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 affilates.
  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
  30 #include <linux/clocksource.h>
  31 #include <linux/interrupt.h>
  32 #include <linux/kvm.h>
  33 #include <linux/fs.h>
  34 #include <linux/vmalloc.h>
  35 #include <linux/module.h>
  36 #include <linux/mman.h>
  37 #include <linux/highmem.h>
  38 #include <linux/iommu.h>
  39 #include <linux/intel-iommu.h>
  40 #include <linux/cpufreq.h>
  41 #include <linux/user-return-notifier.h>
  42 #include <linux/srcu.h>
  43 #include <linux/slab.h>
  44 #include <linux/perf_event.h>
  45 #include <linux/uaccess.h>
  46 #include <trace/events/kvm.h>
  47
  48 #define CREATE_TRACE_POINTS
  49 #include "trace.h"
  50
  51 #include <asm/debugreg.h>
  52 #include <asm/msr.h>
  53 #include <asm/desc.h>
  54 #include <asm/mtrr.h>
  55 #include <asm/mce.h>
  56 #include <asm/i387.h>
  57 #include <asm/xcr.h>
  58 #include <asm/pvclock.h>
  59 #include <asm/div64.h>
  60
  61 #define MAX_IO_MSRS 256
  62 #define CR0_RESERVED_BITS                                               \
  63         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  64                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  65                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  66 #define CR4_RESERVED_BITS                                               \
  67         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  68                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  69                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  70                           | X86_CR4_OSXSAVE \
  71                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  72
  73 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  74
  75 #define KVM_MAX_MCE_BANKS 32
  76 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
  77
  78 /* EFER defaults:
  79  * - enable syscall per default because its emulated by KVM
  80  * - enable LME and LMA per default on 64 bit KVM
  81  */
  82 #ifdef CONFIG_X86_64
  83 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  84 #else
  85 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  86 #endif
  87
  88 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  89 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  90
  91 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  92 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  93                                     struct kvm_cpuid_entry2 __user *entries);
  94
  95 struct kvm_x86_ops *kvm_x86_ops;
  96 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  97
  98 int ignore_msrs = 0;
  99 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
 100
 101 #define KVM_NR_SHARED_MSRS 16
 102
 103 struct kvm_shared_msrs_global {
 104         int nr;
 105         u32 msrs[KVM_NR_SHARED_MSRS];
 106 };
 107
 108 struct kvm_shared_msrs {
 109         struct user_return_notifier urn;
 110         bool registered;
 111         struct kvm_shared_msr_values {
 112                 u64 host;
 113                 u64 curr;
 114         } values[KVM_NR_SHARED_MSRS];
 115 };
 116
 117 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 118 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
 119
 120 struct kvm_stats_debugfs_item debugfs_entries[] = {
 121         { "pf_fixed", VCPU_STAT(pf_fixed) },
 122         { "pf_guest", VCPU_STAT(pf_guest) },
 123         { "tlb_flush", VCPU_STAT(tlb_flush) },
 124         { "invlpg", VCPU_STAT(invlpg) },
 125         { "exits", VCPU_STAT(exits) },
 126         { "io_exits", VCPU_STAT(io_exits) },
 127         { "mmio_exits", VCPU_STAT(mmio_exits) },
 128         { "signal_exits", VCPU_STAT(signal_exits) },
 129         { "irq_window", VCPU_STAT(irq_window_exits) },
 130         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 131         { "halt_exits", VCPU_STAT(halt_exits) },
 132         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 133         { "hypercalls", VCPU_STAT(hypercalls) },
 134         { "request_irq", VCPU_STAT(request_irq_exits) },
 135         { "irq_exits", VCPU_STAT(irq_exits) },
 136         { "host_state_reload", VCPU_STAT(host_state_reload) },
 137         { "efer_reload", VCPU_STAT(efer_reload) },
 138         { "fpu_reload", VCPU_STAT(fpu_reload) },
 139         { "insn_emulation", VCPU_STAT(insn_emulation) },
 140         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 141         { "irq_injections", VCPU_STAT(irq_injections) },
 142         { "nmi_injections", VCPU_STAT(nmi_injections) },
 143         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 144         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 145         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 146         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 147         { "mmu_flooded", VM_STAT(mmu_flooded) },
 148         { "mmu_recycled", VM_STAT(mmu_recycled) },
 149         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 150         { "mmu_unsync", VM_STAT(mmu_unsync) },
 151         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 152         { "largepages", VM_STAT(lpages) },
 153         { NULL }
 154 };
 155
 156 u64 __read_mostly host_xcr0;
 157
 158 static inline u32 bit(int bitno)
 159 {
 160         return 1 << (bitno & 31);
 161 }
 162
 163 static void kvm_on_user_return(struct user_return_notifier *urn)
 164 {
 165         unsigned slot;
 166         struct kvm_shared_msrs *locals
 167                 = container_of(urn, struct kvm_shared_msrs, urn);
 168         struct kvm_shared_msr_values *values;
 169
 170         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 171                 values = &locals->values[slot];
 172                 if (values->host != values->curr) {
 173                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 174                         values->curr = values->host;
 175                 }
 176         }
 177         locals->registered = false;
 178         user_return_notifier_unregister(urn);
 179 }
 180
 181 static void shared_msr_update(unsigned slot, u32 msr)
 182 {
 183         struct kvm_shared_msrs *smsr;
 184         u64 value;
 185
 186         smsr = &__get_cpu_var(shared_msrs);
 187         /* only read, and nobody should modify it at this time,
 188          * so don't need lock */
 189         if (slot >= shared_msrs_global.nr) {
 190                 printk(KERN_ERR "kvm: invalid MSR slot!");
 191                 return;
 192         }
 193         rdmsrl_safe(msr, &value);
 194         smsr->values[slot].host = value;
 195         smsr->values[slot].curr = value;
 196 }
 197
 198 void kvm_define_shared_msr(unsigned slot, u32 msr)
 199 {
 200         if (slot >= shared_msrs_global.nr)
 201                 shared_msrs_global.nr = slot + 1;
 202         shared_msrs_global.msrs[slot] = msr;
 203         /* we need ensured the shared_msr_global have been updated */
 204         smp_wmb();
 205 }
 206 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 207
 208 static void kvm_shared_msr_cpu_online(void)
 209 {
 210         unsigned i;
 211
 212         for (i = 0; i < shared_msrs_global.nr; ++i)
 213                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 214 }
 215
 216 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 217 {
 218         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 219
 220         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 221                 return;
 222         smsr->values[slot].curr = value;
 223         wrmsrl(shared_msrs_global.msrs[slot], value);
 224         if (!smsr->registered) {
 225                 smsr->urn.on_user_return = kvm_on_user_return;
 226                 user_return_notifier_register(&smsr->urn);
 227                 smsr->registered = true;
 228         }
 229 }
 230 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 231
 232 static void drop_user_return_notifiers(void *ignore)
 233 {
 234         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 235
 236         if (smsr->registered)
 237                 kvm_on_user_return(&smsr->urn);
 238 }
 239
 240 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 241 {
 242         if (irqchip_in_kernel(vcpu->kvm))
 243                 return vcpu->arch.apic_base;
 244         else
 245                 return vcpu->arch.apic_base;
 246 }
 247 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 248
 249 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 250 {
 251         /* TODO: reserve bits check */
 252         if (irqchip_in_kernel(vcpu->kvm))
 253                 kvm_lapic_set_base(vcpu, data);
 254         else
 255                 vcpu->arch.apic_base = data;
 256 }
 257 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 258
 259 #define EXCPT_BENIGN            0
 260 #define EXCPT_CONTRIBUTORY      1
 261 #define EXCPT_PF                2
 262
 263 static int exception_class(int vector)
 264 {
 265         switch (vector) {
 266         case PF_VECTOR:
 267                 return EXCPT_PF;
 268         case DE_VECTOR:
 269         case TS_VECTOR:
 270         case NP_VECTOR:
 271         case SS_VECTOR:
 272         case GP_VECTOR:
 273                 return EXCPT_CONTRIBUTORY;
 274         default:
 275                 break;
 276         }
 277         return EXCPT_BENIGN;
 278 }
 279
 280 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 281                 unsigned nr, bool has_error, u32 error_code,
 282                 bool reinject)
 283 {
 284         u32 prev_nr;
 285         int class1, class2;
 286
 287         if (!vcpu->arch.exception.pending) {
 288         queue:
 289                 vcpu->arch.exception.pending = true;
 290                 vcpu->arch.exception.has_error_code = has_error;
 291                 vcpu->arch.exception.nr = nr;
 292                 vcpu->arch.exception.error_code = error_code;
 293                 vcpu->arch.exception.reinject = reinject;
 294                 return;
 295         }
 296
 297         /* to check exception */
 298         prev_nr = vcpu->arch.exception.nr;
 299         if (prev_nr == DF_VECTOR) {
 300                 /* triple fault -> shutdown */
 301                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
 302                 return;
 303         }
 304         class1 = exception_class(prev_nr);
 305         class2 = exception_class(nr);
 306         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 307                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 308                 /* generate double fault per SDM Table 5-5 */
 309                 vcpu->arch.exception.pending = true;
 310                 vcpu->arch.exception.has_error_code = true;
 311                 vcpu->arch.exception.nr = DF_VECTOR;
 312                 vcpu->arch.exception.error_code = 0;
 313         } else
 314                 /* replace previous exception with a new one in a hope
 315                    that instruction re-execution will regenerate lost
 316                    exception */
 317                 goto queue;
 318 }
 319
 320 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 321 {
 322         kvm_multiple_exception(vcpu, nr, false, 0, false);
 323 }
 324 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 325
 326 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 327 {
 328         kvm_multiple_exception(vcpu, nr, false, 0, true);
 329 }
 330 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 331
 332 void kvm_inject_page_fault(struct kvm_vcpu *vcpu)
 333 {
 334         unsigned error_code = vcpu->arch.fault.error_code;
 335
 336         ++vcpu->stat.pf_guest;
 337         vcpu->arch.cr2 = vcpu->arch.fault.address;
 338         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 339 }
 340
 341 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 342 {
 343         vcpu->arch.nmi_pending = 1;
 344 }
 345 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 346
 347 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 348 {
 349         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 350 }
 351 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 352
 353 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 354 {
 355         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 356 }
 357 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 358
 359 /*
 360  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 361  * a #GP and return false.
 362  */
 363 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 364 {
 365         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 366                 return true;
 367         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 368         return false;
 369 }
 370 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 371
 372 /*
 373  * This function will be used to read from the physical memory of the currently
 374  * running guest. The difference to kvm_read_guest_page is that this function
 375  * can read from guest physical or from the guest's guest physical memory.
 376  */
 377 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 378                             gfn_t ngfn, void *data, int offset, int len,
 379                             u32 access)
 380 {
 381         gfn_t real_gfn;
 382         gpa_t ngpa;
 383
 384         ngpa     = gfn_to_gpa(ngfn);
 385         real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
 386         if (real_gfn == UNMAPPED_GVA)
 387                 return -EFAULT;
 388
 389         real_gfn = gpa_to_gfn(real_gfn);
 390
 391         return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
 392 }
 393 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
 394
 395 /*
 396  * Load the pae pdptrs.  Return true is they are all valid.
 397  */
 398 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 399 {
 400         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 401         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 402         int i;
 403         int ret;
 404         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 405
 406         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 407                                   offset * sizeof(u64), sizeof(pdpte));
 408         if (ret < 0) {
 409                 ret = 0;
 410                 goto out;
 411         }
 412         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 413                 if (is_present_gpte(pdpte[i]) &&
 414                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 415                         ret = 0;
 416                         goto out;
 417                 }
 418         }
 419         ret = 1;
 420
 421         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 422         __set_bit(VCPU_EXREG_PDPTR,
 423                   (unsigned long *)&vcpu->arch.regs_avail);
 424         __set_bit(VCPU_EXREG_PDPTR,
 425                   (unsigned long *)&vcpu->arch.regs_dirty);
 426 out:
 427
 428         return ret;
 429 }
 430 EXPORT_SYMBOL_GPL(load_pdptrs);
 431
 432 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 433 {
 434         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 435         bool changed = true;
 436         int r;
 437
 438         if (is_long_mode(vcpu) || !is_pae(vcpu))
 439                 return false;
 440
 441         if (!test_bit(VCPU_EXREG_PDPTR,
 442                       (unsigned long *)&vcpu->arch.regs_avail))
 443                 return true;
 444
 445         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 446         if (r < 0)
 447                 goto out;
 448         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 449 out:
 450
 451         return changed;
 452 }
 453
 454 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 455 {
 456         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 457         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
 458                                     X86_CR0_CD | X86_CR0_NW;
 459
 460         cr0 |= X86_CR0_ET;
 461
 462 #ifdef CONFIG_X86_64
 463         if (cr0 & 0xffffffff00000000UL)
 464                 return 1;
 465 #endif
 466
 467         cr0 &= ~CR0_RESERVED_BITS;
 468
 469         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 470                 return 1;
 471
 472         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 473                 return 1;
 474
 475         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 476 #ifdef CONFIG_X86_64
 477                 if ((vcpu->arch.efer & EFER_LME)) {
 478                         int cs_db, cs_l;
 479
 480                         if (!is_pae(vcpu))
 481                                 return 1;
 482                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 483                         if (cs_l)
 484                                 return 1;
 485                 } else
 486 #endif
 487                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3))
 488                         return 1;
 489         }
 490
 491         kvm_x86_ops->set_cr0(vcpu, cr0);
 492
 493         if ((cr0 ^ old_cr0) & update_bits)
 494                 kvm_mmu_reset_context(vcpu);
 495         return 0;
 496 }
 497 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 498
 499 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 500 {
 501         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 502 }
 503 EXPORT_SYMBOL_GPL(kvm_lmsw);
 504
 505 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 506 {
 507         u64 xcr0;
 508
 509         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
 510         if (index != XCR_XFEATURE_ENABLED_MASK)
 511                 return 1;
 512         xcr0 = xcr;
 513         if (kvm_x86_ops->get_cpl(vcpu) != 0)
 514                 return 1;
 515         if (!(xcr0 & XSTATE_FP))
 516                 return 1;
 517         if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
 518                 return 1;
 519         if (xcr0 & ~host_xcr0)
 520                 return 1;
 521         vcpu->arch.xcr0 = xcr0;
 522         vcpu->guest_xcr0_loaded = 0;
 523         return 0;
 524 }
 525
 526 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 527 {
 528         if (__kvm_set_xcr(vcpu, index, xcr)) {
 529                 kvm_inject_gp(vcpu, 0);
 530                 return 1;
 531         }
 532         return 0;
 533 }
 534 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 535
 536 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
 537 {
 538         struct kvm_cpuid_entry2 *best;
 539
 540         best = kvm_find_cpuid_entry(vcpu, 1, 0);
 541         return best && (best->ecx & bit(X86_FEATURE_XSAVE));
 542 }
 543
 544 static void update_cpuid(struct kvm_vcpu *vcpu)
 545 {
 546         struct kvm_cpuid_entry2 *best;
 547
 548         best = kvm_find_cpuid_entry(vcpu, 1, 0);
 549         if (!best)
 550                 return;
 551
 552         /* Update OSXSAVE bit */
 553         if (cpu_has_xsave && best->function == 0x1) {
 554                 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
 555                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
 556                         best->ecx |= bit(X86_FEATURE_OSXSAVE);
 557         }
 558 }
 559
 560 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 561 {
 562         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 563         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 564
 565         if (cr4 & CR4_RESERVED_BITS)
 566                 return 1;
 567
 568         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
 569                 return 1;
 570
 571         if (is_long_mode(vcpu)) {
 572                 if (!(cr4 & X86_CR4_PAE))
 573                         return 1;
 574         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 575                    && ((cr4 ^ old_cr4) & pdptr_bits)
 576                    && !load_pdptrs(vcpu, vcpu->arch.cr3))
 577                 return 1;
 578
 579         if (cr4 & X86_CR4_VMXE)
 580                 return 1;
 581
 582         kvm_x86_ops->set_cr4(vcpu, cr4);
 583
 584         if ((cr4 ^ old_cr4) & pdptr_bits)
 585                 kvm_mmu_reset_context(vcpu);
 586
 587         if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
 588                 update_cpuid(vcpu);
 589
 590         return 0;
 591 }
 592 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 593
 594 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 595 {
 596         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 597                 kvm_mmu_sync_roots(vcpu);
 598                 kvm_mmu_flush_tlb(vcpu);
 599                 return 0;
 600         }
 601
 602         if (is_long_mode(vcpu)) {
 603                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 604                         return 1;
 605         } else {
 606                 if (is_pae(vcpu)) {
 607                         if (cr3 & CR3_PAE_RESERVED_BITS)
 608                                 return 1;
 609                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3))
 610                                 return 1;
 611                 }
 612                 /*
 613                  * We don't check reserved bits in nonpae mode, because
 614                  * this isn't enforced, and VMware depends on this.
 615                  */
 616         }
 617
 618         /*
 619          * Does the new cr3 value map to physical memory? (Note, we
 620          * catch an invalid cr3 even in real-mode, because it would
 621          * cause trouble later on when we turn on paging anyway.)
 622          *
 623          * A real CPU would silently accept an invalid cr3 and would
 624          * attempt to use it - with largely undefined (and often hard
 625          * to debug) behavior on the guest side.
 626          */
 627         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 628                 return 1;
 629         vcpu->arch.cr3 = cr3;
 630         vcpu->arch.mmu.new_cr3(vcpu);
 631         return 0;
 632 }
 633 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 634
 635 int __kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 636 {
 637         if (cr8 & CR8_RESERVED_BITS)
 638                 return 1;
 639         if (irqchip_in_kernel(vcpu->kvm))
 640                 kvm_lapic_set_tpr(vcpu, cr8);
 641         else
 642                 vcpu->arch.cr8 = cr8;
 643         return 0;
 644 }
 645
 646 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 647 {
 648         if (__kvm_set_cr8(vcpu, cr8))
 649                 kvm_inject_gp(vcpu, 0);
 650 }
 651 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 652
 653 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 654 {
 655         if (irqchip_in_kernel(vcpu->kvm))
 656                 return kvm_lapic_get_cr8(vcpu);
 657         else
 658                 return vcpu->arch.cr8;
 659 }
 660 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 661
 662 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 663 {
 664         switch (dr) {
 665         case 0 ... 3:
 666                 vcpu->arch.db[dr] = val;
 667                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 668                         vcpu->arch.eff_db[dr] = val;
 669                 break;
 670         case 4:
 671                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 672                         return 1; /* #UD */
 673                 /* fall through */
 674         case 6:
 675                 if (val & 0xffffffff00000000ULL)
 676                         return -1; /* #GP */
 677                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
 678                 break;
 679         case 5:
 680                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 681                         return 1; /* #UD */
 682                 /* fall through */
 683         default: /* 7 */
 684                 if (val & 0xffffffff00000000ULL)
 685                         return -1; /* #GP */
 686                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 687                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 688                         kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
 689                         vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
 690                 }
 691                 break;
 692         }
 693
 694         return 0;
 695 }
 696
 697 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 698 {
 699         int res;
 700
 701         res = __kvm_set_dr(vcpu, dr, val);
 702         if (res > 0)
 703                 kvm_queue_exception(vcpu, UD_VECTOR);
 704         else if (res < 0)
 705                 kvm_inject_gp(vcpu, 0);
 706
 707         return res;
 708 }
 709 EXPORT_SYMBOL_GPL(kvm_set_dr);
 710
 711 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 712 {
 713         switch (dr) {
 714         case 0 ... 3:
 715                 *val = vcpu->arch.db[dr];
 716                 break;
 717         case 4:
 718                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 719                         return 1;
 720                 /* fall through */
 721         case 6:
 722                 *val = vcpu->arch.dr6;
 723                 break;
 724         case 5:
 725                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 726                         return 1;
 727                 /* fall through */
 728         default: /* 7 */
 729                 *val = vcpu->arch.dr7;
 730                 break;
 731         }
 732
 733         return 0;
 734 }
 735
 736 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 737 {
 738         if (_kvm_get_dr(vcpu, dr, val)) {
 739                 kvm_queue_exception(vcpu, UD_VECTOR);
 740                 return 1;
 741         }
 742         return 0;
 743 }
 744 EXPORT_SYMBOL_GPL(kvm_get_dr);
 745
 746 /*
 747  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 748  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 749  *
 750  * This list is modified at module load time to reflect the
 751  * capabilities of the host cpu. This capabilities test skips MSRs that are
 752  * kvm-specific. Those are put in the beginning of the list.
 753  */
 754
 755 #define KVM_SAVE_MSRS_BEGIN     7
 756 static u32 msrs_to_save[] = {
 757         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 758         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 759         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 760         HV_X64_MSR_APIC_ASSIST_PAGE,
 761         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 762         MSR_STAR,
 763 #ifdef CONFIG_X86_64
 764         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 765 #endif
 766         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 767 };
 768
 769 static unsigned num_msrs_to_save;
 770
 771 static u32 emulated_msrs[] = {
 772         MSR_IA32_MISC_ENABLE,
 773         MSR_IA32_MCG_STATUS,
 774         MSR_IA32_MCG_CTL,
 775 };
 776
 777 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
 778 {
 779         u64 old_efer = vcpu->arch.efer;
 780
 781         if (efer & efer_reserved_bits)
 782                 return 1;
 783
 784         if (is_paging(vcpu)
 785             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
 786                 return 1;
 787
 788         if (efer & EFER_FFXSR) {
 789                 struct kvm_cpuid_entry2 *feat;
 790
 791                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 792                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
 793                         return 1;
 794         }
 795
 796         if (efer & EFER_SVME) {
 797                 struct kvm_cpuid_entry2 *feat;
 798
 799                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 800                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
 801                         return 1;
 802         }
 803
 804         efer &= ~EFER_LMA;
 805         efer |= vcpu->arch.efer & EFER_LMA;
 806
 807         kvm_x86_ops->set_efer(vcpu, efer);
 808
 809         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 810         kvm_mmu_reset_context(vcpu);
 811
 812         /* Update reserved bits */
 813         if ((efer ^ old_efer) & EFER_NX)
 814                 kvm_mmu_reset_context(vcpu);
 815
 816         return 0;
 817 }
 818
 819 void kvm_enable_efer_bits(u64 mask)
 820 {
 821        efer_reserved_bits &= ~mask;
 822 }
 823 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 824
 825
 826 /*
 827  * Writes msr value into into the appropriate "register".
 828  * Returns 0 on success, non-0 otherwise.
 829  * Assumes vcpu_load() was already called.
 830  */
 831 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 832 {
 833         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 834 }
 835
 836 /*
 837  * Adapt set_msr() to msr_io()'s calling convention
 838  */
 839 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 840 {
 841         return kvm_set_msr(vcpu, index, *data);
 842 }
 843
 844 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 845 {
 846         int version;
 847         int r;
 848         struct pvclock_wall_clock wc;
 849         struct timespec boot;
 850
 851         if (!wall_clock)
 852                 return;
 853
 854         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
 855         if (r)
 856                 return;
 857
 858         if (version & 1)
 859                 ++version;  /* first time write, random junk */
 860
 861         ++version;
 862
 863         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 864
 865         /*
 866          * The guest calculates current wall clock time by adding
 867          * system time (updated by kvm_write_guest_time below) to the
 868          * wall clock specified here.  guest system time equals host
 869          * system time for us, thus we must fill in host boot time here.
 870          */
 871         getboottime(&boot);
 872
 873         wc.sec = boot.tv_sec;
 874         wc.nsec = boot.tv_nsec;
 875         wc.version = version;
 876
 877         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 878
 879         version++;
 880         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 881 }
 882
 883 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 884 {
 885         uint32_t quotient, remainder;
 886
 887         /* Don't try to replace with do_div(), this one calculates
 888          * "(dividend << 32) / divisor" */
 889         __asm__ ( "divl %4"
 890                   : "=a" (quotient), "=d" (remainder)
 891                   : "0" (0), "1" (dividend), "r" (divisor) );
 892         return quotient;
 893 }
 894
 895 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 896 {
 897         uint64_t nsecs = 1000000000LL;
 898         int32_t  shift = 0;
 899         uint64_t tps64;
 900         uint32_t tps32;
 901
 902         tps64 = tsc_khz * 1000LL;
 903         while (tps64 > nsecs*2) {
 904                 tps64 >>= 1;
 905                 shift--;
 906         }
 907
 908         tps32 = (uint32_t)tps64;
 909         while (tps32 <= (uint32_t)nsecs) {
 910                 tps32 <<= 1;
 911                 shift++;
 912         }
 913
 914         hv_clock->tsc_shift = shift;
 915         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 916
 917         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 918                  __func__, tsc_khz, hv_clock->tsc_shift,
 919                  hv_clock->tsc_to_system_mul);
 920 }
 921
 922 static inline u64 get_kernel_ns(void)
 923 {
 924         struct timespec ts;
 925
 926         WARN_ON(preemptible());
 927         ktime_get_ts(&ts);
 928         monotonic_to_bootbased(&ts);
 929         return timespec_to_ns(&ts);
 930 }
 931
 932 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 933
 934 static inline int kvm_tsc_changes_freq(void)
 935 {
 936         int cpu = get_cpu();
 937         int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
 938                   cpufreq_quick_get(cpu) != 0;
 939         put_cpu();
 940         return ret;
 941 }
 942
 943 static inline u64 nsec_to_cycles(u64 nsec)
 944 {
 945         u64 ret;
 946
 947         WARN_ON(preemptible());
 948         if (kvm_tsc_changes_freq())
 949                 printk_once(KERN_WARNING
 950                  "kvm: unreliable cycle conversion on adjustable rate TSC\n");
 951         ret = nsec * __get_cpu_var(cpu_tsc_khz);
 952         do_div(ret, USEC_PER_SEC);
 953         return ret;
 954 }
 955
 956 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
 957 {
 958         struct kvm *kvm = vcpu->kvm;
 959         u64 offset, ns, elapsed;
 960         unsigned long flags;
 961         s64 sdiff;
 962
 963         spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
 964         offset = data - native_read_tsc();
 965         ns = get_kernel_ns();
 966         elapsed = ns - kvm->arch.last_tsc_nsec;
 967         sdiff = data - kvm->arch.last_tsc_write;
 968         if (sdiff < 0)
 969                 sdiff = -sdiff;
 970
 971         /*
 972          * Special case: close write to TSC within 5 seconds of
 973          * another CPU is interpreted as an attempt to synchronize
 974          * The 5 seconds is to accomodate host load / swapping as
 975          * well as any reset of TSC during the boot process.
 976          *
 977          * In that case, for a reliable TSC, we can match TSC offsets,
 978          * or make a best guest using elapsed value.
 979          */
 980         if (sdiff < nsec_to_cycles(5ULL * NSEC_PER_SEC) &&
 981             elapsed < 5ULL * NSEC_PER_SEC) {
 982                 if (!check_tsc_unstable()) {
 983                         offset = kvm->arch.last_tsc_offset;
 984                         pr_debug("kvm: matched tsc offset for %llu\n", data);
 985                 } else {
 986                         u64 delta = nsec_to_cycles(elapsed);
 987                         offset += delta;
 988                         pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
 989                 }
 990                 ns = kvm->arch.last_tsc_nsec;
 991         }
 992         kvm->arch.last_tsc_nsec = ns;
 993         kvm->arch.last_tsc_write = data;
 994         kvm->arch.last_tsc_offset = offset;
 995         kvm_x86_ops->write_tsc_offset(vcpu, offset);
 996         spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
 997
 998         /* Reset of TSC must disable overshoot protection below */
 999         vcpu->arch.hv_clock.tsc_timestamp = 0;
1000 }
1001 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1002
1003 static int kvm_write_guest_time(struct kvm_vcpu *v)
1004 {
1005         unsigned long flags;
1006         struct kvm_vcpu_arch *vcpu = &v->arch;
1007         void *shared_kaddr;
1008         unsigned long this_tsc_khz;
1009         s64 kernel_ns, max_kernel_ns;
1010         u64 tsc_timestamp;
1011
1012         if ((!vcpu->time_page))
1013                 return 0;
1014
1015         /* Keep irq disabled to prevent changes to the clock */
1016         local_irq_save(flags);
1017         kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1018         kernel_ns = get_kernel_ns();
1019         this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1020         local_irq_restore(flags);
1021
1022         if (unlikely(this_tsc_khz == 0)) {
1023                 kvm_make_request(KVM_REQ_KVMCLOCK_UPDATE, v);
1024                 return 1;
1025         }
1026
1027         /*
1028          * Time as measured by the TSC may go backwards when resetting the base
1029          * tsc_timestamp.  The reason for this is that the TSC resolution is
1030          * higher than the resolution of the other clock scales.  Thus, many
1031          * possible measurments of the TSC correspond to one measurement of any
1032          * other clock, and so a spread of values is possible.  This is not a
1033          * problem for the computation of the nanosecond clock; with TSC rates
1034          * around 1GHZ, there can only be a few cycles which correspond to one
1035          * nanosecond value, and any path through this code will inevitably
1036          * take longer than that.  However, with the kernel_ns value itself,
1037          * the precision may be much lower, down to HZ granularity.  If the
1038          * first sampling of TSC against kernel_ns ends in the low part of the
1039          * range, and the second in the high end of the range, we can get:
1040          *
1041          * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1042          *
1043          * As the sampling errors potentially range in the thousands of cycles,
1044          * it is possible such a time value has already been observed by the
1045          * guest.  To protect against this, we must compute the system time as
1046          * observed by the guest and ensure the new system time is greater.
1047          */
1048         max_kernel_ns = 0;
1049         if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1050                 max_kernel_ns = vcpu->last_guest_tsc -
1051                                 vcpu->hv_clock.tsc_timestamp;
1052                 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1053                                     vcpu->hv_clock.tsc_to_system_mul,
1054                                     vcpu->hv_clock.tsc_shift);
1055                 max_kernel_ns += vcpu->last_kernel_ns;
1056         }
1057
1058         if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1059                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
1060                 vcpu->hw_tsc_khz = this_tsc_khz;
1061         }
1062
1063         if (max_kernel_ns > kernel_ns)
1064                 kernel_ns = max_kernel_ns;
1065
1066         /* With all the info we got, fill in the values */
1067         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1068         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1069         vcpu->last_kernel_ns = kernel_ns;
1070         vcpu->hv_clock.flags = 0;
1071
1072         /*
1073          * The interface expects us to write an even number signaling that the
1074          * update is finished. Since the guest won't see the intermediate
1075          * state, we just increase by 2 at the end.
1076          */
1077         vcpu->hv_clock.version += 2;
1078
1079         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1080
1081         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1082                sizeof(vcpu->hv_clock));
1083
1084         kunmap_atomic(shared_kaddr, KM_USER0);
1085
1086         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1087         return 0;
1088 }
1089
1090 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
1091 {
1092         struct kvm_vcpu_arch *vcpu = &v->arch;
1093
1094         if (!vcpu->time_page)
1095                 return 0;
1096         kvm_make_request(KVM_REQ_KVMCLOCK_UPDATE, v);
1097         return 1;
1098 }
1099
1100 static bool msr_mtrr_valid(unsigned msr)
1101 {
1102         switch (msr) {
1103         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1104         case MSR_MTRRfix64K_00000:
1105         case MSR_MTRRfix16K_80000:
1106         case MSR_MTRRfix16K_A0000:
1107         case MSR_MTRRfix4K_C0000:
1108         case MSR_MTRRfix4K_C8000:
1109         case MSR_MTRRfix4K_D0000:
1110         case MSR_MTRRfix4K_D8000:
1111         case MSR_MTRRfix4K_E0000:
1112         case MSR_MTRRfix4K_E8000:
1113         case MSR_MTRRfix4K_F0000:
1114         case MSR_MTRRfix4K_F8000:
1115         case MSR_MTRRdefType:
1116         case MSR_IA32_CR_PAT:
1117                 return true;
1118         case 0x2f8:
1119                 return true;
1120         }
1121         return false;
1122 }
1123
1124 static bool valid_pat_type(unsigned t)
1125 {
1126         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1127 }
1128
1129 static bool valid_mtrr_type(unsigned t)
1130 {
1131         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1132 }
1133
1134 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1135 {
1136         int i;
1137
1138         if (!msr_mtrr_valid(msr))
1139                 return false;
1140
1141         if (msr == MSR_IA32_CR_PAT) {
1142                 for (i = 0; i < 8; i++)
1143                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
1144                                 return false;
1145                 return true;
1146         } else if (msr == MSR_MTRRdefType) {
1147                 if (data & ~0xcff)
1148                         return false;
1149                 return valid_mtrr_type(data & 0xff);
1150         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1151                 for (i = 0; i < 8 ; i++)
1152                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1153                                 return false;
1154                 return true;
1155         }
1156
1157         /* variable MTRRs */
1158         return valid_mtrr_type(data & 0xff);
1159 }
1160
1161 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1162 {
1163         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1164
1165         if (!mtrr_valid(vcpu, msr, data))
1166                 return 1;
1167
1168         if (msr == MSR_MTRRdefType) {
1169                 vcpu->arch.mtrr_state.def_type = data;
1170                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1171         } else if (msr == MSR_MTRRfix64K_00000)
1172                 p[0] = data;
1173         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1174                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1175         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1176                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1177         else if (msr == MSR_IA32_CR_PAT)
1178                 vcpu->arch.pat = data;
1179         else {  /* Variable MTRRs */
1180                 int idx, is_mtrr_mask;
1181                 u64 *pt;
1182
1183                 idx = (msr - 0x200) / 2;
1184                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1185                 if (!is_mtrr_mask)
1186                         pt =
1187                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1188                 else
1189                         pt =
1190                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1191                 *pt = data;
1192         }
1193
1194         kvm_mmu_reset_context(vcpu);
1195         return 0;
1196 }
1197
1198 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1199 {
1200         u64 mcg_cap = vcpu->arch.mcg_cap;
1201         unsigned bank_num = mcg_cap & 0xff;
1202
1203         switch (msr) {
1204         case MSR_IA32_MCG_STATUS:
1205                 vcpu->arch.mcg_status = data;
1206                 break;
1207         case MSR_IA32_MCG_CTL:
1208                 if (!(mcg_cap & MCG_CTL_P))
1209                         return 1;
1210                 if (data != 0 && data != ~(u64)0)
1211                         return -1;
1212                 vcpu->arch.mcg_ctl = data;
1213                 break;
1214         default:
1215                 if (msr >= MSR_IA32_MC0_CTL &&
1216                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1217                         u32 offset = msr - MSR_IA32_MC0_CTL;
1218                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1219                          * some Linux kernels though clear bit 10 in bank 4 to
1220                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1221                          * this to avoid an uncatched #GP in the guest
1222                          */
1223                         if ((offset & 0x3) == 0 &&
1224                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1225                                 return -1;
1226                         vcpu->arch.mce_banks[offset] = data;
1227                         break;
1228                 }
1229                 return 1;
1230         }
1231         return 0;
1232 }
1233
1234 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1235 {
1236         struct kvm *kvm = vcpu->kvm;
1237         int lm = is_long_mode(vcpu);
1238         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1239                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1240         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1241                 : kvm->arch.xen_hvm_config.blob_size_32;
1242         u32 page_num = data & ~PAGE_MASK;
1243         u64 page_addr = data & PAGE_MASK;
1244         u8 *page;
1245         int r;
1246
1247         r = -E2BIG;
1248         if (page_num >= blob_size)
1249                 goto out;
1250         r = -ENOMEM;
1251         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1252         if (!page)
1253                 goto out;
1254         r = -EFAULT;
1255         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1256                 goto out_free;
1257         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1258                 goto out_free;
1259         r = 0;
1260 out_free:
1261         kfree(page);
1262 out:
1263         return r;
1264 }
1265
1266 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1267 {
1268         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1269 }
1270
1271 static bool kvm_hv_msr_partition_wide(u32 msr)
1272 {
1273         bool r = false;
1274         switch (msr) {
1275         case HV_X64_MSR_GUEST_OS_ID:
1276         case HV_X64_MSR_HYPERCALL:
1277                 r = true;
1278                 break;
1279         }
1280
1281         return r;
1282 }
1283
1284 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1285 {
1286         struct kvm *kvm = vcpu->kvm;
1287
1288         switch (msr) {
1289         case HV_X64_MSR_GUEST_OS_ID:
1290                 kvm->arch.hv_guest_os_id = data;
1291                 /* setting guest os id to zero disables hypercall page */
1292                 if (!kvm->arch.hv_guest_os_id)
1293                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1294                 break;
1295         case HV_X64_MSR_HYPERCALL: {
1296                 u64 gfn;
1297                 unsigned long addr;
1298                 u8 instructions[4];
1299
1300                 /* if guest os id is not set hypercall should remain disabled */
1301                 if (!kvm->arch.hv_guest_os_id)
1302                         break;
1303                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1304                         kvm->arch.hv_hypercall = data;
1305                         break;
1306                 }
1307                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1308                 addr = gfn_to_hva(kvm, gfn);
1309                 if (kvm_is_error_hva(addr))
1310                         return 1;
1311                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1312                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1313                 if (copy_to_user((void __user *)addr, instructions, 4))
1314                         return 1;
1315                 kvm->arch.hv_hypercall = data;
1316                 break;
1317         }
1318         default:
1319                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1320                           "data 0x%llx\n", msr, data);
1321                 return 1;
1322         }
1323         return 0;
1324 }
1325
1326 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1327 {
1328         switch (msr) {
1329         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1330                 unsigned long addr;
1331
1332                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1333                         vcpu->arch.hv_vapic = data;
1334                         break;
1335                 }
1336                 addr = gfn_to_hva(vcpu->kvm, data >>
1337                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1338                 if (kvm_is_error_hva(addr))
1339                         return 1;
1340                 if (clear_user((void __user *)addr, PAGE_SIZE))
1341                         return 1;
1342                 vcpu->arch.hv_vapic = data;
1343                 break;
1344         }
1345         case HV_X64_MSR_EOI:
1346                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1347         case HV_X64_MSR_ICR:
1348                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1349         case HV_X64_MSR_TPR:
1350                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1351         default:
1352                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1353                           "data 0x%llx\n", msr, data);
1354                 return 1;
1355         }
1356
1357         return 0;
1358 }
1359
1360 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1361 {
1362         switch (msr) {
1363         case MSR_EFER:
1364                 return set_efer(vcpu, data);
1365         case MSR_K7_HWCR:
1366                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1367                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
1368                 if (data != 0) {
1369                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1370                                 data);
1371                         return 1;
1372                 }
1373                 break;
1374         case MSR_FAM10H_MMIO_CONF_BASE:
1375                 if (data != 0) {
1376                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1377                                 "0x%llx\n", data);
1378                         return 1;
1379                 }
1380                 break;
1381         case MSR_AMD64_NB_CFG:
1382                 break;
1383         case MSR_IA32_DEBUGCTLMSR:
1384                 if (!data) {
1385                         /* We support the non-activated case already */
1386                         break;
1387                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1388                         /* Values other than LBR and BTF are vendor-specific,
1389                            thus reserved and should throw a #GP */
1390                         return 1;
1391                 }
1392                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1393                         __func__, data);
1394                 break;
1395         case MSR_IA32_UCODE_REV:
1396         case MSR_IA32_UCODE_WRITE:
1397         case MSR_VM_HSAVE_PA:
1398         case MSR_AMD64_PATCH_LOADER:
1399                 break;
1400         case 0x200 ... 0x2ff:
1401                 return set_msr_mtrr(vcpu, msr, data);
1402         case MSR_IA32_APICBASE:
1403                 kvm_set_apic_base(vcpu, data);
1404                 break;
1405         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1406                 return kvm_x2apic_msr_write(vcpu, msr, data);
1407         case MSR_IA32_MISC_ENABLE:
1408                 vcpu->arch.ia32_misc_enable_msr = data;
1409                 break;
1410         case MSR_KVM_WALL_CLOCK_NEW:
1411         case MSR_KVM_WALL_CLOCK:
1412                 vcpu->kvm->arch.wall_clock = data;
1413                 kvm_write_wall_clock(vcpu->kvm, data);
1414                 break;
1415         case MSR_KVM_SYSTEM_TIME_NEW:
1416         case MSR_KVM_SYSTEM_TIME: {
1417                 if (vcpu->arch.time_page) {
1418                         kvm_release_page_dirty(vcpu->arch.time_page);
1419                         vcpu->arch.time_page = NULL;
1420                 }
1421
1422                 vcpu->arch.time = data;
1423
1424                 /* we verify if the enable bit is set... */
1425                 if (!(data & 1))
1426                         break;
1427
1428                 /* ...but clean it before doing the actual write */
1429                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1430
1431                 vcpu->arch.time_page =
1432                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1433
1434                 if (is_error_page(vcpu->arch.time_page)) {
1435                         kvm_release_page_clean(vcpu->arch.time_page);
1436                         vcpu->arch.time_page = NULL;
1437                 }
1438
1439                 kvm_request_guest_time_update(vcpu);
1440                 break;
1441         }
1442         case MSR_IA32_MCG_CTL:
1443         case MSR_IA32_MCG_STATUS:
1444         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1445                 return set_msr_mce(vcpu, msr, data);
1446
1447         /* Performance counters are not protected by a CPUID bit,
1448          * so we should check all of them in the generic path for the sake of
1449          * cross vendor migration.
1450          * Writing a zero into the event select MSRs disables them,
1451          * which we perfectly emulate ;-). Any other value should be at least
1452          * reported, some guests depend on them.
1453          */
1454         case MSR_P6_EVNTSEL0:
1455         case MSR_P6_EVNTSEL1:
1456         case MSR_K7_EVNTSEL0:
1457         case MSR_K7_EVNTSEL1:
1458         case MSR_K7_EVNTSEL2:
1459         case MSR_K7_EVNTSEL3:
1460                 if (data != 0)
1461                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1462                                 "0x%x data 0x%llx\n", msr, data);
1463                 break;
1464         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1465          * so we ignore writes to make it happy.
1466          */
1467         case MSR_P6_PERFCTR0:
1468         case MSR_P6_PERFCTR1:
1469         case MSR_K7_PERFCTR0:
1470         case MSR_K7_PERFCTR1:
1471         case MSR_K7_PERFCTR2:
1472         case MSR_K7_PERFCTR3:
1473                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1474                         "0x%x data 0x%llx\n", msr, data);
1475                 break;
1476         case MSR_K7_CLK_CTL:
1477                 /*
1478                  * Ignore all writes to this no longer documented MSR.
1479                  * Writes are only relevant for old K7 processors,
1480                  * all pre-dating SVM, but a recommended workaround from
1481                  * AMD for these chips. It is possible to speicify the
1482                  * affected processor models on the command line, hence
1483                  * the need to ignore the workaround.
1484                  */
1485                 break;
1486         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1487                 if (kvm_hv_msr_partition_wide(msr)) {
1488                         int r;
1489                         mutex_lock(&vcpu->kvm->lock);
1490                         r = set_msr_hyperv_pw(vcpu, msr, data);
1491                         mutex_unlock(&vcpu->kvm->lock);
1492                         return r;
1493                 } else
1494                         return set_msr_hyperv(vcpu, msr, data);
1495                 break;
1496         default:
1497                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1498                         return xen_hvm_config(vcpu, data);
1499                 if (!ignore_msrs) {
1500                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1501                                 msr, data);
1502                         return 1;
1503                 } else {
1504                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1505                                 msr, data);
1506                         break;
1507                 }
1508         }
1509         return 0;
1510 }
1511 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1512
1513
1514 /*
1515  * Reads an msr value (of 'msr_index') into 'pdata'.
1516  * Returns 0 on success, non-0 otherwise.
1517  * Assumes vcpu_load() was already called.
1518  */
1519 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1520 {
1521         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1522 }
1523
1524 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1525 {
1526         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1527
1528         if (!msr_mtrr_valid(msr))
1529                 return 1;
1530
1531         if (msr == MSR_MTRRdefType)
1532                 *pdata = vcpu->arch.mtrr_state.def_type +
1533                          (vcpu->arch.mtrr_state.enabled << 10);
1534         else if (msr == MSR_MTRRfix64K_00000)
1535                 *pdata = p[0];
1536         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1537                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1538         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1539                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1540         else if (msr == MSR_IA32_CR_PAT)
1541                 *pdata = vcpu->arch.pat;
1542         else {  /* Variable MTRRs */
1543                 int idx, is_mtrr_mask;
1544                 u64 *pt;
1545
1546                 idx = (msr - 0x200) / 2;
1547                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1548                 if (!is_mtrr_mask)
1549                         pt =
1550                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1551                 else
1552                         pt =
1553                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1554                 *pdata = *pt;
1555         }
1556
1557         return 0;
1558 }
1559
1560 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1561 {
1562         u64 data;
1563         u64 mcg_cap = vcpu->arch.mcg_cap;
1564         unsigned bank_num = mcg_cap & 0xff;
1565
1566         switch (msr) {
1567         case MSR_IA32_P5_MC_ADDR:
1568         case MSR_IA32_P5_MC_TYPE:
1569                 data = 0;
1570                 break;
1571         case MSR_IA32_MCG_CAP:
1572                 data = vcpu->arch.mcg_cap;
1573                 break;
1574         case MSR_IA32_MCG_CTL:
1575                 if (!(mcg_cap & MCG_CTL_P))
1576                         return 1;
1577                 data = vcpu->arch.mcg_ctl;
1578                 break;
1579         case MSR_IA32_MCG_STATUS:
1580                 data = vcpu->arch.mcg_status;
1581                 break;
1582         default:
1583                 if (msr >= MSR_IA32_MC0_CTL &&
1584                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1585                         u32 offset = msr - MSR_IA32_MC0_CTL;
1586                         data = vcpu->arch.mce_banks[offset];
1587                         break;
1588                 }
1589                 return 1;
1590         }
1591         *pdata = data;
1592         return 0;
1593 }
1594
1595 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1596 {
1597         u64 data = 0;
1598         struct kvm *kvm = vcpu->kvm;
1599
1600         switch (msr) {
1601         case HV_X64_MSR_GUEST_OS_ID:
1602                 data = kvm->arch.hv_guest_os_id;
1603                 break;
1604         case HV_X64_MSR_HYPERCALL:
1605                 data = kvm->arch.hv_hypercall;
1606                 break;
1607         default:
1608                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1609                 return 1;
1610         }
1611
1612         *pdata = data;
1613         return 0;
1614 }
1615
1616 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1617 {
1618         u64 data = 0;
1619
1620         switch (msr) {
1621         case HV_X64_MSR_VP_INDEX: {
1622                 int r;
1623                 struct kvm_vcpu *v;
1624                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1625                         if (v == vcpu)
1626                                 data = r;
1627                 break;
1628         }
1629         case HV_X64_MSR_EOI:
1630                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1631         case HV_X64_MSR_ICR:
1632                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1633         case HV_X64_MSR_TPR:
1634                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1635         default:
1636                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1637                 return 1;
1638         }
1639         *pdata = data;
1640         return 0;
1641 }
1642
1643 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1644 {
1645         u64 data;
1646
1647         switch (msr) {
1648         case MSR_IA32_PLATFORM_ID:
1649         case MSR_IA32_UCODE_REV:
1650         case MSR_IA32_EBL_CR_POWERON:
1651         case MSR_IA32_DEBUGCTLMSR:
1652         case MSR_IA32_LASTBRANCHFROMIP:
1653         case MSR_IA32_LASTBRANCHTOIP:
1654         case MSR_IA32_LASTINTFROMIP:
1655         case MSR_IA32_LASTINTTOIP:
1656         case MSR_K8_SYSCFG:
1657         case MSR_K7_HWCR:
1658         case MSR_VM_HSAVE_PA:
1659         case MSR_P6_PERFCTR0:
1660         case MSR_P6_PERFCTR1:
1661         case MSR_P6_EVNTSEL0:
1662         case MSR_P6_EVNTSEL1:
1663         case MSR_K7_EVNTSEL0:
1664         case MSR_K7_PERFCTR0:
1665         case MSR_K8_INT_PENDING_MSG:
1666         case MSR_AMD64_NB_CFG:
1667         case MSR_FAM10H_MMIO_CONF_BASE:
1668                 data = 0;
1669                 break;
1670         case MSR_MTRRcap:
1671                 data = 0x500 | KVM_NR_VAR_MTRR;
1672                 break;
1673         case 0x200 ... 0x2ff:
1674                 return get_msr_mtrr(vcpu, msr, pdata);
1675         case 0xcd: /* fsb frequency */
1676                 data = 3;
1677                 break;
1678                 /*
1679                  * MSR_EBC_FREQUENCY_ID
1680                  * Conservative value valid for even the basic CPU models.
1681                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1682                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1683                  * and 266MHz for model 3, or 4. Set Core Clock
1684                  * Frequency to System Bus Frequency Ratio to 1 (bits
1685                  * 31:24) even though these are only valid for CPU
1686                  * models > 2, however guests may end up dividing or
1687                  * multiplying by zero otherwise.
1688                  */
1689         case MSR_EBC_FREQUENCY_ID:
1690                 data = 1 << 24;
1691                 break;
1692         case MSR_IA32_APICBASE:
1693                 data = kvm_get_apic_base(vcpu);
1694                 break;
1695         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1696                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1697                 break;
1698         case MSR_IA32_MISC_ENABLE:
1699                 data = vcpu->arch.ia32_misc_enable_msr;
1700                 break;
1701         case MSR_IA32_PERF_STATUS:
1702                 /* TSC increment by tick */
1703                 data = 1000ULL;
1704                 /* CPU multiplier */
1705                 data |= (((uint64_t)4ULL) << 40);
1706                 break;
1707         case MSR_EFER:
1708                 data = vcpu->arch.efer;
1709                 break;
1710         case MSR_KVM_WALL_CLOCK:
1711         case MSR_KVM_WALL_CLOCK_NEW:
1712                 data = vcpu->kvm->arch.wall_clock;
1713                 break;
1714         case MSR_KVM_SYSTEM_TIME:
1715         case MSR_KVM_SYSTEM_TIME_NEW:
1716                 data = vcpu->arch.time;
1717                 break;
1718         case MSR_IA32_P5_MC_ADDR:
1719         case MSR_IA32_P5_MC_TYPE:
1720         case MSR_IA32_MCG_CAP:
1721         case MSR_IA32_MCG_CTL:
1722         case MSR_IA32_MCG_STATUS:
1723         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1724                 return get_msr_mce(vcpu, msr, pdata);
1725         case MSR_K7_CLK_CTL:
1726                 /*
1727                  * Provide expected ramp-up count for K7. All other
1728                  * are set to zero, indicating minimum divisors for
1729                  * every field.
1730                  *
1731                  * This prevents guest kernels on AMD host with CPU
1732                  * type 6, model 8 and higher from exploding due to
1733                  * the rdmsr failing.
1734                  */
1735                 data = 0x20000000;
1736                 break;
1737         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1738                 if (kvm_hv_msr_partition_wide(msr)) {
1739                         int r;
1740                         mutex_lock(&vcpu->kvm->lock);
1741                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1742                         mutex_unlock(&vcpu->kvm->lock);
1743                         return r;
1744                 } else
1745                         return get_msr_hyperv(vcpu, msr, pdata);
1746                 break;
1747         default:
1748                 if (!ignore_msrs) {
1749                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1750                         return 1;
1751                 } else {
1752                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1753                         data = 0;
1754                 }
1755                 break;
1756         }
1757         *pdata = data;
1758         return 0;
1759 }
1760 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1761
1762 /*
1763  * Read or write a bunch of msrs. All parameters are kernel addresses.
1764  *
1765  * @return number of msrs set successfully.
1766  */
1767 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1768                     struct kvm_msr_entry *entries,
1769                     int (*do_msr)(struct kvm_vcpu *vcpu,
1770                                   unsigned index, u64 *data))
1771 {
1772         int i, idx;
1773
1774         idx = srcu_read_lock(&vcpu->kvm->srcu);
1775         for (i = 0; i < msrs->nmsrs; ++i)
1776                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1777                         break;
1778         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1779
1780         return i;
1781 }
1782
1783 /*
1784  * Read or write a bunch of msrs. Parameters are user addresses.
1785  *
1786  * @return number of msrs set successfully.
1787  */
1788 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1789                   int (*do_msr)(struct kvm_vcpu *vcpu,
1790                                 unsigned index, u64 *data),
1791                   int writeback)
1792 {
1793         struct kvm_msrs msrs;
1794         struct kvm_msr_entry *entries;
1795         int r, n;
1796         unsigned size;
1797
1798         r = -EFAULT;
1799         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1800                 goto out;
1801
1802         r = -E2BIG;
1803         if (msrs.nmsrs >= MAX_IO_MSRS)
1804                 goto out;
1805
1806         r = -ENOMEM;
1807         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1808         entries = kmalloc(size, GFP_KERNEL);
1809         if (!entries)
1810                 goto out;
1811
1812         r = -EFAULT;
1813         if (copy_from_user(entries, user_msrs->entries, size))
1814                 goto out_free;
1815
1816         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1817         if (r < 0)
1818                 goto out_free;
1819
1820         r = -EFAULT;
1821         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1822                 goto out_free;
1823
1824         r = n;
1825
1826 out_free:
1827         kfree(entries);
1828 out:
1829         return r;
1830 }
1831
1832 int kvm_dev_ioctl_check_extension(long ext)
1833 {
1834         int r;
1835
1836         switch (ext) {
1837         case KVM_CAP_IRQCHIP:
1838         case KVM_CAP_HLT:
1839         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1840         case KVM_CAP_SET_TSS_ADDR:
1841         case KVM_CAP_EXT_CPUID:
1842         case KVM_CAP_CLOCKSOURCE:
1843         case KVM_CAP_PIT:
1844         case KVM_CAP_NOP_IO_DELAY:
1845         case KVM_CAP_MP_STATE:
1846         case KVM_CAP_SYNC_MMU:
1847         case KVM_CAP_REINJECT_CONTROL:
1848         case KVM_CAP_IRQ_INJECT_STATUS:
1849         case KVM_CAP_ASSIGN_DEV_IRQ:
1850         case KVM_CAP_IRQFD:
1851         case KVM_CAP_IOEVENTFD:
1852         case KVM_CAP_PIT2:
1853         case KVM_CAP_PIT_STATE2:
1854         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1855         case KVM_CAP_XEN_HVM:
1856         case KVM_CAP_ADJUST_CLOCK:
1857         case KVM_CAP_VCPU_EVENTS:
1858         case KVM_CAP_HYPERV:
1859         case KVM_CAP_HYPERV_VAPIC:
1860         case KVM_CAP_HYPERV_SPIN:
1861         case KVM_CAP_PCI_SEGMENT:
1862         case KVM_CAP_DEBUGREGS:
1863         case KVM_CAP_X86_ROBUST_SINGLESTEP:
1864         case KVM_CAP_XSAVE:
1865                 r = 1;
1866                 break;
1867         case KVM_CAP_COALESCED_MMIO:
1868                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1869                 break;
1870         case KVM_CAP_VAPIC:
1871                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1872                 break;
1873         case KVM_CAP_NR_VCPUS:
1874                 r = KVM_MAX_VCPUS;
1875                 break;
1876         case KVM_CAP_NR_MEMSLOTS:
1877                 r = KVM_MEMORY_SLOTS;
1878                 break;
1879         case KVM_CAP_PV_MMU:    /* obsolete */
1880                 r = 0;
1881                 break;
1882         case KVM_CAP_IOMMU:
1883                 r = iommu_found();
1884                 break;
1885         case KVM_CAP_MCE:
1886                 r = KVM_MAX_MCE_BANKS;
1887                 break;
1888         case KVM_CAP_XCRS:
1889                 r = cpu_has_xsave;
1890                 break;
1891         default:
1892                 r = 0;
1893                 break;
1894         }
1895         return r;
1896
1897 }
1898
1899 long kvm_arch_dev_ioctl(struct file *filp,
1900                         unsigned int ioctl, unsigned long arg)
1901 {
1902         void __user *argp = (void __user *)arg;
1903         long r;
1904
1905         switch (ioctl) {
1906         case KVM_GET_MSR_INDEX_LIST: {
1907                 struct kvm_msr_list __user *user_msr_list = argp;
1908                 struct kvm_msr_list msr_list;
1909                 unsigned n;
1910
1911                 r = -EFAULT;
1912                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1913                         goto out;
1914                 n = msr_list.nmsrs;
1915                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1916                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1917                         goto out;
1918                 r = -E2BIG;
1919                 if (n < msr_list.nmsrs)
1920                         goto out;
1921                 r = -EFAULT;
1922                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1923                                  num_msrs_to_save * sizeof(u32)))
1924                         goto out;
1925                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1926                                  &emulated_msrs,
1927                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1928                         goto out;
1929                 r = 0;
1930                 break;
1931         }
1932         case KVM_GET_SUPPORTED_CPUID: {
1933                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1934                 struct kvm_cpuid2 cpuid;
1935
1936                 r = -EFAULT;
1937                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1938                         goto out;
1939                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1940                                                       cpuid_arg->entries);
1941                 if (r)
1942                         goto out;
1943
1944                 r = -EFAULT;
1945                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1946                         goto out;
1947                 r = 0;
1948                 break;
1949         }
1950         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1951                 u64 mce_cap;
1952
1953                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1954                 r = -EFAULT;
1955                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1956                         goto out;
1957                 r = 0;
1958                 break;
1959         }
1960         default:
1961                 r = -EINVAL;
1962         }
1963 out:
1964         return r;
1965 }
1966
1967 static void wbinvd_ipi(void *garbage)
1968 {
1969         wbinvd();
1970 }
1971
1972 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
1973 {
1974         return vcpu->kvm->arch.iommu_domain &&
1975                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
1976 }
1977
1978 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1979 {
1980         /* Address WBINVD may be executed by guest */
1981         if (need_emulate_wbinvd(vcpu)) {
1982                 if (kvm_x86_ops->has_wbinvd_exit())
1983                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
1984                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
1985                         smp_call_function_single(vcpu->cpu,
1986                                         wbinvd_ipi, NULL, 1);
1987         }
1988
1989         kvm_x86_ops->vcpu_load(vcpu, cpu);
1990         if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
1991                 /* Make sure TSC doesn't go backwards */
1992                 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
1993                                 native_read_tsc() - vcpu->arch.last_host_tsc;
1994                 if (tsc_delta < 0)
1995                         mark_tsc_unstable("KVM discovered backwards TSC");
1996                 if (check_tsc_unstable())
1997                         kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
1998                 kvm_migrate_timers(vcpu);
1999                 vcpu->cpu = cpu;
2000         }
2001 }
2002
2003 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2004 {
2005         kvm_x86_ops->vcpu_put(vcpu);
2006         kvm_put_guest_fpu(vcpu);
2007         vcpu->arch.last_host_tsc = native_read_tsc();
2008 }
2009
2010 static int is_efer_nx(void)
2011 {
2012         unsigned long long efer = 0;
2013
2014         rdmsrl_safe(MSR_EFER, &efer);
2015         return efer & EFER_NX;
2016 }
2017
2018 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2019 {
2020         int i;
2021         struct kvm_cpuid_entry2 *e, *entry;
2022
2023         entry = NULL;
2024         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2025                 e = &vcpu->arch.cpuid_entries[i];
2026                 if (e->function == 0x80000001) {
2027                         entry = e;
2028                         break;
2029                 }
2030         }
2031         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2032                 entry->edx &= ~(1 << 20);
2033                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2034         }
2035 }
2036
2037 /* when an old userspace process fills a new kernel module */
2038 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2039                                     struct kvm_cpuid *cpuid,
2040                                     struct kvm_cpuid_entry __user *entries)
2041 {
2042         int r, i;
2043         struct kvm_cpuid_entry *cpuid_entries;
2044
2045         r = -E2BIG;
2046         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2047                 goto out;
2048         r = -ENOMEM;
2049         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2050         if (!cpuid_entries)
2051                 goto out;
2052         r = -EFAULT;
2053         if (copy_from_user(cpuid_entries, entries,
2054                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2055                 goto out_free;
2056         for (i = 0; i < cpuid->nent; i++) {
2057                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2058                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2059                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2060                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2061                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2062                 vcpu->arch.cpuid_entries[i].index = 0;
2063                 vcpu->arch.cpuid_entries[i].flags = 0;
2064                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2065                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2066                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2067         }
2068         vcpu->arch.cpuid_nent = cpuid->nent;
2069         cpuid_fix_nx_cap(vcpu);
2070         r = 0;
2071         kvm_apic_set_version(vcpu);
2072         kvm_x86_ops->cpuid_update(vcpu);
2073         update_cpuid(vcpu);
2074
2075 out_free:
2076         vfree(cpuid_entries);
2077 out:
2078         return r;
2079 }
2080
2081 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2082                                      struct kvm_cpuid2 *cpuid,
2083                                      struct kvm_cpuid_entry2 __user *entries)
2084 {
2085         int r;
2086
2087         r = -E2BIG;
2088         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2089                 goto out;
2090         r = -EFAULT;
2091         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2092                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2093                 goto out;
2094         vcpu->arch.cpuid_nent = cpuid->nent;
2095         kvm_apic_set_version(vcpu);
2096         kvm_x86_ops->cpuid_update(vcpu);
2097         update_cpuid(vcpu);
2098         return 0;
2099
2100 out:
2101         return r;
2102 }
2103
2104 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2105                                      struct kvm_cpuid2 *cpuid,
2106                                      struct kvm_cpuid_entry2 __user *entries)
2107 {
2108         int r;
2109
2110         r = -E2BIG;
2111         if (cpuid->nent < vcpu->arch.cpuid_nent)
2112                 goto out;
2113         r = -EFAULT;
2114         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2115                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2116                 goto out;
2117         return 0;
2118
2119 out:
2120         cpuid->nent = vcpu->arch.cpuid_nent;
2121         return r;
2122 }
2123
2124 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2125                            u32 index)
2126 {
2127         entry->function = function;
2128         entry->index = index;
2129         cpuid_count(entry->function, entry->index,
2130                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2131         entry->flags = 0;
2132 }
2133
2134 #define F(x) bit(X86_FEATURE_##x)
2135
2136 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2137                          u32 index, int *nent, int maxnent)
2138 {
2139         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2140 #ifdef CONFIG_X86_64
2141         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2142                                 ? F(GBPAGES) : 0;
2143         unsigned f_lm = F(LM);
2144 #else
2145         unsigned f_gbpages = 0;
2146         unsigned f_lm = 0;
2147 #endif
2148         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2149
2150         /* cpuid 1.edx */
2151         const u32 kvm_supported_word0_x86_features =
2152                 F(FPU) | F(VME) | F(DE) | F(PSE) |
2153                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2154                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2155                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2156                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2157                 0 /* Reserved, DS, ACPI */ | F(MMX) |
2158                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2159                 0 /* HTT, TM, Reserved, PBE */;
2160         /* cpuid 0x80000001.edx */
2161         const u32 kvm_supported_word1_x86_features =
2162                 F(FPU) | F(VME) | F(DE) | F(PSE) |
2163                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2164                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2165                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2166                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2167                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2168                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2169                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2170         /* cpuid 1.ecx */
2171         const u32 kvm_supported_word4_x86_features =
2172                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2173                 0 /* DS-CPL, VMX, SMX, EST */ |
2174                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2175                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2176                 0 /* Reserved, DCA */ | F(XMM4_1) |
2177                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2178                 0 /* Reserved, AES */ | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX);
2179         /* cpuid 0x80000001.ecx */
2180         const u32 kvm_supported_word6_x86_features =
2181                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
2182                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2183                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
2184                 0 /* SKINIT */ | 0 /* WDT */;
2185
2186         /* all calls to cpuid_count() should be made on the same cpu */
2187         get_cpu();
2188         do_cpuid_1_ent(entry, function, index);
2189         ++*nent;
2190
2191         switch (function) {
2192         case 0:
2193                 entry->eax = min(entry->eax, (u32)0xd);
2194                 break;
2195         case 1:
2196                 entry->edx &= kvm_supported_word0_x86_features;
2197                 entry->ecx &= kvm_supported_word4_x86_features;
2198                 /* we support x2apic emulation even if host does not support
2199                  * it since we emulate x2apic in software */
2200                 entry->ecx |= F(X2APIC);
2201                 break;
2202         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2203          * may return different values. This forces us to get_cpu() before
2204          * issuing the first command, and also to emulate this annoying behavior
2205          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2206         case 2: {
2207                 int t, times = entry->eax & 0xff;
2208
2209                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2210                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2211                 for (t = 1; t < times && *nent < maxnent; ++t) {
2212                         do_cpuid_1_ent(&entry[t], function, 0);
2213                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2214                         ++*nent;
2215                 }
2216                 break;
2217         }
2218         /* function 4 and 0xb have additional index. */
2219         case 4: {
2220                 int i, cache_type;
2221
2222                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2223                 /* read more entries until cache_type is zero */
2224                 for (i = 1; *nent < maxnent; ++i) {
2225                         cache_type = entry[i - 1].eax & 0x1f;
2226                         if (!cache_type)
2227                                 break;
2228                         do_cpuid_1_ent(&entry[i], function, i);
2229                         entry[i].flags |=
2230                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2231                         ++*nent;
2232                 }
2233                 break;
2234         }
2235         case 0xb: {
2236                 int i, level_type;
2237
2238                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2239                 /* read more entries until level_type is zero */
2240                 for (i = 1; *nent < maxnent; ++i) {
2241                         level_type = entry[i - 1].ecx & 0xff00;
2242                         if (!level_type)
2243                                 break;
2244                         do_cpuid_1_ent(&entry[i], function, i);
2245                         entry[i].flags |=
2246                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2247                         ++*nent;
2248                 }
2249                 break;
2250         }
2251         case 0xd: {
2252                 int i;
2253
2254                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2255                 for (i = 1; *nent < maxnent; ++i) {
2256                         if (entry[i - 1].eax == 0 && i != 2)
2257                                 break;
2258                         do_cpuid_1_ent(&entry[i], function, i);
2259                         entry[i].flags |=
2260                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2261                         ++*nent;
2262                 }
2263                 break;
2264         }
2265         case KVM_CPUID_SIGNATURE: {
2266                 char signature[12] = "KVMKVMKVM\0\0";
2267                 u32 *sigptr = (u32 *)signature;
2268                 entry->eax = 0;
2269                 entry->ebx = sigptr[0];
2270                 entry->ecx = sigptr[1];
2271                 entry->edx = sigptr[2];
2272                 break;
2273         }
2274         case KVM_CPUID_FEATURES:
2275                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2276                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
2277                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
2278                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2279                 entry->ebx = 0;
2280                 entry->ecx = 0;
2281                 entry->edx = 0;
2282                 break;
2283         case 0x80000000:
2284                 entry->eax = min(entry->eax, 0x8000001a);
2285                 break;
2286         case 0x80000001:
2287                 entry->edx &= kvm_supported_word1_x86_features;
2288                 entry->ecx &= kvm_supported_word6_x86_features;
2289                 break;
2290         }
2291
2292         kvm_x86_ops->set_supported_cpuid(function, entry);
2293
2294         put_cpu();
2295 }
2296
2297 #undef F
2298
2299 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2300                                      struct kvm_cpuid_entry2 __user *entries)
2301 {
2302         struct kvm_cpuid_entry2 *cpuid_entries;
2303         int limit, nent = 0, r = -E2BIG;
2304         u32 func;
2305
2306         if (cpuid->nent < 1)
2307                 goto out;
2308         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2309                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2310         r = -ENOMEM;
2311         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2312         if (!cpuid_entries)
2313                 goto out;
2314
2315         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2316         limit = cpuid_entries[0].eax;
2317         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2318                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2319                              &nent, cpuid->nent);
2320         r = -E2BIG;
2321         if (nent >= cpuid->nent)
2322                 goto out_free;
2323
2324         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2325         limit = cpuid_entries[nent - 1].eax;
2326         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2327                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2328                              &nent, cpuid->nent);
2329
2330
2331
2332         r = -E2BIG;
2333         if (nent >= cpuid->nent)
2334                 goto out_free;
2335
2336         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2337                      cpuid->nent);
2338
2339         r = -E2BIG;
2340         if (nent >= cpuid->nent)
2341                 goto out_free;
2342
2343         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2344                      cpuid->nent);
2345
2346         r = -E2BIG;
2347         if (nent >= cpuid->nent)
2348                 goto out_free;
2349
2350         r = -EFAULT;
2351         if (copy_to_user(entries, cpuid_entries,
2352                          nent * sizeof(struct kvm_cpuid_entry2)))
2353                 goto out_free;
2354         cpuid->nent = nent;
2355         r = 0;
2356
2357 out_free:
2358         vfree(cpuid_entries);
2359 out:
2360         return r;
2361 }
2362
2363 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2364                                     struct kvm_lapic_state *s)
2365 {
2366         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2367
2368         return 0;
2369 }
2370
2371 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2372                                     struct kvm_lapic_state *s)
2373 {
2374         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2375         kvm_apic_post_state_restore(vcpu);
2376         update_cr8_intercept(vcpu);
2377
2378         return 0;
2379 }
2380
2381 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2382                                     struct kvm_interrupt *irq)
2383 {
2384         if (irq->irq < 0 || irq->irq >= 256)
2385                 return -EINVAL;
2386         if (irqchip_in_kernel(vcpu->kvm))
2387                 return -ENXIO;
2388
2389         kvm_queue_interrupt(vcpu, irq->irq, false);
2390
2391         return 0;
2392 }
2393
2394 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2395 {
2396         kvm_inject_nmi(vcpu);
2397
2398         return 0;
2399 }
2400
2401 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2402                                            struct kvm_tpr_access_ctl *tac)
2403 {
2404         if (tac->flags)
2405                 return -EINVAL;
2406         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2407         return 0;
2408 }
2409
2410 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2411                                         u64 mcg_cap)
2412 {
2413         int r;
2414         unsigned bank_num = mcg_cap & 0xff, bank;
2415
2416         r = -EINVAL;
2417         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2418                 goto out;
2419         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2420                 goto out;
2421         r = 0;
2422         vcpu->arch.mcg_cap = mcg_cap;
2423         /* Init IA32_MCG_CTL to all 1s */
2424         if (mcg_cap & MCG_CTL_P)
2425                 vcpu->arch.mcg_ctl = ~(u64)0;
2426         /* Init IA32_MCi_CTL to all 1s */
2427         for (bank = 0; bank < bank_num; bank++)
2428                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2429 out:
2430         return r;
2431 }
2432
2433 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2434                                       struct kvm_x86_mce *mce)
2435 {
2436         u64 mcg_cap = vcpu->arch.mcg_cap;
2437         unsigned bank_num = mcg_cap & 0xff;
2438         u64 *banks = vcpu->arch.mce_banks;
2439
2440         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2441                 return -EINVAL;
2442         /*
2443          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2444          * reporting is disabled
2445          */
2446         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2447             vcpu->arch.mcg_ctl != ~(u64)0)
2448                 return 0;
2449         banks += 4 * mce->bank;
2450         /*
2451          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2452          * reporting is disabled for the bank
2453          */
2454         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2455                 return 0;
2456         if (mce->status & MCI_STATUS_UC) {
2457                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2458                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2459                         printk(KERN_DEBUG "kvm: set_mce: "
2460                                "injects mce exception while "
2461                                "previous one is in progress!\n");
2462                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2463                         return 0;
2464                 }
2465                 if (banks[1] & MCI_STATUS_VAL)
2466                         mce->status |= MCI_STATUS_OVER;
2467                 banks[2] = mce->addr;
2468                 banks[3] = mce->misc;
2469                 vcpu->arch.mcg_status = mce->mcg_status;
2470                 banks[1] = mce->status;
2471                 kvm_queue_exception(vcpu, MC_VECTOR);
2472         } else if (!(banks[1] & MCI_STATUS_VAL)
2473                    || !(banks[1] & MCI_STATUS_UC)) {
2474                 if (banks[1] & MCI_STATUS_VAL)
2475                         mce->status |= MCI_STATUS_OVER;
2476                 banks[2] = mce->addr;
2477                 banks[3] = mce->misc;
2478                 banks[1] = mce->status;
2479         } else
2480                 banks[1] |= MCI_STATUS_OVER;
2481         return 0;
2482 }
2483
2484 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2485                                                struct kvm_vcpu_events *events)
2486 {
2487         events->exception.injected =
2488                 vcpu->arch.exception.pending &&
2489                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2490         events->exception.nr = vcpu->arch.exception.nr;
2491         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2492         events->exception.error_code = vcpu->arch.exception.error_code;
2493
2494         events->interrupt.injected =
2495                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2496         events->interrupt.nr = vcpu->arch.interrupt.nr;
2497         events->interrupt.soft = 0;
2498         events->interrupt.shadow =
2499                 kvm_x86_ops->get_interrupt_shadow(vcpu,
2500                         KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2501
2502         events->nmi.injected = vcpu->arch.nmi_injected;
2503         events->nmi.pending = vcpu->arch.nmi_pending;
2504         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2505
2506         events->sipi_vector = vcpu->arch.sipi_vector;
2507
2508         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2509                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2510                          | KVM_VCPUEVENT_VALID_SHADOW);
2511 }
2512
2513 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2514                                               struct kvm_vcpu_events *events)
2515 {
2516         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2517                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2518                               | KVM_VCPUEVENT_VALID_SHADOW))
2519                 return -EINVAL;
2520
2521         vcpu->arch.exception.pending = events->exception.injected;
2522         vcpu->arch.exception.nr = events->exception.nr;
2523         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2524         vcpu->arch.exception.error_code = events->exception.error_code;
2525
2526         vcpu->arch.interrupt.pending = events->interrupt.injected;
2527         vcpu->arch.interrupt.nr = events->interrupt.nr;
2528         vcpu->arch.interrupt.soft = events->interrupt.soft;
2529         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2530                 kvm_pic_clear_isr_ack(vcpu->kvm);
2531         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2532                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2533                                                   events->interrupt.shadow);
2534
2535         vcpu->arch.nmi_injected = events->nmi.injected;
2536         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2537                 vcpu->arch.nmi_pending = events->nmi.pending;
2538         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2539
2540         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2541                 vcpu->arch.sipi_vector = events->sipi_vector;
2542
2543         return 0;
2544 }
2545
2546 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2547                                              struct kvm_debugregs *dbgregs)
2548 {
2549         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2550         dbgregs->dr6 = vcpu->arch.dr6;
2551         dbgregs->dr7 = vcpu->arch.dr7;
2552         dbgregs->flags = 0;
2553 }
2554
2555 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2556                                             struct kvm_debugregs *dbgregs)
2557 {
2558         if (dbgregs->flags)
2559                 return -EINVAL;
2560
2561         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2562         vcpu->arch.dr6 = dbgregs->dr6;
2563         vcpu->arch.dr7 = dbgregs->dr7;
2564
2565         return 0;
2566 }
2567
2568 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2569                                          struct kvm_xsave *guest_xsave)
2570 {
2571         if (cpu_has_xsave)
2572                 memcpy(guest_xsave->region,
2573                         &vcpu->arch.guest_fpu.state->xsave,
2574                         xstate_size);
2575         else {
2576                 memcpy(guest_xsave->region,
2577                         &vcpu->arch.guest_fpu.state->fxsave,
2578                         sizeof(struct i387_fxsave_struct));
2579                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2580                         XSTATE_FPSSE;
2581         }
2582 }
2583
2584 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2585                                         struct kvm_xsave *guest_xsave)
2586 {
2587         u64 xstate_bv =
2588                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2589
2590         if (cpu_has_xsave)
2591                 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2592                         guest_xsave->region, xstate_size);
2593         else {
2594                 if (xstate_bv & ~XSTATE_FPSSE)
2595                         return -EINVAL;
2596                 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2597                         guest_xsave->region, sizeof(struct i387_fxsave_struct));
2598         }
2599         return 0;
2600 }
2601
2602 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2603                                         struct kvm_xcrs *guest_xcrs)
2604 {
2605         if (!cpu_has_xsave) {
2606                 guest_xcrs->nr_xcrs = 0;
2607                 return;
2608         }
2609
2610         guest_xcrs->nr_xcrs = 1;
2611         guest_xcrs->flags = 0;
2612         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2613         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2614 }
2615
2616 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2617                                        struct kvm_xcrs *guest_xcrs)
2618 {
2619         int i, r = 0;
2620
2621         if (!cpu_has_xsave)
2622                 return -EINVAL;
2623
2624         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2625                 return -EINVAL;
2626
2627         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2628                 /* Only support XCR0 currently */
2629                 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2630                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2631                                 guest_xcrs->xcrs[0].value);
2632                         break;
2633                 }
2634         if (r)
2635                 r = -EINVAL;
2636         return r;
2637 }
2638
2639 long kvm_arch_vcpu_ioctl(struct file *filp,
2640                          unsigned int ioctl, unsigned long arg)
2641 {
2642         struct kvm_vcpu *vcpu = filp->private_data;
2643         void __user *argp = (void __user *)arg;
2644         int r;
2645         union {
2646                 struct kvm_lapic_state *lapic;
2647                 struct kvm_xsave *xsave;
2648                 struct kvm_xcrs *xcrs;
2649                 void *buffer;
2650         } u;
2651
2652         u.buffer = NULL;
2653         switch (ioctl) {
2654         case KVM_GET_LAPIC: {
2655                 r = -EINVAL;
2656                 if (!vcpu->arch.apic)
2657                         goto out;
2658                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2659
2660                 r = -ENOMEM;
2661                 if (!u.lapic)
2662                         goto out;
2663                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2664                 if (r)
2665                         goto out;
2666                 r = -EFAULT;
2667                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2668                         goto out;
2669                 r = 0;
2670                 break;
2671         }
2672         case KVM_SET_LAPIC: {
2673                 r = -EINVAL;
2674                 if (!vcpu->arch.apic)
2675                         goto out;
2676                 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2677                 r = -ENOMEM;
2678                 if (!u.lapic)
2679                         goto out;
2680                 r = -EFAULT;
2681                 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2682                         goto out;
2683                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2684                 if (r)
2685                         goto out;
2686                 r = 0;
2687                 break;
2688         }
2689         case KVM_INTERRUPT: {
2690                 struct kvm_interrupt irq;
2691
2692                 r = -EFAULT;
2693                 if (copy_from_user(&irq, argp, sizeof irq))
2694                         goto out;
2695                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2696                 if (r)
2697                         goto out;
2698                 r = 0;
2699                 break;
2700         }
2701         case KVM_NMI: {
2702                 r = kvm_vcpu_ioctl_nmi(vcpu);
2703                 if (r)
2704                         goto out;
2705                 r = 0;
2706                 break;
2707         }
2708         case KVM_SET_CPUID: {
2709                 struct kvm_cpuid __user *cpuid_arg = argp;
2710                 struct kvm_cpuid cpuid;
2711
2712                 r = -EFAULT;
2713                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2714                         goto out;
2715                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2716                 if (r)
2717                         goto out;
2718                 break;
2719         }
2720         case KVM_SET_CPUID2: {
2721                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2722                 struct kvm_cpuid2 cpuid;
2723
2724                 r = -EFAULT;
2725                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2726                         goto out;
2727                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2728                                               cpuid_arg->entries);
2729                 if (r)
2730                         goto out;
2731                 break;
2732         }
2733         case KVM_GET_CPUID2: {
2734                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2735                 struct kvm_cpuid2 cpuid;
2736
2737                 r = -EFAULT;
2738                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2739                         goto out;
2740                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2741                                               cpuid_arg->entries);
2742                 if (r)
2743                         goto out;
2744                 r = -EFAULT;
2745                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2746                         goto out;
2747                 r = 0;
2748                 break;
2749         }
2750         case KVM_GET_MSRS:
2751                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2752                 break;
2753         case KVM_SET_MSRS:
2754                 r = msr_io(vcpu, argp, do_set_msr, 0);
2755                 break;
2756         case KVM_TPR_ACCESS_REPORTING: {
2757                 struct kvm_tpr_access_ctl tac;
2758
2759                 r = -EFAULT;
2760                 if (copy_from_user(&tac, argp, sizeof tac))
2761                         goto out;
2762                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2763                 if (r)
2764                         goto out;
2765                 r = -EFAULT;
2766                 if (copy_to_user(argp, &tac, sizeof tac))
2767                         goto out;
2768                 r = 0;
2769                 break;
2770         };
2771         case KVM_SET_VAPIC_ADDR: {
2772                 struct kvm_vapic_addr va;
2773
2774                 r = -EINVAL;
2775                 if (!irqchip_in_kernel(vcpu->kvm))
2776                         goto out;
2777                 r = -EFAULT;
2778                 if (copy_from_user(&va, argp, sizeof va))
2779                         goto out;
2780                 r = 0;
2781                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2782                 break;
2783         }
2784         case KVM_X86_SETUP_MCE: {
2785                 u64 mcg_cap;
2786
2787                 r = -EFAULT;
2788                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2789                         goto out;
2790                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2791                 break;
2792         }
2793         case KVM_X86_SET_MCE: {
2794                 struct kvm_x86_mce mce;
2795
2796                 r = -EFAULT;
2797                 if (copy_from_user(&mce, argp, sizeof mce))
2798                         goto out;
2799                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2800                 break;
2801         }
2802         case KVM_GET_VCPU_EVENTS: {
2803                 struct kvm_vcpu_events events;
2804
2805                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2806
2807                 r = -EFAULT;
2808                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2809                         break;
2810                 r = 0;
2811                 break;
2812         }
2813         case KVM_SET_VCPU_EVENTS: {
2814                 struct kvm_vcpu_events events;
2815
2816                 r = -EFAULT;
2817                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2818                         break;
2819
2820                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2821                 break;
2822         }
2823         case KVM_GET_DEBUGREGS: {
2824                 struct kvm_debugregs dbgregs;
2825
2826                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2827
2828                 r = -EFAULT;
2829                 if (copy_to_user(argp, &dbgregs,
2830                                  sizeof(struct kvm_debugregs)))
2831                         break;
2832                 r = 0;
2833                 break;
2834         }
2835         case KVM_SET_DEBUGREGS: {
2836                 struct kvm_debugregs dbgregs;
2837
2838                 r = -EFAULT;
2839                 if (copy_from_user(&dbgregs, argp,
2840                                    sizeof(struct kvm_debugregs)))
2841                         break;
2842
2843                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2844                 break;
2845         }
2846         case KVM_GET_XSAVE: {
2847                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2848                 r = -ENOMEM;
2849                 if (!u.xsave)
2850                         break;
2851
2852                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2853
2854                 r = -EFAULT;
2855                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2856                         break;
2857                 r = 0;
2858                 break;
2859         }
2860         case KVM_SET_XSAVE: {
2861                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2862                 r = -ENOMEM;
2863                 if (!u.xsave)
2864                         break;
2865
2866                 r = -EFAULT;
2867                 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
2868                         break;
2869
2870                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2871                 break;
2872         }
2873         case KVM_GET_XCRS: {
2874                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2875                 r = -ENOMEM;
2876                 if (!u.xcrs)
2877                         break;
2878
2879                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2880
2881                 r = -EFAULT;
2882                 if (copy_to_user(argp, u.xcrs,
2883                                  sizeof(struct kvm_xcrs)))
2884                         break;
2885                 r = 0;
2886                 break;
2887         }
2888         case KVM_SET_XCRS: {
2889                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2890                 r = -ENOMEM;
2891                 if (!u.xcrs)
2892                         break;
2893
2894                 r = -EFAULT;
2895                 if (copy_from_user(u.xcrs, argp,
2896                                    sizeof(struct kvm_xcrs)))
2897                         break;
2898
2899                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2900                 break;
2901         }
2902         default:
2903                 r = -EINVAL;
2904         }
2905 out:
2906         kfree(u.buffer);
2907         return r;
2908 }
2909
2910 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2911 {
2912         int ret;
2913
2914         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2915                 return -1;
2916         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2917         return ret;
2918 }
2919
2920 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2921                                               u64 ident_addr)
2922 {
2923         kvm->arch.ept_identity_map_addr = ident_addr;
2924         return 0;
2925 }
2926
2927 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2928                                           u32 kvm_nr_mmu_pages)
2929 {
2930         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2931                 return -EINVAL;
2932
2933         mutex_lock(&kvm->slots_lock);
2934         spin_lock(&kvm->mmu_lock);
2935
2936         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2937         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2938
2939         spin_unlock(&kvm->mmu_lock);
2940         mutex_unlock(&kvm->slots_lock);
2941         return 0;
2942 }
2943
2944 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2945 {
2946         return kvm->arch.n_max_mmu_pages;
2947 }
2948
2949 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2950 {
2951         int r;
2952
2953         r = 0;
2954         switch (chip->chip_id) {
2955         case KVM_IRQCHIP_PIC_MASTER:
2956                 memcpy(&chip->chip.pic,
2957                         &pic_irqchip(kvm)->pics[0],
2958                         sizeof(struct kvm_pic_state));
2959                 break;
2960         case KVM_IRQCHIP_PIC_SLAVE:
2961                 memcpy(&chip->chip.pic,
2962                         &pic_irqchip(kvm)->pics[1],
2963                         sizeof(struct kvm_pic_state));
2964                 break;
2965         case KVM_IRQCHIP_IOAPIC:
2966                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2967                 break;
2968         default:
2969                 r = -EINVAL;
2970                 break;
2971         }
2972         return r;
2973 }
2974
2975 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2976 {
2977         int r;
2978
2979         r = 0;
2980         switch (chip->chip_id) {
2981         case KVM_IRQCHIP_PIC_MASTER:
2982                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2983                 memcpy(&pic_irqchip(kvm)->pics[0],
2984                         &chip->chip.pic,
2985                         sizeof(struct kvm_pic_state));
2986                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2987                 break;
2988         case KVM_IRQCHIP_PIC_SLAVE:
2989                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2990                 memcpy(&pic_irqchip(kvm)->pics[1],
2991                         &chip->chip.pic,
2992                         sizeof(struct kvm_pic_state));
2993                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2994                 break;
2995         case KVM_IRQCHIP_IOAPIC:
2996                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2997                 break;
2998         default:
2999                 r = -EINVAL;
3000                 break;
3001         }
3002         kvm_pic_update_irq(pic_irqchip(kvm));
3003         return r;
3004 }
3005
3006 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3007 {
3008         int r = 0;
3009
3010         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3011         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3012         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3013         return r;
3014 }
3015
3016 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3017 {
3018         int r = 0;
3019
3020         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3021         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3022         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3023         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3024         return r;
3025 }
3026
3027 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3028 {
3029         int r = 0;
3030
3031         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3032         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3033                 sizeof(ps->channels));
3034         ps->flags = kvm->arch.vpit->pit_state.flags;
3035         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3036         return r;
3037 }
3038
3039 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3040 {
3041         int r = 0, start = 0;
3042         u32 prev_legacy, cur_legacy;
3043         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3044         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3045         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3046         if (!prev_legacy && cur_legacy)
3047                 start = 1;
3048         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3049                sizeof(kvm->arch.vpit->pit_state.channels));
3050         kvm->arch.vpit->pit_state.flags = ps->flags;
3051         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3052         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3053         return r;
3054 }
3055
3056 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3057                                  struct kvm_reinject_control *control)
3058 {
3059         if (!kvm->arch.vpit)
3060                 return -ENXIO;
3061         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3062         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3063         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3064         return 0;
3065 }
3066
3067 /*
3068  * Get (and clear) the dirty memory log for a memory slot.
3069  */
3070 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3071                                       struct kvm_dirty_log *log)
3072 {
3073         int r, i;
3074         struct kvm_memory_slot *memslot;
3075         unsigned long n;
3076         unsigned long is_dirty = 0;
3077
3078         mutex_lock(&kvm->slots_lock);
3079
3080         r = -EINVAL;
3081         if (log->slot >= KVM_MEMORY_SLOTS)
3082                 goto out;
3083
3084         memslot = &kvm->memslots->memslots[log->slot];
3085         r = -ENOENT;
3086         if (!memslot->dirty_bitmap)
3087                 goto out;
3088
3089         n = kvm_dirty_bitmap_bytes(memslot);
3090
3091         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3092                 is_dirty = memslot->dirty_bitmap[i];
3093
3094         /* If nothing is dirty, don't bother messing with page tables. */
3095         if (is_dirty) {
3096                 struct kvm_memslots *slots, *old_slots;
3097                 unsigned long *dirty_bitmap;
3098
3099                 spin_lock(&kvm->mmu_lock);
3100                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3101                 spin_unlock(&kvm->mmu_lock);
3102
3103                 r = -ENOMEM;
3104                 dirty_bitmap = vmalloc(n);
3105                 if (!dirty_bitmap)
3106                         goto out;
3107                 memset(dirty_bitmap, 0, n);
3108
3109                 r = -ENOMEM;
3110                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3111                 if (!slots) {
3112                         vfree(dirty_bitmap);
3113                         goto out;
3114                 }
3115                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3116                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3117
3118                 old_slots = kvm->memslots;
3119                 rcu_assign_pointer(kvm->memslots, slots);
3120                 synchronize_srcu_expedited(&kvm->srcu);
3121                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3122                 kfree(old_slots);
3123
3124                 r = -EFAULT;
3125                 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n)) {
3126                         vfree(dirty_bitmap);
3127                         goto out;
3128                 }
3129                 vfree(dirty_bitmap);
3130         } else {
3131                 r = -EFAULT;
3132                 if (clear_user(log->dirty_bitmap, n))
3133                         goto out;
3134         }
3135
3136         r = 0;
3137 out:
3138         mutex_unlock(&kvm->slots_lock);
3139         return r;
3140 }
3141
3142 long kvm_arch_vm_ioctl(struct file *filp,
3143                        unsigned int ioctl, unsigned long arg)
3144 {
3145         struct kvm *kvm = filp->private_data;
3146         void __user *argp = (void __user *)arg;
3147         int r = -ENOTTY;
3148         /*
3149          * This union makes it completely explicit to gcc-3.x
3150          * that these two variables' stack usage should be
3151          * combined, not added together.
3152          */
3153         union {
3154                 struct kvm_pit_state ps;
3155                 struct kvm_pit_state2 ps2;
3156                 struct kvm_pit_config pit_config;
3157         } u;
3158
3159         switch (ioctl) {
3160         case KVM_SET_TSS_ADDR:
3161                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3162                 if (r < 0)
3163                         goto out;
3164                 break;
3165         case KVM_SET_IDENTITY_MAP_ADDR: {
3166                 u64 ident_addr;
3167
3168                 r = -EFAULT;
3169                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3170                         goto out;
3171                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3172                 if (r < 0)
3173                         goto out;
3174                 break;
3175         }
3176         case KVM_SET_NR_MMU_PAGES:
3177                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3178                 if (r)
3179                         goto out;
3180                 break;
3181         case KVM_GET_NR_MMU_PAGES:
3182                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3183                 break;
3184         case KVM_CREATE_IRQCHIP: {
3185                 struct kvm_pic *vpic;
3186
3187                 mutex_lock(&kvm->lock);
3188                 r = -EEXIST;
3189                 if (kvm->arch.vpic)
3190                         goto create_irqchip_unlock;
3191                 r = -ENOMEM;
3192                 vpic = kvm_create_pic(kvm);
3193                 if (vpic) {
3194                         r = kvm_ioapic_init(kvm);
3195                         if (r) {
3196                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3197                                                           &vpic->dev);
3198                                 kfree(vpic);
3199                                 goto create_irqchip_unlock;
3200                         }
3201                 } else
3202                         goto create_irqchip_unlock;
3203                 smp_wmb();
3204                 kvm->arch.vpic = vpic;
3205                 smp_wmb();
3206                 r = kvm_setup_default_irq_routing(kvm);
3207                 if (r) {
3208                         mutex_lock(&kvm->irq_lock);
3209                         kvm_ioapic_destroy(kvm);
3210                         kvm_destroy_pic(kvm);
3211                         mutex_unlock(&kvm->irq_lock);
3212                 }
3213         create_irqchip_unlock:
3214                 mutex_unlock(&kvm->lock);
3215                 break;
3216         }
3217         case KVM_CREATE_PIT:
3218                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3219                 goto create_pit;
3220         case KVM_CREATE_PIT2:
3221                 r = -EFAULT;
3222                 if (copy_from_user(&u.pit_config, argp,
3223                                    sizeof(struct kvm_pit_config)))
3224                         goto out;
3225         create_pit:
3226                 mutex_lock(&kvm->slots_lock);
3227                 r = -EEXIST;
3228                 if (kvm->arch.vpit)
3229                         goto create_pit_unlock;
3230                 r = -ENOMEM;
3231                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3232                 if (kvm->arch.vpit)
3233                         r = 0;
3234         create_pit_unlock:
3235                 mutex_unlock(&kvm->slots_lock);
3236                 break;
3237         case KVM_IRQ_LINE_STATUS:
3238         case KVM_IRQ_LINE: {
3239                 struct kvm_irq_level irq_event;
3240
3241                 r = -EFAULT;
3242                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3243                         goto out;
3244                 r = -ENXIO;
3245                 if (irqchip_in_kernel(kvm)) {
3246                         __s32 status;
3247                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3248                                         irq_event.irq, irq_event.level);
3249                         if (ioctl == KVM_IRQ_LINE_STATUS) {
3250                                 r = -EFAULT;
3251                                 irq_event.status = status;
3252                                 if (copy_to_user(argp, &irq_event,
3253                                                         sizeof irq_event))
3254                                         goto out;
3255                         }
3256                         r = 0;
3257                 }
3258                 break;
3259         }
3260         case KVM_GET_IRQCHIP: {
3261                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3262                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3263
3264                 r = -ENOMEM;
3265                 if (!chip)
3266                         goto out;
3267                 r = -EFAULT;
3268                 if (copy_from_user(chip, argp, sizeof *chip))
3269                         goto get_irqchip_out;
3270                 r = -ENXIO;
3271                 if (!irqchip_in_kernel(kvm))
3272                         goto get_irqchip_out;
3273                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3274                 if (r)
3275                         goto get_irqchip_out;
3276                 r = -EFAULT;
3277                 if (copy_to_user(argp, chip, sizeof *chip))
3278                         goto get_irqchip_out;
3279                 r = 0;
3280         get_irqchip_out:
3281                 kfree(chip);
3282                 if (r)
3283                         goto out;
3284                 break;
3285         }
3286         case KVM_SET_IRQCHIP: {
3287                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3288                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3289
3290                 r = -ENOMEM;
3291                 if (!chip)
3292                         goto out;
3293                 r = -EFAULT;
3294                 if (copy_from_user(chip, argp, sizeof *chip))
3295                         goto set_irqchip_out;
3296                 r = -ENXIO;
3297                 if (!irqchip_in_kernel(kvm))
3298                         goto set_irqchip_out;
3299                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3300                 if (r)
3301                         goto set_irqchip_out;
3302                 r = 0;
3303         set_irqchip_out:
3304                 kfree(chip);
3305                 if (r)
3306                         goto out;
3307                 break;
3308         }
3309         case KVM_GET_PIT: {
3310                 r = -EFAULT;
3311                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3312                         goto out;
3313                 r = -ENXIO;
3314                 if (!kvm->arch.vpit)
3315                         goto out;
3316                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3317                 if (r)
3318                         goto out;
3319                 r = -EFAULT;
3320                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3321                         goto out;
3322                 r = 0;
3323                 break;
3324         }
3325         case KVM_SET_PIT: {
3326                 r = -EFAULT;
3327                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3328                         goto out;
3329                 r = -ENXIO;
3330                 if (!kvm->arch.vpit)
3331                         goto out;
3332                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3333                 if (r)
3334                         goto out;
3335                 r = 0;
3336                 break;
3337         }
3338         case KVM_GET_PIT2: {
3339                 r = -ENXIO;
3340                 if (!kvm->arch.vpit)
3341                         goto out;
3342                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3343                 if (r)
3344                         goto out;
3345                 r = -EFAULT;
3346                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3347                         goto out;
3348                 r = 0;
3349                 break;
3350         }
3351         case KVM_SET_PIT2: {
3352                 r = -EFAULT;
3353                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3354                         goto out;
3355                 r = -ENXIO;
3356                 if (!kvm->arch.vpit)
3357                         goto out;
3358                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3359                 if (r)
3360                         goto out;
3361                 r = 0;
3362                 break;
3363         }
3364         case KVM_REINJECT_CONTROL: {
3365                 struct kvm_reinject_control control;
3366                 r =  -EFAULT;
3367                 if (copy_from_user(&control, argp, sizeof(control)))
3368                         goto out;
3369                 r = kvm_vm_ioctl_reinject(kvm, &control);
3370                 if (r)
3371                         goto out;
3372                 r = 0;
3373                 break;
3374         }
3375         case KVM_XEN_HVM_CONFIG: {
3376                 r = -EFAULT;
3377                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3378                                    sizeof(struct kvm_xen_hvm_config)))
3379                         goto out;
3380                 r = -EINVAL;
3381                 if (kvm->arch.xen_hvm_config.flags)
3382                         goto out;
3383                 r = 0;
3384                 break;
3385         }
3386         case KVM_SET_CLOCK: {
3387                 struct kvm_clock_data user_ns;
3388                 u64 now_ns;
3389                 s64 delta;
3390
3391                 r = -EFAULT;
3392                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3393                         goto out;
3394
3395                 r = -EINVAL;
3396                 if (user_ns.flags)
3397                         goto out;
3398
3399                 r = 0;
3400                 now_ns = get_kernel_ns();
3401                 delta = user_ns.clock - now_ns;
3402                 kvm->arch.kvmclock_offset = delta;
3403                 break;
3404         }
3405         case KVM_GET_CLOCK: {
3406                 struct kvm_clock_data user_ns;
3407                 u64 now_ns;
3408
3409                 now_ns = get_kernel_ns();
3410                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3411                 user_ns.flags = 0;
3412
3413                 r = -EFAULT;
3414                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3415                         goto out;
3416                 r = 0;
3417                 break;
3418         }
3419
3420         default:
3421                 ;
3422         }
3423 out:
3424         return r;
3425 }
3426
3427 static void kvm_init_msr_list(void)
3428 {
3429         u32 dummy[2];
3430         unsigned i, j;
3431
3432         /* skip the first msrs in the list. KVM-specific */
3433         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3434                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3435                         continue;
3436                 if (j < i)
3437                         msrs_to_save[j] = msrs_to_save[i];
3438                 j++;
3439         }
3440         num_msrs_to_save = j;
3441 }
3442
3443 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3444                            const void *v)
3445 {
3446         if (vcpu->arch.apic &&
3447             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3448                 return 0;
3449
3450         return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3451 }
3452
3453 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3454 {
3455         if (vcpu->arch.apic &&
3456             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3457                 return 0;
3458
3459         return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3460 }
3461
3462 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3463                         struct kvm_segment *var, int seg)
3464 {
3465         kvm_x86_ops->set_segment(vcpu, var, seg);
3466 }
3467
3468 void kvm_get_segment(struct kvm_vcpu *vcpu,
3469                      struct kvm_segment *var, int seg)
3470 {
3471         kvm_x86_ops->get_segment(vcpu, var, seg);
3472 }
3473
3474 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3475 {
3476         return gpa;
3477 }
3478
3479 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3480 {
3481         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3482         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, error);
3483 }
3484
3485  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3486 {
3487         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3488         access |= PFERR_FETCH_MASK;
3489         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, error);
3490 }
3491
3492 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3493 {
3494         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3495         access |= PFERR_WRITE_MASK;
3496         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, error);
3497 }
3498
3499 /* uses this to access any guest's mapped memory without checking CPL */
3500 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3501 {
3502         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, error);
3503 }
3504
3505 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3506                                       struct kvm_vcpu *vcpu, u32 access,
3507                                       u32 *error)
3508 {
3509         void *data = val;
3510         int r = X86EMUL_CONTINUE;
3511
3512         while (bytes) {
3513                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3514                                                             error);
3515                 unsigned offset = addr & (PAGE_SIZE-1);
3516                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3517                 int ret;
3518
3519                 if (gpa == UNMAPPED_GVA) {
3520                         r = X86EMUL_PROPAGATE_FAULT;
3521                         goto out;
3522                 }
3523                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3524                 if (ret < 0) {
3525                         r = X86EMUL_IO_NEEDED;
3526                         goto out;
3527                 }
3528
3529                 bytes -= toread;
3530                 data += toread;
3531                 addr += toread;
3532         }
3533 out:
3534         return r;
3535 }
3536
3537 /* used for instruction fetching */
3538 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3539                                 struct kvm_vcpu *vcpu, u32 *error)
3540 {
3541         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3542         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3543                                           access | PFERR_FETCH_MASK, error);
3544 }
3545
3546 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3547                                struct kvm_vcpu *vcpu, u32 *error)
3548 {
3549         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3550         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3551                                           error);
3552 }
3553
3554 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3555                                struct kvm_vcpu *vcpu, u32 *error)
3556 {
3557         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3558 }
3559
3560 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3561                                        unsigned int bytes,
3562                                        struct kvm_vcpu *vcpu,
3563                                        u32 *error)
3564 {
3565         void *data = val;
3566         int r = X86EMUL_CONTINUE;
3567
3568         while (bytes) {
3569                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3570                                                              PFERR_WRITE_MASK,
3571                                                              error);
3572                 unsigned offset = addr & (PAGE_SIZE-1);
3573                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3574                 int ret;
3575
3576                 if (gpa == UNMAPPED_GVA) {
3577                         r = X86EMUL_PROPAGATE_FAULT;
3578                         goto out;
3579                 }
3580                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3581                 if (ret < 0) {
3582                         r = X86EMUL_IO_NEEDED;
3583                         goto out;
3584                 }
3585
3586                 bytes -= towrite;
3587                 data += towrite;
3588                 addr += towrite;
3589         }
3590 out:
3591         return r;
3592 }
3593
3594 static int emulator_read_emulated(unsigned long addr,
3595                                   void *val,
3596                                   unsigned int bytes,
3597                                   unsigned int *error_code,
3598                                   struct kvm_vcpu *vcpu)
3599 {
3600         gpa_t                 gpa;
3601
3602         if (vcpu->mmio_read_completed) {
3603                 memcpy(val, vcpu->mmio_data, bytes);
3604                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3605                                vcpu->mmio_phys_addr, *(u64 *)val);
3606                 vcpu->mmio_read_completed = 0;
3607                 return X86EMUL_CONTINUE;
3608         }
3609
3610         gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, error_code);
3611
3612         if (gpa == UNMAPPED_GVA)
3613                 return X86EMUL_PROPAGATE_FAULT;
3614
3615         /* For APIC access vmexit */
3616         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3617                 goto mmio;
3618
3619         if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3620                                 == X86EMUL_CONTINUE)
3621                 return X86EMUL_CONTINUE;
3622
3623 mmio:
3624         /*
3625          * Is this MMIO handled locally?
3626          */
3627         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3628                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3629                 return X86EMUL_CONTINUE;
3630         }
3631
3632         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3633
3634         vcpu->mmio_needed = 1;
3635         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3636         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3637         vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3638         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3639
3640         return X86EMUL_IO_NEEDED;
3641 }
3642
3643 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3644                           const void *val, int bytes)
3645 {
3646         int ret;
3647
3648         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3649         if (ret < 0)
3650                 return 0;
3651         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3652         return 1;
3653 }
3654
3655 static int emulator_write_emulated_onepage(unsigned long addr,
3656                                            const void *val,
3657                                            unsigned int bytes,
3658                                            unsigned int *error_code,
3659                                            struct kvm_vcpu *vcpu)
3660 {
3661         gpa_t                 gpa;
3662
3663         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error_code);
3664
3665         if (gpa == UNMAPPED_GVA)
3666                 return X86EMUL_PROPAGATE_FAULT;
3667
3668         /* For APIC access vmexit */
3669         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3670                 goto mmio;
3671
3672         if (emulator_write_phys(vcpu, gpa, val, bytes))
3673                 return X86EMUL_CONTINUE;
3674
3675 mmio:
3676         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3677         /*
3678          * Is this MMIO handled locally?
3679          */
3680         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3681                 return X86EMUL_CONTINUE;
3682
3683         vcpu->mmio_needed = 1;
3684         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3685         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3686         vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3687         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3688         memcpy(vcpu->run->mmio.data, val, bytes);
3689
3690         return X86EMUL_CONTINUE;
3691 }
3692
3693 int emulator_write_emulated(unsigned long addr,
3694                             const void *val,
3695                             unsigned int bytes,
3696                             unsigned int *error_code,
3697                             struct kvm_vcpu *vcpu)
3698 {
3699         /* Crossing a page boundary? */
3700         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3701                 int rc, now;
3702
3703                 now = -addr & ~PAGE_MASK;
3704                 rc = emulator_write_emulated_onepage(addr, val, now, error_code,
3705                                                      vcpu);
3706                 if (rc != X86EMUL_CONTINUE)
3707                         return rc;
3708                 addr += now;
3709                 val += now;
3710                 bytes -= now;
3711         }
3712         return emulator_write_emulated_onepage(addr, val, bytes, error_code,
3713                                                vcpu);
3714 }
3715
3716 #define CMPXCHG_TYPE(t, ptr, old, new) \
3717         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3718
3719 #ifdef CONFIG_X86_64
3720 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3721 #else
3722 #  define CMPXCHG64(ptr, old, new) \
3723         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3724 #endif
3725
3726 static int emulator_cmpxchg_emulated(unsigned long addr,
3727                                      const void *old,
3728                                      const void *new,
3729                                      unsigned int bytes,
3730                                      unsigned int *error_code,
3731                                      struct kvm_vcpu *vcpu)
3732 {
3733         gpa_t gpa;
3734         struct page *page;
3735         char *kaddr;
3736         bool exchanged;
3737
3738         /* guests cmpxchg8b have to be emulated atomically */
3739         if (bytes > 8 || (bytes & (bytes - 1)))
3740                 goto emul_write;
3741
3742         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3743
3744         if (gpa == UNMAPPED_GVA ||
3745             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3746                 goto emul_write;
3747
3748         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3749                 goto emul_write;
3750
3751         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3752         if (is_error_page(page)) {
3753                 kvm_release_page_clean(page);
3754                 goto emul_write;
3755         }
3756
3757         kaddr = kmap_atomic(page, KM_USER0);
3758         kaddr += offset_in_page(gpa);
3759         switch (bytes) {
3760         case 1:
3761                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3762                 break;
3763         case 2:
3764                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3765                 break;
3766         case 4:
3767                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3768                 break;
3769         case 8:
3770                 exchanged = CMPXCHG64(kaddr, old, new);
3771                 break;
3772         default:
3773                 BUG();
3774         }
3775         kunmap_atomic(kaddr, KM_USER0);
3776         kvm_release_page_dirty(page);
3777
3778         if (!exchanged)
3779                 return X86EMUL_CMPXCHG_FAILED;
3780
3781         kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3782
3783         return X86EMUL_CONTINUE;
3784
3785 emul_write:
3786         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3787
3788         return emulator_write_emulated(addr, new, bytes, error_code, vcpu);
3789 }
3790
3791 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3792 {
3793         /* TODO: String I/O for in kernel device */
3794         int r;
3795
3796         if (vcpu->arch.pio.in)
3797                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3798                                     vcpu->arch.pio.size, pd);
3799         else
3800                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3801                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
3802                                      pd);
3803         return r;
3804 }
3805
3806
3807 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
3808                              unsigned int count, struct kvm_vcpu *vcpu)
3809 {
3810         if (vcpu->arch.pio.count)
3811                 goto data_avail;
3812
3813         trace_kvm_pio(0, port, size, 1);
3814
3815         vcpu->arch.pio.port = port;
3816         vcpu->arch.pio.in = 1;
3817         vcpu->arch.pio.count  = count;
3818         vcpu->arch.pio.size = size;
3819
3820         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3821         data_avail:
3822                 memcpy(val, vcpu->arch.pio_data, size * count);
3823                 vcpu->arch.pio.count = 0;
3824                 return 1;
3825         }
3826
3827         vcpu->run->exit_reason = KVM_EXIT_IO;
3828         vcpu->run->io.direction = KVM_EXIT_IO_IN;
3829         vcpu->run->io.size = size;
3830         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3831         vcpu->run->io.count = count;
3832         vcpu->run->io.port = port;
3833
3834         return 0;
3835 }
3836
3837 static int emulator_pio_out_emulated(int size, unsigned short port,
3838                               const void *val, unsigned int count,
3839                               struct kvm_vcpu *vcpu)
3840 {
3841         trace_kvm_pio(1, port, size, 1);
3842
3843         vcpu->arch.pio.port = port;
3844         vcpu->arch.pio.in = 0;
3845         vcpu->arch.pio.count = count;
3846         vcpu->arch.pio.size = size;
3847
3848         memcpy(vcpu->arch.pio_data, val, size * count);
3849
3850         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3851                 vcpu->arch.pio.count = 0;
3852                 return 1;
3853         }
3854
3855         vcpu->run->exit_reason = KVM_EXIT_IO;
3856         vcpu->run->io.direction = KVM_EXIT_IO_OUT;
3857         vcpu->run->io.size = size;
3858         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3859         vcpu->run->io.count = count;
3860         vcpu->run->io.port = port;
3861
3862         return 0;
3863 }
3864
3865 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3866 {
3867         return kvm_x86_ops->get_segment_base(vcpu, seg);
3868 }
3869
3870 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3871 {
3872         kvm_mmu_invlpg(vcpu, address);
3873         return X86EMUL_CONTINUE;
3874 }
3875
3876 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
3877 {
3878         if (!need_emulate_wbinvd(vcpu))
3879                 return X86EMUL_CONTINUE;
3880
3881         if (kvm_x86_ops->has_wbinvd_exit()) {
3882                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
3883                                 wbinvd_ipi, NULL, 1);
3884                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
3885         }
3886         wbinvd();
3887         return X86EMUL_CONTINUE;
3888 }
3889 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
3890
3891 int emulate_clts(struct kvm_vcpu *vcpu)
3892 {
3893         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3894         kvm_x86_ops->fpu_activate(vcpu);
3895         return X86EMUL_CONTINUE;
3896 }
3897
3898 int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu)
3899 {
3900         return _kvm_get_dr(vcpu, dr, dest);
3901 }
3902
3903 int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu)
3904 {
3905
3906         return __kvm_set_dr(vcpu, dr, value);
3907 }
3908
3909 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3910 {
3911         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3912 }
3913
3914 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
3915 {
3916         unsigned long value;
3917
3918         switch (cr) {
3919         case 0:
3920                 value = kvm_read_cr0(vcpu);
3921                 break;
3922         case 2:
3923                 value = vcpu->arch.cr2;
3924                 break;
3925         case 3:
3926                 value = vcpu->arch.cr3;
3927                 break;
3928         case 4:
3929                 value = kvm_read_cr4(vcpu);
3930                 break;
3931         case 8:
3932                 value = kvm_get_cr8(vcpu);
3933                 break;
3934         default:
3935                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3936                 return 0;
3937         }
3938
3939         return value;
3940 }
3941
3942 static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
3943 {
3944         int res = 0;
3945
3946         switch (cr) {
3947         case 0:
3948                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
3949                 break;
3950         case 2:
3951                 vcpu->arch.cr2 = val;
3952                 break;
3953         case 3:
3954                 res = kvm_set_cr3(vcpu, val);
3955                 break;
3956         case 4:
3957                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
3958                 break;
3959         case 8:
3960                 res = __kvm_set_cr8(vcpu, val & 0xfUL);
3961                 break;
3962         default:
3963                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3964                 res = -1;
3965         }
3966
3967         return res;
3968 }
3969
3970 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
3971 {
3972         return kvm_x86_ops->get_cpl(vcpu);
3973 }
3974
3975 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
3976 {
3977         kvm_x86_ops->get_gdt(vcpu, dt);
3978 }
3979
3980 static void emulator_get_idt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
3981 {
3982         kvm_x86_ops->get_idt(vcpu, dt);
3983 }
3984
3985 static unsigned long emulator_get_cached_segment_base(int seg,
3986                                                       struct kvm_vcpu *vcpu)
3987 {
3988         return get_segment_base(vcpu, seg);
3989 }
3990
3991 static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg,
3992                                            struct kvm_vcpu *vcpu)
3993 {
3994         struct kvm_segment var;
3995
3996         kvm_get_segment(vcpu, &var, seg);
3997
3998         if (var.unusable)
3999                 return false;
4000
4001         if (var.g)
4002                 var.limit >>= 12;
4003         set_desc_limit(desc, var.limit);
4004         set_desc_base(desc, (unsigned long)var.base);
4005         desc->type = var.type;
4006         desc->s = var.s;
4007         desc->dpl = var.dpl;
4008         desc->p = var.present;
4009         desc->avl = var.avl;
4010         desc->l = var.l;
4011         desc->d = var.db;
4012         desc->g = var.g;
4013
4014         return true;
4015 }
4016
4017 static void emulator_set_cached_descriptor(struct desc_struct *desc, int seg,
4018                                            struct kvm_vcpu *vcpu)
4019 {
4020         struct kvm_segment var;
4021
4022         /* needed to preserve selector */
4023         kvm_get_segment(vcpu, &var, seg);
4024
4025         var.base = get_desc_base(desc);
4026         var.limit = get_desc_limit(desc);
4027         if (desc->g)
4028                 var.limit = (var.limit << 12) | 0xfff;
4029         var.type = desc->type;
4030         var.present = desc->p;
4031         var.dpl = desc->dpl;
4032         var.db = desc->d;
4033         var.s = desc->s;
4034         var.l = desc->l;
4035         var.g = desc->g;
4036         var.avl = desc->avl;
4037         var.present = desc->p;
4038         var.unusable = !var.present;
4039         var.padding = 0;
4040
4041         kvm_set_segment(vcpu, &var, seg);
4042         return;
4043 }
4044
4045 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
4046 {
4047         struct kvm_segment kvm_seg;
4048
4049         kvm_get_segment(vcpu, &kvm_seg, seg);
4050         return kvm_seg.selector;
4051 }
4052
4053 static void emulator_set_segment_selector(u16 sel, int seg,
4054                                           struct kvm_vcpu *vcpu)
4055 {
4056         struct kvm_segment kvm_seg;
4057
4058         kvm_get_segment(vcpu, &kvm_seg, seg);
4059         kvm_seg.selector = sel;
4060         kvm_set_segment(vcpu, &kvm_seg, seg);
4061 }
4062
4063 static struct x86_emulate_ops emulate_ops = {
4064         .read_std            = kvm_read_guest_virt_system,
4065         .write_std           = kvm_write_guest_virt_system,
4066         .fetch               = kvm_fetch_guest_virt,
4067         .read_emulated       = emulator_read_emulated,
4068         .write_emulated      = emulator_write_emulated,
4069         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
4070         .pio_in_emulated     = emulator_pio_in_emulated,
4071         .pio_out_emulated    = emulator_pio_out_emulated,
4072         .get_cached_descriptor = emulator_get_cached_descriptor,
4073         .set_cached_descriptor = emulator_set_cached_descriptor,
4074         .get_segment_selector = emulator_get_segment_selector,
4075         .set_segment_selector = emulator_set_segment_selector,
4076         .get_cached_segment_base = emulator_get_cached_segment_base,
4077         .get_gdt             = emulator_get_gdt,
4078         .get_idt             = emulator_get_idt,
4079         .get_cr              = emulator_get_cr,
4080         .set_cr              = emulator_set_cr,
4081         .cpl                 = emulator_get_cpl,
4082         .get_dr              = emulator_get_dr,
4083         .set_dr              = emulator_set_dr,
4084         .set_msr             = kvm_set_msr,
4085         .get_msr             = kvm_get_msr,
4086 };
4087
4088 static void cache_all_regs(struct kvm_vcpu *vcpu)
4089 {
4090         kvm_register_read(vcpu, VCPU_REGS_RAX);
4091         kvm_register_read(vcpu, VCPU_REGS_RSP);
4092         kvm_register_read(vcpu, VCPU_REGS_RIP);
4093         vcpu->arch.regs_dirty = ~0;
4094 }
4095
4096 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4097 {
4098         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4099         /*
4100          * an sti; sti; sequence only disable interrupts for the first
4101          * instruction. So, if the last instruction, be it emulated or
4102          * not, left the system with the INT_STI flag enabled, it
4103          * means that the last instruction is an sti. We should not
4104          * leave the flag on in this case. The same goes for mov ss
4105          */
4106         if (!(int_shadow & mask))
4107                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4108 }
4109
4110 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4111 {
4112         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4113         if (ctxt->exception == PF_VECTOR)
4114                 kvm_inject_page_fault(vcpu);
4115         else if (ctxt->error_code_valid)
4116                 kvm_queue_exception_e(vcpu, ctxt->exception, ctxt->error_code);
4117         else
4118                 kvm_queue_exception(vcpu, ctxt->exception);
4119 }
4120
4121 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4122 {
4123         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4124         int cs_db, cs_l;
4125
4126         cache_all_regs(vcpu);
4127
4128         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4129
4130         vcpu->arch.emulate_ctxt.vcpu = vcpu;
4131         vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
4132         vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4133         vcpu->arch.emulate_ctxt.mode =
4134                 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4135                 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4136                 ? X86EMUL_MODE_VM86 : cs_l
4137                 ? X86EMUL_MODE_PROT64 : cs_db
4138                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4139         memset(c, 0, sizeof(struct decode_cache));
4140         memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4141 }
4142
4143 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4144 {
4145         ++vcpu->stat.insn_emulation_fail;
4146         trace_kvm_emulate_insn_failed(vcpu);
4147         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4148         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4149         vcpu->run->internal.ndata = 0;
4150         kvm_queue_exception(vcpu, UD_VECTOR);
4151         return EMULATE_FAIL;
4152 }
4153
4154 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4155 {
4156         gpa_t gpa;
4157
4158         if (tdp_enabled)
4159                 return false;
4160
4161         /*
4162          * if emulation was due to access to shadowed page table
4163          * and it failed try to unshadow page and re-entetr the
4164          * guest to let CPU execute the instruction.
4165          */
4166         if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4167                 return true;
4168
4169         gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4170
4171         if (gpa == UNMAPPED_GVA)
4172                 return true; /* let cpu generate fault */
4173
4174         if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4175                 return true;
4176
4177         return false;
4178 }
4179
4180 int emulate_instruction(struct kvm_vcpu *vcpu,
4181                         unsigned long cr2,
4182                         u16 error_code,
4183                         int emulation_type)
4184 {
4185         int r;
4186         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4187
4188         kvm_clear_exception_queue(vcpu);
4189         vcpu->arch.mmio_fault_cr2 = cr2;
4190         /*
4191          * TODO: fix emulate.c to use guest_read/write_register
4192          * instead of direct ->regs accesses, can save hundred cycles
4193          * on Intel for instructions that don't read/change RSP, for
4194          * for example.
4195          */
4196         cache_all_regs(vcpu);
4197
4198         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4199                 init_emulate_ctxt(vcpu);
4200                 vcpu->arch.emulate_ctxt.interruptibility = 0;
4201                 vcpu->arch.emulate_ctxt.exception = -1;
4202                 vcpu->arch.emulate_ctxt.perm_ok = false;
4203
4204                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt);
4205                 trace_kvm_emulate_insn_start(vcpu);
4206
4207                 /* Only allow emulation of specific instructions on #UD
4208                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
4209                 if (emulation_type & EMULTYPE_TRAP_UD) {
4210                         if (!c->twobyte)
4211                                 return EMULATE_FAIL;
4212                         switch (c->b) {
4213                         case 0x01: /* VMMCALL */
4214                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
4215                                         return EMULATE_FAIL;
4216                                 break;
4217                         case 0x34: /* sysenter */
4218                         case 0x35: /* sysexit */
4219                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4220                                         return EMULATE_FAIL;
4221                                 break;
4222                         case 0x05: /* syscall */
4223                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4224                                         return EMULATE_FAIL;
4225                                 break;
4226                         default:
4227                                 return EMULATE_FAIL;
4228                         }
4229
4230                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
4231                                 return EMULATE_FAIL;
4232                 }
4233
4234                 ++vcpu->stat.insn_emulation;
4235                 if (r)  {
4236                         if (reexecute_instruction(vcpu, cr2))
4237                                 return EMULATE_DONE;
4238                         if (emulation_type & EMULTYPE_SKIP)
4239                                 return EMULATE_FAIL;
4240                         return handle_emulation_failure(vcpu);
4241                 }
4242         }
4243
4244         if (emulation_type & EMULTYPE_SKIP) {
4245                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4246                 return EMULATE_DONE;
4247         }
4248
4249         /* this is needed for vmware backdor interface to work since it
4250            changes registers values  during IO operation */
4251         memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4252
4253 restart:
4254         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4255
4256         if (r == EMULATION_FAILED) {
4257                 if (reexecute_instruction(vcpu, cr2))
4258                         return EMULATE_DONE;
4259
4260                 return handle_emulation_failure(vcpu);
4261         }
4262
4263         if (vcpu->arch.emulate_ctxt.exception >= 0) {
4264                 inject_emulated_exception(vcpu);
4265                 r = EMULATE_DONE;
4266         } else if (vcpu->arch.pio.count) {
4267                 if (!vcpu->arch.pio.in)
4268                         vcpu->arch.pio.count = 0;
4269                 r = EMULATE_DO_MMIO;
4270         } else if (vcpu->mmio_needed) {
4271                 if (vcpu->mmio_is_write)
4272                         vcpu->mmio_needed = 0;
4273                 r = EMULATE_DO_MMIO;
4274         } else if (r == EMULATION_RESTART)
4275                 goto restart;
4276         else
4277                 r = EMULATE_DONE;
4278
4279         toggle_interruptibility(vcpu, vcpu->arch.emulate_ctxt.interruptibility);
4280         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4281         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4282         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4283
4284         return r;
4285 }
4286 EXPORT_SYMBOL_GPL(emulate_instruction);
4287
4288 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4289 {
4290         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4291         int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
4292         /* do not return to emulator after return from userspace */
4293         vcpu->arch.pio.count = 0;
4294         return ret;
4295 }
4296 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4297
4298 static void tsc_bad(void *info)
4299 {
4300         __get_cpu_var(cpu_tsc_khz) = 0;
4301 }
4302
4303 static void tsc_khz_changed(void *data)
4304 {
4305         struct cpufreq_freqs *freq = data;
4306         unsigned long khz = 0;
4307
4308         if (data)
4309                 khz = freq->new;
4310         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4311                 khz = cpufreq_quick_get(raw_smp_processor_id());
4312         if (!khz)
4313                 khz = tsc_khz;
4314         __get_cpu_var(cpu_tsc_khz) = khz;
4315 }
4316
4317 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4318                                      void *data)
4319 {
4320         struct cpufreq_freqs *freq = data;
4321         struct kvm *kvm;
4322         struct kvm_vcpu *vcpu;
4323         int i, send_ipi = 0;
4324
4325         /*
4326          * We allow guests to temporarily run on slowing clocks,
4327          * provided we notify them after, or to run on accelerating
4328          * clocks, provided we notify them before.  Thus time never
4329          * goes backwards.
4330          *
4331          * However, we have a problem.  We can't atomically update
4332          * the frequency of a given CPU from this function; it is
4333          * merely a notifier, which can be called from any CPU.
4334          * Changing the TSC frequency at arbitrary points in time
4335          * requires a recomputation of local variables related to
4336          * the TSC for each VCPU.  We must flag these local variables
4337          * to be updated and be sure the update takes place with the
4338          * new frequency before any guests proceed.
4339          *
4340          * Unfortunately, the combination of hotplug CPU and frequency
4341          * change creates an intractable locking scenario; the order
4342          * of when these callouts happen is undefined with respect to
4343          * CPU hotplug, and they can race with each other.  As such,
4344          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4345          * undefined; you can actually have a CPU frequency change take
4346          * place in between the computation of X and the setting of the
4347          * variable.  To protect against this problem, all updates of
4348          * the per_cpu tsc_khz variable are done in an interrupt
4349          * protected IPI, and all callers wishing to update the value
4350          * must wait for a synchronous IPI to complete (which is trivial
4351          * if the caller is on the CPU already).  This establishes the
4352          * necessary total order on variable updates.
4353          *
4354          * Note that because a guest time update may take place
4355          * anytime after the setting of the VCPU's request bit, the
4356          * correct TSC value must be set before the request.  However,
4357          * to ensure the update actually makes it to any guest which
4358          * starts running in hardware virtualization between the set
4359          * and the acquisition of the spinlock, we must also ping the
4360          * CPU after setting the request bit.
4361          *
4362          */
4363
4364         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4365                 return 0;
4366         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4367                 return 0;
4368
4369         smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4370
4371         spin_lock(&kvm_lock);
4372         list_for_each_entry(kvm, &vm_list, vm_list) {
4373                 kvm_for_each_vcpu(i, vcpu, kvm) {
4374                         if (vcpu->cpu != freq->cpu)
4375                                 continue;
4376                         if (!kvm_request_guest_time_update(vcpu))
4377                                 continue;
4378                         if (vcpu->cpu != smp_processor_id())
4379                                 send_ipi = 1;
4380                 }
4381         }
4382         spin_unlock(&kvm_lock);
4383
4384         if (freq->old < freq->new && send_ipi) {
4385                 /*
4386                  * We upscale the frequency.  Must make the guest
4387                  * doesn't see old kvmclock values while running with
4388                  * the new frequency, otherwise we risk the guest sees
4389                  * time go backwards.
4390                  *
4391                  * In case we update the frequency for another cpu
4392                  * (which might be in guest context) send an interrupt
4393                  * to kick the cpu out of guest context.  Next time
4394                  * guest context is entered kvmclock will be updated,
4395                  * so the guest will not see stale values.
4396                  */
4397                 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4398         }
4399         return 0;
4400 }
4401
4402 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4403         .notifier_call  = kvmclock_cpufreq_notifier
4404 };
4405
4406 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4407                                         unsigned long action, void *hcpu)
4408 {
4409         unsigned int cpu = (unsigned long)hcpu;
4410
4411         switch (action) {
4412                 case CPU_ONLINE:
4413                 case CPU_DOWN_FAILED:
4414                         smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4415                         break;
4416                 case CPU_DOWN_PREPARE:
4417                         smp_call_function_single(cpu, tsc_bad, NULL, 1);
4418                         break;
4419         }
4420         return NOTIFY_OK;
4421 }
4422
4423 static struct notifier_block kvmclock_cpu_notifier_block = {
4424         .notifier_call  = kvmclock_cpu_notifier,
4425         .priority = -INT_MAX
4426 };
4427
4428 static void kvm_timer_init(void)
4429 {
4430         int cpu;
4431
4432         register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4433         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4434                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4435                                           CPUFREQ_TRANSITION_NOTIFIER);
4436         }
4437         for_each_online_cpu(cpu)
4438                 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4439 }
4440
4441 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4442
4443 static int kvm_is_in_guest(void)
4444 {
4445         return percpu_read(current_vcpu) != NULL;
4446 }
4447
4448 static int kvm_is_user_mode(void)
4449 {
4450         int user_mode = 3;
4451
4452         if (percpu_read(current_vcpu))
4453                 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4454
4455         return user_mode != 0;
4456 }
4457
4458 static unsigned long kvm_get_guest_ip(void)
4459 {
4460         unsigned long ip = 0;
4461
4462         if (percpu_read(current_vcpu))
4463                 ip = kvm_rip_read(percpu_read(current_vcpu));
4464
4465         return ip;
4466 }
4467
4468 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4469         .is_in_guest            = kvm_is_in_guest,
4470         .is_user_mode           = kvm_is_user_mode,
4471         .get_guest_ip           = kvm_get_guest_ip,
4472 };
4473
4474 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4475 {
4476         percpu_write(current_vcpu, vcpu);
4477 }
4478 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4479
4480 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4481 {
4482         percpu_write(current_vcpu, NULL);
4483 }
4484 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4485
4486 int kvm_arch_init(void *opaque)
4487 {
4488         int r;
4489         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4490
4491         if (kvm_x86_ops) {
4492                 printk(KERN_ERR "kvm: already loaded the other module\n");
4493                 r = -EEXIST;
4494                 goto out;
4495         }
4496
4497         if (!ops->cpu_has_kvm_support()) {
4498                 printk(KERN_ERR "kvm: no hardware support\n");
4499                 r = -EOPNOTSUPP;
4500                 goto out;
4501         }
4502         if (ops->disabled_by_bios()) {
4503                 printk(KERN_ERR "kvm: disabled by bios\n");
4504                 r = -EOPNOTSUPP;
4505                 goto out;
4506         }
4507
4508         r = kvm_mmu_module_init();
4509         if (r)
4510                 goto out;
4511
4512         kvm_init_msr_list();
4513
4514         kvm_x86_ops = ops;
4515         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4516         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
4517         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4518                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
4519
4520         kvm_timer_init();
4521
4522         perf_register_guest_info_callbacks(&kvm_guest_cbs);
4523
4524         if (cpu_has_xsave)
4525                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4526
4527         return 0;
4528
4529 out:
4530         return r;
4531 }
4532
4533 void kvm_arch_exit(void)
4534 {
4535         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4536
4537         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4538                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4539                                             CPUFREQ_TRANSITION_NOTIFIER);
4540         unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4541         kvm_x86_ops = NULL;
4542         kvm_mmu_module_exit();
4543 }
4544
4545 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4546 {
4547         ++vcpu->stat.halt_exits;
4548         if (irqchip_in_kernel(vcpu->kvm)) {
4549                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4550                 return 1;
4551         } else {
4552                 vcpu->run->exit_reason = KVM_EXIT_HLT;
4553                 return 0;
4554         }
4555 }
4556 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4557
4558 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4559                            unsigned long a1)
4560 {
4561         if (is_long_mode(vcpu))
4562                 return a0;
4563         else
4564                 return a0 | ((gpa_t)a1 << 32);
4565 }
4566
4567 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4568 {
4569         u64 param, ingpa, outgpa, ret;
4570         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4571         bool fast, longmode;
4572         int cs_db, cs_l;
4573
4574         /*
4575          * hypercall generates UD from non zero cpl and real mode
4576          * per HYPER-V spec
4577          */
4578         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4579                 kvm_queue_exception(vcpu, UD_VECTOR);
4580                 return 0;
4581         }
4582
4583         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4584         longmode = is_long_mode(vcpu) && cs_l == 1;
4585
4586         if (!longmode) {
4587                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4588                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4589                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4590                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4591                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4592                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4593         }
4594 #ifdef CONFIG_X86_64
4595         else {
4596                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4597                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4598                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4599         }
4600 #endif
4601
4602         code = param & 0xffff;
4603         fast = (param >> 16) & 0x1;
4604         rep_cnt = (param >> 32) & 0xfff;
4605         rep_idx = (param >> 48) & 0xfff;
4606
4607         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4608
4609         switch (code) {
4610         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4611                 kvm_vcpu_on_spin(vcpu);
4612                 break;
4613         default:
4614                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4615                 break;
4616         }
4617
4618         ret = res | (((u64)rep_done & 0xfff) << 32);
4619         if (longmode) {
4620                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4621         } else {
4622                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4623                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4624         }
4625
4626         return 1;
4627 }
4628
4629 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4630 {
4631         unsigned long nr, a0, a1, a2, a3, ret;
4632         int r = 1;
4633
4634         if (kvm_hv_hypercall_enabled(vcpu->kvm))
4635                 return kvm_hv_hypercall(vcpu);
4636
4637         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4638         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4639         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4640         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4641         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4642
4643         trace_kvm_hypercall(nr, a0, a1, a2, a3);
4644
4645         if (!is_long_mode(vcpu)) {
4646                 nr &= 0xFFFFFFFF;
4647                 a0 &= 0xFFFFFFFF;
4648                 a1 &= 0xFFFFFFFF;
4649                 a2 &= 0xFFFFFFFF;
4650                 a3 &= 0xFFFFFFFF;
4651         }
4652
4653         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4654                 ret = -KVM_EPERM;
4655                 goto out;
4656         }
4657
4658         switch (nr) {
4659         case KVM_HC_VAPIC_POLL_IRQ:
4660                 ret = 0;
4661                 break;
4662         case KVM_HC_MMU_OP:
4663                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4664                 break;
4665         default:
4666                 ret = -KVM_ENOSYS;
4667                 break;
4668         }
4669 out:
4670         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4671         ++vcpu->stat.hypercalls;
4672         return r;
4673 }
4674 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4675
4676 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4677 {
4678         char instruction[3];
4679         unsigned long rip = kvm_rip_read(vcpu);
4680
4681         /*
4682          * Blow out the MMU to ensure that no other VCPU has an active mapping
4683          * to ensure that the updated hypercall appears atomically across all
4684          * VCPUs.
4685          */
4686         kvm_mmu_zap_all(vcpu->kvm);
4687
4688         kvm_x86_ops->patch_hypercall(vcpu, instruction);
4689
4690         return emulator_write_emulated(rip, instruction, 3, NULL, vcpu);
4691 }
4692
4693 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4694 {
4695         struct desc_ptr dt = { limit, base };
4696
4697         kvm_x86_ops->set_gdt(vcpu, &dt);
4698 }
4699
4700 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4701 {
4702         struct desc_ptr dt = { limit, base };
4703
4704         kvm_x86_ops->set_idt(vcpu, &dt);
4705 }
4706
4707 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4708 {
4709         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4710         int j, nent = vcpu->arch.cpuid_nent;
4711
4712         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4713         /* when no next entry is found, the current entry[i] is reselected */
4714         for (j = i + 1; ; j = (j + 1) % nent) {
4715                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4716                 if (ej->function == e->function) {
4717                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4718                         return j;
4719                 }
4720         }
4721         return 0; /* silence gcc, even though control never reaches here */
4722 }
4723
4724 /* find an entry with matching function, matching index (if needed), and that
4725  * should be read next (if it's stateful) */
4726 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4727         u32 function, u32 index)
4728 {
4729         if (e->function != function)
4730                 return 0;
4731         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4732                 return 0;
4733         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4734             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4735                 return 0;
4736         return 1;
4737 }
4738
4739 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4740                                               u32 function, u32 index)
4741 {
4742         int i;
4743         struct kvm_cpuid_entry2 *best = NULL;
4744
4745         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4746                 struct kvm_cpuid_entry2 *e;
4747
4748                 e = &vcpu->arch.cpuid_entries[i];
4749                 if (is_matching_cpuid_entry(e, function, index)) {
4750                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4751                                 move_to_next_stateful_cpuid_entry(vcpu, i);
4752                         best = e;
4753                         break;
4754                 }
4755                 /*
4756                  * Both basic or both extended?
4757                  */
4758                 if (((e->function ^ function) & 0x80000000) == 0)
4759                         if (!best || e->function > best->function)
4760                                 best = e;
4761         }
4762         return best;
4763 }
4764 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4765
4766 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4767 {
4768         struct kvm_cpuid_entry2 *best;
4769
4770         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
4771         if (!best || best->eax < 0x80000008)
4772                 goto not_found;
4773         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4774         if (best)
4775                 return best->eax & 0xff;
4776 not_found:
4777         return 36;
4778 }
4779
4780 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4781 {
4782         u32 function, index;
4783         struct kvm_cpuid_entry2 *best;
4784
4785         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4786         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4787         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4788         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4789         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4790         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4791         best = kvm_find_cpuid_entry(vcpu, function, index);
4792         if (best) {
4793                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4794                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4795                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4796                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4797         }
4798         kvm_x86_ops->skip_emulated_instruction(vcpu);
4799         trace_kvm_cpuid(function,
4800                         kvm_register_read(vcpu, VCPU_REGS_RAX),
4801                         kvm_register_read(vcpu, VCPU_REGS_RBX),
4802                         kvm_register_read(vcpu, VCPU_REGS_RCX),
4803                         kvm_register_read(vcpu, VCPU_REGS_RDX));
4804 }
4805 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4806
4807 /*
4808  * Check if userspace requested an interrupt window, and that the
4809  * interrupt window is open.
4810  *
4811  * No need to exit to userspace if we already have an interrupt queued.
4812  */
4813 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4814 {
4815         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4816                 vcpu->run->request_interrupt_window &&
4817                 kvm_arch_interrupt_allowed(vcpu));
4818 }
4819
4820 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4821 {
4822         struct kvm_run *kvm_run = vcpu->run;
4823
4824         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4825         kvm_run->cr8 = kvm_get_cr8(vcpu);
4826         kvm_run->apic_base = kvm_get_apic_base(vcpu);
4827         if (irqchip_in_kernel(vcpu->kvm))
4828                 kvm_run->ready_for_interrupt_injection = 1;
4829         else
4830                 kvm_run->ready_for_interrupt_injection =
4831                         kvm_arch_interrupt_allowed(vcpu) &&
4832                         !kvm_cpu_has_interrupt(vcpu) &&
4833                         !kvm_event_needs_reinjection(vcpu);
4834 }
4835
4836 static void vapic_enter(struct kvm_vcpu *vcpu)
4837 {
4838         struct kvm_lapic *apic = vcpu->arch.apic;
4839         struct page *page;
4840
4841         if (!apic || !apic->vapic_addr)
4842                 return;
4843
4844         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4845
4846         vcpu->arch.apic->vapic_page = page;
4847 }
4848
4849 static void vapic_exit(struct kvm_vcpu *vcpu)
4850 {
4851         struct kvm_lapic *apic = vcpu->arch.apic;
4852         int idx;
4853
4854         if (!apic || !apic->vapic_addr)
4855                 return;
4856
4857         idx = srcu_read_lock(&vcpu->kvm->srcu);
4858         kvm_release_page_dirty(apic->vapic_page);
4859         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4860         srcu_read_unlock(&vcpu->kvm->srcu, idx);
4861 }
4862
4863 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4864 {
4865         int max_irr, tpr;
4866
4867         if (!kvm_x86_ops->update_cr8_intercept)
4868                 return;
4869
4870         if (!vcpu->arch.apic)
4871                 return;
4872
4873         if (!vcpu->arch.apic->vapic_addr)
4874                 max_irr = kvm_lapic_find_highest_irr(vcpu);
4875         else
4876                 max_irr = -1;
4877
4878         if (max_irr != -1)
4879                 max_irr >>= 4;
4880
4881         tpr = kvm_lapic_get_cr8(vcpu);
4882
4883         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4884 }
4885
4886 static void inject_pending_event(struct kvm_vcpu *vcpu)
4887 {
4888         /* try to reinject previous events if any */
4889         if (vcpu->arch.exception.pending) {
4890                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
4891                                         vcpu->arch.exception.has_error_code,
4892                                         vcpu->arch.exception.error_code);
4893                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4894                                           vcpu->arch.exception.has_error_code,
4895                                           vcpu->arch.exception.error_code,
4896                                           vcpu->arch.exception.reinject);
4897                 return;
4898         }
4899
4900         if (vcpu->arch.nmi_injected) {
4901                 kvm_x86_ops->set_nmi(vcpu);
4902                 return;
4903         }
4904
4905         if (vcpu->arch.interrupt.pending) {
4906                 kvm_x86_ops->set_irq(vcpu);
4907                 return;
4908         }
4909
4910         /* try to inject new event if pending */
4911         if (vcpu->arch.nmi_pending) {
4912                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4913                         vcpu->arch.nmi_pending = false;
4914                         vcpu->arch.nmi_injected = true;
4915                         kvm_x86_ops->set_nmi(vcpu);
4916                 }
4917         } else if (kvm_cpu_has_interrupt(vcpu)) {
4918                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4919                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4920                                             false);
4921                         kvm_x86_ops->set_irq(vcpu);
4922                 }
4923         }
4924 }
4925
4926 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
4927 {
4928         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
4929                         !vcpu->guest_xcr0_loaded) {
4930                 /* kvm_set_xcr() also depends on this */
4931                 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
4932                 vcpu->guest_xcr0_loaded = 1;
4933         }
4934 }
4935
4936 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
4937 {
4938         if (vcpu->guest_xcr0_loaded) {
4939                 if (vcpu->arch.xcr0 != host_xcr0)
4940                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
4941                 vcpu->guest_xcr0_loaded = 0;
4942         }
4943 }
4944
4945 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4946 {
4947         int r;
4948         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4949                 vcpu->run->request_interrupt_window;
4950
4951         if (vcpu->requests) {
4952                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
4953                         kvm_mmu_unload(vcpu);
4954                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
4955                         __kvm_migrate_timers(vcpu);
4956                 if (kvm_check_request(KVM_REQ_KVMCLOCK_UPDATE, vcpu)) {
4957                         r = kvm_write_guest_time(vcpu);
4958                         if (unlikely(r))
4959                                 goto out;
4960                 }
4961                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
4962                         kvm_mmu_sync_roots(vcpu);
4963                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
4964                         kvm_x86_ops->tlb_flush(vcpu);
4965                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
4966                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4967                         r = 0;
4968                         goto out;
4969                 }
4970                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
4971                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4972                         r = 0;
4973                         goto out;
4974                 }
4975                 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
4976                         vcpu->fpu_active = 0;
4977                         kvm_x86_ops->fpu_deactivate(vcpu);
4978                 }
4979         }
4980
4981         r = kvm_mmu_reload(vcpu);
4982         if (unlikely(r))
4983                 goto out;
4984
4985         preempt_disable();
4986
4987         kvm_x86_ops->prepare_guest_switch(vcpu);
4988         if (vcpu->fpu_active)
4989                 kvm_load_guest_fpu(vcpu);
4990         kvm_load_guest_xcr0(vcpu);
4991
4992         atomic_set(&vcpu->guest_mode, 1);
4993         smp_wmb();
4994
4995         local_irq_disable();
4996
4997         if (!atomic_read(&vcpu->guest_mode) || vcpu->requests
4998             || need_resched() || signal_pending(current)) {
4999                 atomic_set(&vcpu->guest_mode, 0);
5000                 smp_wmb();
5001                 local_irq_enable();
5002                 preempt_enable();
5003                 r = 1;
5004                 goto out;
5005         }
5006
5007         inject_pending_event(vcpu);
5008
5009         /* enable NMI/IRQ window open exits if needed */
5010         if (vcpu->arch.nmi_pending)
5011                 kvm_x86_ops->enable_nmi_window(vcpu);
5012         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5013                 kvm_x86_ops->enable_irq_window(vcpu);
5014
5015         if (kvm_lapic_enabled(vcpu)) {
5016                 update_cr8_intercept(vcpu);
5017                 kvm_lapic_sync_to_vapic(vcpu);
5018         }
5019
5020         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5021
5022         kvm_guest_enter();
5023
5024         if (unlikely(vcpu->arch.switch_db_regs)) {
5025                 set_debugreg(0, 7);
5026                 set_debugreg(vcpu->arch.eff_db[0], 0);
5027                 set_debugreg(vcpu->arch.eff_db[1], 1);
5028                 set_debugreg(vcpu->arch.eff_db[2], 2);
5029                 set_debugreg(vcpu->arch.eff_db[3], 3);
5030         }
5031
5032         trace_kvm_entry(vcpu->vcpu_id);
5033         kvm_x86_ops->run(vcpu);
5034
5035         /*
5036          * If the guest has used debug registers, at least dr7
5037          * will be disabled while returning to the host.
5038          * If we don't have active breakpoints in the host, we don't
5039          * care about the messed up debug address registers. But if
5040          * we have some of them active, restore the old state.
5041          */
5042         if (hw_breakpoint_active())
5043                 hw_breakpoint_restore();
5044
5045         kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5046
5047         atomic_set(&vcpu->guest_mode, 0);
5048         smp_wmb();
5049         local_irq_enable();
5050
5051         ++vcpu->stat.exits;
5052
5053         /*
5054          * We must have an instruction between local_irq_enable() and
5055          * kvm_guest_exit(), so the timer interrupt isn't delayed by
5056          * the interrupt shadow.  The stat.exits increment will do nicely.
5057          * But we need to prevent reordering, hence this barrier():
5058          */
5059         barrier();
5060
5061         kvm_guest_exit();
5062
5063         preempt_enable();
5064
5065         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5066
5067         /*
5068          * Profile KVM exit RIPs:
5069          */
5070         if (unlikely(prof_on == KVM_PROFILING)) {
5071                 unsigned long rip = kvm_rip_read(vcpu);
5072                 profile_hit(KVM_PROFILING, (void *)rip);
5073         }
5074
5075
5076         kvm_lapic_sync_from_vapic(vcpu);
5077
5078         r = kvm_x86_ops->handle_exit(vcpu);
5079 out:
5080         return r;
5081 }
5082
5083
5084 static int __vcpu_run(struct kvm_vcpu *vcpu)
5085 {
5086         int r;
5087         struct kvm *kvm = vcpu->kvm;
5088
5089         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5090                 pr_debug("vcpu %d received sipi with vector # %x\n",
5091                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
5092                 kvm_lapic_reset(vcpu);
5093                 r = kvm_arch_vcpu_reset(vcpu);
5094                 if (r)
5095                         return r;
5096                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5097         }
5098
5099         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5100         vapic_enter(vcpu);
5101
5102         r = 1;
5103         while (r > 0) {
5104                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
5105                         r = vcpu_enter_guest(vcpu);
5106                 else {
5107                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5108                         kvm_vcpu_block(vcpu);
5109                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5110                         if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5111                         {
5112                                 switch(vcpu->arch.mp_state) {
5113                                 case KVM_MP_STATE_HALTED:
5114                                         vcpu->arch.mp_state =
5115                                                 KVM_MP_STATE_RUNNABLE;
5116                                 case KVM_MP_STATE_RUNNABLE:
5117                                         break;
5118                                 case KVM_MP_STATE_SIPI_RECEIVED:
5119                                 default:
5120                                         r = -EINTR;
5121                                         break;
5122                                 }
5123                         }
5124                 }
5125
5126                 if (r <= 0)
5127                         break;
5128
5129                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5130                 if (kvm_cpu_has_pending_timer(vcpu))
5131                         kvm_inject_pending_timer_irqs(vcpu);
5132
5133                 if (dm_request_for_irq_injection(vcpu)) {
5134                         r = -EINTR;
5135                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5136                         ++vcpu->stat.request_irq_exits;
5137                 }
5138                 if (signal_pending(current)) {
5139                         r = -EINTR;
5140                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5141                         ++vcpu->stat.signal_exits;
5142                 }
5143                 if (need_resched()) {
5144                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5145                         kvm_resched(vcpu);
5146                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5147                 }
5148         }
5149
5150         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5151
5152         vapic_exit(vcpu);
5153
5154         return r;
5155 }
5156
5157 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5158 {
5159         int r;
5160         sigset_t sigsaved;
5161
5162         if (vcpu->sigset_active)
5163                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5164
5165         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5166                 kvm_vcpu_block(vcpu);
5167                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5168                 r = -EAGAIN;
5169                 goto out;
5170         }
5171
5172         /* re-sync apic's tpr */
5173         if (!irqchip_in_kernel(vcpu->kvm))
5174                 kvm_set_cr8(vcpu, kvm_run->cr8);
5175
5176         if (vcpu->arch.pio.count || vcpu->mmio_needed) {
5177                 if (vcpu->mmio_needed) {
5178                         memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
5179                         vcpu->mmio_read_completed = 1;
5180                         vcpu->mmio_needed = 0;
5181                 }
5182                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5183                 r = emulate_instruction(vcpu, 0, 0, EMULTYPE_NO_DECODE);
5184                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5185                 if (r != EMULATE_DONE) {
5186                         r = 0;
5187                         goto out;
5188                 }
5189         }
5190         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5191                 kvm_register_write(vcpu, VCPU_REGS_RAX,
5192                                      kvm_run->hypercall.ret);
5193
5194         r = __vcpu_run(vcpu);
5195
5196 out:
5197         post_kvm_run_save(vcpu);
5198         if (vcpu->sigset_active)
5199                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5200
5201         return r;
5202 }
5203
5204 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5205 {
5206         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5207         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5208         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5209         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5210         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5211         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5212         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5213         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5214 #ifdef CONFIG_X86_64
5215         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5216         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5217         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5218         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5219         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5220         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5221         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5222         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5223 #endif
5224
5225         regs->rip = kvm_rip_read(vcpu);
5226         regs->rflags = kvm_get_rflags(vcpu);
5227
5228         return 0;
5229 }
5230
5231 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5232 {
5233         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5234         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5235         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5236         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5237         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5238         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5239         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5240         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5241 #ifdef CONFIG_X86_64
5242         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5243         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5244         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5245         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5246         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5247         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5248         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5249         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5250 #endif
5251
5252         kvm_rip_write(vcpu, regs->rip);
5253         kvm_set_rflags(vcpu, regs->rflags);
5254
5255         vcpu->arch.exception.pending = false;
5256
5257         return 0;
5258 }
5259
5260 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5261 {
5262         struct kvm_segment cs;
5263
5264         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5265         *db = cs.db;
5266         *l = cs.l;
5267 }
5268 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5269
5270 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5271                                   struct kvm_sregs *sregs)
5272 {
5273         struct desc_ptr dt;
5274
5275         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5276         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5277         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5278         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5279         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5280         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5281
5282         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5283         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5284
5285         kvm_x86_ops->get_idt(vcpu, &dt);
5286         sregs->idt.limit = dt.size;
5287         sregs->idt.base = dt.address;
5288         kvm_x86_ops->get_gdt(vcpu, &dt);
5289         sregs->gdt.limit = dt.size;
5290         sregs->gdt.base = dt.address;
5291
5292         sregs->cr0 = kvm_read_cr0(vcpu);
5293         sregs->cr2 = vcpu->arch.cr2;
5294         sregs->cr3 = vcpu->arch.cr3;
5295         sregs->cr4 = kvm_read_cr4(vcpu);
5296         sregs->cr8 = kvm_get_cr8(vcpu);
5297         sregs->efer = vcpu->arch.efer;
5298         sregs->apic_base = kvm_get_apic_base(vcpu);
5299
5300         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5301
5302         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5303                 set_bit(vcpu->arch.interrupt.nr,
5304                         (unsigned long *)sregs->interrupt_bitmap);
5305
5306         return 0;
5307 }
5308
5309 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5310                                     struct kvm_mp_state *mp_state)
5311 {
5312         mp_state->mp_state = vcpu->arch.mp_state;
5313         return 0;
5314 }
5315
5316 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5317                                     struct kvm_mp_state *mp_state)
5318 {
5319         vcpu->arch.mp_state = mp_state->mp_state;
5320         return 0;
5321 }
5322
5323 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5324                     bool has_error_code, u32 error_code)
5325 {
5326         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5327         int ret;
5328
5329         init_emulate_ctxt(vcpu);
5330
5331         ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5332                                    tss_selector, reason, has_error_code,
5333                                    error_code);
5334
5335         if (ret)
5336                 return EMULATE_FAIL;
5337
5338         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5339         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5340         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5341         return EMULATE_DONE;
5342 }
5343 EXPORT_SYMBOL_GPL(kvm_task_switch);
5344
5345 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5346                                   struct kvm_sregs *sregs)
5347 {
5348         int mmu_reset_needed = 0;
5349         int pending_vec, max_bits;
5350         struct desc_ptr dt;
5351
5352         dt.size = sregs->idt.limit;
5353         dt.address = sregs->idt.base;
5354         kvm_x86_ops->set_idt(vcpu, &dt);
5355         dt.size = sregs->gdt.limit;
5356         dt.address = sregs->gdt.base;
5357         kvm_x86_ops->set_gdt(vcpu, &dt);
5358
5359         vcpu->arch.cr2 = sregs->cr2;
5360         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5361         vcpu->arch.cr3 = sregs->cr3;
5362
5363         kvm_set_cr8(vcpu, sregs->cr8);
5364
5365         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5366         kvm_x86_ops->set_efer(vcpu, sregs->efer);
5367         kvm_set_apic_base(vcpu, sregs->apic_base);
5368
5369         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5370         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5371         vcpu->arch.cr0 = sregs->cr0;
5372
5373         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5374         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5375         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5376                 load_pdptrs(vcpu, vcpu->arch.cr3);
5377                 mmu_reset_needed = 1;
5378         }
5379
5380         if (mmu_reset_needed)
5381                 kvm_mmu_reset_context(vcpu);
5382
5383         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5384         pending_vec = find_first_bit(
5385                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5386         if (pending_vec < max_bits) {
5387                 kvm_queue_interrupt(vcpu, pending_vec, false);
5388                 pr_debug("Set back pending irq %d\n", pending_vec);
5389                 if (irqchip_in_kernel(vcpu->kvm))
5390                         kvm_pic_clear_isr_ack(vcpu->kvm);
5391         }
5392
5393         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5394         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5395         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5396         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5397         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5398         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5399
5400         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5401         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5402
5403         update_cr8_intercept(vcpu);
5404
5405         /* Older userspace won't unhalt the vcpu on reset. */
5406         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5407             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5408             !is_protmode(vcpu))
5409                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5410
5411         return 0;
5412 }
5413
5414 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5415                                         struct kvm_guest_debug *dbg)
5416 {
5417         unsigned long rflags;
5418         int i, r;
5419
5420         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5421                 r = -EBUSY;
5422                 if (vcpu->arch.exception.pending)
5423                         goto out;
5424                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5425                         kvm_queue_exception(vcpu, DB_VECTOR);
5426                 else
5427                         kvm_queue_exception(vcpu, BP_VECTOR);
5428         }
5429
5430         /*
5431          * Read rflags as long as potentially injected trace flags are still
5432          * filtered out.
5433          */
5434         rflags = kvm_get_rflags(vcpu);
5435
5436         vcpu->guest_debug = dbg->control;
5437         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5438                 vcpu->guest_debug = 0;
5439
5440         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5441                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5442                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5443                 vcpu->arch.switch_db_regs =
5444                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5445         } else {
5446                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5447                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5448                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5449         }
5450
5451         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5452                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5453                         get_segment_base(vcpu, VCPU_SREG_CS);
5454
5455         /*
5456          * Trigger an rflags update that will inject or remove the trace
5457          * flags.
5458          */
5459         kvm_set_rflags(vcpu, rflags);
5460
5461         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5462
5463         r = 0;
5464
5465 out:
5466
5467         return r;
5468 }
5469
5470 /*
5471  * Translate a guest virtual address to a guest physical address.
5472  */
5473 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5474                                     struct kvm_translation *tr)
5475 {
5476         unsigned long vaddr = tr->linear_address;
5477         gpa_t gpa;
5478         int idx;
5479
5480         idx = srcu_read_lock(&vcpu->kvm->srcu);
5481         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5482         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5483         tr->physical_address = gpa;
5484         tr->valid = gpa != UNMAPPED_GVA;
5485         tr->writeable = 1;
5486         tr->usermode = 0;
5487
5488         return 0;
5489 }
5490
5491 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5492 {
5493         struct i387_fxsave_struct *fxsave =
5494                         &vcpu->arch.guest_fpu.state->fxsave;
5495
5496         memcpy(fpu->fpr, fxsave->st_space, 128);
5497         fpu->fcw = fxsave->cwd;
5498         fpu->fsw = fxsave->swd;
5499         fpu->ftwx = fxsave->twd;
5500         fpu->last_opcode = fxsave->fop;
5501         fpu->last_ip = fxsave->rip;
5502         fpu->last_dp = fxsave->rdp;
5503         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5504
5505         return 0;
5506 }
5507
5508 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5509 {
5510         struct i387_fxsave_struct *fxsave =
5511                         &vcpu->arch.guest_fpu.state->fxsave;
5512
5513         memcpy(fxsave->st_space, fpu->fpr, 128);
5514         fxsave->cwd = fpu->fcw;
5515         fxsave->swd = fpu->fsw;
5516         fxsave->twd = fpu->ftwx;
5517         fxsave->fop = fpu->last_opcode;
5518         fxsave->rip = fpu->last_ip;
5519         fxsave->rdp = fpu->last_dp;
5520         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5521
5522         return 0;
5523 }
5524
5525 int fx_init(struct kvm_vcpu *vcpu)
5526 {
5527         int err;
5528
5529         err = fpu_alloc(&vcpu->arch.guest_fpu);
5530         if (err)
5531                 return err;
5532
5533         fpu_finit(&vcpu->arch.guest_fpu);
5534
5535         /*
5536          * Ensure guest xcr0 is valid for loading
5537          */
5538         vcpu->arch.xcr0 = XSTATE_FP;
5539
5540         vcpu->arch.cr0 |= X86_CR0_ET;
5541
5542         return 0;
5543 }
5544 EXPORT_SYMBOL_GPL(fx_init);
5545
5546 static void fx_free(struct kvm_vcpu *vcpu)
5547 {
5548         fpu_free(&vcpu->arch.guest_fpu);
5549 }
5550
5551 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5552 {
5553         if (vcpu->guest_fpu_loaded)
5554                 return;
5555
5556         /*
5557          * Restore all possible states in the guest,
5558          * and assume host would use all available bits.
5559          * Guest xcr0 would be loaded later.
5560          */
5561         kvm_put_guest_xcr0(vcpu);
5562         vcpu->guest_fpu_loaded = 1;
5563         unlazy_fpu(current);
5564         fpu_restore_checking(&vcpu->arch.guest_fpu);
5565         trace_kvm_fpu(1);
5566 }
5567
5568 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5569 {
5570         kvm_put_guest_xcr0(vcpu);
5571
5572         if (!vcpu->guest_fpu_loaded)
5573                 return;
5574
5575         vcpu->guest_fpu_loaded = 0;
5576         fpu_save_init(&vcpu->arch.guest_fpu);
5577         ++vcpu->stat.fpu_reload;
5578         kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
5579         trace_kvm_fpu(0);
5580 }
5581
5582 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5583 {
5584         if (vcpu->arch.time_page) {
5585                 kvm_release_page_dirty(vcpu->arch.time_page);
5586                 vcpu->arch.time_page = NULL;
5587         }
5588
5589         free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5590         fx_free(vcpu);
5591         kvm_x86_ops->vcpu_free(vcpu);
5592 }
5593
5594 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5595                                                 unsigned int id)
5596 {
5597         if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5598                 printk_once(KERN_WARNING
5599                 "kvm: SMP vm created on host with unstable TSC; "
5600                 "guest TSC will not be reliable\n");
5601         return kvm_x86_ops->vcpu_create(kvm, id);
5602 }
5603
5604 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5605 {
5606         int r;
5607
5608         vcpu->arch.mtrr_state.have_fixed = 1;
5609         vcpu_load(vcpu);
5610         r = kvm_arch_vcpu_reset(vcpu);
5611         if (r == 0)
5612                 r = kvm_mmu_setup(vcpu);
5613         vcpu_put(vcpu);
5614         if (r < 0)
5615                 goto free_vcpu;
5616
5617         return 0;
5618 free_vcpu:
5619         kvm_x86_ops->vcpu_free(vcpu);
5620         return r;
5621 }
5622
5623 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5624 {
5625         vcpu_load(vcpu);
5626         kvm_mmu_unload(vcpu);
5627         vcpu_put(vcpu);
5628
5629         fx_free(vcpu);
5630         kvm_x86_ops->vcpu_free(vcpu);
5631 }
5632
5633 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5634 {
5635         vcpu->arch.nmi_pending = false;
5636         vcpu->arch.nmi_injected = false;
5637
5638         vcpu->arch.switch_db_regs = 0;
5639         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5640         vcpu->arch.dr6 = DR6_FIXED_1;
5641         vcpu->arch.dr7 = DR7_FIXED_1;
5642
5643         return kvm_x86_ops->vcpu_reset(vcpu);
5644 }
5645
5646 int kvm_arch_hardware_enable(void *garbage)
5647 {
5648         struct kvm *kvm;
5649         struct kvm_vcpu *vcpu;
5650         int i;
5651
5652         kvm_shared_msr_cpu_online();
5653         list_for_each_entry(kvm, &vm_list, vm_list)
5654                 kvm_for_each_vcpu(i, vcpu, kvm)
5655                         if (vcpu->cpu == smp_processor_id())
5656                                 kvm_request_guest_time_update(vcpu);
5657         return kvm_x86_ops->hardware_enable(garbage);
5658 }
5659
5660 void kvm_arch_hardware_disable(void *garbage)
5661 {
5662         kvm_x86_ops->hardware_disable(garbage);
5663         drop_user_return_notifiers(garbage);
5664 }
5665
5666 int kvm_arch_hardware_setup(void)
5667 {
5668         return kvm_x86_ops->hardware_setup();
5669 }
5670
5671 void kvm_arch_hardware_unsetup(void)
5672 {
5673         kvm_x86_ops->hardware_unsetup();
5674 }
5675
5676 void kvm_arch_check_processor_compat(void *rtn)
5677 {
5678         kvm_x86_ops->check_processor_compatibility(rtn);
5679 }
5680
5681 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5682 {
5683         struct page *page;
5684         struct kvm *kvm;
5685         int r;
5686
5687         BUG_ON(vcpu->kvm == NULL);
5688         kvm = vcpu->kvm;
5689
5690         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
5691         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
5692         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5693         vcpu->arch.mmu.translate_gpa = translate_gpa;
5694         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5695                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5696         else
5697                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5698
5699         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5700         if (!page) {
5701                 r = -ENOMEM;
5702                 goto fail;
5703         }
5704         vcpu->arch.pio_data = page_address(page);
5705
5706         r = kvm_mmu_create(vcpu);
5707         if (r < 0)
5708                 goto fail_free_pio_data;
5709
5710         if (irqchip_in_kernel(kvm)) {
5711                 r = kvm_create_lapic(vcpu);
5712                 if (r < 0)
5713                         goto fail_mmu_destroy;
5714         }
5715
5716         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5717                                        GFP_KERNEL);
5718         if (!vcpu->arch.mce_banks) {
5719                 r = -ENOMEM;
5720                 goto fail_free_lapic;
5721         }
5722         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5723
5724         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
5725                 goto fail_free_mce_banks;
5726
5727         return 0;
5728 fail_free_mce_banks:
5729         kfree(vcpu->arch.mce_banks);
5730 fail_free_lapic:
5731         kvm_free_lapic(vcpu);
5732 fail_mmu_destroy:
5733         kvm_mmu_destroy(vcpu);
5734 fail_free_pio_data:
5735         free_page((unsigned long)vcpu->arch.pio_data);
5736 fail:
5737         return r;
5738 }
5739
5740 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5741 {
5742         int idx;
5743
5744         kfree(vcpu->arch.mce_banks);
5745         kvm_free_lapic(vcpu);
5746         idx = srcu_read_lock(&vcpu->kvm->srcu);
5747         kvm_mmu_destroy(vcpu);
5748         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5749         free_page((unsigned long)vcpu->arch.pio_data);
5750 }
5751
5752 struct  kvm *kvm_arch_create_vm(void)
5753 {
5754         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5755
5756         if (!kvm)
5757                 return ERR_PTR(-ENOMEM);
5758
5759         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5760         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5761
5762         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5763         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5764
5765         spin_lock_init(&kvm->arch.tsc_write_lock);
5766
5767         return kvm;
5768 }
5769
5770 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5771 {
5772         vcpu_load(vcpu);
5773         kvm_mmu_unload(vcpu);
5774         vcpu_put(vcpu);
5775 }
5776
5777 static void kvm_free_vcpus(struct kvm *kvm)
5778 {
5779         unsigned int i;
5780         struct kvm_vcpu *vcpu;
5781
5782         /*
5783          * Unpin any mmu pages first.
5784          */
5785         kvm_for_each_vcpu(i, vcpu, kvm)
5786                 kvm_unload_vcpu_mmu(vcpu);
5787         kvm_for_each_vcpu(i, vcpu, kvm)
5788                 kvm_arch_vcpu_free(vcpu);
5789
5790         mutex_lock(&kvm->lock);
5791         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5792                 kvm->vcpus[i] = NULL;
5793
5794         atomic_set(&kvm->online_vcpus, 0);
5795         mutex_unlock(&kvm->lock);
5796 }
5797
5798 void kvm_arch_sync_events(struct kvm *kvm)
5799 {
5800         kvm_free_all_assigned_devices(kvm);
5801         kvm_free_pit(kvm);
5802 }
5803
5804 void kvm_arch_destroy_vm(struct kvm *kvm)
5805 {
5806         kvm_iommu_unmap_guest(kvm);
5807         kfree(kvm->arch.vpic);
5808         kfree(kvm->arch.vioapic);
5809         kvm_free_vcpus(kvm);
5810         kvm_free_physmem(kvm);
5811         if (kvm->arch.apic_access_page)
5812                 put_page(kvm->arch.apic_access_page);
5813         if (kvm->arch.ept_identity_pagetable)
5814                 put_page(kvm->arch.ept_identity_pagetable);
5815         cleanup_srcu_struct(&kvm->srcu);
5816         kfree(kvm);
5817 }
5818
5819 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5820                                 struct kvm_memory_slot *memslot,
5821                                 struct kvm_memory_slot old,
5822                                 struct kvm_userspace_memory_region *mem,
5823                                 int user_alloc)
5824 {
5825         int npages = memslot->npages;
5826         int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
5827
5828         /* Prevent internal slot pages from being moved by fork()/COW. */
5829         if (memslot->id >= KVM_MEMORY_SLOTS)
5830                 map_flags = MAP_SHARED | MAP_ANONYMOUS;
5831
5832         /*To keep backward compatibility with older userspace,
5833          *x86 needs to hanlde !user_alloc case.
5834          */
5835         if (!user_alloc) {
5836                 if (npages && !old.rmap) {
5837                         unsigned long userspace_addr;
5838
5839                         down_write(&current->mm->mmap_sem);
5840                         userspace_addr = do_mmap(NULL, 0,
5841                                                  npages * PAGE_SIZE,
5842                                                  PROT_READ | PROT_WRITE,
5843                                                  map_flags,
5844                                                  0);
5845                         up_write(&current->mm->mmap_sem);
5846
5847                         if (IS_ERR((void *)userspace_addr))
5848                                 return PTR_ERR((void *)userspace_addr);
5849
5850                         memslot->userspace_addr = userspace_addr;
5851                 }
5852         }
5853
5854
5855         return 0;
5856 }
5857
5858 void kvm_arch_commit_memory_region(struct kvm *kvm,
5859                                 struct kvm_userspace_memory_region *mem,
5860                                 struct kvm_memory_slot old,
5861                                 int user_alloc)
5862 {
5863
5864         int npages = mem->memory_size >> PAGE_SHIFT;
5865
5866         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5867                 int ret;
5868
5869                 down_write(&current->mm->mmap_sem);
5870                 ret = do_munmap(current->mm, old.userspace_addr,
5871                                 old.npages * PAGE_SIZE);
5872                 up_write(&current->mm->mmap_sem);
5873                 if (ret < 0)
5874                         printk(KERN_WARNING
5875                                "kvm_vm_ioctl_set_memory_region: "
5876                                "failed to munmap memory\n");
5877         }
5878
5879         spin_lock(&kvm->mmu_lock);
5880         if (!kvm->arch.n_requested_mmu_pages) {
5881                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5882                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5883         }
5884
5885         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5886         spin_unlock(&kvm->mmu_lock);
5887 }
5888
5889 void kvm_arch_flush_shadow(struct kvm *kvm)
5890 {
5891         kvm_mmu_zap_all(kvm);
5892         kvm_reload_remote_mmus(kvm);
5893 }
5894
5895 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5896 {
5897         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5898                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5899                 || vcpu->arch.nmi_pending ||
5900                 (kvm_arch_interrupt_allowed(vcpu) &&
5901                  kvm_cpu_has_interrupt(vcpu));
5902 }
5903
5904 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5905 {
5906         int me;
5907         int cpu = vcpu->cpu;
5908
5909         if (waitqueue_active(&vcpu->wq)) {
5910                 wake_up_interruptible(&vcpu->wq);
5911                 ++vcpu->stat.halt_wakeup;
5912         }
5913
5914         me = get_cpu();
5915         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5916                 if (atomic_xchg(&vcpu->guest_mode, 0))
5917                         smp_send_reschedule(cpu);
5918         put_cpu();
5919 }
5920
5921 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5922 {
5923         return kvm_x86_ops->interrupt_allowed(vcpu);
5924 }
5925
5926 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
5927 {
5928         unsigned long current_rip = kvm_rip_read(vcpu) +
5929                 get_segment_base(vcpu, VCPU_SREG_CS);
5930
5931         return current_rip == linear_rip;
5932 }
5933 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
5934
5935 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5936 {
5937         unsigned long rflags;
5938
5939         rflags = kvm_x86_ops->get_rflags(vcpu);
5940         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5941                 rflags &= ~X86_EFLAGS_TF;
5942         return rflags;
5943 }
5944 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5945
5946 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5947 {
5948         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5949             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
5950                 rflags |= X86_EFLAGS_TF;
5951         kvm_x86_ops->set_rflags(vcpu, rflags);
5952 }
5953 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5954
5955 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5956 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5957 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5958 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5959 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5960 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5961 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5962 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5963 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5964 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5965 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
5966 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);