_timestamp;
1689
1690         if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1691                 &guest_hv_clock, sizeof(guest_hv_clock))))
1692                 return 0;
1693
1694         /* This VCPU is paused, but it's legal for a guest to read another
1695          * VCPU's kvmclock, so we really have to follow the specification where
1696          * it says that version is odd if data is being modified, and even after
1697          * it is consistent.
1698          *
1699          * Version field updates must be kept separate.  This is because
1700          * kvm_write_guest_cached might use a "rep movs" instruction, and
1701          * writes within a string instruction are weakly ordered.  So there
1702          * are three writes overall.
1703          *
1704          * As a small optimization, only write the version field in the first
1705          * and third write.  The vcpu->pv_time cache is still valid, because the
1706          * version field is the first in the struct.
1707          */
1708         BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1709
1710         vcpu->hv_clock.version = guest_hv_clock.version + 1;
1711         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1712                                 &vcpu->hv_clock,
1713                                 sizeof(vcpu->hv_clock.version));
1714
1715         smp_wmb();
1716
1717         /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1718         pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1719
1720         if (vcpu->pvclock_set_guest_stopped_request) {
1721                 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1722                 vcpu->pvclock_set_guest_stopped_request = false;
1723         }
1724
1725         pvclock_flags |= PVCLOCK_COUNTS_FROM_ZERO;
1726
1727         /* If the host uses TSC clocksource, then it is stable */
1728         if (use_master_clock)
1729                 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1730
1731         vcpu->hv_clock.flags = pvclock_flags;
1732
1733         trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1734
1735         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1736                                 &vcpu->hv_clock,
1737                                 sizeof(vcpu->hv_clock));
1738
1739         smp_wmb();
1740
1741         vcpu->hv_clock.version++;
1742         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1743                                 &vcpu->hv_clock,
1744                                 sizeof(vcpu->hv_clock.version));
1745         return 0;
1746 }
1747
1748 /*
1749  * kvmclock updates which are isolated to a given vcpu, such as
1750  * vcpu->cpu migration, should not allow system_timestamp from
1751  * the rest of the vcpus to remain static. Otherwise ntp frequency
1752  * correction applies to one vcpu's system_timestamp but not
1753  * the others.
1754  *
1755  * So in those cases, request a kvmclock update for all vcpus.
1756  * We need to rate-limit these requests though, as they can
1757  * considerably slow guests that have a large number of vcpus.
1758  * The time for a remote vcpu to update its kvmclock is bound
1759  * by the delay we use to rate-limit the updates.
1760  */
1761
1762 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1763
1764 static void kvmclock_update_fn(struct work_struct *work)
1765 {
1766         int i;
1767         struct delayed_work *dwork = to_delayed_work(work);
1768         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1769                                            kvmclock_update_work);
1770         struct kvm *kvm = container_of(ka, struct kvm, arch);
1771         struct kvm_vcpu *vcpu;
1772
1773         kvm_for_each_vcpu(i, vcpu, kvm) {
1774                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1775                 kvm_vcpu_kick(vcpu);
1776         }
1777 }
1778
1779 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1780 {
1781         struct kvm *kvm = v->kvm;
1782
1783         kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1784         schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1785                                         KVMCLOCK_UPDATE_DELAY);
1786 }
1787
1788 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1789
1790 static void kvmclock_sync_fn(struct work_struct *work)
1791 {
1792         struct delayed_work *dwork = to_delayed_work(work);
1793         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1794                                            kvmclock_sync_work);
1795         struct kvm *kvm = container_of(ka, struct kvm, arch);
1796
1797         if (!kvmclock_periodic_sync)
1798                 return;
1799
1800         schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1801         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1802                                         KVMCLOCK_SYNC_PERIOD);
1803 }
1804
1805 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1806 {
1807         u64 mcg_cap = vcpu->arch.mcg_cap;
1808         unsigned bank_num = mcg_cap & 0xff;
1809
1810         switch (msr) {
1811         case MSR_IA32_MCG_STATUS:
1812                 vcpu->arch.mcg_status = data;
1813                 break;
1814         case MSR_IA32_MCG_CTL:
1815                 if (!(mcg_cap & MCG_CTL_P))
1816                         return 1;
1817                 if (data != 0 && data != ~(u64)0)
1818                         return -1;
1819                 vcpu->arch.mcg_ctl = data;
1820                 break;
1821         default:
1822                 if (msr >= MSR_IA32_MC0_CTL &&
1823                     msr < MSR_IA32_MCx_CTL(bank_num)) {
1824                         u32 offset = msr - MSR_IA32_MC0_CTL;
1825                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1826                          * some Linux kernels though clear bit 10 in bank 4 to
1827                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1828                          * this to avoid an uncatched #GP in the guest
1829                          */
1830                         if ((offset & 0x3) == 0 &&
1831                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1832                                 return -1;
1833                         vcpu->arch.mce_banks[offset] = data;
1834                         break;
1835                 }
1836                 return 1;
1837         }
1838         return 0;
1839 }
1840
1841 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1842 {
1843         struct kvm *kvm = vcpu->kvm;
1844         int lm = is_long_mode(vcpu);
1845         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1846                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1847         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1848                 : kvm->arch.xen_hvm_config.blob_size_32;
1849         u32 page_num = data & ~PAGE_MASK;
1850         u64 page_addr = data & PAGE_MASK;
1851         u8 *page;
1852         int r;
1853
1854         r = -E2BIG;
1855         if (page_num >= blob_size)
1856                 goto out;
1857         r = -ENOMEM;
1858         page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1859         if (IS_ERR(page)) {
1860                 r = PTR_ERR(page);
1861                 goto out;
1862         }
1863         if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1864                 goto out_free;
1865         r = 0;
1866 out_free:
1867         kfree(page);
1868 out:
1869         return r;
1870 }
1871
1872 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1873 {
1874         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1875 }
1876
1877 static bool kvm_hv_msr_partition_wide(u32 msr)
1878 {
1879         bool r = false;
1880         switch (msr) {
1881         case HV_X64_MSR_GUEST_OS_ID:
1882         case HV_X64_MSR_HYPERCALL:
1883         case HV_X64_MSR_REFERENCE_TSC:
1884         case HV_X64_MSR_TIME_REF_COUNT:
1885                 r = true;
1886                 break;
1887         }
1888
1889         return r;
1890 }
1891
1892 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1893 {
1894         struct kvm *kvm = vcpu->kvm;
1895
1896         switch (msr) {
1897         case HV_X64_MSR_GUEST_OS_ID:
1898                 kvm->arch.hv_guest_os_id = data;
1899                 /* setting guest os id to zero disables hypercall page */
1900                 if (!kvm->arch.hv_guest_os_id)
1901                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1902                 break;
1903         case HV_X64_MSR_HYPERCALL: {
1904                 u64 gfn;
1905                 unsigned long addr;
1906                 u8 instructions[4];
1907
1908                 /* if guest os id is not set hypercall should remain disabled */
1909                 if (!kvm->arch.hv_guest_os_id)
1910                         break;
1911                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1912                         kvm->arch.hv_hypercall = data;
1913                         break;
1914                 }
1915                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1916                 addr = gfn_to_hva(kvm, gfn);
1917                 if (kvm_is_error_hva(addr))
1918                         return 1;
1919                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1920                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1921                 if (__copy_to_user((void __user *)addr, instructions, 4))
1922                         return 1;
1923                 kvm->arch.hv_hypercall = data;
1924                 mark_page_dirty(kvm, gfn);
1925                 break;
1926         }
1927         case HV_X64_MSR_REFERENCE_TSC: {
1928                 u64 gfn;
1929                 HV_REFERENCE_TSC_PAGE tsc_ref;
1930                 memset(&tsc_ref, 0, sizeof(tsc_ref));
1931                 kvm->arch.hv_tsc_page = data;
1932                 if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1933                         break;
1934                 gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1935                 if (kvm_write_guest(kvm, gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
1936                         &tsc_ref, sizeof(tsc_ref)))
1937                         return 1;
1938                 mark_page_dirty(kvm, gfn);
1939                 break;
1940         }
1941         default:
1942                 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1943                             "data 0x%llx\n", msr, data);
1944                 return 1;
1945         }
1946         return 0;
1947 }
1948
1949 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1950 {
1951         switch (msr) {
1952         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1953                 u64 gfn;
1954                 unsigned long addr;
1955
1956                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1957                         vcpu->arch.hv_vapic = data;
1958                         if (kvm_lapic_enable_pv_eoi(vcpu, 0))
1959                                 return 1;
1960                         break;
1961                 }
1962                 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
1963                 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1964                 if (kvm_is_error_hva(addr))
1965                         return 1;
1966                 if (__clear_user((void __user *)addr, PAGE_SIZE))
1967                         return 1;
1968                 vcpu->arch.hv_vapic = data;
1969                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
1970                 if (kvm_lapic_enable_pv_eoi(vcpu, gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
1971                         return 1;
1972                 break;
1973         }
1974         case HV_X64_MSR_EOI:
1975                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1976         case HV_X64_MSR_ICR:
1977                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1978         case HV_X64_MSR_TPR:
1979                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1980         default:
1981                 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1982                             "data 0x%llx\n", msr, data);
1983                 return 1;
1984         }
1985
1986         return 0;
1987 }
1988
1989 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1990 {
1991         gpa_t gpa = data & ~0x3f;
1992
1993         /* Bits 2:5 are reserved, Should be zero */
1994         if (data & 0x3c)
1995                 return 1;
1996
1997         vcpu->arch.apf.msr_val = data;
1998
1999         if (!(data & KVM_ASYNC_PF_ENABLED)) {
2000                 kvm_clear_async_pf_completion_queue(vcpu);
2001                 kvm_async_pf_hash_reset(vcpu);
2002                 return 0;
2003         }
2004
2005         if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2006                                         sizeof(u32)))
2007                 return 1;
2008
2009         vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2010         kvm_async_pf_wakeup_all(vcpu);
2011         return 0;
2012 }
2013
2014 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2015 {
2016         vcpu->arch.pv_time_enabled = false;
2017 }
2018
2019 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2020 {
2021         u64 delta;
2022
2023         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2024                 return;
2025
2026         delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2027         vcpu->arch.st.last_steal = current->sched_info.run_delay;
2028         vcpu->arch.st.accum_steal = delta;
2029 }
2030
2031 static void record_steal_time(struct kvm_vcpu *vcpu)
2032 {
2033         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2034                 return;
2035
2036         if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2037                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2038                 return;
2039
2040         vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2041         vcpu->arch.st.steal.version += 2;
2042         vcpu->arch.st.accum_steal = 0;
2043
2044         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2045                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2046 }
2047
2048 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2049 {
2050         bool pr = false;
2051         u32 msr = msr_info->index;
2052         u64 data = msr_info->data;
2053
2054         switch (msr) {
2055         case MSR_AMD64_NB_CFG:
2056         case MSR_IA32_UCODE_REV:
2057         case MSR_IA32_UCODE_WRITE:
2058         case MSR_VM_HSAVE_PA:
2059         case MSR_AMD64_PATCH_LOADER:
2060         case MSR_AMD64_BU_CFG2:
2061                 break;
2062
2063         case MSR_EFER:
2064                 return set_efer(vcpu, data);
2065         case MSR_K7_HWCR:
2066                 data &= ~(u64)0x40;     /* ignore flush filter disable */
2067                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
2068                 data &= ~(u64)0x8;      /* ignore TLB cache disable */
2069                 data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2070                 if (data != 0) {
2071                         vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2072                                     data);
2073                         return 1;
2074                 }
2075                 break;
2076         case MSR_FAM10H_MMIO_CONF_BASE:
2077                 if (data != 0) {
2078                         vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2079                                     "0x%llx\n", data);
2080                         return 1;
2081                 }
2082                 break;
2083         case MSR_IA32_DEBUGCTLMSR:
2084                 if (!data) {
2085                         /* We support the non-activated case already */
2086                         break;
2087                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2088                         /* Values other than LBR and BTF are vendor-specific,
2089                            thus reserved and should throw a #GP */
2090                         return 1;
2091                 }
2092                 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2093                             __func__, data);
2094                 break;
2095         case 0x200 ... 0x2ff:
2096                 return kvm_mtrr_set_msr(vcpu, msr, data);
2097         case MSR_IA32_APICBASE:
2098                 return kvm_set_apic_base(vcpu, msr_info);
2099         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2100                 return kvm_x2apic_msr_write(vcpu, msr, data);
2101         case MSR_IA32_TSCDEADLINE:
2102                 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2103                 break;
2104         case MSR_IA32_TSC_ADJUST:
2105                 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2106                         if (!msr_info->host_initiated) {
2107                                 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2108                                 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2109                         }
2110                         vcpu->arch.ia32_tsc_adjust_msr = data;
2111                 }
2112                 break;
2113         case MSR_IA32_MISC_ENABLE:
2114                 vcpu->arch.ia32_misc_enable_msr = data;
2115                 break;
2116         case MSR_IA32_SMBASE:
2117                 if (!msr_info->host_initiated)
2118                         return 1;
2119                 vcpu->arch.smbase = data;
2120                 break;
2121         case MSR_KVM_WALL_CLOCK_NEW:
2122         case MSR_KVM_WALL_CLOCK:
2123                 vcpu->kvm->arch.wall_clock = data;
2124                 kvm_write_wall_clock(vcpu->kvm, data);
2125                 break;
2126         case MSR_KVM_SYSTEM_TIME_NEW:
2127         case MSR_KVM_SYSTEM_TIME: {
2128                 u64 gpa_offset;
2129                 struct kvm_arch *ka = &vcpu->kvm->arch;
2130
2131                 kvmclock_reset(vcpu);
2132
2133                 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2134                         bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2135
2136                         if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2137                                 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2138                                         &vcpu->requests);
2139
2140                         ka->boot_vcpu_runs_old_kvmclock = tmp;
2141
2142                         ka->kvmclock_offset = -get_kernel_ns();
2143                 }
2144
2145                 vcpu->arch.time = data;
2146                 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2147
2148                 /* we verify if the enable bit is set... */
2149                 if (!(data & 1))
2150                         break;
2151
2152                 gpa_offset = data & ~(PAGE_MASK | 1);
2153
2154                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2155                      &vcpu->arch.pv_time, data & ~1ULL,
2156                      sizeof(struct pvclock_vcpu_time_info)))
2157                         vcpu->arch.pv_time_enabled = false;
2158                 else
2159                         vcpu->arch.pv_time_enabled = true;
2160
2161                 break;
2162         }
2163         case MSR_KVM_ASYNC_PF_EN:
2164                 if (kvm_pv_enable_async_pf(vcpu, data))
2165                         return 1;
2166                 break;
2167         case MSR_KVM_STEAL_TIME:
2168
2169                 if (unlikely(!sched_info_on()))
2170                         return 1;
2171
2172                 if (data & KVM_STEAL_RESERVED_MASK)
2173                         return 1;
2174
2175                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2176                                                 data & KVM_STEAL_VALID_BITS,
2177                                                 sizeof(struct kvm_steal_time)))
2178                         return 1;
2179
2180                 vcpu->arch.st.msr_val = data;
2181
2182                 if (!(data & KVM_MSR_ENABLED))
2183                         break;
2184
2185                 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2186
2187                 preempt_disable();
2188                 accumulate_steal_time(vcpu);
2189                 preempt_enable();
2190
2191                 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2192
2193                 break;
2194         case MSR_KVM_PV_EOI_EN:
2195                 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2196                         return 1;
2197                 break;
2198
2199         case MSR_IA32_MCG_CTL:
2200         case MSR_IA32_MCG_STATUS:
2201         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2202                 return set_msr_mce(vcpu, msr, data);
2203
2204         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2205         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2206                 pr = true; /* fall through */
2207         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2208         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2209                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2210                         return kvm_pmu_set_msr(vcpu, msr_info);
2211
2212                 if (pr || data != 0)
2213                         vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2214                                     "0x%x data 0x%llx\n", msr, data);
2215                 break;
2216         case MSR_K7_CLK_CTL:
2217                 /*
2218                  * Ignore all writes to this no longer documented MSR.
2219                  * Writes are only relevant for old K7 processors,
2220                  * all pre-dating SVM, but a recommended workaround from
2221                  * AMD for these chips. It is possible to specify the
2222                  * affected processor models on the command line, hence
2223                  * the need to ignore the workaround.
2224                  */
2225                 break;
2226         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2227                 if (kvm_hv_msr_partition_wide(msr)) {
2228                         int r;
2229                         mutex_lock(&vcpu->kvm->lock);
2230                         r = set_msr_hyperv_pw(vcpu, msr, data);
2231                         mutex_unlock(&vcpu->kvm->lock);
2232                         return r;
2233                 } else
2234                         return set_msr_hyperv(vcpu, msr, data);
2235                 break;
2236         case MSR_IA32_BBL_CR_CTL3:
2237                 /* Drop writes to this legacy MSR -- see rdmsr
2238                  * counterpart for further detail.
2239                  */
2240                 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2241                 break;
2242         case MSR_AMD64_OSVW_ID_LENGTH:
2243                 if (!guest_cpuid_has_osvw(vcpu))
2244                         return 1;
2245                 vcpu->arch.osvw.length = data;
2246                 break;
2247         case MSR_AMD64_OSVW_STATUS:
2248                 if (!guest_cpuid_has_osvw(vcpu))
2249                         return 1;
2250                 vcpu->arch.osvw.status = data;
2251                 break;
2252         default:
2253                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2254                         return xen_hvm_config(vcpu, data);
2255                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2256                         return kvm_pmu_set_msr(vcpu, msr_info);
2257                 if (!ignore_msrs) {
2258                         vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2259                                     msr, data);
2260                         return 1;
2261                 } else {
2262                         vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2263                                     msr, data);
2264                         break;
2265                 }
2266         }
2267         return 0;
2268 }
2269 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2270
2271
2272 /*
2273  * Reads an msr value (of 'msr_index') into 'pdata'.
2274  * Returns 0 on success, non-0 otherwise.
2275  * Assumes vcpu_load() was already called.
2276  */
2277 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2278 {
2279         return kvm_x86_ops->get_msr(vcpu, msr);
2280 }
2281 EXPORT_SYMBOL_GPL(kvm_get_msr);
2282
2283 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2284 {
2285         u64 data;
2286         u64 mcg_cap = vcpu->arch.mcg_cap;
2287         unsigned bank_num = mcg_cap & 0xff;
2288
2289         switch (msr) {
2290         case MSR_IA32_P5_MC_ADDR:
2291         case MSR_IA32_P5_MC_TYPE:
2292                 data = 0;
2293                 break;
2294         case MSR_IA32_MCG_CAP:
2295                 data = vcpu->arch.mcg_cap;
2296                 break;
2297         case MSR_IA32_MCG_CTL:
2298                 if (!(mcg_cap & MCG_CTL_P))
2299                         return 1;
2300                 data = vcpu->arch.mcg_ctl;
2301                 break;
2302         case MSR_IA32_MCG_STATUS:
2303                 data = vcpu->arch.mcg_status;
2304                 break;
2305         default:
2306                 if (msr >= MSR_IA32_MC0_CTL &&
2307                     msr < MSR_IA32_MCx_CTL(bank_num)) {
2308                         u32 offset = msr - MSR_IA32_MC0_CTL;
2309                         data = vcpu->arch.mce_banks[offset];
2310                         break;
2311                 }
2312                 return 1;
2313         }
2314         *pdata = data;
2315         return 0;
2316 }
2317
2318 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2319 {
2320         u64 data = 0;
2321         struct kvm *kvm = vcpu->kvm;
2322
2323         switch (msr) {
2324         case HV_X64_MSR_GUEST_OS_ID:
2325                 data = kvm->arch.hv_guest_os_id;
2326                 break;
2327         case HV_X64_MSR_HYPERCALL:
2328                 data = kvm->arch.hv_hypercall;
2329                 break;
2330         case HV_X64_MSR_TIME_REF_COUNT: {
2331                 data =
2332                      div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2333                 break;
2334         }
2335         case HV_X64_MSR_REFERENCE_TSC:
2336                 data = kvm->arch.hv_tsc_page;
2337                 break;
2338         default:
2339                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2340                 return 1;
2341         }
2342
2343         *pdata = data;
2344         return 0;
2345 }
2346
2347 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2348 {
2349         u64 data = 0;
2350
2351         switch (msr) {
2352         case HV_X64_MSR_VP_INDEX: {
2353                 int r;
2354                 struct kvm_vcpu *v;
2355                 kvm_for_each_vcpu(r, v, vcpu->kvm) {
2356                         if (v == vcpu) {
2357                                 data = r;
2358                                 break;
2359                         }
2360                 }
2361                 break;
2362         }
2363         case HV_X64_MSR_EOI:
2364                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2365         case HV_X64_MSR_ICR:
2366                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2367         case HV_X64_MSR_TPR:
2368                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2369         case HV_X64_MSR_APIC_ASSIST_PAGE:
2370                 data = vcpu->arch.hv_vapic;
2371                 break;
2372         default:
2373                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2374                 return 1;
2375         }
2376         *pdata = data;
2377         return 0;
2378 }
2379
2380 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2381 {
2382         switch (msr_info->index) {
2383         case MSR_IA32_PLATFORM_ID:
2384         case MSR_IA32_EBL_CR_POWERON:
2385         case MSR_IA32_DEBUGCTLMSR:
2386         case MSR_IA32_LASTBRANCHFROMIP:
2387         case MSR_IA32_LASTBRANCHTOIP:
2388         case MSR_IA32_LASTINTFROMIP:
2389         case MSR_IA32_LASTINTTOIP:
2390         case MSR_K8_SYSCFG:
2391         case MSR_K7_HWCR:
2392         case MSR_VM_HSAVE_PA:
2393         case MSR_K8_INT_PENDING_MSG:
2394         case MSR_AMD64_NB_CFG:
2395         case MSR_FAM10H_MMIO_CONF_BASE:
2396         case MSR_AMD64_BU_CFG2:
2397                 msr_info->data = 0;
2398                 break;
2399         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2400         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2401         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2402         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2403                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2404                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2405                 msr_info->data = 0;
2406                 break;
2407         case MSR_IA32_UCODE_REV:
2408                 msr_info->data = 0x100000000ULL;
2409                 break;
2410         case MSR_MTRRcap:
2411         case 0x200 ... 0x2ff:
2412                 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2413         case 0xcd: /* fsb frequency */
2414                 msr_info->data = 3;
2415                 break;
2416                 /*
2417                  * MSR_EBC_FREQUENCY_ID
2418                  * Conservative value valid for even the basic CPU models.
2419                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2420                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2421                  * and 266MHz for model 3, or 4. Set Core Clock
2422                  * Frequency to System Bus Frequency Ratio to 1 (bits
2423                  * 31:24) even though these are only valid for CPU
2424                  * models > 2, however guests may end up dividing or
2425                  * multiplying by zero otherwise.
2426                  */
2427         case MSR_EBC_FREQUENCY_ID:
2428                 msr_info->data = 1 << 24;
2429                 break;
2430         case MSR_IA32_APICBASE:
2431                 msr_info->data = kvm_get_apic_base(vcpu);
2432                 break;
2433         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2434                 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2435                 break;
2436         case MSR_IA32_TSCDEADLINE:
2437                 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2438                 break;
2439         case MSR_IA32_TSC_ADJUST:
2440                 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2441                 break;
2442         case MSR_IA32_MISC_ENABLE:
2443                 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2444                 break;
2445         case MSR_IA32_SMBASE:
2446                 if (!msr_info->host_initiated)
2447                         return 1;
2448                 msr_info->data = vcpu->arch.smbase;
2449                 break;
2450         case MSR_IA32_PERF_STATUS:
2451                 /* TSC increment by tick */
2452                 msr_info->data = 1000ULL;
2453                 /* CPU multiplier */
2454                 msr_info->data |= (((uint64_t)4ULL) << 40);
2455                 break;
2456         case MSR_EFER:
2457                 msr_info->data = vcpu->arch.efer;
2458                 break;
2459         case MSR_KVM_WALL_CLOCK:
2460         case MSR_KVM_WALL_CLOCK_NEW:
2461                 msr_info->data = vcpu->kvm->arch.wall_clock;
2462                 break;
2463         case MSR_KVM_SYSTEM_TIME:
2464         case MSR_KVM_SYSTEM_TIME_NEW:
2465                 msr_info->data = vcpu->arch.time;
2466                 break;
2467         case MSR_KVM_ASYNC_PF_EN:
2468                 msr_info->data = vcpu->arch.apf.msr_val;
2469                 break;
2470         case MSR_KVM_STEAL_TIME:
2471                 msr_info->data = vcpu->arch.st.msr_val;
2472                 break;
2473         case MSR_KVM_PV_EOI_EN:
2474                 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2475                 break;
2476         case MSR_IA32_P5_MC_ADDR:
2477         case MSR_IA32_P5_MC_TYPE:
2478         case MSR_IA32_MCG_CAP:
2479         case MSR_IA32_MCG_CTL:
2480         case MSR_IA32_MCG_STATUS:
2481         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2482                 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2483         case MSR_K7_CLK_CTL:
2484                 /*
2485                  * Provide expected ramp-up count for K7. All other
2486                  * are set to zero, indicating minimum divisors for
2487                  * every field.
2488                  *
2489                  * This prevents guest kernels on AMD host with CPU
2490                  * type 6, model 8 and higher from exploding due to
2491                  * the rdmsr failing.
2492                  */
2493                 msr_info->data = 0x20000000;
2494                 break;
2495         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2496                 if (kvm_hv_msr_partition_wide(msr_info->index)) {
2497                         int r;
2498                         mutex_lock(&vcpu->kvm->lock);
2499                         r = get_msr_hyperv_pw(vcpu, msr_info->index, &msr_info->data);
2500                         mutex_unlock(&vcpu->kvm->lock);
2501                         return r;
2502                 } else
2503                         return get_msr_hyperv(vcpu, msr_info->index, &msr_info->data);
2504                 break;
2505         case MSR_IA32_BBL_CR_CTL3:
2506                 /* This legacy MSR exists but isn't fully documented in current
2507                  * silicon.  It is however accessed by winxp in very narrow
2508                  * scenarios where it sets bit #19, itself documented as
2509                  * a "reserved" bit.  Best effort attempt to source coherent
2510                  * read data here should the balance of the register be
2511                  * interpreted by the guest:
2512                  *
2513                  * L2 cache control register 3: 64GB range, 256KB size,
2514                  * enabled, latency 0x1, configured
2515                  */
2516                 msr_info->data = 0xbe702111;
2517                 break;
2518         case MSR_AMD64_OSVW_ID_LENGTH:
2519                 if (!guest_cpuid_has_osvw(vcpu))
2520                         return 1;
2521                 msr_info->data = vcpu->arch.osvw.length;
2522                 break;
2523         case MSR_AMD64_OSVW_STATUS:
2524                 if (!guest_cpuid_has_osvw(vcpu))
2525                         return 1;
2526                 msr_info->data = vcpu->arch.osvw.status;
2527                 break;
2528         default:
2529                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2530                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2531                 if (!ignore_msrs) {
2532                         vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2533                         return 1;
2534                 } else {
2535                         vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2536                         msr_info->data = 0;
2537                 }
2538                 break;
2539         }
2540         return 0;
2541 }
2542 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2543
2544 /*
2545  * Read or write a bunch of msrs. All parameters are kernel addresses.
2546  *
2547  * @return number of msrs set successfully.
2548  */
2549 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2550                     struct kvm_msr_entry *entries,
2551                     int (*do_msr)(struct kvm_vcpu *vcpu,
2552                                   unsigned index, u64 *data))
2553 {
2554         int i, idx;
2555
2556         idx = srcu_read_lock(&vcpu->kvm->srcu);
2557         for (i = 0; i < msrs->nmsrs; ++i)
2558                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2559                         break;
2560         srcu_read_unlock(&vcpu->kvm->srcu, idx);
2561
2562         return i;
2563 }
2564
2565 /*
2566  * Read or write a bunch of msrs. Parameters are user addresses.
2567  *
2568  * @return number of msrs set successfully.
2569  */
2570 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2571                   int (*do_msr)(struct kvm_vcpu *vcpu,
2572                                 unsigned index, u64 *data),
2573                   int writeback)
2574 {
2575         struct kvm_msrs msrs;
2576         struct kvm_msr_entry *entries;
2577         int r, n;
2578         unsigned size;
2579
2580         r = -EFAULT;
2581         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2582                 goto out;
2583
2584         r = -E2BIG;
2585         if (msrs.nmsrs >= MAX_IO_MSRS)
2586                 goto out;
2587
2588         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2589         entries = memdup_user(user_msrs->entries, size);
2590         if (IS_ERR(entries)) {
2591                 r = PTR_ERR(entries);
2592                 goto out;
2593         }
2594
2595         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2596         if (r < 0)
2597                 goto out_free;
2598
2599         r = -EFAULT;
2600         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2601                 goto out_free;
2602
2603         r = n;
2604
2605 out_free:
2606         kfree(entries);
2607 out:
2608         return r;
2609 }
2610
2611 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2612 {
2613         int r;
2614
2615         switch (ext) {
2616         case KVM_CAP_IRQCHIP:
2617         case KVM_CAP_HLT:
2618         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2619         case KVM_CAP_SET_TSS_ADDR:
2620         case KVM_CAP_EXT_CPUID:
2621         case KVM_CAP_EXT_EMUL_CPUID:
2622         case KVM_CAP_CLOCKSOURCE:
2623         case KVM_CAP_PIT:
2624         case KVM_CAP_NOP_IO_DELAY:
2625         case KVM_CAP_MP_STATE:
2626         case KVM_CAP_SYNC_MMU:
2627         case KVM_CAP_USER_NMI:
2628         case KVM_CAP_REINJECT_CONTROL:
2629         case KVM_CAP_IRQ_INJECT_STATUS:
2630         case KVM_CAP_IOEVENTFD:
2631         case KVM_CAP_IOEVENTFD_NO_LENGTH:
2632         case KVM_CAP_PIT2:
2633         case KVM_CAP_PIT_STATE2:
2634         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2635         case KVM_CAP_XEN_HVM:
2636         case KVM_CAP_ADJUST_CLOCK:
2637         case KVM_CAP_VCPU_EVENTS:
2638         case KVM_CAP_HYPERV:
2639         case KVM_CAP_HYPERV_VAPIC:
2640         case KVM_CAP_HYPERV_SPIN:
2641         case KVM_CAP_PCI_SEGMENT:
2642         case KVM_CAP_DEBUGREGS:
2643         case KVM_CAP_X86_ROBUST_SINGLESTEP:
2644         case KVM_CAP_XSAVE:
2645         case KVM_CAP_ASYNC_PF:
2646         case KVM_CAP_GET_TSC_KHZ:
2647         case KVM_CAP_KVMCLOCK_CTRL:
2648         case KVM_CAP_READONLY_MEM:
2649         case KVM_CAP_HYPERV_TIME:
2650         case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2651         case KVM_CAP_TSC_DEADLINE_TIMER:
2652         case KVM_CAP_ENABLE_CAP_VM:
2653         case KVM_CAP_DISABLE_QUIRKS:
2654 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2655         case KVM_CAP_ASSIGN_DEV_IRQ:
2656         case KVM_CAP_PCI_2_3:
2657 #endif
2658                 r = 1;
2659                 break;
2660         case KVM_CAP_X86_SMM:
2661                 /* SMBASE is usually relocated above 1M on modern chipsets,
2662                  * and SMM handlers might indeed rely on 4G segment limits,
2663                  * so do not report SMM to be available if real mode is
2664                  * emulated via vm86 mode.  Still, do not go to great lengths
2665                  * to avoid userspace's usage of the feature, because it is a
2666                  * fringe case that is not enabled except via specific settings
2667                  * of the module parameters.
2668                  */
2669                 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2670                 break;
2671         case KVM_CAP_COALESCED_MMIO:
2672                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2673                 break;
2674         case KVM_CAP_VAPIC:
2675                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2676                 break;
2677         case KVM_CAP_NR_VCPUS:
2678                 r = KVM_SOFT_MAX_VCPUS;
2679                 break;
2680         case KVM_CAP_MAX_VCPUS:
2681                 r = KVM_MAX_VCPUS;
2682                 break;
2683         case KVM_CAP_NR_MEMSLOTS:
2684                 r = KVM_USER_MEM_SLOTS;
2685                 break;
2686         case KVM_CAP_PV_MMU:    /* obsolete */
2687                 r = 0;
2688                 break;
2689 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2690         case KVM_CAP_IOMMU:
2691                 r = iommu_present(&pci_bus_type);
2692                 break;
2693 #endif
2694         case KVM_CAP_MCE:
2695                 r = KVM_MAX_MCE_BANKS;
2696                 break;
2697         case KVM_CAP_XCRS:
2698                 r = cpu_has_xsave;
2699                 break;
2700         case KVM_CAP_TSC_CONTROL:
2701                 r = kvm_has_tsc_control;
2702                 break;
2703         default:
2704                 r = 0;
2705                 break;
2706         }
2707         return r;
2708
2709 }
2710
2711 long kvm_arch_dev_ioctl(struct file *filp,
2712                         unsigned int ioctl, unsigned long arg)
2713 {
2714         void __user *argp = (void __user *)arg;
2715         long r;
2716
2717         switch (ioctl) {
2718         case KVM_GET_MSR_INDEX_LIST: {
2719                 struct kvm_msr_list __user *user_msr_list = argp;
2720                 struct kvm_msr_list msr_list;
2721                 unsigned n;
2722
2723                 r = -EFAULT;
2724                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2725                         goto out;
2726                 n = msr_list.nmsrs;
2727                 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2728                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2729                         goto out;
2730                 r = -E2BIG;
2731                 if (n < msr_list.nmsrs)
2732                         goto out;
2733                 r = -EFAULT;
2734                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2735                                  num_msrs_to_save * sizeof(u32)))
2736                         goto out;
2737                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2738                                  &emulated_msrs,
2739                                  num_emulated_msrs * sizeof(u32)))
2740                         goto out;
2741                 r = 0;
2742                 break;
2743         }
2744         case KVM_GET_SUPPORTED_CPUID:
2745         case KVM_GET_EMULATED_CPUID: {
2746                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2747                 struct kvm_cpuid2 cpuid;
2748
2749                 r = -EFAULT;
2750                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2751                         goto out;
2752
2753                 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2754                                             ioctl);
2755                 if (r)
2756                         goto out;
2757
2758                 r = -EFAULT;
2759                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2760                         goto out;
2761                 r = 0;
2762                 break;
2763         }
2764         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2765                 u64 mce_cap;
2766
2767                 mce_cap = KVM_MCE_CAP_SUPPORTED;
2768                 r = -EFAULT;
2769                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2770                         goto out;
2771                 r = 0;
2772                 break;
2773         }
2774         default:
2775                 r = -EINVAL;
2776         }
2777 out:
2778         return r;
2779 }
2780
2781 static void wbinvd_ipi(void *garbage)
2782 {
2783         wbinvd();
2784 }
2785
2786 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2787 {
2788         return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2789 }
2790
2791 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2792 {
2793         /* Address WBINVD may be executed by guest */
2794         if (need_emulate_wbinvd(vcpu)) {
2795                 if (kvm_x86_ops->has_wbinvd_exit())
2796                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2797                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2798                         smp_call_function_single(vcpu->cpu,
2799                                         wbinvd_ipi, NULL, 1);
2800         }
2801
2802         kvm_x86_ops->vcpu_load(vcpu, cpu);
2803
2804         /* Apply any externally detected TSC adjustments (due to suspend) */
2805         if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2806                 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2807                 vcpu->arch.tsc_offset_adjustment = 0;
2808                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2809         }
2810
2811         if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2812                 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2813                                 native_read_tsc() - vcpu->arch.last_host_tsc;
2814                 if (tsc_delta < 0)
2815                         mark_tsc_unstable("KVM discovered backwards TSC");
2816                 if (check_tsc_unstable()) {
2817                         u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2818                                                 vcpu->arch.last_guest_tsc);
2819                         kvm_x86_ops->write_tsc_offset(vcpu, offset);
2820                         vcpu->arch.tsc_catchup = 1;
2821                 }
2822                 /*
2823                  * On a host with synchronized TSC, there is no need to update
2824                  * kvmclock on vcpu->cpu migration
2825                  */
2826                 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2827                         kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2828                 if (vcpu->cpu != cpu)
2829                         kvm_migrate_timers(vcpu);
2830                 vcpu->cpu = cpu;
2831         }
2832
2833         accumulate_steal_time(vcpu);
2834         kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2835 }
2836
2837 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2838 {
2839         kvm_x86_ops->vcpu_put(vcpu);
2840         kvm_put_guest_fpu(vcpu);
2841         vcpu->arch.last_host_tsc = native_read_tsc();
2842 }
2843
2844 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2845                                     struct kvm_lapic_state *s)
2846 {
2847         kvm_x86_ops->sync_pir_to_irr(vcpu);
2848         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2849
2850         return 0;
2851 }
2852
2853 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2854                                     struct kvm_lapic_state *s)
2855 {
2856         kvm_apic_post_state_restore(vcpu, s);
2857         update_cr8_intercept(vcpu);
2858
2859         return 0;
2860 }
2861
2862 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2863                                     struct kvm_interrupt *irq)
2864 {
2865         if (irq->irq >= KVM_NR_INTERRUPTS)
2866                 return -EINVAL;
2867         if (irqchip_in_kernel(vcpu->kvm))
2868                 return -ENXIO;
2869
2870         kvm_queue_interrupt(vcpu, irq->irq, false);
2871         kvm_make_request(KVM_REQ_EVENT, vcpu);
2872
2873         return 0;
2874 }
2875
2876 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2877 {
2878         kvm_inject_nmi(vcpu);
2879
2880         return 0;
2881 }
2882
2883 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2884 {
2885         kvm_make_request(KVM_REQ_SMI, vcpu);
2886
2887         return 0;
2888 }
2889
2890 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2891                                            struct kvm_tpr_access_ctl *tac)
2892 {
2893         if (tac->flags)
2894                 return -EINVAL;
2895         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2896         return 0;
2897 }
2898
2899 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2900                                         u64 mcg_cap)
2901 {
2902         int r;
2903         unsigned bank_num = mcg_cap & 0xff, bank;
2904
2905         r = -EINVAL;
2906         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2907                 goto out;
2908         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2909                 goto out;
2910         r = 0;
2911         vcpu->arch.mcg_cap = mcg_cap;
2912         /* Init IA32_MCG_CTL to all 1s */
2913         if (mcg_cap & MCG_CTL_P)
2914                 vcpu->arch.mcg_ctl = ~(u64)0;
2915         /* Init IA32_MCi_CTL to all 1s */
2916         for (bank = 0; bank < bank_num; bank++)
2917                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2918 out:
2919         return r;
2920 }
2921
2922 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2923                                       struct kvm_x86_mce *mce)
2924 {
2925         u64 mcg_cap = vcpu->arch.mcg_cap;
2926         unsigned bank_num = mcg_cap & 0xff;
2927         u64 *banks = vcpu->arch.mce_banks;
2928
2929         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2930                 return -EINVAL;
2931         /*
2932          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2933          * reporting is disabled
2934          */
2935         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2936             vcpu->arch.mcg_ctl != ~(u64)0)
2937                 return 0;
2938         banks += 4 * mce->bank;
2939         /*
2940          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2941          * reporting is disabled for the bank
2942          */
2943         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2944                 return 0;
2945         if (mce->status & MCI_STATUS_UC) {
2946                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2947                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2948                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2949                         return 0;
2950                 }
2951                 if (banks[1] & MCI_STATUS_VAL)
2952                         mce->status |= MCI_STATUS_OVER;
2953                 banks[2] = mce->addr;
2954                 banks[3] = mce->misc;
2955                 vcpu->arch.mcg_status = mce->mcg_status;
2956                 banks[1] = mce->status;
2957                 kvm_queue_exception(vcpu, MC_VECTOR);
2958         } else if (!(banks[1] & MCI_STATUS_VAL)
2959                    || !(banks[1] & MCI_STATUS_UC)) {
2960                 if (banks[1] & MCI_STATUS_VAL)
2961                         mce->status |= MCI_STATUS_OVER;
2962                 banks[2] = mce->addr;
2963                 banks[3] = mce->misc;
2964                 banks[1] = mce->status;
2965         } else
2966                 banks[1] |= MCI_STATUS_OVER;
2967         return 0;
2968 }
2969
2970 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2971                                                struct kvm_vcpu_events *events)
2972 {
2973         process_nmi(vcpu);
2974         events->exception.injected =
2975                 vcpu->arch.exception.pending &&
2976                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2977         events->exception.nr = vcpu->arch.exception.nr;
2978         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2979         events->exception.pad = 0;
2980         events->exception.error_code = vcpu->arch.exception.error_code;
2981
2982         events->interrupt.injected =
2983                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2984         events->interrupt.nr = vcpu->arch.interrupt.nr;
2985         events->interrupt.soft = 0;
2986         events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2987
2988         events->nmi.injected = vcpu->arch.nmi_injected;
2989         events->nmi.pending = vcpu->arch.nmi_pending != 0;
2990         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2991         events->nmi.pad = 0;
2992
2993         events->sipi_vector = 0; /* never valid when reporting to user space */
2994
2995         events->smi.smm = is_smm(vcpu);
2996         events->smi.pending = vcpu->arch.smi_pending;
2997         events->smi.smm_inside_nmi =
2998                 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2999         events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3000
3001         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3002                          | KVM_VCPUEVENT_VALID_SHADOW
3003                          | KVM_VCPUEVENT_VALID_SMM);
3004         memset(&events->reserved, 0, sizeof(events->reserved));
3005 }
3006
3007 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3008                                               struct kvm_vcpu_events *events)
3009 {
3010         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3011                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3012                               | KVM_VCPUEVENT_VALID_SHADOW
3013                               | KVM_VCPUEVENT_VALID_SMM))
3014                 return -EINVAL;
3015
3016         process_nmi(vcpu);
3017         vcpu->arch.exception.pending = events->exception.injected;
3018         vcpu->arch.exception.nr = events->exception.nr;
3019         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3020         vcpu->arch.exception.error_code = events->exception.error_code;
3021
3022         vcpu->arch.interrupt.pending = events->interrupt.injected;
3023         vcpu->arch.interrupt.nr = events->interrupt.nr;
3024         vcpu->arch.interrupt.soft = events->interrupt.soft;
3025         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3026                 kvm_x86_ops->set_interrupt_shadow(vcpu,
3027                                                   events->interrupt.shadow);
3028
3029         vcpu->arch.nmi_injected = events->nmi.injected;
3030         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3031                 vcpu->arch.nmi_pending = events->nmi.pending;
3032         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3033
3034         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3035             kvm_vcpu_has_lapic(vcpu))
3036                 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3037
3038         if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3039                 if (events->smi.smm)
3040                         vcpu->arch.hflags |= HF_SMM_MASK;
3041                 else
3042                         vcpu->arch.hflags &= ~HF_SMM_MASK;
3043                 vcpu->arch.smi_pending = events->smi.pending;
3044                 if (events->smi.smm_inside_nmi)
3045                         vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3046                 else
3047                         vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3048                 if (kvm_vcpu_has_lapic(vcpu)) {
3049                         if (events->smi.latched_init)
3050                                 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3051                         else
3052                                 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3053                 }
3054         }
3055
3056         kvm_make_request(KVM_REQ_EVENT, vcpu);
3057
3058         return 0;
3059 }
3060
3061 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3062                                              struct kvm_debugregs *dbgregs)
3063 {
3064         unsigned long val;
3065
3066         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3067         kvm_get_dr(vcpu, 6, &val);
3068         dbgregs->dr6 = val;
3069         dbgregs->dr7 = vcpu->arch.dr7;
3070         dbgregs->flags = 0;
3071         memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3072 }
3073
3074 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3075                                             struct kvm_debugregs *dbgregs)
3076 {
3077         if (dbgregs->flags)
3078                 return -EINVAL;
3079
3080         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3081         kvm_update_dr0123(vcpu);
3082         vcpu->arch.dr6 = dbgregs->dr6;
3083         kvm_update_dr6(vcpu);
3084         vcpu->arch.dr7 = dbgregs->dr7;
3085         kvm_update_dr7(vcpu);
3086
3087         return 0;
3088 }
3089
3090 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3091
3092 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3093 {
3094         struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3095         u64 xstate_bv = xsave->header.xfeatures;
3096         u64 valid;
3097
3098         /*
3099          * Copy legacy XSAVE area, to avoid complications with CPUID
3100          * leaves 0 and 1 in the loop below.
3101          */
3102         memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3103
3104         /* Set XSTATE_BV */
3105         *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3106
3107         /*
3108          * Copy each region from the possibly compacted offset to the
3109          * non-compacted offset.
3110          */
3111         valid = xstate_bv & ~XSTATE_FPSSE;
3112         while (valid) {
3113                 u64 feature = valid & -valid;
3114                 int index = fls64(feature) - 1;
3115                 void *src = get_xsave_addr(xsave, feature);
3116
3117                 if (src) {
3118                         u32 size, offset, ecx, edx;
3119                         cpuid_count(XSTATE_CPUID, index,
3120                                     &size, &offset, &ecx, &edx);
3121                         memcpy(dest + offset, src, size);
3122                 }
3123
3124                 valid -= feature;
3125         }
3126 }
3127
3128 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3129 {
3130         struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3131         u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3132         u64 valid;
3133
3134         /*
3135          * Copy legacy XSAVE area, to avoid complications with CPUID
3136          * leaves 0 and 1 in the loop below.
3137          */
3138         memcpy(xsave, src, XSAVE_HDR_OFFSET);
3139
3140         /* Set XSTATE_BV and possibly XCOMP_BV.  */
3141         xsave->header.xfeatures = xstate_bv;
3142         if (cpu_has_xsaves)
3143                 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3144
3145         /*
3146          * Copy each region from the non-compacted offset to the
3147          * possibly compacted offset.
3148          */
3149         valid = xstate_bv & ~XSTATE_FPSSE;
3150         while (valid) {
3151                 u64 feature = valid & -valid;
3152                 int index = fls64(feature) - 1;
3153                 void *dest = get_xsave_addr(xsave, feature);
3154
3155                 if (dest) {
3156                         u32 size, offset, ecx, edx;
3157                         cpuid_count(XSTATE_CPUID, index,
3158                                     &size, &offset, &ecx, &edx);
3159                         memcpy(dest, src + offset, size);
3160                 }
3161
3162                 valid -= feature;
3163         }
3164 }
3165
3166 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3167                                          struct kvm_xsave *guest_xsave)
3168 {
3169         if (cpu_has_xsave) {
3170                 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3171                 fill_xsave((u8 *) guest_xsave->region, vcpu);
3172         } else {
3173                 memcpy(guest_xsave->region,
3174                         &vcpu->arch.guest_fpu.state.fxsave,
3175                         sizeof(struct fxregs_state));
3176                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3177                         XSTATE_FPSSE;
3178         }
3179 }
3180
3181 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3182                                         struct kvm_xsave *guest_xsave)
3183 {
3184         u64 xstate_bv =
3185                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3186
3187         if (cpu_has_xsave) {
3188                 /*
3189                  * Here we allow setting states that are not present in
3190                  * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3191                  * with old userspace.
3192                  */
3193                 if (xstate_bv & ~kvm_supported_xcr0())
3194                         return -EINVAL;
3195                 load_xsave(vcpu, (u8 *)guest_xsave->region);
3196         } else {
3197                 if (xstate_bv & ~XSTATE_FPSSE)
3198                         return -EINVAL;
3199                 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3200                         guest_xsave->region, sizeof(struct fxregs_state));
3201         }
3202         return 0;
3203 }
3204
3205 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3206                                         struct kvm_xcrs *guest_xcrs)
3207 {
3208         if (!cpu_has_xsave) {
3209                 guest_xcrs->nr_xcrs = 0;
3210                 return;
3211         }
3212
3213         guest_xcrs->nr_xcrs = 1;
3214         guest_xcrs->flags = 0;
3215         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3216         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3217 }
3218
3219 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3220                                        struct kvm_xcrs *guest_xcrs)
3221 {
3222         int i, r = 0;
3223
3224         if (!cpu_has_xsave)
3225                 return -EINVAL;
3226
3227         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3228                 return -EINVAL;
3229
3230         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3231                 /* Only support XCR0 currently */
3232                 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3233                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3234                                 guest_xcrs->xcrs[i].value);
3235                         break;
3236                 }
3237         if (r)
3238                 r = -EINVAL;
3239         return r;
3240 }
3241
3242 /*
3243  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3244  * stopped by the hypervisor.  This function will be called from the host only.
3245  * EINVAL is returned when the host attempts to set the flag for a guest that
3246  * does not support pv clocks.
3247  */
3248 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3249 {
3250         if (!vcpu->arch.pv_time_enabled)
3251                 return -EINVAL;
3252         vcpu->arch.pvclock_set_guest_stopped_request = true;
3253         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3254         return 0;
3255 }
3256
3257 long kvm_arch_vcpu_ioctl(struct file *filp,
3258                          unsigned int ioctl, unsigned long arg)
3259 {
3260         struct kvm_vcpu *vcpu = filp->private_data;
3261         void __user *argp = (void __user *)arg;
3262         int r;
3263         union {
3264                 struct kvm_lapic_state *lapic;
3265                 struct kvm_xsave *xsave;
3266                 struct kvm_xcrs *xcrs;
3267                 void *buffer;
3268         } u;
3269
3270         u.buffer = NULL;
3271         switch (ioctl) {
3272         case KVM_GET_LAPIC: {
3273                 r = -EINVAL;
3274                 if (!vcpu->arch.apic)
3275                         goto out;
3276                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3277
3278                 r = -ENOMEM;
3279                 if (!u.lapic)
3280                         goto out;
3281                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3282                 if (r)
3283                         goto out;
3284                 r = -EFAULT;
3285                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3286                         goto out;
3287                 r = 0;
3288                 break;
3289         }
3290         case KVM_SET_LAPIC: {
3291                 r = -EINVAL;
3292                 if (!vcpu->arch.apic)
3293                         goto out;
3294                 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3295                 if (IS_ERR(u.lapic))
3296                         return PTR_ERR(u.lapic);
3297
3298                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3299                 break;
3300         }
3301         case KVM_INTERRUPT: {
3302                 struct kvm_interrupt irq;
3303
3304                 r = -EFAULT;
3305                 if (copy_from_user(&irq, argp, sizeof irq))
3306                         goto out;
3307                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3308                 break;
3309         }
3310         case KVM_NMI: {
3311                 r = kvm_vcpu_ioctl_nmi(vcpu);
3312                 break;
3313         }
3314         case KVM_SMI: {
3315                 r = kvm_vcpu_ioctl_smi(vcpu);
3316                 break;
3317         }
3318         case KVM_SET_CPUID: {
3319                 struct kvm_cpuid __user *cpuid_arg = argp;
3320                 struct kvm_cpuid cpuid;
3321
3322                 r = -EFAULT;
3323                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3324                         goto out;
3325                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3326                 break;
3327         }
3328         case KVM_SET_CPUID2: {
3329                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3330                 struct kvm_cpuid2 cpuid;
3331
3332                 r = -EFAULT;
3333                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3334                         goto out;
3335                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3336                                               cpuid_arg->entries);
3337                 break;
3338         }
3339         case KVM_GET_CPUID2: {
3340                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3341                 struct kvm_cpuid2 cpuid;
3342
3343                 r = -EFAULT;
3344                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3345                         goto out;
3346                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3347                                               cpuid_arg->entries);
3348                 if (r)
3349                         goto out;
3350                 r = -EFAULT;
3351                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3352                         goto out;
3353                 r = 0;
3354                 break;
3355         }
3356         case KVM_GET_MSRS:
3357                 r = msr_io(vcpu, argp, do_get_msr, 1);
3358                 break;
3359         case KVM_SET_MSRS:
3360                 r = msr_io(vcpu, argp, do_set_msr, 0);
3361                 break;
3362         case KVM_TPR_ACCESS_REPORTING: {
3363                 struct kvm_tpr_access_ctl tac;
3364
3365                 r = -EFAULT;
3366                 if (copy_from_user(&tac, argp, sizeof tac))
3367                         goto out;
3368                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3369                 if (r)
3370                         goto out;
3371                 r = -EFAULT;
3372                 if (copy_to_user(argp, &tac, sizeof tac))
3373                         goto out;
3374                 r = 0;
3375                 break;
3376         };
3377         case KVM_SET_VAPIC_ADDR: {
3378                 struct kvm_vapic_addr va;
3379
3380                 r = -EINVAL;
3381                 if (!irqchip_in_kernel(vcpu->kvm))
3382                         goto out;
3383                 r = -EFAULT;
3384                 if (copy_from_user(&va, argp, sizeof va))
3385                         goto out;
3386                 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3387                 break;
3388         }
3389         case KVM_X86_SETUP_MCE: {
3390                 u64 mcg_cap;
3391
3392                 r = -EFAULT;
3393                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3394                         goto out;
3395                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3396                 break;
3397         }
3398         case KVM_X86_SET_MCE: {
3399                 struct kvm_x86_mce mce;
3400
3401                 r = -EFAULT;
3402                 if (copy_from_user(&mce, argp, sizeof mce))
3403                         goto out;
3404                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3405                 break;
3406         }
3407         case KVM_GET_VCPU_EVENTS: {
3408                 struct kvm_vcpu_events events;
3409
3410                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3411
3412                 r = -EFAULT;
3413                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3414                         break;
3415                 r = 0;
3416                 break;
3417         }
3418         case KVM_SET_VCPU_EVENTS: {
3419                 struct kvm_vcpu_events events;
3420
3421                 r = -EFAULT;
3422                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3423                         break;
3424
3425                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3426                 break;
3427         }
3428         case KVM_GET_DEBUGREGS: {
3429                 struct kvm_debugregs dbgregs;
3430
3431                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3432
3433                 r = -EFAULT;
3434                 if (copy_to_user(argp, &dbgregs,
3435                                  sizeof(struct kvm_debugregs)))
3436                         break;
3437                 r = 0;
3438                 break;
3439         }
3440         case KVM_SET_DEBUGREGS: {
3441                 struct kvm_debugregs dbgregs;
3442
3443                 r = -EFAULT;
3444                 if (copy_from_user(&dbgregs, argp,
3445                                    sizeof(struct kvm_debugregs)))
3446                         break;
3447
3448                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3449                 break;
3450         }
3451         case KVM_GET_XSAVE: {
3452                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3453                 r = -ENOMEM;
3454                 if (!u.xsave)
3455                         break;
3456
3457                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3458
3459                 r = -EFAULT;
3460                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3461                         break;
3462                 r = 0;
3463                 break;
3464         }
3465         case KVM_SET_XSAVE: {
3466                 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3467                 if (IS_ERR(u.xsave))
3468                         return PTR_ERR(u.xsave);
3469
3470                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3471                 break;
3472         }
3473         case KVM_GET_XCRS: {
3474                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3475                 r = -ENOMEM;
3476                 if (!u.xcrs)
3477                         break;
3478
3479                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3480
3481                 r = -EFAULT;
3482                 if (copy_to_user(argp, u.xcrs,
3483                                  sizeof(struct kvm_xcrs)))
3484                         break;
3485                 r = 0;
3486                 break;
3487         }
3488         case KVM_SET_XCRS: {
3489                 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3490                 if (IS_ERR(u.xcrs))
3491                         return PTR_ERR(u.xcrs);
3492
3493                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3494                 break;
3495         }
3496         case KVM_SET_TSC_KHZ: {
3497                 u32 user_tsc_khz;
3498
3499                 r = -EINVAL;
3500                 user_tsc_khz = (u32)arg;
3501
3502                 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3503                         goto out;
3504
3505                 if (user_tsc_khz == 0)
3506                         user_tsc_khz = tsc_khz;
3507
3508                 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3509
3510                 r = 0;
3511                 goto out;
3512         }
3513         case KVM_GET_TSC_KHZ: {
3514                 r = vcpu->arch.virtual_tsc_khz;
3515                 goto out;
3516         }
3517         case KVM_KVMCLOCK_CTRL: {
3518                 r = kvm_set_guest_paused(vcpu);
3519                 goto out;
3520         }
3521         default:
3522                 r = -EINVAL;
3523         }
3524 out:
3525         kfree(u.buffer);
3526         return r;
3527 }
3528
3529 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3530 {
3531         return VM_FAULT_SIGBUS;
3532 }
3533
3534 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3535 {
3536         int ret;
3537
3538         if (addr > (unsigned int)(-3 * PAGE_SIZE))
3539                 return -EINVAL;
3540         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3541         return ret;
3542 }
3543
3544 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3545                                               u64 ident_addr)
3546 {
3547         kvm->arch.ept_identity_map_addr = ident_addr;
3548         return 0;
3549 }
3550
3551 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3552                                           u32 kvm_nr_mmu_pages)
3553 {
3554         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3555                 return -EINVAL;
3556
3557         mutex_lock(&kvm->slots_lock);
3558
3559         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3560         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3561
3562         mutex_unlock(&kvm->slots_lock);
3563         return 0;
3564 }
3565
3566 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3567 {
3568         return kvm->arch.n_max_mmu_pages;
3569 }
3570
3571 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3572 {
3573         int r;
3574
3575         r = 0;
3576         switch (chip->chip_id) {
3577         case KVM_IRQCHIP_PIC_MASTER:
3578                 memcpy(&chip->chip.pic,
3579                         &pic_irqchip(kvm)->pics[0],
3580                         sizeof(struct kvm_pic_state));
3581                 break;
3582         case KVM_IRQCHIP_PIC_SLAVE:
3583                 memcpy(&chip->chip.pic,
3584                         &pic_irqchip(kvm)->pics[1],
3585                         sizeof(struct kvm_pic_state));
3586                 break;
3587         case KVM_IRQCHIP_IOAPIC:
3588                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3589                 break;
3590         default:
3591                 r = -EINVAL;
3592                 break;
3593         }
3594         return r;
3595 }
3596
3597 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3598 {
3599         int r;
3600
3601         r = 0;
3602         switch (chip->chip_id) {
3603         case KVM_IRQCHIP_PIC_MASTER:
3604                 spin_lock(&pic_irqchip(kvm)->lock);
3605                 memcpy(&pic_irqchip(kvm)->pics[0],
3606                         &chip->chip.pic,
3607                         sizeof(struct kvm_pic_state));
3608                 spin_unlock(&pic_irqchip(kvm)->lock);
3609                 break;
3610         case KVM_IRQCHIP_PIC_SLAVE:
3611                 spin_lock(&pic_irqchip(kvm)->lock);
3612                 memcpy(&pic_irqchip(kvm)->pics[1],
3613                         &chip->chip.pic,
3614                         sizeof(struct kvm_pic_state));
3615                 spin_unlock(&pic_irqchip(kvm)->lock);
3616                 break;
3617         case KVM_IRQCHIP_IOAPIC:
3618                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3619                 break;
3620         default:
3621                 r = -EINVAL;
3622                 break;
3623         }
3624         kvm_pic_update_irq(pic_irqchip(kvm));
3625         return r;
3626 }
3627
3628 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3629 {
3630         int r = 0;
3631
3632         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3633         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3634         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3635         return r;
3636 }
3637
3638 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3639 {
3640         int r = 0;
3641
3642         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3643         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3644         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3645         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3646         return r;
3647 }
3648
3649 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3650 {
3651         int r = 0;
3652
3653         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3654         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3655                 sizeof(ps->channels));
3656         ps->flags = kvm->arch.vpit->pit_state.flags;
3657         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3658         memset(&ps->reserved, 0, sizeof(ps->reserved));
3659         return r;
3660 }
3661
3662 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3663 {
3664         int r = 0, start = 0;
3665         u32 prev_legacy, cur_legacy;
3666         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3667         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3668         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3669         if (!prev_legacy && cur_legacy)
3670                 start = 1;
3671         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3672                sizeof(kvm->arch.vpit->pit_state.channels));
3673         kvm->arch.vpit->pit_state.flags = ps->flags;
3674         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3675         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3676         return r;
3677 }
3678
3679 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3680                                  struct kvm_reinject_control *control)
3681 {
3682         if (!kvm->arch.vpit)
3683                 return -ENXIO;
3684         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3685         kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3686         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3687         return 0;
3688 }
3689
3690 /**
3691  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3692  * @kvm: kvm instance
3693  * @log: slot id and address to which we copy the log
3694  *
3695  * Steps 1-4 below provide general overview of dirty page logging. See
3696  * kvm_get_dirty_log_protect() function description for additional details.
3697  *
3698  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3699  * always flush the TLB (step 4) even if previous step failed  and the dirty
3700  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3701  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3702  * writes will be marked dirty for next log read.
3703  *
3704  *   1. Take a snapshot of the bit and clear it if needed.
3705  *   2. Write protect the corresponding page.
3706  *   3. Copy the snapshot to the userspace.
3707  *   4. Flush TLB's if needed.
3708  */
3709 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3710 {
3711         bool is_dirty = false;
3712         int r;
3713
3714         mutex_lock(&kvm->slots_lock);
3715
3716         /*
3717          * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3718          */
3719         if (kvm_x86_ops->flush_log_dirty)
3720                 kvm_x86_ops->flush_log_dirty(kvm);
3721
3722         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3723
3724         /*
3725          * All the TLBs can be flushed out of mmu lock, see the comments in
3726          * kvm_mmu_slot_remove_write_access().
3727          */
3728         lockdep_assert_held(&kvm->slots_lock);
3729         if (is_dirty)
3730                 kvm_flush_remote_tlbs(kvm);
3731
3732         mutex_unlock(&kvm->slots_lock);
3733         return r;
3734 }
3735
3736 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3737                         bool line_status)
3738 {
3739         if (!irqchip_in_kernel(kvm))
3740                 return -ENXIO;
3741
3742         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3743                                         irq_event->irq, irq_event->level,
3744                                         line_status);
3745         return 0;
3746 }
3747
3748 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3749                                    struct kvm_enable_cap *cap)
3750 {
3751         int r;
3752
3753         if (cap->flags)
3754                 return -EINVAL;
3755
3756         switch (cap->cap) {
3757         case KVM_CAP_DISABLE_QUIRKS:
3758                 kvm->arch.disabled_quirks = cap->args[0];
3759                 r = 0;
3760                 break;
3761         default:
3762                 r = -EINVAL;
3763                 break;
3764         }
3765         return r;
3766 }
3767
3768 long kvm_arch_vm_ioctl(struct file *filp,
3769                        unsigned int ioctl, unsigned long arg)
3770 {
3771         struct kvm *kvm = filp->private_data;
3772         void __user *argp = (void __user *)arg;
3773         int r = -ENOTTY;
3774         /*
3775          * This union makes it completely explicit to gcc-3.x
3776          * that these two variables' stack usage should be
3777          * combined, not added together.
3778          */
3779         union {
3780                 struct kvm_pit_state ps;
3781                 struct kvm_pit_state2 ps2;
3782                 struct kvm_pit_config pit_config;
3783         } u;
3784
3785         switch (ioctl) {
3786         case KVM_SET_TSS_ADDR:
3787                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3788                 break;
3789         case KVM_SET_IDENTITY_MAP_ADDR: {
3790                 u64 ident_addr;
3791
3792                 r = -EFAULT;
3793                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3794                         goto out;
3795                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3796                 break;
3797         }
3798         case KVM_SET_NR_MMU_PAGES:
3799                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3800                 break;
3801         case KVM_GET_NR_MMU_PAGES:
3802                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3803                 break;
3804         case KVM_CREATE_IRQCHIP: {
3805                 struct kvm_pic *vpic;
3806
3807                 mutex_lock(&kvm->lock);
3808                 r = -EEXIST;
3809                 if (kvm->arch.vpic)
3810                         goto create_irqchip_unlock;
3811                 r = -EINVAL;
3812                 if (atomic_read(&kvm->online_vcpus))
3813                         goto create_irqchip_unlock;
3814                 r = -ENOMEM;
3815                 vpic = kvm_create_pic(kvm);
3816                 if (vpic) {
3817                         r = kvm_ioapic_init(kvm);
3818                         if (r) {
3819                                 mutex_lock(&kvm->slots_lock);
3820                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3821                                                           &vpic->dev_master);
3822                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3823                                                           &vpic->dev_slave);
3824                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3825                                                           &vpic->dev_eclr);
3826                                 mutex_unlock(&kvm->slots_lock);
3827                                 kfree(vpic);
3828                                 goto create_irqchip_unlock;
3829                         }
3830                 } else
3831                         goto create_irqchip_unlock;
3832                 smp_wmb();
3833                 kvm->arch.vpic = vpic;
3834                 smp_wmb();
3835                 r = kvm_setup_default_irq_routing(kvm);
3836                 if (r) {
3837                         mutex_lock(&kvm->slots_lock);
3838                         mutex_lock(&kvm->irq_lock);
3839                         kvm_ioapic_destroy(kvm);
3840                         kvm_destroy_pic(kvm);
3841                         mutex_unlock(&kvm->irq_lock);
3842                         mutex_unlock(&kvm->slots_lock);
3843                 }
3844         create_irqchip_unlock:
3845                 mutex_unlock(&kvm->lock);
3846                 break;
3847         }
3848         case KVM_CREATE_PIT:
3849                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3850                 goto create_pit;
3851         case KVM_CREATE_PIT2:
3852                 r = -EFAULT;
3853                 if (copy_from_user(&u.pit_config, argp,
3854                                    sizeof(struct kvm_pit_config)))
3855                         goto out;
3856         create_pit:
3857                 mutex_lock(&kvm->slots_lock);
3858                 r = -EEXIST;
3859                 if (kvm->arch.vpit)
3860                         goto create_pit_unlock;
3861                 r = -ENOMEM;
3862                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3863                 if (kvm->arch.vpit)
3864                         r = 0;
3865         create_pit_unlock:
3866                 mutex_unlock(&kvm->slots_lock);
3867                 break;
3868         case KVM_GET_IRQCHIP: {
3869                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3870                 struct kvm_irqchip *chip;
3871
3872                 chip = memdup_user(argp, sizeof(*chip));
3873                 if (IS_ERR(chip)) {
3874                         r = PTR_ERR(chip);
3875                         goto out;
3876                 }
3877
3878                 r = -ENXIO;
3879                 if (!irqchip_in_kernel(kvm))
3880                         goto get_irqchip_out;
3881                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3882                 if (r)
3883                         goto get_irqchip_out;
3884                 r = -EFAULT;
3885                 if (copy_to_user(argp, chip, sizeof *chip))
3886                         goto get_irqchip_out;
3887                 r = 0;
3888         get_irqchip_out:
3889                 kfree(chip);
3890                 break;
3891         }
3892         case KVM_SET_IRQCHIP: {
3893                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3894                 struct kvm_irqchip *chip;
3895
3896                 chip = memdup_user(argp, sizeof(*chip));
3897                 if (IS_ERR(chip)) {
3898                         r = PTR_ERR(chip);
3899                         goto out;
3900                 }
3901
3902                 r = -ENXIO;
3903                 if (!irqchip_in_kernel(kvm))
3904                         goto set_irqchip_out;
3905                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3906                 if (r)
3907                         goto set_irqchip_out;
3908                 r = 0;
3909         set_irqchip_out:
3910                 kfree(chip);
3911                 break;
3912         }
3913         case KVM_GET_PIT: {
3914                 r = -EFAULT;
3915                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3916                         goto out;
3917                 r = -ENXIO;
3918                 if (!kvm->arch.vpit)
3919                         goto out;
3920                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3921                 if (r)
3922                         goto out;
3923                 r = -EFAULT;
3924                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3925                         goto out;
3926                 r = 0;
3927                 break;
3928         }
3929         case KVM_SET_PIT: {
3930                 r = -EFAULT;
3931                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3932                         goto out;
3933                 r = -ENXIO;
3934                 if (!kvm->arch.vpit)
3935                         goto out;
3936                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3937                 break;
3938         }
3939         case KVM_GET_PIT2: {
3940                 r = -ENXIO;
3941                 if (!kvm->arch.vpit)
3942                         goto out;
3943                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3944                 if (r)
3945                         goto out;
3946                 r = -EFAULT;
3947                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3948                         goto out;
3949                 r = 0;
3950                 break;
3951         }
3952         case KVM_SET_PIT2: {
3953                 r = -EFAULT;
3954                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3955                         goto out;
3956                 r = -ENXIO;
3957                 if (!kvm->arch.vpit)
3958                         goto out;
3959                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3960                 break;
3961         }
3962         case KVM_REINJECT_CONTROL: {
3963                 struct kvm_reinject_control control;
3964                 r =  -EFAULT;
3965                 if (copy_from_user(&control, argp, sizeof(control)))
3966                         goto out;
3967                 r = kvm_vm_ioctl_reinject(kvm, &control);
3968                 break;
3969         }
3970         case KVM_XEN_HVM_CONFIG: {
3971                 r = -EFAULT;
3972                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3973                                    sizeof(struct kvm_xen_hvm_config)))
3974                         goto out;
3975                 r = -EINVAL;
3976                 if (kvm->arch.xen_hvm_config.flags)
3977                         goto out;
3978                 r = 0;
3979                 break;
3980         }
3981         case KVM_SET_CLOCK: {
3982                 struct kvm_clock_data user_ns;
3983                 u64 now_ns;
3984                 s64 delta;
3985
3986                 r = -EFAULT;
3987                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3988                         goto out;
3989
3990                 r = -EINVAL;
3991                 if (user_ns.flags)
3992                         goto out;
3993
3994                 r = 0;
3995                 local_irq_disable();
3996                 now_ns = get_kernel_ns();
3997                 delta = user_ns.clock - now_ns;
3998                 local_irq_enable();
3999                 kvm->arch.kvmclock_offset = delta;
4000                 kvm_gen_update_masterclock(kvm);
4001                 break;
4002         }
4003         case KVM_GET_CLOCK: {
4004                 struct kvm_clock_data user_ns;
4005                 u64 now_ns;
4006
4007                 local_irq_disable();
4008                 now_ns = get_kernel_ns();
4009                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4010                 local_irq_enable();
4011                 user_ns.flags = 0;
4012                 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4013
4014                 r = -EFAULT;
4015                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4016                         goto out;
4017                 r = 0;
4018                 break;
4019         }
4020         case KVM_ENABLE_CAP: {
4021                 struct kvm_enable_cap cap;
4022
4023                 r = -EFAULT;
4024                 if (copy_from_user(&cap, argp, sizeof(cap)))
4025                         goto out;
4026                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4027                 break;
4028         }
4029         default:
4030                 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4031         }
4032 out:
4033         return r;
4034 }
4035
4036 static void kvm_init_msr_list(void)
4037 {
4038         u32 dummy[2];
4039         unsigned i, j;
4040
4041         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4042                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4043                         continue;
4044
4045                 /*
4046                  * Even MSRs that are valid in the host may not be exposed
4047                  * to the guests in some cases.  We could work around this
4048                  * in VMX with the generic MSR save/load machinery, but it
4049                  * is not really worthwhile since it will really only
4050                  * happen with nested virtualization.
4051                  */
4052                 switch (msrs_to_save[i]) {
4053                 case MSR_IA32_BNDCFGS:
4054                         if (!kvm_x86_ops->mpx_supported())
4055                                 continue;
4056                         break;
4057                 default:
4058                         break;
4059                 }
4060
4061                 if (j < i)
4062                         msrs_to_save[j] = msrs_to_save[i];
4063                 j++;
4064         }
4065         num_msrs_to_save = j;
4066
4067         for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4068                 switch (emulated_msrs[i]) {
4069                 case MSR_IA32_SMBASE:
4070                         if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4071                                 continue;
4072                         break;
4073                 default:
4074                         break;
4075                 }
4076
4077                 if (j < i)
4078                         emulated_msrs[j] = emulated_msrs[i];
4079                 j++;
4080         }
4081         num_emulated_msrs = j;
4082 }
4083
4084 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4085                            const void *v)
4086 {
4087         int handled = 0;
4088         int n;
4089
4090         do {
4091                 n = min(len, 8);
4092                 if (!(vcpu->arch.apic &&
4093                       !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4094                     && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4095                         break;
4096                 handled += n;
4097                 addr += n;
4098                 len -= n;
4099                 v += n;
4100         } while (len);
4101
4102         return handled;
4103 }
4104
4105 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4106 {
4107         int handled = 0;
4108         int n;
4109
4110         do {
4111                 n = min(len, 8);
4112                 if (!(vcpu->arch.apic &&
4113                       !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4114                                          addr, n, v))
4115                     && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4116                         break;
4117                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4118                 handled += n;
4119                 addr += n;
4120                 len -= n;
4121                 v += n;
4122         } while (len);
4123
4124         return handled;
4125 }
4126
4127 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4128                         struct kvm_segment *var, int seg)
4129 {
4130         kvm_x86_ops->set_segment(vcpu, var, seg);
4131 }
4132
4133 void kvm_get_segment(struct kvm_vcpu *vcpu,
4134                      struct kvm_segment *var, int seg)
4135 {
4136         kvm_x86_ops->get_segment(vcpu, var, seg);
4137 }
4138
4139 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4140                            struct x86_exception *exception)
4141 {
4142         gpa_t t_gpa;
4143
4144         BUG_ON(!mmu_is_nested(vcpu));
4145
4146         /* NPT walks are always user-walks */
4147         access |= PFERR_USER_MASK;
4148         t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4149
4150         return t_gpa;
4151 }
4152
4153 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4154                               struct x86_exception *exception)
4155 {
4156         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4157         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4158 }
4159
4160  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4161                                 struct x86_exception *exception)
4162 {
4163         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4164         access |= PFERR_FETCH_MASK;
4165         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4166 }
4167
4168 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4169                                struct x86_exception *exception)
4170 {
4171         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4172         access |= PFERR_WRITE_MASK;
4173         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4174 }
4175
4176 /* uses this to access any guest's mapped memory without checking CPL */
4177 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4178                                 struct x86_exception *exception)
4179 {
4180         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4181 }
4182
4183 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4184                                       struct kvm_vcpu *vcpu, u32 access,
4185                                       struct x86_exception *exception)
4186 {
4187         void *data = val;
4188         int r = X86EMUL_CONTINUE;
4189
4190         while (bytes) {
4191                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4192                                                             exception);
4193                 unsigned offset = addr & (PAGE_SIZE-1);
4194                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4195                 int ret;
4196
4197                 if (gpa == UNMAPPED_GVA)
4198                         return X86EMUL_PROPAGATE_FAULT;
4199                 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4200                                                offset, toread);
4201                 if (ret < 0) {
4202                         r = X86EMUL_IO_NEEDED;
4203                         goto out;
4204                 }
4205
4206                 bytes -= toread;
4207                 data += toread;
4208                 addr += toread;
4209         }
4210 out:
4211         return r;
4212 }
4213
4214 /* used for instruction fetching */
4215 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4216                                 gva_t addr, void *val, unsigned int bytes,
4217                                 struct x86_exception *exception)
4218 {
4219         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4220         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4221         unsigned offset;
4222         int ret;
4223
4224         /* Inline kvm_read_guest_virt_helper for speed.  */
4225         gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4226                                                     exception);
4227         if (unlikely(gpa == UNMAPPED_GVA))
4228                 return X86EMUL_PROPAGATE_FAULT;
4229
4230         offset = addr & (PAGE_SIZE-1);
4231         if (WARN_ON(offset + bytes > PAGE_SIZE))
4232                 bytes = (unsigned)PAGE_SIZE - offset;
4233         ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4234                                        offset, bytes);
4235         if (unlikely(ret < 0))
4236                 return X86EMUL_IO_NEEDED;
4237
4238         return X86EMUL_CONTINUE;
4239 }
4240
4241 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4242                                gva_t addr, void *val, unsigned int bytes,
4243                                struct x86_exception *exception)
4244 {
4245         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4246         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4247
4248         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4249                                           exception);
4250 }
4251 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4252
4253 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4254                                       gva_t addr, void *val, unsigned int bytes,
4255                                       struct x86_exception *exception)
4256 {
4257         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4258         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4259 }
4260
4261 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4262                                        gva_t addr, void *val,
4263                                        unsigned int bytes,
4264                                        struct x86_exception *exception)
4265 {
4266         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4267         void *data = val;
4268         int r = X86EMUL_CONTINUE;
4269
4270         while (bytes) {
4271                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4272                                                              PFERR_WRITE_MASK,
4273                                                              exception);
4274                 unsigned offset = addr & (PAGE_SIZE-1);
4275                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4276                 int ret;
4277
4278                 if (gpa == UNMAPPED_GVA)
4279                         return X86EMUL_PROPAGATE_FAULT;
4280                 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4281                 if (ret < 0) {
4282                         r = X86EMUL_IO_NEEDED;
4283                         goto out;
4284                 }
4285
4286                 bytes -= towrite;
4287                 data += towrite;
4288                 addr += towrite;
4289         }
4290 out:
4291         return r;
4292 }
4293 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4294
4295 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4296                                 gpa_t *gpa, struct x86_exception *exception,
4297                                 bool write)
4298 {
4299         u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4300                 | (write ? PFERR_WRITE_MASK : 0);
4301
4302         if (vcpu_match_mmio_gva(vcpu, gva)
4303             && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4304                                  vcpu->arch.access, access)) {
4305                 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4306                                         (gva & (PAGE_SIZE - 1));
4307                 trace_vcpu_match_mmio(gva, *gpa, write, false);
4308                 return 1;
4309         }
4310
4311         *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4312
4313         if (*gpa == UNMAPPED_GVA)
4314                 return -1;
4315
4316         /* For APIC access vmexit */
4317         if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4318                 return 1;
4319
4320         if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4321                 trace_vcpu_match_mmio(gva, *gpa, write, true);
4322                 return 1;
4323         }
4324
4325         return 0;
4326 }
4327
4328 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4329                         const void *val, int bytes)
4330 {
4331         int ret;
4332
4333         ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4334         if (ret < 0)
4335                 return 0;
4336         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4337         return 1;
4338 }
4339
4340 struct read_write_emulator_ops {
4341         int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4342                                   int bytes);
4343         int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4344                                   void *val, int bytes);
4345         int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4346                                int bytes, void *val);
4347         int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4348                                     void *val, int bytes);
4349         bool write;
4350 };
4351
4352 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4353 {
4354         if (vcpu->mmio_read_completed) {
4355                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4356                                vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4357                 vcpu->mmio_read_completed = 0;
4358                 return 1;
4359         }
4360
4361         return 0;
4362 }
4363
4364 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4365                         void *val, int bytes)
4366 {
4367         return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4368 }
4369
4370 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4371                          void *val, int bytes)
4372 {
4373         return emulator_write_phys(vcpu, gpa, val, bytes);
4374 }
4375
4376 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4377 {
4378         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4379         return vcpu_mmio_write(vcpu, gpa, bytes, val);
4380 }
4381
4382 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4383                           void *val, int bytes)
4384 {
4385         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4386         return X86EMUL_IO_NEEDED;
4387 }
4388
4389 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4390                            void *val, int bytes)
4391 {
4392         struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4393
4394         memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4395         return X86EMUL_CONTINUE;
4396 }
4397
4398 static const struct read_write_emulator_ops read_emultor = {
4399         .read_write_prepare = read_prepare,
4400         .read_write_emulate = read_emulate,
4401         .read_write_mmio = vcpu_mmio_read,
4402         .read_write_exit_mmio = read_exit_mmio,
4403 };
4404
4405 static const struct read_write_emulator_ops write_emultor = {
4406         .read_write_emulate = write_emulate,
4407         .read_write_mmio = write_mmio,
4408         .read_write_exit_mmio = write_exit_mmio,
4409         .write = true,
4410 };
4411
4412 static int emulator_read_write_onepage(unsigned long addr, void *val,
4413                                        unsigned int bytes,
4414                                        struct x86_exception *exception,
4415                                        struct kvm_vcpu *vcpu,
4416                                        const struct read_write_emulator_ops *ops)
4417 {
4418         gpa_t gpa;
4419         int handled, ret;
4420         bool write = ops->write;
4421         struct kvm_mmio_fragment *frag;
4422
4423         ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4424
4425         if (ret < 0)
4426                 return X86EMUL_PROPAGATE_FAULT;
4427
4428         /* For APIC access vmexit */
4429         if (ret)
4430                 goto mmio;
4431
4432         if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4433                 return X86EMUL_CONTINUE;
4434
4435 mmio:
4436         /*
4437          * Is this MMIO handled locally?
4438          */
4439         handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4440         if (handled == bytes)
4441                 return X86EMUL_CONTINUE;
4442
4443         gpa += handled;
4444         bytes -= handled;
4445         val += handled;
4446
4447         WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4448         frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4449         frag->gpa = gpa;
4450         frag->data = val;
4451         frag->len = bytes;
4452         return X86EMUL_CONTINUE;
4453 }
4454
4455 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4456                         unsigned long addr,
4457                         void *val, unsigned int bytes,
4458                         struct x86_exception *exception,
4459                         const struct read_write_emulator_ops *ops)
4460 {
4461         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4462         gpa_t gpa;
4463         int rc;
4464
4465         if (ops->read_write_prepare &&
4466                   ops->read_write_prepare(vcpu, val, bytes))
4467                 return X86EMUL_CONTINUE;
4468
4469         vcpu->mmio_nr_fragments = 0;
4470
4471         /* Crossing a page boundary? */
4472         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4473                 int now;
4474
4475                 now = -addr & ~PAGE_MASK;
4476                 rc = emulator_read_write_onepage(addr, val, now, exception,
4477                                                  vcpu, ops);
4478
4479                 if (rc != X86EMUL_CONTINUE)
4480                         return rc;
4481                 addr += now;
4482                 if (ctxt->mode != X86EMUL_MODE_PROT64)
4483                         addr = (u32)addr;
4484                 val += now;
4485                 bytes -= now;
4486         }
4487
4488         rc = emulator_read_write_onepage(addr, val, bytes, exception,
4489                                          vcpu, ops);
4490         if (rc != X86EMUL_CONTINUE)
4491                 return rc;
4492
4493         if (!vcpu->mmio_nr_fragments)
4494                 return rc;
4495
4496         gpa = vcpu->mmio_fragments[0].gpa;
4497
4498         vcpu->mmio_needed = 1;
4499         vcpu->mmio_cur_fragment = 0;
4500
4501         vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4502         vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4503         vcpu->run->exit_reason = KVM_EXIT_MMIO;
4504         vcpu->run->mmio.phys_addr = gpa;
4505
4506         return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4507 }
4508
4509 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4510                                   unsigned long addr,
4511                                   void *val,
4512                                   unsigned int bytes,
4513                                   struct x86_exception *exception)
4514 {
4515         return emulator_read_write(ctxt, addr, val, bytes,
4516                                    exception, &read_emultor);
4517 }
4518
4519 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4520                             unsigned long addr,
4521                             const void *val,
4522                             unsigned int bytes,
4523                             struct x86_exception *exception)
4524 {
4525         return emulator_read_write(ctxt, addr, (void *)val, bytes,
4526                                    exception, &write_emultor);
4527 }
4528
4529 #define CMPXCHG_TYPE(t, ptr, old, new) \
4530         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4531
4532 #ifdef CONFIG_X86_64
4533 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4534 #else
4535 #  define CMPXCHG64(ptr, old, new) \
4536         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4537 #endif
4538
4539 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4540                                      unsigned long addr,
4541                                      const void *old,
4542                                      const void *new,
4543                                      unsigned int bytes,
4544                                      struct x86_exception *exception)
4545 {
4546         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4547         gpa_t gpa;
4548         struct page *page;
4549         char *kaddr;
4550         bool exchanged;
4551
4552         /* guests cmpxchg8b have to be emulated atomically */
4553         if (bytes > 8 || (bytes & (bytes - 1)))
4554                 goto emul_write;
4555
4556         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4557
4558         if (gpa == UNMAPPED_GVA ||
4559             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4560                 goto emul_write;
4561
4562         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4563                 goto emul_write;
4564
4565         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4566         if (is_error_page(page))
4567                 goto emul_write;
4568
4569         kaddr = kmap_atomic(page);
4570         kaddr += offset_in_page(gpa);
4571         switch (bytes) {
4572         case 1:
4573                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4574                 break;
4575         case 2:
4576                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4577                 break;
4578         case 4:
4579                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4580                 break;
4581         case 8:
4582                 exchanged = CMPXCHG64(kaddr, old, new);
4583                 break;
4584         default:
4585                 BUG();
4586         }
4587         kunmap_atomic(kaddr);
4588         kvm_release_page_dirty(page);
4589
4590         if (!exchanged)
4591                 return X86EMUL_CMPXCHG_FAILED;
4592
4593         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4594         kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4595
4596         return X86EMUL_CONTINUE;
4597
4598 emul_write:
4599         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4600
4601         return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4602 }
4603
4604 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4605 {
4606         /* TODO: String I/O for in kernel device */
4607         int r;
4608
4609         if (vcpu->arch.pio.in)
4610                 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4611                                     vcpu->arch.pio.size, pd);
4612         else
4613                 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4614                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
4615                                      pd);
4616         return r;
4617 }
4618
4619 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4620                                unsigned short port, void *val,
4621                                unsigned int count, bool in)
4622 {
4623         vcpu->arch.pio.port = port;
4624         vcpu->arch.pio.in = in;
4625         vcpu->arch.pio.count  = count;
4626         vcpu->arch.pio.size = size;
4627
4628         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4629                 vcpu->arch.pio.count = 0;
4630                 return 1;
4631         }
4632
4633         vcpu->run->exit_reason = KVM_EXIT_IO;
4634         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4635         vcpu->run->io.size = size;
4636         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4637         vcpu->run->io.count = count;
4638         vcpu->run->io.port = port;
4639
4640         return 0;
4641 }
4642
4643 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4644                                     int size, unsigned short port, void *val,
4645                                     unsigned int count)
4646 {
4647         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4648         int ret;
4649
4650         if (vcpu->arch.pio.count)
4651                 goto data_avail;
4652
4653         ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4654         if (ret) {
4655 data_avail:
4656                 memcpy(val, vcpu->arch.pio_data, size * count);
4657                 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4658                 vcpu->arch.pio.count = 0;
4659                 return 1;
4660         }
4661
4662         return 0;
4663 }
4664
4665 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4666                                      int size, unsigned short port,
4667                                      const void *val, unsigned int count)
4668 {
4669         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4670
4671         memcpy(vcpu->arch.pio_data, val, size * count);
4672         trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4673         return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4674 }
4675
4676 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4677 {
4678         return kvm_x86_ops->get_segment_base(vcpu, seg);
4679 }
4680
4681 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4682 {
4683         kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4684 }
4685
4686 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4687 {
4688         if (!need_emulate_wbinvd(vcpu))
4689                 return X86EMUL_CONTINUE;
4690
4691         if (kvm_x86_ops->has_wbinvd_exit()) {
4692                 int cpu = get_cpu();
4693
4694                 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4695                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4696                                 wbinvd_ipi, NULL, 1);
4697                 put_cpu();
4698                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4699         } else
4700                 wbinvd();
4701         return X86EMUL_CONTINUE;
4702 }
4703
4704 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4705 {
4706         kvm_x86_ops->skip_emulated_instruction(vcpu);
4707         return kvm_emulate_wbinvd_noskip(vcpu);
4708 }
4709 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4710
4711
4712
4713 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4714 {
4715         kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4716 }
4717
4718 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4719                            unsigned long *dest)
4720 {
4721         return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4722 }
4723
4724 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4725                            unsigned long value)
4726 {
4727
4728         return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4729 }
4730
4731 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4732 {
4733         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4734 }
4735
4736 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4737 {
4738         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4739         unsigned long value;
4740
4741         switch (cr) {
4742         case 0:
4743                 value = kvm_read_cr0(vcpu);
4744                 break;
4745         case 2:
4746                 value = vcpu->arch.cr2;
4747                 break;
4748         case 3:
4749                 value = kvm_read_cr3(vcpu);
4750                 break;
4751         case 4:
4752                 value = kvm_read_cr4(vcpu);
4753                 break;
4754         case 8:
4755                 value = kvm_get_cr8(vcpu);
4756                 break;
4757         default:
4758                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4759                 return 0;
4760         }
4761
4762         return value;
4763 }
4764
4765 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4766 {
4767         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4768         int res = 0;
4769
4770         switch (cr) {
4771         case 0:
4772                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4773                 break;
4774         case 2:
4775                 vcpu->arch.cr2 = val;
4776                 break;
4777         case 3:
4778                 res = kvm_set_cr3(vcpu, val);
4779                 break;
4780         case 4:
4781                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4782                 break;
4783         case 8:
4784                 res = kvm_set_cr8(vcpu, val);
4785                 break;
4786         default:
4787                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4788                 res = -1;
4789         }
4790
4791         return res;
4792 }
4793
4794 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4795 {
4796         return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4797 }
4798
4799 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4800 {
4801         kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4802 }
4803
4804 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4805 {
4806         kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4807 }
4808
4809 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4810 {
4811         kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4812 }
4813
4814 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4815 {
4816         kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4817 }
4818
4819 static unsigned long emulator_get_cached_segment_base(
4820         struct x86_emulate_ctxt *ctxt, int seg)
4821 {
4822         return get_segment_base(emul_to_vcpu(ctxt), seg);
4823 }
4824
4825 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4826                                  struct desc_struct *desc, u32 *base3,
4827                                  int seg)
4828 {
4829         struct kvm_segment var;
4830
4831         kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4832         *selector = var.selector;
4833
4834         if (var.unusable) {
4835                 memset(desc, 0, sizeof(*desc));
4836                 return false;
4837         }
4838
4839         if (var.g)
4840                 var.limit >>= 12;
4841         set_desc_limit(desc, var.limit);
4842         set_desc_base(desc, (unsigned long)var.base);
4843 #ifdef CONFIG_X86_64
4844         if (base3)
4845                 *base3 = var.base >> 32;
4846 #endif
4847         desc->type = var.type;
4848         desc->s = var.s;
4849         desc->dpl = var.dpl;
4850         desc->p = var.present;
4851         desc->avl = var.avl;
4852         desc->l = var.l;
4853         desc->d = var.db;
4854         desc->g = var.g;
4855
4856         return true;
4857 }
4858
4859 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4860                                  struct desc_struct *desc, u32 base3,
4861                                  int seg)
4862 {
4863         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4864         struct kvm_segment var;
4865
4866         var.selector = selector;
4867         var.base = get_desc_base(desc);
4868 #ifdef CONFIG_X86_64
4869         var.base |= ((u64)base3) << 32;
4870 #endif
4871         var.limit = get_desc_limit(desc);
4872         if (desc->g)
4873                 var.limit = (var.limit << 12) | 0xfff;
4874         var.type = desc->type;
4875         var.dpl = desc->dpl;
4876         var.db = desc->d;
4877         var.s = desc->s;
4878         var.l = desc->l;
4879         var.g = desc->g;
4880         var.avl = desc->avl;
4881         var.present = desc->p;
4882         var.unusable = !var.present;
4883         var.padding = 0;
4884
4885         kvm_set_segment(vcpu, &var, seg);
4886         return;
4887 }
4888
4889 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4890                             u32 msr_index, u64 *pdata)
4891 {
4892         struct msr_data msr;
4893         int r;
4894
4895         msr.index = msr_index;
4896         msr.host_initiated = false;
4897         r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4898         if (r)
4899                 return r;
4900
4901         *pdata = msr.data;
4902         return 0;
4903 }
4904
4905 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4906                             u32 msr_index, u64 data)
4907 {
4908         struct msr_data msr;
4909
4910         msr.data = data;
4911         msr.index = msr_index;
4912         msr.host_initiated = false;
4913         return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4914 }
4915
4916 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4917 {
4918         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4919
4920         return vcpu->arch.smbase;
4921 }
4922
4923 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4924 {
4925         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4926
4927         vcpu->arch.smbase = smbase;
4928 }
4929
4930 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4931                               u32 pmc)
4932 {
4933         return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4934 }
4935
4936 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4937                              u32 pmc, u64 *pdata)
4938 {
4939         return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4940 }
4941
4942 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4943 {
4944         emul_to_vcpu(ctxt)->arch.halt_request = 1;
4945 }
4946
4947 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4948 {
4949         preempt_disable();
4950         kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4951         /*
4952          * CR0.TS may reference the host fpu state, not the guest fpu state,
4953          * so it may be clear at this point.
4954          */
4955         clts();
4956 }
4957
4958 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4959 {
4960         preempt_enable();
4961 }
4962
4963 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4964                               struct x86_instruction_info *info,
4965                               enum x86_intercept_stage stage)
4966 {
4967         return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4968 }
4969
4970 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4971                                u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4972 {
4973         kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4974 }
4975
4976 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4977 {
4978         return kvm_register_read(emul_to_vcpu(ctxt), reg);
4979 }
4980
4981 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4982 {
4983         kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4984 }
4985
4986 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
4987 {
4988         kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
4989 }
4990
4991 static const struct x86_emulate_ops emulate_ops = {
4992         .read_gpr            = emulator_read_gpr,
4993         .write_gpr           = emulator_write_gpr,
4994         .read_std            = kvm_read_guest_virt_system,
4995         .write_std           = kvm_write_guest_virt_system,
4996         .fetch               = kvm_fetch_guest_virt,
4997         .read_emulated       = emulator_read_emulated,
4998         .write_emulated      = emulator_write_emulated,
4999         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
5000         .invlpg              = emulator_invlpg,
5001         .pio_in_emulated     = emulator_pio_in_emulated,
5002         .pio_out_emulated    = emulator_pio_out_emulated,
5003         .get_segment         = emulator_get_segment,
5004         .set_segment         = emulator_set_segment,
5005         .get_cached_segment_base = emulator_get_cached_segment_base,
5006         .get_gdt             = emulator_get_gdt,
5007         .get_idt             = emulator_get_idt,
5008         .set_gdt             = emulator_set_gdt,
5009         .set_idt             = emulator_set_idt,
5010         .get_cr              = emulator_get_cr,
5011         .set_cr              = emulator_set_cr,
5012         .cpl                 = emulator_get_cpl,
5013         .get_dr              = emulator_get_dr,
5014         .set_dr              = emulator_set_dr,
5015         .get_smbase          = emulator_get_smbase,
5016         .set_smbase          = emulator_set_smbase,
5017         .set_msr             = emulator_set_msr,
5018         .get_msr             = emulator_get_msr,
5019         .check_pmc           = emulator_check_pmc,
5020         .read_pmc            = emulator_read_pmc,
5021         .halt                = emulator_halt,
5022         .wbinvd              = emulator_wbinvd,
5023         .fix_hypercall       = emulator_fix_hypercall,
5024         .get_fpu             = emulator_get_fpu,
5025         .put_fpu             = emulator_put_fpu,
5026         .intercept           = emulator_intercept,
5027         .get_cpuid           = emulator_get_cpuid,
5028         .set_nmi_mask        = emulator_set_nmi_mask,
5029 };
5030
5031 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5032 {
5033         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5034         /*
5035          * an sti; sti; sequence only disable interrupts for the first
5036          * instruction. So, if the last instruction, be it emulated or
5037          * not, left the system with the INT_STI flag enabled, it
5038          * means that the last instruction is an sti. We should not
5039          * leave the flag on in this case. The same goes for mov ss
5040          */
5041         if (int_shadow & mask)
5042                 mask = 0;
5043         if (unlikely(int_shadow || mask)) {
5044                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5045                 if (!mask)
5046                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5047         }
5048 }
5049
5050 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5051 {
5052         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5053         if (ctxt->exception.vector == PF_VECTOR)
5054                 return kvm_propagate_fault(vcpu, &ctxt->exception);
5055
5056         if (ctxt->exception.error_code_valid)
5057                 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5058                                       ctxt->exception.error_code);
5059         else
5060                 kvm_queue_exception(vcpu, ctxt->exception.vector);
5061         return false;
5062 }
5063
5064 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5065 {
5066         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5067         int cs_db, cs_l;
5068
5069         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5070
5071         ctxt->eflags = kvm_get_rflags(vcpu);
5072         ctxt->eip = kvm_rip_read(vcpu);
5073         ctxt->mode = (!is_protmode(vcpu))               ? X86EMUL_MODE_REAL :
5074                      (ctxt->eflags & X86_EFLAGS_VM)     ? X86EMUL_MODE_VM86 :
5075                      (cs_l && is_long_mode(vcpu))       ? X86EMUL_MODE_PROT64 :
5076                      cs_db                              ? X86EMUL_MODE_PROT32 :
5077                                                           X86EMUL_MODE_PROT16;
5078         BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5079         BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5080         BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5081         ctxt->emul_flags = vcpu->arch.hflags;
5082
5083         init_decode_cache(ctxt);
5084         vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5085 }
5086
5087 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5088 {
5089         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5090         int ret;
5091
5092         init_emulate_ctxt(vcpu);
5093
5094         ctxt->op_bytes = 2;
5095         ctxt->ad_bytes = 2;
5096         ctxt->_eip = ctxt->eip + inc_eip;
5097         ret = emulate_int_real(ctxt, irq);
5098
5099         if (ret != X86EMUL_CONTINUE)
5100                 return EMULATE_FAIL;
5101
5102         ctxt->eip = ctxt->_eip;
5103         kvm_rip_write(vcpu, ctxt->eip);
5104         kvm_set_rflags(vcpu, ctxt->eflags);
5105
5106         if (irq == NMI_VECTOR)
5107                 vcpu->arch.nmi_pending = 0;
5108         else
5109                 vcpu->arch.interrupt.pending = false;
5110
5111         return EMULATE_DONE;
5112 }
5113 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5114
5115 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5116 {
5117         int r = EMULATE_DONE;
5118
5119         ++vcpu->stat.insn_emulation_fail;
5120         trace_kvm_emulate_insn_failed(vcpu);
5121         if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5122                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5123                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5124                 vcpu->run->internal.ndata = 0;
5125                 r = EMULATE_FAIL;
5126         }
5127         kvm_queue_exception(vcpu, UD_VECTOR);
5128
5129         return r;
5130 }
5131
5132 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5133                                   bool write_fault_to_shadow_pgtable,
5134                                   int emulation_type)
5135 {
5136         gpa_t gpa = cr2;
5137         pfn_t pfn;
5138
5139         if (emulation_type & EMULTYPE_NO_REEXECUTE)
5140                 return false;
5141
5142         if (!vcpu->arch.mmu.direct_map) {
5143                 /*
5144                  * Write permission should be allowed since only
5145                  * write access need to be emulated.
5146                  */
5147                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5148
5149                 /*
5150                  * If the mapping is invalid in guest, let cpu retry
5151                  * it to generate fault.
5152                  */
5153                 if (gpa == UNMAPPED_GVA)
5154                         return true;
5155         }
5156
5157         /*
5158          * Do not retry the unhandleable instruction if it faults on the
5159          * readonly host memory, otherwise it will goto a infinite loop:
5160          * retry instruction -> write #PF -> emulation fail -> retry
5161          * instruction -> ...
5162          */
5163         pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5164
5165         /*
5166          * If the instruction failed on the error pfn, it can not be fixed,
5167          * report the error to userspace.
5168          */
5169         if (is_error_noslot_pfn(pfn))
5170                 return false;
5171
5172         kvm_release_pfn_clean(pfn);
5173
5174         /* The instructions are well-emulated on direct mmu. */
5175         if (vcpu->arch.mmu.direct_map) {
5176                 unsigned int indirect_shadow_pages;
5177
5178                 spin_lock(&vcpu->kvm->mmu_lock);
5179                 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5180                 spin_unlock(&vcpu->kvm->mmu_lock);
5181
5182                 if (indirect_shadow_pages)
5183                         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5184
5185                 return true;
5186         }
5187
5188         /*
5189          * if emulation was due to access to shadowed page table
5190          * and it failed try to unshadow page and re-enter the
5191          * guest to let CPU execute the instruction.
5192          */
5193         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5194
5195         /*
5196          * If the access faults on its page table, it can not
5197          * be fixed by unprotecting shadow page and it should
5198          * be reported to userspace.
5199          */
5200         return !write_fault_to_shadow_pgtable;
5201 }
5202
5203 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5204                               unsigned long cr2,  int emulation_type)
5205 {
5206         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5207         unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5208
5209         last_retry_eip = vcpu->arch.last_retry_eip;
5210         last_retry_addr = vcpu->arch.last_retry_addr;
5211
5212         /*
5213          * If the emulation is caused by #PF and it is non-page_table
5214          * writing instruction, it means the VM-EXIT is caused by shadow
5215          * page protected, we can zap the shadow page and retry this
5216          * instruction directly.
5217          *
5218          * Note: if the guest uses a non-page-table modifying instruction
5219          * on the PDE that points to the instruction, then we will unmap
5220          * the instruction and go to an infinite loop. So, we cache the
5221          * last retried eip and the last fault address, if we meet the eip
5222          * and the address again, we can break out of the potential infinite
5223          * loop.
5224          */
5225         vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5226
5227         if (!(emulation_type & EMULTYPE_RETRY))
5228                 return false;
5229
5230         if (x86_page_table_writing_insn(ctxt))
5231                 return false;
5232
5233         if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5234                 return false;
5235
5236         vcpu->arch.last_retry_eip = ctxt->eip;
5237         vcpu->arch.last_retry_addr = cr2;
5238
5239         if (!vcpu->arch.mmu.direct_map)
5240                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5241
5242         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5243
5244         return true;
5245 }
5246
5247 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5248 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5249
5250 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5251 {
5252         if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5253                 /* This is a good place to trace that we are exiting SMM.  */
5254                 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5255
5256                 if (unlikely(vcpu->arch.smi_pending)) {
5257                         kvm_make_request(KVM_REQ_SMI, vcpu);
5258                         vcpu->arch.smi_pending = 0;
5259                 } else {
5260                         /* Process a latched INIT, if any.  */
5261                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5262                 }
5263         }
5264
5265         kvm_mmu_reset_context(vcpu);
5266 }
5267
5268 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5269 {
5270         unsigned changed = vcpu->arch.hflags ^ emul_flags;
5271
5272         vcpu->arch.hflags = emul_flags;
5273
5274         if (changed & HF_SMM_MASK)
5275                 kvm_smm_changed(vcpu);
5276 }
5277
5278 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5279                                 unsigned long *db)
5280 {
5281         u32 dr6 = 0;
5282         int i;
5283         u32 enable, rwlen;
5284
5285         enable = dr7;
5286         rwlen = dr7 >> 16;
5287         for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5288                 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5289                         dr6 |= (1 << i);
5290         return dr6;
5291 }
5292
5293 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5294 {
5295         struct kvm_run *kvm_run = vcpu->run;
5296
5297         /*
5298          * rflags is the old, "raw" value of the flags.  The new value has
5299          * not been saved yet.
5300          *
5301          * This is correct even for TF set by the guest, because "the
5302          * processor will not generate this exception after the instruction
5303          * that sets the TF flag".
5304          */
5305         if (unlikely(rflags & X86_EFLAGS_TF)) {
5306                 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5307                         kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5308                                                   DR6_RTM;
5309                         kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5310                         kvm_run->debug.arch.exception = DB_VECTOR;
5311                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
5312                         *r = EMULATE_USER_EXIT;
5313                 } else {
5314                         vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5315                         /*
5316                          * "Certain debug exceptions may clear bit 0-3.  The
5317                          * remaining contents of the DR6 register are never
5318                          * cleared by the processor".
5319                          */
5320                         vcpu->arch.dr6 &= ~15;
5321                         vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5322                         kvm_queue_exception(vcpu, DB_VECTOR);
5323                 }
5324         }
5325 }
5326
5327 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5328 {
5329         if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5330             (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5331                 struct kvm_run *kvm_run = vcpu->run;
5332                 unsigned long eip = kvm_get_linear_rip(vcpu);
5333                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5334                                            vcpu->arch.guest_debug_dr7,
5335                                            vcpu->arch.eff_db);
5336
5337                 if (dr6 != 0) {
5338                         kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5339                         kvm_run->debug.arch.pc = eip;
5340                         kvm_run->debug.arch.exception = DB_VECTOR;
5341                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
5342                         *r = EMULATE_USER_EXIT;
5343                         return true;
5344                 }
5345         }
5346
5347         if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5348             !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5349                 unsigned long eip = kvm_get_linear_rip(vcpu);
5350                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5351                                            vcpu->arch.dr7,
5352                                            vcpu->arch.db);
5353
5354                 if (dr6 != 0) {
5355                         vcpu->arch.dr6 &= ~15;
5356                         vcpu->arch.dr6 |= dr6 | DR6_RTM;
5357                         kvm_queue_exception(vcpu, DB_VECTOR);
5358                         *r = EMULATE_DONE;
5359                         return true;
5360                 }
5361         }
5362
5363         return false;
5364 }
5365
5366 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5367                             unsigned long cr2,
5368                             int emulation_type,
5369                             void *insn,
5370                             int insn_len)
5371 {
5372         int r;
5373         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5374         bool writeback = true;
5375         bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5376
5377         /*
5378          * Clear write_fault_to_shadow_pgtable here to ensure it is
5379          * never reused.
5380          */
5381         vcpu->arch.write_fault_to_shadow_pgtable = false;
5382         kvm_clear_exception_queue(vcpu);
5383
5384         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5385                 init_emulate_ctxt(vcpu);
5386
5387                 /*
5388                  * We will reenter on the same instruction since
5389                  * we do not set complete_userspace_io.  This does not
5390                  * handle watchpoints yet, those would be handled in
5391                  * the emulate_ops.
5392                  */
5393                 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5394                         return r;
5395
5396                 ctxt->interruptibility = 0;
5397                 ctxt->have_exception = false;
5398                 ctxt->exception.vector = -1;
5399                 ctxt->perm_ok = false;
5400
5401                 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5402
5403                 r = x86_decode_insn(ctxt, insn, insn_len);
5404
5405                 trace_kvm_emulate_insn_start(vcpu);
5406                 ++vcpu->stat.insn_emulation;
5407                 if (r != EMULATION_OK)  {
5408                         if (emulation_type & EMULTYPE_TRAP_UD)
5409                                 return EMULATE_FAIL;
5410                         if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5411                                                 emulation_type))
5412                                 return EMULATE_DONE;
5413                         if (emulation_type & EMULTYPE_SKIP)
5414                                 return EMULATE_FAIL;
5415                         return handle_emulation_failure(vcpu);
5416                 }
5417         }
5418
5419         if (emulation_type & EMULTYPE_SKIP) {
5420                 kvm_rip_write(vcpu, ctxt->_eip);
5421                 if (ctxt->eflags & X86_EFLAGS_RF)
5422                         kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5423                 return EMULATE_DONE;
5424         }
5425
5426         if (retry_instruction(ctxt, cr2, emulation_type))
5427                 return EMULATE_DONE;
5428
5429         /* this is needed for vmware backdoor interface to work since it
5430            changes registers values  during IO operation */
5431         if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5432                 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5433                 emulator_invalidate_register_cache(ctxt);
5434         }
5435
5436 restart:
5437         r = x86_emulate_insn(ctxt);
5438
5439         if (r == EMULATION_INTERCEPTED)
5440                 return EMULATE_DONE;
5441
5442         if (r == EMULATION_FAILED) {
5443                 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5444                                         emulation_type))
5445                         return EMULATE_DONE;
5446
5447                 return handle_emulation_failure(vcpu);
5448         }
5449
5450         if (ctxt->have_exception) {
5451                 r = EMULATE_DONE;
5452                 if (inject_emulated_exception(vcpu))
5453                         return r;
5454         } else if (vcpu->arch.pio.count) {
5455                 if (!vcpu->arch.pio.in) {
5456                         /* FIXME: return into emulator if single-stepping.  */
5457                         vcpu->arch.pio.count = 0;
5458                 } else {
5459                         writeback = false;
5460                         vcpu->arch.complete_userspace_io = complete_emulated_pio;
5461                 }
5462                 r = EMULATE_USER_EXIT;
5463         } else if (vcpu->mmio_needed) {
5464                 if (!vcpu->mmio_is_write)
5465                         writeback = false;
5466                 r = EMULATE_USER_EXIT;
5467                 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5468         } else if (r == EMULATION_RESTART)
5469                 goto restart;
5470         else
5471                 r = EMULATE_DONE;
5472
5473         if (writeback) {
5474                 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5475                 toggle_interruptibility(vcpu, ctxt->interruptibility);
5476                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5477                 if (vcpu->arch.hflags != ctxt->emul_flags)
5478                         kvm_set_hflags(vcpu, ctxt->emul_flags);
5479                 kvm_rip_write(vcpu, ctxt->eip);
5480                 if (r == EMULATE_DONE)
5481                         kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5482                 if (!ctxt->have_exception ||
5483                     exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5484                         __kvm_set_rflags(vcpu, ctxt->eflags);
5485
5486                 /*
5487                  * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5488                  * do nothing, and it will be requested again as soon as
5489                  * the shadow expires.  But we still need to check here,
5490                  * because POPF has no interrupt shadow.
5491                  */
5492                 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5493                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5494         } else
5495                 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5496
5497         return r;
5498 }
5499 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5500
5501 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5502 {
5503         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5504         int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5505                                             size, port, &val, 1);
5506         /* do not return to emulator after return from userspace */
5507         vcpu->arch.pio.count = 0;
5508         return ret;
5509 }
5510 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5511
5512 static void tsc_bad(void *info)
5513 {
5514         __this_cpu_write(cpu_tsc_khz, 0);
5515 }
5516
5517 static void tsc_khz_changed(void *data)
5518 {
5519         struct cpufreq_freqs *freq = data;
5520         unsigned long khz = 0;
5521
5522         if (data)
5523                 khz = freq->new;
5524         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5525                 khz = cpufreq_quick_get(raw_smp_processor_id());
5526         if (!khz)
5527                 khz = tsc_khz;
5528         __this_cpu_write(cpu_tsc_khz, khz);
5529 }
5530
5531 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5532                                      void *data)
5533 {
5534         struct cpufreq_freqs *freq = data;
5535         struct kvm *kvm;
5536         struct kvm_vcpu *vcpu;
5537         int i, send_ipi = 0;
5538
5539         /*
5540          * We allow guests to temporarily run on slowing clocks,
5541          * provided we notify them after, or to run on accelerating
5542          * clocks, provided we notify them before.  Thus time never
5543          * goes backwards.
5544          *
5545          * However, we have a problem.  We can't atomically update
5546          * the frequency of a given CPU from this function; it is
5547          * merely a notifier, which can be called from any CPU.
5548          * Changing the TSC frequency at arbitrary points in time
5549          * requires a recomputation of local variables related to
5550          * the TSC for each VCPU.  We must flag these local variables
5551          * to be updated and be sure the update takes place with the
5552          * new frequency before any guests proceed.
5553          *
5554          * Unfortunately, the combination of hotplug CPU and frequency
5555          * change creates an intractable locking scenario; the order
5556          * of when these callouts happen is undefined with respect to
5557          * CPU hotplug, and they can race with each other.  As such,
5558          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5559          * undefined; you can actually have a CPU frequency change take
5560          * place in between the computation of X and the setting of the
5561          * variable.  To protect against this problem, all updates of
5562          * the per_cpu tsc_khz variable are done in an interrupt
5563          * protected IPI, and all callers wishing to update the value
5564          * must wait for a synchronous IPI to complete (which is trivial
5565          * if the caller is on the CPU already).  This establishes the
5566          * necessary total order on variable updates.
5567          *
5568          * Note that because a guest time update may take place
5569          * anytime after the setting of the VCPU's request bit, the
5570          * correct TSC value must be set before the request.  However,
5571          * to ensure the update actually makes it to any guest which
5572          * starts running in hardware virtualization between the set
5573          * and the acquisition of the spinlock, we must also ping the
5574          * CPU after setting the request bit.
5575          *
5576          */
5577
5578         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5579                 return 0;
5580         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5581                 return 0;
5582
5583         smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5584
5585         spin_lock(&kvm_lock);
5586         list_for_each_entry(kvm, &vm_list, vm_list) {
5587                 kvm_for_each_vcpu(i, vcpu, kvm) {
5588                         if (vcpu->cpu != freq->cpu)
5589                                 continue;
5590                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5591                         if (vcpu->cpu != smp_processor_id())
5592                                 send_ipi = 1;
5593                 }
5594         }
5595         spin_unlock(&kvm_lock);
5596
5597         if (freq->old < freq->new && send_ipi) {
5598                 /*
5599                  * We upscale the frequency.  Must make the guest
5600                  * doesn't see old kvmclock values while running with
5601                  * the new frequency, otherwise we risk the guest sees
5602                  * time go backwards.
5603                  *
5604                  * In case we update the frequency for another cpu
5605                  * (which might be in guest context) send an interrupt
5606                  * to kick the cpu out of guest context.  Next time
5607                  * guest context is entered kvmclock will be updated,
5608                  * so the guest will not see stale values.
5609                  */
5610                 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5611         }
5612         return 0;
5613 }
5614
5615 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5616         .notifier_call  = kvmclock_cpufreq_notifier
5617 };
5618
5619 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5620                                         unsigned long action, void *hcpu)
5621 {
5622         unsigned int cpu = (unsigned long)hcpu;
5623
5624         switch (action) {
5625                 case CPU_ONLINE:
5626                 case CPU_DOWN_FAILED:
5627                         smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5628                         break;
5629                 case CPU_DOWN_PREPARE:
5630                         smp_call_function_single(cpu, tsc_bad, NULL, 1);
5631                         break;
5632         }
5633         return NOTIFY_OK;
5634 }
5635
5636 static struct notifier_block kvmclock_cpu_notifier_block = {
5637         .notifier_call  = kvmclock_cpu_notifier,
5638         .priority = -INT_MAX
5639 };
5640
5641 static void kvm_timer_init(void)
5642 {
5643         int cpu;
5644
5645         max_tsc_khz = tsc_khz;
5646
5647         cpu_notifier_register_begin();
5648         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5649 #ifdef CONFIG_CPU_FREQ
5650                 struct cpufreq_policy policy;
5651                 memset(&policy, 0, sizeof(policy));
5652                 cpu = get_cpu();
5653                 cpufreq_get_policy(&policy, cpu);
5654                 if (policy.cpuinfo.max_freq)
5655                         max_tsc_khz = policy.cpuinfo.max_freq;
5656                 put_cpu();
5657 #endif
5658                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5659                                           CPUFREQ_TRANSITION_NOTIFIER);
5660         }
5661         pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5662         for_each_online_cpu(cpu)
5663                 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5664
5665         __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5666         cpu_notifier_register_done();
5667
5668 }
5669
5670 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5671
5672 int kvm_is_in_guest(void)
5673 {
5674         return __this_cpu_read(current_vcpu) != NULL;
5675 }
5676
5677 static int kvm_is_user_mode(void)
5678 {
5679         int user_mode = 3;
5680
5681         if (__this_cpu_read(current_vcpu))
5682                 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5683
5684         return user_mode != 0;
5685 }
5686
5687 static unsigned long kvm_get_guest_ip(void)
5688 {
5689         unsigned long ip = 0;
5690
5691         if (__this_cpu_read(current_vcpu))
5692                 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5693
5694         return ip;
5695 }
5696
5697 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5698         .is_in_guest            = kvm_is_in_guest,
5699         .is_user_mode           = kvm_is_user_mode,
5700         .get_guest_ip           = kvm_get_guest_ip,
5701 };
5702
5703 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5704 {
5705         __this_cpu_write(current_vcpu, vcpu);
5706 }
5707 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5708
5709 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5710 {
5711         __this_cpu_write(current_vcpu, NULL);
5712 }
5713 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5714
5715 static void kvm_set_mmio_spte_mask(void)
5716 {
5717         u64 mask;
5718         int maxphyaddr = boot_cpu_data.x86_phys_bits;
5719
5720         /*
5721          * Set the reserved bits and the present bit of an paging-structure
5722          * entry to generate page fault with PFER.RSV = 1.
5723          */
5724          /* Mask the reserved physical address bits. */
5725         mask = rsvd_bits(maxphyaddr, 51);
5726
5727         /* Bit 62 is always reserved for 32bit host. */
5728         mask |= 0x3ull << 62;
5729
5730         /* Set the present bit. */
5731         mask |= 1ull;
5732
5733 #ifdef CONFIG_X86_64
5734         /*
5735          * If reserved bit is not supported, clear the present bit to disable
5736          * mmio page fault.
5737          */
5738         if (maxphyaddr == 52)
5739                 mask &= ~1ull;
5740 #endif
5741
5742         kvm_mmu_set_mmio_spte_mask(mask);
5743 }
5744
5745 #ifdef CONFIG_X86_64
5746 static void pvclock_gtod_update_fn(struct work_struct *work)
5747 {
5748         struct kvm *kvm;
5749
5750         struct kvm_vcpu *vcpu;
5751         int i;
5752
5753         spin_lock(&kvm_lock);
5754         list_for_each_entry(kvm, &vm_list, vm_list)
5755                 kvm_for_each_vcpu(i, vcpu, kvm)
5756                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5757         atomic_set(&kvm_guest_has_master_clock, 0);
5758         spin_unlock(&kvm_lock);
5759 }
5760
5761 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5762
5763 /*
5764  * Notification about pvclock gtod data update.
5765  */
5766 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5767                                void *priv)
5768 {
5769         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5770         struct timekeeper *tk = priv;
5771
5772         update_pvclock_gtod(tk);
5773
5774         /* disable master clock if host does not trust, or does not
5775          * use, TSC clocksource
5776          */
5777         if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5778             atomic_read(&kvm_guest_has_master_clock) != 0)
5779                 queue_work(system_long_wq, &pvclock_gtod_work);
5780
5781         return 0;
5782 }
5783
5784 static struct notifier_block pvclock_gtod_notifier = {
5785         .notifier_call = pvclock_gtod_notify,
5786 };
5787 #endif
5788
5789 int kvm_arch_init(void *opaque)
5790 {
5791         int r;
5792         struct kvm_x86_ops *ops = opaque;
5793
5794         if (kvm_x86_ops) {
5795                 printk(KERN_ERR "kvm: already loaded the other module\n");
5796                 r = -EEXIST;
5797                 goto out;
5798         }
5799
5800         if (!ops->cpu_has_kvm_support()) {
5801                 printk(KERN_ERR "kvm: no hardware support\n");
5802                 r = -EOPNOTSUPP;
5803                 goto out;
5804         }
5805         if (ops->disabled_by_bios()) {
5806                 printk(KERN_ERR "kvm: disabled by bios\n");
5807                 r = -EOPNOTSUPP;
5808                 goto out;
5809         }
5810
5811         r = -ENOMEM;
5812         shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5813         if (!shared_msrs) {
5814                 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5815                 goto out;
5816         }
5817
5818         r = kvm_mmu_module_init();
5819         if (r)
5820                 goto out_free_percpu;
5821
5822         kvm_set_mmio_spte_mask();
5823
5824         kvm_x86_ops = ops;
5825
5826         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5827                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
5828
5829         kvm_timer_init();
5830
5831         perf_register_guest_info_callbacks(&kvm_guest_cbs);
5832
5833         if (cpu_has_xsave)
5834                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5835
5836         kvm_lapic_init();
5837 #ifdef CONFIG_X86_64
5838         pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5839 #endif
5840
5841         return 0;
5842
5843 out_free_percpu:
5844         free_percpu(shared_msrs);
5845 out:
5846         return r;
5847 }
5848
5849 void kvm_arch_exit(void)
5850 {
5851         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5852
5853         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5854                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5855                                             CPUFREQ_TRANSITION_NOTIFIER);
5856         unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5857 #ifdef CONFIG_X86_64
5858         pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5859 #endif
5860         kvm_x86_ops = NULL;
5861         kvm_mmu_module_exit();
5862         free_percpu(shared_msrs);
5863 }
5864
5865 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5866 {
5867         ++vcpu->stat.halt_exits;
5868         if (irqchip_in_kernel(vcpu->kvm)) {
5869                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5870                 return 1;
5871         } else {
5872                 vcpu->run->exit_reason = KVM_EXIT_HLT;
5873                 return 0;
5874         }
5875 }
5876 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5877
5878 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5879 {
5880         kvm_x86_ops->skip_emulated_instruction(vcpu);
5881         return kvm_vcpu_halt(vcpu);
5882 }
5883 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5884
5885 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5886 {
5887         u64 param, ingpa, outgpa, ret;
5888         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5889         bool fast, longmode;
5890
5891         /*
5892          * hypercall generates UD from non zero cpl and real mode
5893          * per HYPER-V spec
5894          */
5895         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5896                 kvm_queue_exception(vcpu, UD_VECTOR);
5897                 return 0;
5898         }
5899
5900         longmode = is_64_bit_mode(vcpu);
5901
5902         if (!longmode) {
5903                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5904                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5905                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5906                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5907                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5908                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5909         }
5910 #ifdef CONFIG_X86_64
5911         else {
5912                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5913                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5914                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5915         }
5916 #endif
5917
5918         code = param & 0xffff;
5919         fast = (param >> 16) & 0x1;
5920         rep_cnt = (param >> 32) & 0xfff;
5921         rep_idx = (param >> 48) & 0xfff;
5922
5923         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5924
5925         switch (code) {
5926         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5927                 kvm_vcpu_on_spin(vcpu);
5928                 break;
5929         default:
5930                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5931                 break;
5932         }
5933
5934         ret = res | (((u64)rep_done & 0xfff) << 32);
5935         if (longmode) {
5936                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5937         } else {
5938                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5939                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5940         }
5941
5942         return 1;
5943 }
5944
5945 /*
5946  * kvm_pv_kick_cpu_op:  Kick a vcpu.
5947  *
5948  * @apicid - apicid of vcpu to be kicked.
5949  */
5950 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5951 {
5952         struct kvm_lapic_irq lapic_irq;
5953
5954         lapic_irq.shorthand = 0;
5955         lapic_irq.dest_mode = 0;
5956         lapic_irq.dest_id = apicid;
5957         lapic_irq.msi_redir_hint = false;
5958
5959         lapic_irq.delivery_mode = APIC_DM_REMRD;
5960         kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5961 }
5962
5963 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5964 {
5965         unsigned long nr, a0, a1, a2, a3, ret;
5966         int op_64_bit, r = 1;
5967
5968         kvm_x86_ops->skip_emulated_instruction(vcpu);
5969
5970         if (kvm_hv_hypercall_enabled(vcpu->kvm))
5971                 return kvm_hv_hypercall(vcpu);
5972
5973         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5974         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5975         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5976         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5977         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5978
5979         trace_kvm_hypercall(nr, a0, a1, a2, a3);
5980
5981         op_64_bit = is_64_bit_mode(vcpu);
5982         if (!op_64_bit) {
5983                 nr &= 0xFFFFFFFF;
5984                 a0 &= 0xFFFFFFFF;
5985                 a1 &= 0xFFFFFFFF;
5986                 a2 &= 0xFFFFFFFF;
5987                 a3 &= 0xFFFFFFFF;
5988         }
5989
5990         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5991                 ret = -KVM_EPERM;
5992                 goto out;
5993         }
5994
5995         switch (nr) {
5996         case KVM_HC_VAPIC_POLL_IRQ:
5997                 ret = 0;
5998                 break;
5999         case KVM_HC_KICK_CPU:
6000                 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6001                 ret = 0;
6002                 break;
6003         default:
6004                 ret = -KVM_ENOSYS;
6005                 break;
6006         }
6007 out:
6008         if (!op_64_bit)
6009                 ret = (u32)ret;
6010         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6011         ++vcpu->stat.hypercalls;
6012         return r;
6013 }
6014 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6015
6016 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6017 {
6018         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6019         char instruction[3];
6020         unsigned long rip = kvm_rip_read(vcpu);
6021
6022         kvm_x86_ops->patch_hypercall(vcpu, instruction);
6023
6024         return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
6025 }
6026
6027 /*
6028  * Check if userspace requested an interrupt window, and that the
6029  * interrupt window is open.
6030  *
6031  * No need to exit to userspace if we already have an interrupt queued.
6032  */
6033 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6034 {
6035         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
6036                 vcpu->run->request_interrupt_window &&
6037                 kvm_arch_interrupt_allowed(vcpu));
6038 }
6039
6040 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6041 {
6042         struct kvm_run *kvm_run = vcpu->run;
6043
6044         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6045         kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6046         kvm_run->cr8 = kvm_get_cr8(vcpu);
6047         kvm_run->apic_base = kvm_get_apic_base(vcpu);
6048         if (irqchip_in_kernel(vcpu->kvm))
6049                 kvm_run->ready_for_interrupt_injection = 1;
6050         else
6051                 kvm_run->ready_for_interrupt_injection =
6052                         kvm_arch_interrupt_allowed(vcpu) &&
6053                         !kvm_cpu_has_interrupt(vcpu) &&
6054                         !kvm_event_needs_reinjection(vcpu);
6055 }
6056
6057 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6058 {
6059         int max_irr, tpr;
6060
6061         if (!kvm_x86_ops->update_cr8_intercept)
6062                 return;
6063
6064         if (!vcpu->arch.apic)
6065                 return;
6066
6067         if (!vcpu->arch.apic->vapic_addr)
6068                 max_irr = kvm_lapic_find_highest_irr(vcpu);
6069         else
6070                 max_irr = -1;
6071
6072         if (max_irr != -1)
6073                 max_irr >>= 4;
6074
6075         tpr = kvm_lapic_get_cr8(vcpu);
6076
6077         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6078 }
6079
6080 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6081 {
6082         int r;
6083
6084         /* try to reinject previous events if any */
6085         if (vcpu->arch.exception.pending) {
6086                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6087                                         vcpu->arch.exception.has_error_code,
6088                                         vcpu->arch.exception.error_code);
6089
6090                 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6091                         __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6092                                              X86_EFLAGS_RF);
6093
6094                 if (vcpu->arch.exception.nr == DB_VECTOR &&
6095                     (vcpu->arch.dr7 & DR7_GD)) {
6096                         vcpu->arch.dr7 &= ~DR7_GD;
6097                         kvm_update_dr7(vcpu);
6098                 }
6099
6100                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6101                                           vcpu->arch.exception.has_error_code,
6102                                           vcpu->arch.exception.error_code,
6103                                           vcpu->arch.exception.reinject);
6104                 return 0;
6105         }
6106
6107         if (vcpu->arch.nmi_injected) {
6108                 kvm_x86_ops->set_nmi(vcpu);
6109                 return 0;
6110         }
6111
6112         if (vcpu->arch.interrupt.pending) {
6113                 kvm_x86_ops->set_irq(vcpu);
6114                 return 0;
6115         }
6116
6117         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6118                 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6119                 if (r != 0)
6120                         return r;
6121         }
6122
6123         /* try to inject new event if pending */
6124         if (vcpu->arch.nmi_pending) {
6125                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
6126                         --vcpu->arch.nmi_pending;
6127                         vcpu->arch.nmi_injected = true;
6128                         kvm_x86_ops->set_nmi(vcpu);
6129                 }
6130         } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6131                 /*
6132                  * Because interrupts can be injected asynchronously, we are
6133                  * calling check_nested_events again here to avoid a race condition.
6134                  * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6135                  * proposal and current concerns.  Perhaps we should be setting
6136                  * KVM_REQ_EVENT only on certain events and not unconditionally?
6137                  */
6138                 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6139                         r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6140                         if (r != 0)
6141                                 return r;
6142                 }
6143                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6144                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6145                                             false);
6146                         kvm_x86_ops->set_irq(vcpu);
6147                 }
6148         }
6149         return 0;
6150 }
6151
6152 static void process_nmi(struct kvm_vcpu *vcpu)
6153 {
6154         unsigned limit = 2;
6155
6156         /*
6157          * x86 is limited to one NMI running, and one NMI pending after it.
6158          * If an NMI is already in progress, limit further NMIs to just one.
6159          * Otherwise, allow two (and we'll inject the first one immediately).
6160          */
6161         if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6162                 limit = 1;
6163
6164         vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6165         vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6166         kvm_make_request(KVM_REQ_EVENT, vcpu);
6167 }
6168
6169 #define put_smstate(type, buf, offset, val)                       \
6170         *(type *)((buf) + (offset) - 0x7e00) = val
6171
6172 static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
6173 {
6174         u32 flags = 0;
6175         flags |= seg->g       << 23;
6176         flags |= seg->db      << 22;
6177         flags |= seg->l       << 21;
6178         flags |= seg->avl     << 20;
6179         flags |= seg->present << 15;
6180         flags |= seg->dpl     << 13;
6181         flags |= seg->s       << 12;
6182         flags |= seg->type    << 8;
6183         return flags;
6184 }
6185
6186 static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6187 {
6188         struct kvm_segment seg;
6189         int offset;
6190
6191         kvm_get_segment(vcpu, &seg, n);
6192         put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6193
6194         if (n < 3)
6195                 offset = 0x7f84 + n * 12;
6196         else
6197                 offset = 0x7f2c + (n - 3) * 12;
6198
6199         put_smstate(u32, buf, offset + 8, seg.base);
6200         put_smstate(u32, buf, offset + 4, seg.limit);
6201         put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
6202 }
6203
6204 static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6205 {
6206         struct kvm_segment seg;
6207         int offset;
6208         u16 flags;
6209
6210         kvm_get_segment(vcpu, &seg, n);
6211         offset = 0x7e00 + n * 16;
6212
6213         flags = process_smi_get_segment_flags(&seg) >> 8;
6214         put_smstate(u16, buf, offset, seg.selector);
6215         put_smstate(u16, buf, offset + 2, flags);
6216         put_smstate(u32, buf, offset + 4, seg.limit);
6217         put_smstate(u64, buf, offset + 8, seg.base);
6218 }
6219
6220 static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6221 {
6222         struct desc_ptr dt;
6223         struct kvm_segment seg;
6224         unsigned long val;
6225         int i;
6226
6227         put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6228         put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6229         put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6230         put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6231
6232         for (i = 0; i < 8; i++)
6233                 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6234
6235         kvm_get_dr(vcpu, 6, &val);
6236         put_smstate(u32, buf, 0x7fcc, (u32)val);
6237         kvm_get_dr(vcpu, 7, &val);
6238         put_smstate(u32, buf, 0x7fc8, (u32)val);
6239
6240         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6241         put_smstate(u32, buf, 0x7fc4, seg.selector);
6242         put_smstate(u32, buf, 0x7f64, seg.base);
6243         put_smstate(u32, buf, 0x7f60, seg.limit);
6244         put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
6245
6246         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6247         put_smstate(u32, buf, 0x7fc0, seg.selector);
6248         put_smstate(u32, buf, 0x7f80, seg.base);
6249         put_smstate(u32, buf, 0x7f7c, seg.limit);
6250         put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
6251
6252         kvm_x86_ops->get_gdt(vcpu, &dt);
6253         put_smstate(u32, buf, 0x7f74, dt.address);
6254         put_smstate(u32, buf, 0x7f70, dt.size);
6255
6256         kvm_x86_ops->get_idt(vcpu, &dt);
6257         put_smstate(u32, buf, 0x7f58, dt.address);
6258         put_smstate(u32, buf, 0x7f54, dt.size);
6259
6260         for (i = 0; i < 6; i++)
6261                 process_smi_save_seg_32(vcpu, buf, i);
6262
6263         put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6264
6265         /* revision id */
6266         put_smstate(u32, buf, 0x7efc, 0x00020000);
6267         put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6268 }
6269
6270 static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6271 {
6272 #ifdef CONFIG_X86_64
6273         struct desc_ptr dt;
6274         struct kvm_segment seg;
6275         unsigned long val;
6276         int i;
6277
6278         for (i = 0; i < 16; i++)
6279                 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6280
6281         put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6282         put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6283
6284         kvm_get_dr(vcpu, 6, &val);
6285         put_smstate(u64, buf, 0x7f68, val);
6286         kvm_get_dr(vcpu, 7, &val);
6287         put_smstate(u64, buf, 0x7f60, val);
6288
6289         put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6290         put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6291         put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6292
6293         put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6294
6295         /* revision id */
6296         put_smstate(u32, buf, 0x7efc, 0x00020064);
6297
6298         put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6299
6300         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6301         put_smstate(u16, buf, 0x7e90, seg.selector);
6302         put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6303         put_smstate(u32, buf, 0x7e94, seg.limit);
6304         put_smstate(u64, buf, 0x7e98, seg.base);
6305
6306         kvm_x86_ops->get_idt(vcpu, &dt);
6307         put_smstate(u32, buf, 0x7e84, dt.size);
6308         put_smstate(u64, buf, 0x7e88, dt.address);
6309
6310         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6311         put_smstate(u16, buf, 0x7e70, seg.selector);
6312         put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6313         put_smstate(u32, buf, 0x7e74, seg.limit);
6314         put_smstate(u64, buf, 0x7e78, seg.base);
6315
6316         kvm_x86_ops->get_gdt(vcpu, &dt);
6317         put_smstate(u32, buf, 0x7e64, dt.size);
6318         put_smstate(u64, buf, 0x7e68, dt.address);
6319
6320         for (i = 0; i < 6; i++)
6321                 process_smi_save_seg_64(vcpu, buf, i);
6322 #else
6323         WARN_ON_ONCE(1);
6324 #endif
6325 }
6326
6327 static void process_smi(struct kvm_vcpu *vcpu)
6328 {
6329         struct kvm_segment cs, ds;
6330         char buf[512];
6331         u32 cr0;
6332
6333         if (is_smm(vcpu)) {
6334                 vcpu->arch.smi_pending = true;
6335                 return;
6336         }
6337
6338         trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6339         vcpu->arch.hflags |= HF_SMM_MASK;
6340         memset(buf, 0, 512);
6341         if (guest_cpuid_has_longmode(vcpu))
6342                 process_smi_save_state_64(vcpu, buf);
6343         else
6344                 process_smi_save_state_32(vcpu, buf);
6345
6346         kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6347
6348         if (kvm_x86_ops->get_nmi_mask(vcpu))
6349                 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6350         else
6351                 kvm_x86_ops->set_nmi_mask(vcpu, true);
6352
6353         kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6354         kvm_rip_write(vcpu, 0x8000);
6355
6356         cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6357         kvm_x86_ops->set_cr0(vcpu, cr0);
6358         vcpu->arch.cr0 = cr0;
6359
6360         kvm_x86_ops->set_cr4(vcpu, 0);
6361
6362         __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6363
6364         cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6365         cs.base = vcpu->arch.smbase;
6366
6367         ds.selector = 0;
6368         ds.base = 0;
6369
6370         cs.limit    = ds.limit = 0xffffffff;
6371         cs.type     = ds.type = 0x3;
6372         cs.dpl      = ds.dpl = 0;
6373         cs.db       = ds.db = 0;
6374         cs.s        = ds.s = 1;
6375         cs.l        = ds.l = 0;
6376         cs.g        = ds.g = 1;
6377         cs.avl      = ds.avl = 0;
6378         cs.present  = ds.present = 1;
6379         cs.unusable = ds.unusable = 0;
6380         cs.padding  = ds.padding = 0;
6381
6382         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6383         kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6384         kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6385         kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6386         kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6387         kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6388
6389         if (guest_cpuid_has_longmode(vcpu))
6390                 kvm_x86_ops->set_efer(vcpu, 0);
6391
6392         kvm_update_cpuid(vcpu);
6393         kvm_mmu_reset_context(vcpu);
6394 }
6395
6396 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6397 {
6398         u64 eoi_exit_bitmap[4];
6399         u32 tmr[8];
6400
6401         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6402                 return;
6403
6404         memset(eoi_exit_bitmap, 0, 32);
6405         memset(tmr, 0, 32);
6406
6407         kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
6408         kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6409         kvm_apic_update_tmr(vcpu, tmr);
6410 }
6411
6412 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6413 {
6414         ++vcpu->stat.tlb_flush;
6415         kvm_x86_ops->tlb_flush(vcpu);
6416 }
6417
6418 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6419 {
6420         struct page *page = NULL;
6421
6422         if (!irqchip_in_kernel(vcpu->kvm))
6423                 return;
6424
6425         if (!kvm_x86_ops->set_apic_access_page_addr)
6426                 return;
6427
6428         page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6429         if (is_error_page(page))
6430                 return;
6431         kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6432
6433         /*
6434          * Do not pin apic access page in memory, the MMU notifier
6435          * will call us again if it is migrated or swapped out.
6436          */
6437         put_page(page);
6438 }
6439 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6440
6441 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6442                                            unsigned long address)
6443 {
6444         /*
6445          * The physical address of apic access page is stored in the VMCS.
6446          * Update it when it becomes invalid.
6447          */
6448         if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6449                 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6450 }
6451
6452 /*
6453  * Returns 1 to let vcpu_run() continue the guest execution loop without
6454  * exiting to the userspace.  Otherwise, the value will be returned to the
6455  * userspace.
6456  */
6457 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6458 {
6459         int r;
6460         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6461                 vcpu->run->request_interrupt_window;
6462         bool req_immediate_exit = false;
6463
6464         if (vcpu->requests) {
6465                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6466                         kvm_mmu_unload(vcpu);
6467                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6468                         __kvm_migrate_timers(vcpu);
6469                 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6470                         kvm_gen_update_masterclock(vcpu->kvm);
6471                 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6472                         kvm_gen_kvmclock_update(vcpu);
6473                 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6474                         r = kvm_guest_time_update(vcpu);
6475                         if (unlikely(r))
6476                                 goto out;
6477                 }
6478                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6479                         kvm_mmu_sync_roots(vcpu);
6480                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6481                         kvm_vcpu_flush_tlb(vcpu);
6482                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6483                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6484                         r = 0;
6485                         goto out;
6486                 }
6487                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6488                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6489                         r = 0;
6490                         goto out;
6491                 }
6492                 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6493                         vcpu->fpu_active = 0;
6494                         kvm_x86_ops->fpu_deactivate(vcpu);
6495                 }
6496                 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6497                         /* Page is swapped out. Do synthetic halt */
6498                         vcpu->arch.apf.halted = true;
6499                         r = 1;
6500                         goto out;
6501                 }
6502                 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6503                         record_steal_time(vcpu);
6504                 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6505                         process_smi(vcpu);
6506                 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6507                         process_nmi(vcpu);
6508                 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6509                         kvm_pmu_handle_event(vcpu);
6510                 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6511                         kvm_pmu_deliver_pmi(vcpu);
6512                 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6513                         vcpu_scan_ioapic(vcpu);
6514                 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6515                         kvm_vcpu_reload_apic_access_page(vcpu);
6516         }
6517
6518         if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6519                 kvm_apic_accept_events(vcpu);
6520                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6521                         r = 1;
6522                         goto out;
6523                 }
6524
6525                 if (inject_pending_event(vcpu, req_int_win) != 0)
6526                         req_immediate_exit = true;
6527                 /* enable NMI/IRQ window open exits if needed */
6528                 else if (vcpu->arch.nmi_pending)
6529                         kvm_x86_ops->enable_nmi_window(vcpu);
6530                 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6531                         kvm_x86_ops->enable_irq_window(vcpu);
6532
6533                 if (kvm_lapic_enabled(vcpu)) {
6534                         /*
6535                          * Update architecture specific hints for APIC
6536                          * virtual interrupt delivery.
6537                          */
6538                         if (kvm_x86_ops->hwapic_irr_update)
6539                                 kvm_x86_ops->hwapic_irr_update(vcpu,
6540                                         kvm_lapic_find_highest_irr(vcpu));
6541                         update_cr8_intercept(vcpu);
6542                         kvm_lapic_sync_to_vapic(vcpu);
6543                 }
6544         }
6545
6546         r = kvm_mmu_reload(vcpu);
6547         if (unlikely(r)) {
6548                 goto cancel_injection;
6549         }
6550
6551         preempt_disable();
6552
6553         kvm_x86_ops->prepare_guest_switch(vcpu);
6554         if (vcpu->fpu_active)
6555                 kvm_load_guest_fpu(vcpu);
6556         kvm_load_guest_xcr0(vcpu);
6557
6558         vcpu->mode = IN_GUEST_MODE;
6559
6560         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6561
6562         /* We should set ->mode before check ->requests,
6563          * see the comment in make_all_cpus_request.
6564          */
6565         smp_mb__after_srcu_read_unlock();
6566
6567         local_irq_disable();
6568
6569         if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6570             || need_resched() || signal_pending(current)) {
6571                 vcpu->mode = OUTSIDE_GUEST_MODE;
6572                 smp_wmb();
6573                 local_irq_enable();
6574                 preempt_enable();
6575                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6576                 r = 1;
6577                 goto cancel_injection;
6578         }
6579
6580         if (req_immediate_exit)
6581                 smp_send_reschedule(vcpu->cpu);
6582
6583         __kvm_guest_enter();
6584
6585         if (unlikely(vcpu->arch.switch_db_regs)) {
6586                 set_debugreg(0, 7);
6587                 set_debugreg(vcpu->arch.eff_db[0], 0);
6588                 set_debugreg(vcpu->arch.eff_db[1], 1);
6589                 set_debugreg(vcpu->arch.eff_db[2], 2);
6590                 set_debugreg(vcpu->arch.eff_db[3], 3);
6591                 set_debugreg(vcpu->arch.dr6, 6);
6592                 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6593         }
6594
6595         trace_kvm_entry(vcpu->vcpu_id);
6596         wait_lapic_expire(vcpu);
6597         kvm_x86_ops->run(vcpu);
6598
6599         /*
6600          * Do this here before restoring debug registers on the host.  And
6601          * since we do this before handling the vmexit, a DR access vmexit
6602          * can (a) read the correct value of the debug registers, (b) set
6603          * KVM_DEBUGREG_WONT_EXIT again.
6604          */
6605         if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6606                 int i;
6607
6608                 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6609                 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6610                 for (i = 0; i < KVM_NR_DB_REGS; i++)
6611                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6612         }
6613
6614         /*
6615          * If the guest has used debug registers, at least dr7
6616          * will be disabled while returning to the host.
6617          * If we don't have active breakpoints in the host, we don't
6618          * care about the messed up debug address registers. But if
6619          * we have some of them active, restore the old state.
6620          */
6621         if (hw_breakpoint_active())
6622                 hw_breakpoint_restore();
6623
6624         vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6625                                                            native_read_tsc());
6626
6627         vcpu->mode = OUTSIDE_GUEST_MODE;
6628         smp_wmb();
6629
6630         /* Interrupt is enabled by handle_external_intr() */
6631         kvm_x86_ops->handle_external_intr(vcpu);
6632
6633         ++vcpu->stat.exits;
6634
6635         /*
6636          * We must have an instruction between local_irq_enable() and
6637          * kvm_guest_exit(), so the timer interrupt isn't delayed by
6638          * the interrupt shadow.  The stat.exits increment will do nicely.
6639          * But we need to prevent reordering, hence this barrier():
6640          */
6641         barrier();
6642
6643         kvm_guest_exit();
6644
6645         preempt_enable();
6646
6647         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6648
6649         /*
6650          * Profile KVM exit RIPs:
6651          */
6652         if (unlikely(prof_on == KVM_PROFILING)) {
6653                 unsigned long rip = kvm_rip_read(vcpu);
6654                 profile_hit(KVM_PROFILING, (void *)rip);
6655         }
6656
6657         if (unlikely(vcpu->arch.tsc_always_catchup))
6658                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6659
6660         if (vcpu->arch.apic_attention)
6661                 kvm_lapic_sync_from_vapic(vcpu);
6662
6663         r = kvm_x86_ops->handle_exit(vcpu);
6664         return r;
6665
6666 cancel_injection:
6667         kvm_x86_ops->cancel_injection(vcpu);
6668         if (unlikely(vcpu->arch.apic_attention))
6669                 kvm_lapic_sync_from_vapic(vcpu);
6670 out:
6671         return r;
6672 }
6673
6674 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6675 {
6676         if (!kvm_arch_vcpu_runnable(vcpu)) {
6677                 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6678                 kvm_vcpu_block(vcpu);
6679                 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6680                 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6681                         return 1;
6682         }
6683
6684         kvm_apic_accept_events(vcpu);
6685         switch(vcpu->arch.mp_state) {
6686         case KVM_MP_STATE_HALTED:
6687                 vcpu->arch.pv.pv_unhalted = false;
6688                 vcpu->arch.mp_state =
6689                         KVM_MP_STATE_RUNNABLE;
6690         case KVM_MP_STATE_RUNNABLE:
6691                 vcpu->arch.apf.halted = false;
6692                 break;
6693         case KVM_MP_STATE_INIT_RECEIVED:
6694                 break;
6695         default:
6696                 return -EINTR;
6697                 break;
6698         }
6699         return 1;
6700 }
6701
6702 static int vcpu_run(struct kvm_vcpu *vcpu)
6703 {
6704         int r;
6705         struct kvm *kvm = vcpu->kvm;
6706
6707         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6708
6709         for (;;) {
6710                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6711                     !vcpu->arch.apf.halted)
6712                         r = vcpu_enter_guest(vcpu);
6713                 else
6714                         r = vcpu_block(kvm, vcpu);
6715                 if (r <= 0)
6716                         break;
6717
6718                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6719                 if (kvm_cpu_has_pending_timer(vcpu))
6720                         kvm_inject_pending_timer_irqs(vcpu);
6721
6722                 if (dm_request_for_irq_injection(vcpu)) {
6723                         r = -EINTR;
6724                         vcpu->run->exit_reason = KVM_EXIT_INTR;
6725                         ++vcpu->stat.request_irq_exits;
6726                         break;
6727                 }
6728
6729                 kvm_check_async_pf_completion(vcpu);
6730
6731                 if (signal_pending(current)) {
6732                         r = -EINTR;
6733                         vcpu->run->exit_reason = KVM_EXIT_INTR;
6734                         ++vcpu->stat.signal_exits;
6735                         break;
6736                 }
6737                 if (need_resched()) {
6738                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6739                         cond_resched();
6740                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6741                 }
6742         }
6743
6744         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6745
6746         return r;
6747 }
6748
6749 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6750 {
6751         int r;
6752         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6753         r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6754         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6755         if (r != EMULATE_DONE)
6756                 return 0;
6757         return 1;
6758 }
6759
6760 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6761 {
6762         BUG_ON(!vcpu->arch.pio.count);
6763
6764         return complete_emulated_io(vcpu);
6765 }
6766
6767 /*
6768  * Implements the following, as a state machine:
6769  *
6770  * read:
6771  *   for each fragment
6772  *     for each mmio piece in the fragment
6773  *       write gpa, len
6774  *       exit
6775  *       copy data
6776  *   execute insn
6777  *
6778  * write:
6779  *   for each fragment
6780  *     for each mmio piece in the fragment
6781  *       write gpa, len
6782  *       copy data
6783  *       exit
6784  */
6785 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6786 {
6787         struct kvm_run *run = vcpu->run;
6788         struct kvm_mmio_fragment *frag;
6789         unsigned len;
6790
6791         BUG_ON(!vcpu->mmio_needed);
6792
6793         /* Complete previous fragment */
6794         frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6795         len = min(8u, frag->len);
6796         if (!vcpu->mmio_is_write)
6797                 memcpy(frag->data, run->mmio.data, len);
6798
6799         if (frag->len <= 8) {
6800                 /* Switch to the next fragment. */
6801                 frag++;
6802                 vcpu->mmio_cur_fragment++;
6803         } else {
6804                 /* Go forward to the next mmio piece. */
6805                 frag->data += len;
6806                 frag->gpa += len;
6807                 frag->len -= len;
6808         }
6809
6810         if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6811                 vcpu->mmio_needed = 0;
6812
6813                 /* FIXME: return into emulator if single-stepping.  */
6814                 if (vcpu->mmio_is_write)
6815                         return 1;
6816                 vcpu->mmio_read_completed = 1;
6817                 return complete_emulated_io(vcpu);
6818         }
6819
6820         run->exit_reason = KVM_EXIT_MMIO;
6821         run->mmio.phys_addr = frag->gpa;
6822         if (vcpu->mmio_is_write)
6823                 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6824         run->mmio.len = min(8u, frag->len);
6825         run->mmio.is_write = vcpu->mmio_is_write;
6826         vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6827         return 0;
6828 }
6829
6830
6831 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6832 {
6833         struct fpu *fpu = &current->thread.fpu;
6834         int r;
6835         sigset_t sigsaved;
6836
6837         fpu__activate_curr(fpu);
6838
6839         if (vcpu->sigset_active)
6840                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6841
6842         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6843                 kvm_vcpu_block(vcpu);
6844                 kvm_apic_accept_events(vcpu);
6845                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6846                 r = -EAGAIN;
6847                 goto out;
6848         }
6849
6850         /* re-sync apic's tpr */
6851         if (!irqchip_in_kernel(vcpu->kvm)) {
6852                 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6853                         r = -EINVAL;
6854                         goto out;
6855                 }
6856         }
6857
6858         if (unlikely(vcpu->arch.complete_userspace_io)) {
6859                 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6860                 vcpu->arch.complete_userspace_io = NULL;
6861                 r = cui(vcpu);
6862                 if (r <= 0)
6863                         goto out;
6864         } else
6865                 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6866
6867         r = vcpu_run(vcpu);
6868
6869 out:
6870         post_kvm_run_save(vcpu);
6871         if (vcpu->sigset_active)
6872                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6873
6874         return r;
6875 }
6876
6877 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6878 {
6879         if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6880                 /*
6881                  * We are here if userspace calls get_regs() in the middle of
6882                  * instruction emulation. Registers state needs to be copied
6883                  * back from emulation context to vcpu. Userspace shouldn't do
6884                  * that usually, but some bad designed PV devices (vmware
6885                  * backdoor interface) need this to work
6886                  */
6887                 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6888                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6889         }
6890         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6891         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6892         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6893         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6894         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6895         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6896         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6897         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6898 #ifdef CONFIG_X86_64
6899         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6900         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6901         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6902         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6903         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6904         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6905         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6906         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6907 #endif
6908
6909         regs->rip = kvm_rip_read(vcpu);
6910         regs->rflags = kvm_get_rflags(vcpu);
6911
6912         return 0;
6913 }
6914
6915 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6916 {
6917         vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6918         vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6919
6920         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6921         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6922         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6923         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6924         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6925         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6926         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6927         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6928 #ifdef CONFIG_X86_64
6929         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6930         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6931         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6932         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6933         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6934         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6935         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6936         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6937 #endif
6938
6939         kvm_rip_write(vcpu, regs->rip);
6940         kvm_set_rflags(vcpu, regs->rflags);
6941
6942         vcpu->arch.exception.pending = false;
6943
6944         kvm_make_request(KVM_REQ_EVENT, vcpu);
6945
6946         return 0;
6947 }
6948
6949 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6950 {
6951         struct kvm_segment cs;
6952
6953         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6954         *db = cs.db;
6955         *l = cs.l;
6956 }
6957 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6958
6959 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6960                                   struct kvm_sregs *sregs)
6961 {
6962         struct desc_ptr dt;
6963
6964         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6965         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6966         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6967         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6968         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6969         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6970
6971         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6972         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6973
6974         kvm_x86_ops->get_idt(vcpu, &dt);
6975         sregs->idt.limit = dt.size;
6976         sregs->idt.base = dt.address;
6977         kvm_x86_ops->get_gdt(vcpu, &dt);
6978         sregs->gdt.limit = dt.size;
6979         sregs->gdt.base = dt.address;
6980
6981         sregs->cr0 = kvm_read_cr0(vcpu);
6982         sregs->cr2 = vcpu->arch.cr2;
6983         sregs->cr3 = kvm_read_cr3(vcpu);
6984         sregs->cr4 = kvm_read_cr4(vcpu);
6985         sregs->cr8 = kvm_get_cr8(vcpu);
6986         sregs->efer = vcpu->arch.efer;
6987         sregs->apic_base = kvm_get_apic_base(vcpu);
6988
6989         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6990
6991         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6992                 set_bit(vcpu->arch.interrupt.nr,
6993                         (unsigned long *)sregs->interrupt_bitmap);
6994
6995         return 0;
6996 }
6997
6998 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6999                                     struct kvm_mp_state *mp_state)
7000 {
7001         kvm_apic_accept_events(vcpu);
7002         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7003                                         vcpu->arch.pv.pv_unhalted)
7004                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7005         else
7006                 mp_state->mp_state = vcpu->arch.mp_state;
7007
7008         return 0;
7009 }
7010
7011 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7012                                     struct kvm_mp_state *mp_state)
7013 {
7014         if (!kvm_vcpu_has_lapic(vcpu) &&
7015             mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7016                 return -EINVAL;
7017
7018         if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7019                 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7020                 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7021         } else
7022                 vcpu->arch.mp_state = mp_state->mp_state;
7023         kvm_make_request(KVM_REQ_EVENT, vcpu);
7024         return 0;
7025 }
7026
7027 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7028                     int reason, bool has_error_code, u32 error_code)
7029 {
7030         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7031         int ret;
7032
7033         init_emulate_ctxt(vcpu);
7034
7035         ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7036                                    has_error_code, error_code);
7037
7038         if (ret)
7039                 return EMULATE_FAIL;
7040
7041         kvm_rip_write(vcpu, ctxt->eip);
7042         kvm_set_rflags(vcpu, ctxt->eflags);
7043         kvm_make_request(KVM_REQ_EVENT, vcpu);
7044         return EMULATE_DONE;
7045 }
7046 EXPORT_SYMBOL_GPL(kvm_task_switch);
7047
7048 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7049                                   struct kvm_sregs *sregs)
7050 {
7051         struct msr_data apic_base_msr;
7052         int mmu_reset_needed = 0;
7053         int pending_vec, max_bits, idx;
7054         struct desc_ptr dt;
7055
7056         if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7057                 return -EINVAL;
7058
7059         dt.size = sregs->idt.limit;
7060         dt.address = sregs->idt.base;
7061         kvm_x86_ops->set_idt(vcpu, &dt);
7062         dt.size = sregs->gdt.limit;
7063         dt.address = sregs->gdt.base;
7064         kvm_x86_ops->set_gdt(vcpu, &dt);
7065
7066         vcpu->arch.cr2 = sregs->cr2;
7067         mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7068         vcpu->arch.cr3 = sregs->cr3;
7069         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7070
7071         kvm_set_cr8(vcpu, sregs->cr8);
7072
7073         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7074         kvm_x86_ops->set_efer(vcpu, sregs->efer);
7075         apic_base_msr.data = sregs->apic_base;
7076         apic_base_msr.host_initiated = true;
7077         kvm_set_apic_base(vcpu, &apic_base_msr);
7078
7079         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7080         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7081         vcpu->arch.cr0 = sregs->cr0;
7082
7083         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7084         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7085         if (sregs->cr4 & X86_CR4_OSXSAVE)
7086                 kvm_update_cpuid(vcpu);
7087
7088         idx = srcu_read_lock(&vcpu->kvm->srcu);
7089         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7090                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7091                 mmu_reset_needed = 1;
7092         }
7093         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7094
7095         if (mmu_reset_needed)
7096                 kvm_mmu_reset_context(vcpu);
7097
7098         max_bits = KVM_NR_INTERRUPTS;
7099         pending_vec = find_first_bit(
7100                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7101         if (pending_vec < max_bits) {
7102                 kvm_queue_interrupt(vcpu, pending_vec, false);
7103                 pr_debug("Set back pending irq %d\n", pending_vec);
7104         }
7105
7106         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7107         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7108         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7109         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7110         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7111         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7112
7113         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7114         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7115
7116         update_cr8_intercept(vcpu);
7117
7118         /* Older userspace won't unhalt the vcpu on reset. */
7119         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7120             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7121             !is_protmode(vcpu))
7122                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7123
7124         kvm_make_request(KVM_REQ_EVENT, vcpu);
7125
7126         return 0;
7127 }
7128
7129 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7130                                         struct kvm_guest_debug *dbg)
7131 {
7132         unsigned long rflags;
7133         int i, r;
7134
7135         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7136                 r = -EBUSY;
7137                 if (vcpu->arch.exception.pending)
7138                         goto out;
7139                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7140                         kvm_queue_exception(vcpu, DB_VECTOR);
7141                 else
7142                         kvm_queue_exception(vcpu, BP_VECTOR);
7143         }
7144
7145         /*
7146          * Read rflags as long as potentially injected trace flags are still
7147          * filtered out.
7148          */
7149         rflags = kvm_get_rflags(vcpu);
7150
7151         vcpu->guest_debug = dbg->control;
7152         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7153                 vcpu->guest_debug = 0;
7154
7155         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7156                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7157                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7158                 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7159         } else {
7160                 for (i = 0; i < KVM_NR_DB_REGS; i++)
7161                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7162         }
7163         kvm_update_dr7(vcpu);
7164
7165         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7166                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7167                         get_segment_base(vcpu, VCPU_SREG_CS);
7168
7169         /*
7170          * Trigger an rflags update that will inject or remove the trace
7171          * flags.
7172          */
7173         kvm_set_rflags(vcpu, rflags);
7174
7175         kvm_x86_ops->update_db_bp_intercept(vcpu);
7176
7177         r = 0;
7178
7179 out:
7180
7181         return r;
7182 }
7183
7184 /*
7185  * Translate a guest virtual address to a guest physical address.
7186  */
7187 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7188                                     struct kvm_translation *tr)
7189 {
7190         unsigned long vaddr = tr->linear_address;
7191         gpa_t gpa;
7192         int idx;
7193
7194         idx = srcu_read_lock(&vcpu->kvm->srcu);
7195         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7196         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7197         tr->physical_address = gpa;
7198         tr->valid = gpa != UNMAPPED_GVA;
7199         tr->writeable = 1;
7200         tr->usermode = 0;
7201
7202         return 0;
7203 }
7204
7205 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7206 {
7207         struct fxregs_state *fxsave =
7208                         &vcpu->arch.guest_fpu.state.fxsave;
7209
7210         memcpy(fpu->fpr, fxsave->st_space, 128);
7211         fpu->fcw = fxsave->cwd;
7212         fpu->fsw = fxsave->swd;
7213         fpu->ftwx = fxsave->twd;
7214         fpu->last_opcode = fxsave->fop;
7215         fpu->last_ip = fxsave->rip;
7216         fpu->last_dp = fxsave->rdp;
7217         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7218
7219         return 0;
7220 }
7221
7222 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7223 {
7224         struct fxregs_state *fxsave =
7225                         &vcpu->arch.guest_fpu.state.fxsave;
7226
7227         memcpy(fxsave->st_space, fpu->fpr, 128);
7228         fxsave->cwd = fpu->fcw;
7229         fxsave->swd = fpu->fsw;
7230         fxsave->twd = fpu->ftwx;
7231         fxsave->fop = fpu->last_opcode;
7232         fxsave->rip = fpu->last_ip;
7233         fxsave->rdp = fpu->last_dp;
7234         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7235
7236         return 0;
7237 }
7238
7239 static void fx_init(struct kvm_vcpu *vcpu)
7240 {
7241         fpstate_init(&vcpu->arch.guest_fpu.state);
7242         if (cpu_has_xsaves)
7243                 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7244                         host_xcr0 | XSTATE_COMPACTION_ENABLED;
7245
7246         /*
7247          * Ensure guest xcr0 is valid for loading
7248          */
7249         vcpu->arch.xcr0 = XSTATE_FP;
7250
7251         vcpu->arch.cr0 |= X86_CR0_ET;
7252 }
7253
7254 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7255 {
7256         if (vcpu->guest_fpu_loaded)
7257                 return;
7258
7259         /*
7260          * Restore all possible states in the guest,
7261          * and assume host would use all available bits.
7262          * Guest xcr0 would be loaded later.
7263          */
7264         kvm_put_guest_xcr0(vcpu);
7265         vcpu->guest_fpu_loaded = 1;
7266         __kernel_fpu_begin();
7267         __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7268         trace_kvm_fpu(1);
7269 }
7270
7271 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7272 {
7273         kvm_put_guest_xcr0(vcpu);
7274
7275         if (!vcpu->guest_fpu_loaded) {
7276                 vcpu->fpu_counter = 0;
7277                 return;
7278         }
7279
7280         vcpu->guest_fpu_loaded = 0;
7281         copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7282         __kernel_fpu_end();
7283         ++vcpu->stat.fpu_reload;
7284         /*
7285          * If using eager FPU mode, or if the guest is a frequent user
7286          * of the FPU, just leave the FPU active for next time.
7287          * Every 255 times fpu_counter rolls over to 0; a guest that uses
7288          * the FPU in bursts will revert to loading it on demand.
7289          */
7290         if (!vcpu->arch.eager_fpu) {
7291                 if (++vcpu->fpu_counter < 5)
7292                         kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7293         }
7294         trace_kvm_fpu(0);
7295 }
7296
7297 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7298 {
7299         kvmclock_reset(vcpu);
7300
7301         free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7302         kvm_x86_ops->vcpu_free(vcpu);
7303 }
7304
7305 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7306                                                 unsigned int id)
7307 {
7308         struct kvm_vcpu *vcpu;
7309
7310         if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7311                 printk_once(KERN_WARNING
7312                 "kvm: SMP vm created on host with unstable TSC; "
7313                 "guest TSC will not be reliable\n");
7314
7315         vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7316
7317         return vcpu;
7318 }
7319
7320 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7321 {
7322         int r;
7323
7324         kvm_vcpu_mtrr_init(vcpu);
7325         r = vcpu_load(vcpu);
7326         if (r)
7327                 return r;
7328         kvm_vcpu_reset(vcpu, false);
7329         kvm_mmu_setup(vcpu);
7330         vcpu_put(vcpu);
7331         return r;
7332 }
7333
7334 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7335 {
7336         struct msr_data msr;
7337         struct kvm *kvm = vcpu->kvm;
7338
7339         if (vcpu_load(vcpu))
7340                 return;
7341         msr.data = 0x0;
7342         msr.index = MSR_IA32_TSC;
7343         msr.host_initiated = true;
7344         kvm_write_tsc(vcpu, &msr);
7345         vcpu_put(vcpu);
7346
7347         if (!kvmclock_periodic_sync)
7348                 return;
7349
7350         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7351                                         KVMCLOCK_SYNC_PERIOD);
7352 }
7353
7354 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7355 {
7356         int r;
7357         vcpu->arch.apf.msr_val = 0;
7358
7359         r = vcpu_load(vcpu);
7360         BUG_ON(r);
7361         kvm_mmu_unload(vcpu);
7362         vcpu_put(vcpu);
7363
7364         kvm_x86_ops->vcpu_free(vcpu);
7365 }
7366
7367 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7368 {
7369         vcpu->arch.hflags = 0;
7370
7371         atomic_set(&vcpu->arch.nmi_queued, 0);
7372         vcpu->arch.nmi_pending = 0;
7373         vcpu->arch.nmi_injected = false;
7374         kvm_clear_interrupt_queue(vcpu);
7375         kvm_clear_exception_queue(vcpu);
7376
7377         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7378         kvm_update_dr0123(vcpu);
7379         vcpu->arch.dr6 = DR6_INIT;
7380         kvm_update_dr6(vcpu);
7381         vcpu->arch.dr7 = DR7_FIXED_1;
7382         kvm_update_dr7(vcpu);
7383
7384         vcpu->arch.cr2 = 0;
7385
7386         kvm_make_request(KVM_REQ_EVENT, vcpu);
7387         vcpu->arch.apf.msr_val = 0;
7388         vcpu->arch.st.msr_val = 0;
7389
7390         kvmclock_reset(vcpu);
7391
7392         kvm_clear_async_pf_completion_queue(vcpu);
7393         kvm_async_pf_hash_reset(vcpu);
7394         vcpu->arch.apf.halted = false;
7395
7396         if (!init_event) {
7397                 kvm_pmu_reset(vcpu);
7398                 vcpu->arch.smbase = 0x30000;
7399         }
7400
7401         memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7402         vcpu->arch.regs_avail = ~0;
7403         vcpu->arch.regs_dirty = ~0;
7404
7405         kvm_x86_ops->vcpu_reset(vcpu, init_event);
7406 }
7407
7408 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7409 {
7410         struct kvm_segment cs;
7411
7412         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7413         cs.selector = vector << 8;
7414         cs.base = vector << 12;
7415         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7416         kvm_rip_write(vcpu, 0);
7417 }
7418
7419 int kvm_arch_hardware_enable(void)
7420 {
7421         struct kvm *kvm;
7422         struct kvm_vcpu *vcpu;
7423         int i;
7424         int ret;
7425         u64 local_tsc;
7426         u64 max_tsc = 0;
7427         bool stable, backwards_tsc = false;
7428
7429         kvm_shared_msr_cpu_online();
7430         ret = kvm_x86_ops->hardware_enable();
7431         if (ret != 0)
7432                 return ret;
7433
7434         local_tsc = native_read_tsc();
7435         stable = !check_tsc_unstable();
7436         list_for_each_entry(kvm, &vm_list, vm_list) {
7437                 kvm_for_each_vcpu(i, vcpu, kvm) {
7438                         if (!stable && vcpu->cpu == smp_processor_id())
7439                                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7440                         if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7441                                 backwards_tsc = true;
7442                                 if (vcpu->arch.last_host_tsc > max_tsc)
7443                                         max_tsc = vcpu->arch.last_host_tsc;
7444                         }
7445                 }
7446         }
7447
7448         /*
7449          * Sometimes, even reliable TSCs go backwards.  This happens on
7450          * platforms that reset TSC during suspend or hibernate actions, but
7451          * maintain synchronization.  We must compensate.  Fortunately, we can
7452          * detect that condition here, which happens early in CPU bringup,
7453          * before any KVM threads can be running.  Unfortunately, we can't
7454          * bring the TSCs fully up to date with real time, as we aren't yet far
7455          * enough into CPU bringup that we know how much real time has actually
7456          * elapsed; our helper function, get_kernel_ns() will be using boot
7457          * variables that haven't been updated yet.
7458          *
7459          * So we simply find the maximum observed TSC above, then record the
7460          * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7461          * the adjustment will be applied.  Note that we accumulate
7462          * adjustments, in case multiple suspend cycles happen before some VCPU
7463          * gets a chance to run again.  In the event that no KVM threads get a
7464          * chance to run, we will miss the entire elapsed period, as we'll have
7465          * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7466          * loose cycle time.  This isn't too big a deal, since the loss will be
7467          * uniform across all VCPUs (not to mention the scenario is extremely
7468          * unlikely). It is possible that a second hibernate recovery happens
7469          * much faster than a first, causing the observed TSC here to be
7470          * smaller; this would require additional padding adjustment, which is
7471          * why we set last_host_tsc to the local tsc observed here.
7472          *
7473          * N.B. - this code below runs only on platforms with reliable TSC,
7474          * as that is the only way backwards_tsc is set above.  Also note
7475          * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7476          * have the same delta_cyc adjustment applied if backwards_tsc
7477          * is detected.  Note further, this adjustment is only done once,
7478          * as we reset last_host_tsc on all VCPUs to stop this from being
7479          * called multiple times (one for each physical CPU bringup).
7480          *
7481          * Platforms with unreliable TSCs don't have to deal with this, they
7482          * will be compensated by the logic in vcpu_load, which sets the TSC to
7483          * catchup mode.  This will catchup all VCPUs to real time, but cannot
7484          * guarantee that they stay in perfect synchronization.
7485          */
7486         if (backwards_tsc) {
7487                 u64 delta_cyc = max_tsc - local_tsc;
7488                 backwards_tsc_observed = true;
7489                 list_for_each_entry(kvm, &vm_list, vm_list) {
7490                         kvm_for_each_vcpu(i, vcpu, kvm) {
7491                                 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7492                                 vcpu->arch.last_host_tsc = local_tsc;
7493                                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7494                         }
7495
7496                         /*
7497                          * We have to disable TSC offset matching.. if you were
7498                          * booting a VM while issuing an S4 host suspend....
7499                          * you may have some problem.  Solving this issue is
7500                          * left as an exercise to the reader.
7501                          */
7502                         kvm->arch.last_tsc_nsec = 0;
7503                         kvm->arch.last_tsc_write = 0;
7504                 }
7505
7506         }
7507         return 0;
7508 }
7509
7510 void kvm_arch_hardware_disable(void)
7511 {
7512         kvm_x86_ops->hardware_disable();
7513         drop_user_return_notifiers();
7514 }
7515
7516 int kvm_arch_hardware_setup(void)
7517 {
7518         int r;
7519
7520         r = kvm_x86_ops->hardware_setup();
7521         if (r != 0)
7522                 return r;
7523
7524         kvm_init_msr_list();
7525         return 0;
7526 }
7527
7528 void kvm_arch_hardware_unsetup(void)
7529 {
7530         kvm_x86_ops->hardware_unsetup();
7531 }
7532
7533 void kvm_arch_check_processor_compat(void *rtn)
7534 {
7535         kvm_x86_ops->check_processor_compatibility(rtn);
7536 }
7537
7538 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7539 {
7540         return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7541 }
7542
7543 struct static_key kvm_no_apic_vcpu __read_mostly;
7544
7545 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7546 {
7547         struct page *page;
7548         struct kvm *kvm;
7549         int r;
7550
7551         BUG_ON(vcpu->kvm == NULL);
7552         kvm = vcpu->kvm;
7553
7554         vcpu->arch.pv.pv_unhalted = false;
7555         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7556         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7557                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7558         else
7559                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7560
7561         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7562         if (!page) {
7563                 r = -ENOMEM;
7564                 goto fail;
7565         }
7566         vcpu->arch.pio_data = page_address(page);
7567
7568         kvm_set_tsc_khz(vcpu, max_tsc_khz);
7569
7570         r = kvm_mmu_create(vcpu);
7571         if (r < 0)
7572                 goto fail_free_pio_data;
7573
7574         if (irqchip_in_kernel(kvm)) {
7575                 r = kvm_create_lapic(vcpu);
7576                 if (r < 0)
7577                         goto fail_mmu_destroy;
7578         } else
7579                 static_key_slow_inc(&kvm_no_apic_vcpu);
7580
7581         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7582                                        GFP_KERNEL);
7583         if (!vcpu->arch.mce_banks) {
7584                 r = -ENOMEM;
7585                 goto fail_free_lapic;
7586         }
7587         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7588
7589         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7590                 r = -ENOMEM;
7591                 goto fail_free_mce_banks;
7592         }
7593
7594         fx_init(vcpu);
7595
7596         vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7597         vcpu->arch.pv_time_enabled = false;
7598
7599         vcpu->arch.guest_supported_xcr0 = 0;
7600         vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7601
7602         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7603
7604         vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7605
7606         kvm_async_pf_hash_reset(vcpu);
7607         kvm_pmu_init(vcpu);
7608
7609         return 0;
7610
7611 fail_free_mce_banks:
7612         kfree(vcpu->arch.mce_banks);
7613 fail_free_lapic:
7614         kvm_free_lapic(vcpu);
7615 fail_mmu_destroy:
7616         kvm_mmu_destroy(vcpu);
7617 fail_free_pio_data:
7618         free_page((unsigned long)vcpu->arch.pio_data);
7619 fail:
7620         return r;
7621 }
7622
7623 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7624 {
7625         int idx;
7626
7627         kvm_pmu_destroy(vcpu);
7628         kfree(vcpu->arch.mce_banks);
7629         kvm_free_lapic(vcpu);
7630         idx = srcu_read_lock(&vcpu->kvm->srcu);
7631         kvm_mmu_destroy(vcpu);
7632         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7633         free_page((unsigned long)vcpu->arch.pio_data);
7634         if (!irqchip_in_kernel(vcpu->kvm))
7635                 static_key_slow_dec(&kvm_no_apic_vcpu);
7636 }
7637
7638 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7639 {
7640         kvm_x86_ops->sched_in(vcpu, cpu);
7641 }
7642
7643 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7644 {
7645         if (type)
7646                 return -EINVAL;
7647
7648         INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7649         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7650         INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7651         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7652         atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7653
7654         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7655         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7656         /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7657         set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7658                 &kvm->arch.irq_sources_bitmap);
7659
7660         raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7661         mutex_init(&kvm->arch.apic_map_lock);
7662         spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7663
7664         pvclock_update_vm_gtod_copy(kvm);
7665
7666         INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7667         INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7668
7669         return 0;
7670 }
7671
7672 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7673 {
7674         int r;
7675         r = vcpu_load(vcpu);
7676         BUG_ON(r);
7677         kvm_mmu_unload(vcpu);
7678         vcpu_put(vcpu);
7679 }
7680
7681 static void kvm_free_vcpus(struct kvm *kvm)
7682 {
7683         unsigned int i;
7684         struct kvm_vcpu *vcpu;
7685
7686         /*
7687          * Unpin any mmu pages first.
7688          */
7689         kvm_for_each_vcpu(i, vcpu, kvm) {
7690                 kvm_clear_async_pf_completion_queue(vcpu);
7691                 kvm_unload_vcpu_mmu(vcpu);
7692         }
7693         kvm_for_each_vcpu(i, vcpu, kvm)
7694                 kvm_arch_vcpu_free(vcpu);
7695
7696         mutex_lock(&kvm->lock);
7697         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7698                 kvm->vcpus[i] = NULL;
7699
7700         atomic_set(&kvm->online_vcpus, 0);
7701         mutex_unlock(&kvm->lock);
7702 }
7703
7704 void kvm_arch_sync_events(struct kvm *kvm)
7705 {
7706         cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7707         cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7708         kvm_free_all_assigned_devices(kvm);
7709         kvm_free_pit(kvm);
7710 }
7711
7712 int __x86_set_memory_region(struct kvm *kvm,
7713                             const struct kvm_userspace_memory_region *mem)
7714 {
7715         int i, r;
7716
7717         /* Called with kvm->slots_lock held.  */
7718         BUG_ON(mem->slot >= KVM_MEM_SLOTS_NUM);
7719
7720         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7721                 struct kvm_userspace_memory_region m = *mem;
7722
7723                 m.slot |= i << 16;
7724                 r = __kvm_set_memory_region(kvm, &m);
7725                 if (r < 0)
7726                         return r;
7727         }
7728
7729         return 0;
7730 }
7731 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7732
7733 int x86_set_memory_region(struct kvm *kvm,
7734                           const struct kvm_userspace_memory_region *mem)
7735 {
7736         int r;
7737
7738         mutex_lock(&kvm->slots_lock);
7739         r = __x86_set_memory_region(kvm, mem);
7740         mutex_unlock(&kvm->slots_lock);
7741
7742         return r;
7743 }
7744 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7745
7746 void kvm_arch_destroy_vm(struct kvm *kvm)
7747 {
7748         if (current->mm == kvm->mm) {
7749                 /*
7750                  * Free memory regions allocated on behalf of userspace,
7751                  * unless the the memory map has changed due to process exit
7752                  * or fd copying.
7753                  */
7754                 struct kvm_userspace_memory_region mem;
7755                 memset(&mem, 0, sizeof(mem));
7756                 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7757                 x86_set_memory_region(kvm, &mem);
7758
7759                 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7760                 x86_set_memory_region(kvm, &mem);
7761
7762                 mem.slot = TSS_PRIVATE_MEMSLOT;
7763                 x86_set_memory_region(kvm, &mem);
7764         }
7765         kvm_iommu_unmap_guest(kvm);
7766         kfree(kvm->arch.vpic);
7767         kfree(kvm->arch.vioapic);
7768         kvm_free_vcpus(kvm);
7769         kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7770 }
7771
7772 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7773                            struct kvm_memory_slot *dont)
7774 {
7775         int i;
7776
7777         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7778                 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7779                         kvfree(free->arch.rmap[i]);
7780                         free->arch.rmap[i] = NULL;
7781                 }
7782                 if (i == 0)
7783                         continue;
7784
7785                 if (!dont || free->arch.lpage_info[i - 1] !=
7786                              dont->arch.lpage_info[i - 1]) {
7787                         kvfree(free->arch.lpage_info[i - 1]);
7788                         free->arch.lpage_info[i - 1] = NULL;
7789                 }
7790         }
7791 }
7792
7793 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7794                             unsigned long npages)
7795 {
7796         int i;
7797
7798         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7799                 unsigned long ugfn;
7800                 int lpages;
7801                 int level = i + 1;
7802
7803                 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7804                                       slot->base_gfn, level) + 1;
7805
7806                 slot->arch.rmap[i] =
7807                         kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7808                 if (!slot->arch.rmap[i])
7809                         goto out_free;
7810                 if (i == 0)
7811                         continue;
7812
7813                 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7814                                         sizeof(*slot->arch.lpage_info[i - 1]));
7815                 if (!slot->arch.lpage_info[i - 1])
7816                         goto out_free;
7817
7818                 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7819                         slot->arch.lpage_info[i - 1][0].write_count = 1;
7820                 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7821                         slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7822                 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7823                 /*
7824                  * If the gfn and userspace address are not aligned wrt each
7825                  * other, or if explicitly asked to, disable large page
7826                  * support for this slot
7827                  */
7828                 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7829                     !kvm_largepages_enabled()) {
7830                         unsigned long j;
7831
7832                         for (j = 0; j < lpages; ++j)
7833                                 slot->arch.lpage_info[i - 1][j].write_count = 1;
7834                 }
7835         }
7836
7837         return 0;
7838
7839 out_free:
7840         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7841                 kvfree(slot->arch.rmap[i]);
7842                 slot->arch.rmap[i] = NULL;
7843                 if (i == 0)
7844                         continue;
7845
7846                 kvfree(slot->arch.lpage_info[i - 1]);
7847                 slot->arch.lpage_info[i - 1] = NULL;
7848         }
7849         return -ENOMEM;
7850 }
7851
7852 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7853 {
7854         /*
7855          * memslots->generation has been incremented.
7856          * mmio generation may have reached its maximum value.
7857          */
7858         kvm_mmu_invalidate_mmio_sptes(kvm, slots);
7859 }
7860
7861 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7862                                 struct kvm_memory_slot *memslot,
7863                                 const struct kvm_userspace_memory_region *mem,
7864                                 enum kvm_mr_change change)
7865 {
7866         /*
7867          * Only private memory slots need to be mapped here since
7868          * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7869          */
7870         if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7871                 unsigned long userspace_addr;
7872
7873                 /*
7874                  * MAP_SHARED to prevent internal slot pages from being moved
7875                  * by fork()/COW.
7876                  */
7877                 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7878                                          PROT_READ | PROT_WRITE,
7879                                          MAP_SHARED | MAP_ANONYMOUS, 0);
7880
7881                 if (IS_ERR((void *)userspace_addr))
7882                         return PTR_ERR((void *)userspace_addr);
7883
7884                 memslot->userspace_addr = userspace_addr;
7885         }
7886
7887         return 0;
7888 }
7889
7890 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7891                                      struct kvm_memory_slot *new)
7892 {
7893         /* Still write protect RO slot */
7894         if (new->flags & KVM_MEM_READONLY) {
7895                 kvm_mmu_slot_remove_write_access(kvm, new);
7896                 return;
7897         }
7898
7899         /*
7900          * Call kvm_x86_ops dirty logging hooks when they are valid.
7901          *
7902          * kvm_x86_ops->slot_disable_log_dirty is called when:
7903          *
7904          *  - KVM_MR_CREATE with dirty logging is disabled
7905          *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7906          *
7907          * The reason is, in case of PML, we need to set D-bit for any slots
7908          * with dirty logging disabled in order to eliminate unnecessary GPA
7909          * logging in PML buffer (and potential PML buffer full VMEXT). This
7910          * guarantees leaving PML enabled during guest's lifetime won't have
7911          * any additonal overhead from PML when guest is running with dirty
7912          * logging disabled for memory slots.
7913          *
7914          * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7915          * to dirty logging mode.
7916          *
7917          * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7918          *
7919          * In case of write protect:
7920          *
7921          * Write protect all pages for dirty logging.
7922          *
7923          * All the sptes including the large sptes which point to this
7924          * slot are set to readonly. We can not create any new large
7925          * spte on this slot until the end of the logging.
7926          *
7927          * See the comments in fast_page_fault().
7928          */
7929         if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7930                 if (kvm_x86_ops->slot_enable_log_dirty)
7931                         kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7932                 else
7933                         kvm_mmu_slot_remove_write_access(kvm, new);
7934         } else {
7935                 if (kvm_x86_ops->slot_disable_log_dirty)
7936                         kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7937         }
7938 }
7939
7940 void kvm_arch_commit_memory_region(struct kvm *kvm,
7941                                 const struct kvm_userspace_memory_region *mem,
7942                                 const struct kvm_memory_slot *old,
7943                                 const struct kvm_memory_slot *new,
7944                                 enum kvm_mr_change change)
7945 {
7946         int nr_mmu_pages = 0;
7947
7948         if (change == KVM_MR_DELETE && old->id >= KVM_USER_MEM_SLOTS) {
7949                 int ret;
7950
7951                 ret = vm_munmap(old->userspace_addr,
7952                                 old->npages * PAGE_SIZE);
7953                 if (ret < 0)
7954                         printk(KERN_WARNING
7955                                "kvm_vm_ioctl_set_memory_region: "
7956                                "failed to munmap memory\n");
7957         }
7958
7959         if (!kvm->arch.n_requested_mmu_pages)
7960                 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7961
7962         if (nr_mmu_pages)
7963                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7964
7965         /*
7966          * Dirty logging tracks sptes in 4k granularity, meaning that large
7967          * sptes have to be split.  If live migration is successful, the guest
7968          * in the source machine will be destroyed and large sptes will be
7969          * created in the destination. However, if the guest continues to run
7970          * in the source machine (for example if live migration fails), small
7971          * sptes will remain around and cause bad performance.
7972          *
7973          * Scan sptes if dirty logging has been stopped, dropping those
7974          * which can be collapsed into a single large-page spte.  Later
7975          * page faults will create the large-page sptes.
7976          */
7977         if ((change != KVM_MR_DELETE) &&
7978                 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7979                 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7980                 kvm_mmu_zap_collapsible_sptes(kvm, new);
7981
7982         /*
7983          * Set up write protection and/or dirty logging for the new slot.
7984          *
7985          * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7986          * been zapped so no dirty logging staff is needed for old slot. For
7987          * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7988          * new and it's also covered when dealing with the new slot.
7989          *
7990          * FIXME: const-ify all uses of struct kvm_memory_slot.
7991          */
7992         if (change != KVM_MR_DELETE)
7993                 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
7994 }
7995
7996 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7997 {
7998         kvm_mmu_invalidate_zap_all_pages(kvm);
7999 }
8000
8001 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8002                                    struct kvm_memory_slot *slot)
8003 {
8004         kvm_mmu_invalidate_zap_all_pages(kvm);
8005 }
8006
8007 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8008 {
8009         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8010                 kvm_x86_ops->check_nested_events(vcpu, false);
8011
8012         return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8013                 !vcpu->arch.apf.halted)
8014                 || !list_empty_careful(&vcpu->async_pf.done)
8015                 || kvm_apic_has_events(vcpu)
8016                 || vcpu->arch.pv.pv_unhalted
8017                 || atomic_read(&vcpu->arch.nmi_queued) ||
8018                 (kvm_arch_interrupt_allowed(vcpu) &&
8019                  kvm_cpu_has_interrupt(vcpu));
8020 }
8021
8022 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8023 {
8024         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8025 }
8026
8027 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8028 {
8029         return kvm_x86_ops->interrupt_allowed(vcpu);
8030 }
8031
8032 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8033 {
8034         if (is_64_bit_mode(vcpu))
8035                 return kvm_rip_read(vcpu);
8036         return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8037                      kvm_rip_read(vcpu));
8038 }
8039 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8040
8041 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8042 {
8043         return kvm_get_linear_rip(vcpu) == linear_rip;
8044 }
8045 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8046
8047 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8048 {
8049         unsigned long rflags;
8050
8051         rflags = kvm_x86_ops->get_rflags(vcpu);
8052         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8053                 rflags &= ~X86_EFLAGS_TF;
8054         return rflags;
8055 }
8056 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8057
8058 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8059 {
8060         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8061             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8062                 rflags |= X86_EFLAGS_TF;
8063         kvm_x86_ops->set_rflags(vcpu, rflags);
8064 }
8065
8066 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8067 {
8068         __kvm_set_rflags(vcpu, rflags);
8069         kvm_make_request(KVM_REQ_EVENT, vcpu);
8070 }
8071 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8072
8073 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8074 {
8075         int r;
8076
8077         if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8078               work->wakeup_all)
8079                 return;
8080
8081         r = kvm_mmu_reload(vcpu);
8082         if (unlikely(r))
8083                 return;
8084
8085         if (!vcpu->arch.mmu.direct_map &&
8086               work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8087                 return;
8088
8089         vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8090 }
8091
8092 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8093 {
8094         return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8095 }
8096
8097 static inline u32 kvm_async_pf_next_probe(u32 key)
8098 {
8099         return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8100 }
8101
8102 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8103 {
8104         u32 key = kvm_async_pf_hash_fn(gfn);
8105
8106         while (vcpu->arch.apf.gfns[key] != ~0)
8107                 key = kvm_async_pf_next_probe(key);
8108
8109         vcpu->arch.apf.gfns[key] = gfn;
8110 }
8111
8112 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8113 {
8114         int i;
8115         u32 key = kvm_async_pf_hash_fn(gfn);
8116
8117         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8118                      (vcpu->arch.apf.gfns[key] != gfn &&
8119                       vcpu->arch.apf.gfns[key] != ~0); i++)
8120                 key = kvm_async_pf_next_probe(key);
8121
8122         return key;
8123 }
8124
8125 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8126 {
8127         return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8128 }
8129
8130 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8131 {
8132         u32 i, j, k;
8133
8134         i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8135         while (true) {
8136                 vcpu->arch.apf.gfns[i] = ~0;
8137                 do {
8138                         j = kvm_async_pf_next_probe(j);
8139                         if (vcpu->arch.apf.gfns[j] == ~0)
8140                                 return;
8141                         k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8142                         /*
8143                          * k lies cyclically in ]i,j]
8144                          * |    i.k.j |
8145                          * |....j i.k.| or  |.k..j i...|
8146                          */
8147                 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8148                 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8149                 i = j;
8150         }
8151 }
8152
8153 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8154 {
8155
8156         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8157                                       sizeof(val));
8158 }
8159
8160 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8161                                      struct kvm_async_pf *work)
8162 {
8163         struct x86_exception fault;
8164
8165         trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8166         kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8167
8168         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8169             (vcpu->arch.apf.send_user_only &&
8170              kvm_x86_ops->get_cpl(vcpu) == 0))
8171                 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8172         else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8173                 fault.vector = PF_VECTOR;
8174                 fault.error_code_valid = true;
8175                 fault.error_code = 0;
8176                 fault.nested_page_fault = false;
8177                 fault.address = work->arch.token;
8178                 kvm_inject_page_fault(vcpu, &fault);
8179         }
8180 }
8181
8182 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8183                                  struct kvm_async_pf *work)
8184 {
8185         struct x86_exception fault;
8186
8187         trace_kvm_async_pf_ready(work->arch.token, work->gva);
8188         if (work->wakeup_all)
8189                 work->arch.token = ~0; /* broadcast wakeup */
8190         else
8191                 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8192
8193         if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8194             !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8195                 fault.vector = PF_VECTOR;
8196                 fault.error_code_valid = true;
8197                 fault.error_code = 0;
8198                 fault.nested_page_fault = false;
8199                 fault.address = work->arch.token;
8200                 kvm_inject_page_fault(vcpu, &fault);
8201         }
8202         vcpu->arch.apf.halted = false;
8203         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8204 }
8205
8206 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8207 {
8208         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8209                 return true;
8210         else
8211                 return !kvm_event_needs_reinjection(vcpu) &&
8212                         kvm_x86_ops->interrupt_allowed(vcpu);
8213 }
8214
8215 void kvm_arch_start_assignment(struct kvm *kvm)
8216 {
8217         atomic_inc(&kvm->arch.assigned_device_count);
8218 }
8219 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8220
8221 void kvm_arch_end_assignment(struct kvm *kvm)
8222 {
8223         atomic_dec(&kvm->arch.assigned_device_count);
8224 }
8225 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8226
8227 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8228 {
8229         return atomic_read(&kvm->arch.assigned_device_count);
8230 }
8231 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8232
8233 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8234 {
8235         atomic_inc(&kvm->arch.noncoherent_dma_count);
8236 }
8237 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8238
8239 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8240 {
8241         atomic_dec(&kvm->arch.noncoherent_dma_count);
8242 }
8243 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8244
8245 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8246 {
8247         return atomic_read(&kvm->arch.noncoherent_dma_count);
8248 }
8249 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8250
8251 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8252 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8253 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8254 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8255 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8256 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8257 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8258 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8259 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8260 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8261 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8262 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8263 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8264 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8265 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);