arch/x86/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  10  *
  11  * Authors:
  12  *   Avi Kivity   <avi@qumranet.com>
  13  *   Yaniv Kamay  <yaniv@qumranet.com>
  14  *   Amit Shah    <amit.shah@qumranet.com>
  15  *   Ben-Ami Yassour <benami@il.ibm.com>
  16  *
  17  * This work is licensed under the terms of the GNU GPL, version 2.  See
  18  * the COPYING file in the top-level directory.
  19  *
  20  */
  21
  22 #include <linux/kvm_host.h>
  23 #include "irq.h"
  24 #include "mmu.h"
  25 #include "i8254.h"
  26 #include "tss.h"
  27 #include "kvm_cache_regs.h"
  28 #include "x86.h"
  29
  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 <linux/hash.h>
  47 #include <trace/events/kvm.h>
  48
  49 #define CREATE_TRACE_POINTS
  50 #include "trace.h"
  51
  52 #include <asm/debugreg.h>
  53 #include <asm/msr.h>
  54 #include <asm/desc.h>
  55 #include <asm/mtrr.h>
  56 #include <asm/mce.h>
  57 #include <asm/i387.h>
  58 #include <asm/xcr.h>
  59 #include <asm/pvclock.h>
  60 #include <asm/div64.h>
  61
  62 #define MAX_IO_MSRS 256
  63 #define CR0_RESERVED_BITS                                               \
  64         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  65                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  66                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  67 #define CR4_RESERVED_BITS                                               \
  68         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  69                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  70                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  71                           | X86_CR4_OSXSAVE \
  72                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  73
  74 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  75
  76 #define KVM_MAX_MCE_BANKS 32
  77 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
  78
  79 /* EFER defaults:
  80  * - enable syscall per default because its emulated by KVM
  81  * - enable LME and LMA per default on 64 bit KVM
  82  */
  83 #ifdef CONFIG_X86_64
  84 static
  85 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
  86 #else
  87 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
  88 #endif
  89
  90 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  91 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  92
  93 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  94 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  95                                     struct kvm_cpuid_entry2 __user *entries);
  96
  97 struct kvm_x86_ops *kvm_x86_ops;
  98 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  99
 100 int ignore_msrs = 0;
 101 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
 102
 103 #define KVM_NR_SHARED_MSRS 16
 104
 105 struct kvm_shared_msrs_global {
 106         int nr;
 107         u32 msrs[KVM_NR_SHARED_MSRS];
 108 };
 109
 110 struct kvm_shared_msrs {
 111         struct user_return_notifier urn;
 112         bool registered;
 113         struct kvm_shared_msr_values {
 114                 u64 host;
 115                 u64 curr;
 116         } values[KVM_NR_SHARED_MSRS];
 117 };
 118
 119 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 120 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
 121
 122 struct kvm_stats_debugfs_item debugfs_entries[] = {
 123         { "pf_fixed", VCPU_STAT(pf_fixed) },
 124         { "pf_guest", VCPU_STAT(pf_guest) },
 125         { "tlb_flush", VCPU_STAT(tlb_flush) },
 126         { "invlpg", VCPU_STAT(invlpg) },
 127         { "exits", VCPU_STAT(exits) },
 128         { "io_exits", VCPU_STAT(io_exits) },
 129         { "mmio_exits", VCPU_STAT(mmio_exits) },
 130         { "signal_exits", VCPU_STAT(signal_exits) },
 131         { "irq_window", VCPU_STAT(irq_window_exits) },
 132         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 133         { "halt_exits", VCPU_STAT(halt_exits) },
 134         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 135         { "hypercalls", VCPU_STAT(hypercalls) },
 136         { "request_irq", VCPU_STAT(request_irq_exits) },
 137         { "irq_exits", VCPU_STAT(irq_exits) },
 138         { "host_state_reload", VCPU_STAT(host_state_reload) },
 139         { "efer_reload", VCPU_STAT(efer_reload) },
 140         { "fpu_reload", VCPU_STAT(fpu_reload) },
 141         { "insn_emulation", VCPU_STAT(insn_emulation) },
 142         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 143         { "irq_injections", VCPU_STAT(irq_injections) },
 144         { "nmi_injections", VCPU_STAT(nmi_injections) },
 145         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 146         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 147         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 148         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 149         { "mmu_flooded", VM_STAT(mmu_flooded) },
 150         { "mmu_recycled", VM_STAT(mmu_recycled) },
 151         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 152         { "mmu_unsync", VM_STAT(mmu_unsync) },
 153         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 154         { "largepages", VM_STAT(lpages) },
 155         { NULL }
 156 };
 157
 158 u64 __read_mostly host_xcr0;
 159
 160 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
 161 {
 162         int i;
 163         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
 164                 vcpu->arch.apf.gfns[i] = ~0;
 165 }
 166
 167 static void kvm_on_user_return(struct user_return_notifier *urn)
 168 {
 169         unsigned slot;
 170         struct kvm_shared_msrs *locals
 171                 = container_of(urn, struct kvm_shared_msrs, urn);
 172         struct kvm_shared_msr_values *values;
 173
 174         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 175                 values = &locals->values[slot];
 176                 if (values->host != values->curr) {
 177                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 178                         values->curr = values->host;
 179                 }
 180         }
 181         locals->registered = false;
 182         user_return_notifier_unregister(urn);
 183 }
 184
 185 static void shared_msr_update(unsigned slot, u32 msr)
 186 {
 187         struct kvm_shared_msrs *smsr;
 188         u64 value;
 189
 190         smsr = &__get_cpu_var(shared_msrs);
 191         /* only read, and nobody should modify it at this time,
 192          * so don't need lock */
 193         if (slot >= shared_msrs_global.nr) {
 194                 printk(KERN_ERR "kvm: invalid MSR slot!");
 195                 return;
 196         }
 197         rdmsrl_safe(msr, &value);
 198         smsr->values[slot].host = value;
 199         smsr->values[slot].curr = value;
 200 }
 201
 202 void kvm_define_shared_msr(unsigned slot, u32 msr)
 203 {
 204         if (slot >= shared_msrs_global.nr)
 205                 shared_msrs_global.nr = slot + 1;
 206         shared_msrs_global.msrs[slot] = msr;
 207         /* we need ensured the shared_msr_global have been updated */
 208         smp_wmb();
 209 }
 210 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 211
 212 static void kvm_shared_msr_cpu_online(void)
 213 {
 214         unsigned i;
 215
 216         for (i = 0; i < shared_msrs_global.nr; ++i)
 217                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 218 }
 219
 220 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 221 {
 222         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 223
 224         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 225                 return;
 226         smsr->values[slot].curr = value;
 227         wrmsrl(shared_msrs_global.msrs[slot], value);
 228         if (!smsr->registered) {
 229                 smsr->urn.on_user_return = kvm_on_user_return;
 230                 user_return_notifier_register(&smsr->urn);
 231                 smsr->registered = true;
 232         }
 233 }
 234 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 235
 236 static void drop_user_return_notifiers(void *ignore)
 237 {
 238         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 239
 240         if (smsr->registered)
 241                 kvm_on_user_return(&smsr->urn);
 242 }
 243
 244 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 245 {
 246         if (irqchip_in_kernel(vcpu->kvm))
 247                 return vcpu->arch.apic_base;
 248         else
 249                 return vcpu->arch.apic_base;
 250 }
 251 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 252
 253 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 254 {
 255         /* TODO: reserve bits check */
 256         if (irqchip_in_kernel(vcpu->kvm))
 257                 kvm_lapic_set_base(vcpu, data);
 258         else
 259                 vcpu->arch.apic_base = data;
 260 }
 261 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 262
 263 #define EXCPT_BENIGN            0
 264 #define EXCPT_CONTRIBUTORY      1
 265 #define EXCPT_PF                2
 266
 267 static int exception_class(int vector)
 268 {
 269         switch (vector) {
 270         case PF_VECTOR:
 271                 return EXCPT_PF;
 272         case DE_VECTOR:
 273         case TS_VECTOR:
 274         case NP_VECTOR:
 275         case SS_VECTOR:
 276         case GP_VECTOR:
 277                 return EXCPT_CONTRIBUTORY;
 278         default:
 279                 break;
 280         }
 281         return EXCPT_BENIGN;
 282 }
 283
 284 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 285                 unsigned nr, bool has_error, u32 error_code,
 286                 bool reinject)
 287 {
 288         u32 prev_nr;
 289         int class1, class2;
 290
 291         kvm_make_request(KVM_REQ_EVENT, vcpu);
 292
 293         if (!vcpu->arch.exception.pending) {
 294         queue:
 295                 vcpu->arch.exception.pending = true;
 296                 vcpu->arch.exception.has_error_code = has_error;
 297                 vcpu->arch.exception.nr = nr;
 298                 vcpu->arch.exception.error_code = error_code;
 299                 vcpu->arch.exception.reinject = reinject;
 300                 return;
 301         }
 302
 303         /* to check exception */
 304         prev_nr = vcpu->arch.exception.nr;
 305         if (prev_nr == DF_VECTOR) {
 306                 /* triple fault -> shutdown */
 307                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
 308                 return;
 309         }
 310         class1 = exception_class(prev_nr);
 311         class2 = exception_class(nr);
 312         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 313                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 314                 /* generate double fault per SDM Table 5-5 */
 315                 vcpu->arch.exception.pending = true;
 316                 vcpu->arch.exception.has_error_code = true;
 317                 vcpu->arch.exception.nr = DF_VECTOR;
 318                 vcpu->arch.exception.error_code = 0;
 319         } else
 320                 /* replace previous exception with a new one in a hope
 321                    that instruction re-execution will regenerate lost
 322                    exception */
 323                 goto queue;
 324 }
 325
 326 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 327 {
 328         kvm_multiple_exception(vcpu, nr, false, 0, false);
 329 }
 330 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 331
 332 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 333 {
 334         kvm_multiple_exception(vcpu, nr, false, 0, true);
 335 }
 336 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 337
 338 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
 339 {
 340         if (err)
 341                 kvm_inject_gp(vcpu, 0);
 342         else
 343                 kvm_x86_ops->skip_emulated_instruction(vcpu);
 344 }
 345 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
 346
 347 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 348 {
 349         ++vcpu->stat.pf_guest;
 350         vcpu->arch.cr2 = fault->address;
 351         kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
 352 }
 353
 354 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 355 {
 356         if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
 357                 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
 358         else
 359                 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
 360 }
 361
 362 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 363 {
 364         kvm_make_request(KVM_REQ_EVENT, vcpu);
 365         vcpu->arch.nmi_pending = 1;
 366 }
 367 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 368
 369 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 370 {
 371         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 372 }
 373 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 374
 375 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 376 {
 377         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 378 }
 379 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 380
 381 /*
 382  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 383  * a #GP and return false.
 384  */
 385 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 386 {
 387         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 388                 return true;
 389         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 390         return false;
 391 }
 392 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 393
 394 /*
 395  * This function will be used to read from the physical memory of the currently
 396  * running guest. The difference to kvm_read_guest_page is that this function
 397  * can read from guest physical or from the guest's guest physical memory.
 398  */
 399 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 400                             gfn_t ngfn, void *data, int offset, int len,
 401                             u32 access)
 402 {
 403         gfn_t real_gfn;
 404         gpa_t ngpa;
 405
 406         ngpa     = gfn_to_gpa(ngfn);
 407         real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
 408         if (real_gfn == UNMAPPED_GVA)
 409                 return -EFAULT;
 410
 411         real_gfn = gpa_to_gfn(real_gfn);
 412
 413         return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
 414 }
 415 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
 416
 417 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
 418                                void *data, int offset, int len, u32 access)
 419 {
 420         return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
 421                                        data, offset, len, access);
 422 }
 423
 424 /*
 425  * Load the pae pdptrs.  Return true is they are all valid.
 426  */
 427 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
 428 {
 429         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 430         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 431         int i;
 432         int ret;
 433         u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
 434
 435         ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
 436                                       offset * sizeof(u64), sizeof(pdpte),
 437                                       PFERR_USER_MASK|PFERR_WRITE_MASK);
 438         if (ret < 0) {
 439                 ret = 0;
 440                 goto out;
 441         }
 442         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 443                 if (is_present_gpte(pdpte[i]) &&
 444                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 445                         ret = 0;
 446                         goto out;
 447                 }
 448         }
 449         ret = 1;
 450
 451         memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
 452         __set_bit(VCPU_EXREG_PDPTR,
 453                   (unsigned long *)&vcpu->arch.regs_avail);
 454         __set_bit(VCPU_EXREG_PDPTR,
 455                   (unsigned long *)&vcpu->arch.regs_dirty);
 456 out:
 457
 458         return ret;
 459 }
 460 EXPORT_SYMBOL_GPL(load_pdptrs);
 461
 462 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 463 {
 464         u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
 465         bool changed = true;
 466         int offset;
 467         gfn_t gfn;
 468         int r;
 469
 470         if (is_long_mode(vcpu) || !is_pae(vcpu))
 471                 return false;
 472
 473         if (!test_bit(VCPU_EXREG_PDPTR,
 474                       (unsigned long *)&vcpu->arch.regs_avail))
 475                 return true;
 476
 477         gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
 478         offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
 479         r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
 480                                        PFERR_USER_MASK | PFERR_WRITE_MASK);
 481         if (r < 0)
 482                 goto out;
 483         changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
 484 out:
 485
 486         return changed;
 487 }
 488
 489 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 490 {
 491         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 492         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
 493                                     X86_CR0_CD | X86_CR0_NW;
 494
 495         cr0 |= X86_CR0_ET;
 496
 497 #ifdef CONFIG_X86_64
 498         if (cr0 & 0xffffffff00000000UL)
 499                 return 1;
 500 #endif
 501
 502         cr0 &= ~CR0_RESERVED_BITS;
 503
 504         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 505                 return 1;
 506
 507         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 508                 return 1;
 509
 510         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 511 #ifdef CONFIG_X86_64
 512                 if ((vcpu->arch.efer & EFER_LME)) {
 513                         int cs_db, cs_l;
 514
 515                         if (!is_pae(vcpu))
 516                                 return 1;
 517                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 518                         if (cs_l)
 519                                 return 1;
 520                 } else
 521 #endif
 522                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 523                                                  kvm_read_cr3(vcpu)))
 524                         return 1;
 525         }
 526
 527         kvm_x86_ops->set_cr0(vcpu, cr0);
 528
 529         if ((cr0 ^ old_cr0) & X86_CR0_PG) {
 530                 kvm_clear_async_pf_completion_queue(vcpu);
 531                 kvm_async_pf_hash_reset(vcpu);
 532         }
 533
 534         if ((cr0 ^ old_cr0) & update_bits)
 535                 kvm_mmu_reset_context(vcpu);
 536         return 0;
 537 }
 538 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 539
 540 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 541 {
 542         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 543 }
 544 EXPORT_SYMBOL_GPL(kvm_lmsw);
 545
 546 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 547 {
 548         u64 xcr0;
 549
 550         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
 551         if (index != XCR_XFEATURE_ENABLED_MASK)
 552                 return 1;
 553         xcr0 = xcr;
 554         if (kvm_x86_ops->get_cpl(vcpu) != 0)
 555                 return 1;
 556         if (!(xcr0 & XSTATE_FP))
 557                 return 1;
 558         if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
 559                 return 1;
 560         if (xcr0 & ~host_xcr0)
 561                 return 1;
 562         vcpu->arch.xcr0 = xcr0;
 563         vcpu->guest_xcr0_loaded = 0;
 564         return 0;
 565 }
 566
 567 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 568 {
 569         if (__kvm_set_xcr(vcpu, index, xcr)) {
 570                 kvm_inject_gp(vcpu, 0);
 571                 return 1;
 572         }
 573         return 0;
 574 }
 575 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 576
 577 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
 578 {
 579         struct kvm_cpuid_entry2 *best;
 580
 581         best = kvm_find_cpuid_entry(vcpu, 1, 0);
 582         return best && (best->ecx & bit(X86_FEATURE_XSAVE));
 583 }
 584
 585 static void update_cpuid(struct kvm_vcpu *vcpu)
 586 {
 587         struct kvm_cpuid_entry2 *best;
 588
 589         best = kvm_find_cpuid_entry(vcpu, 1, 0);
 590         if (!best)
 591                 return;
 592
 593         /* Update OSXSAVE bit */
 594         if (cpu_has_xsave && best->function == 0x1) {
 595                 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
 596                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
 597                         best->ecx |= bit(X86_FEATURE_OSXSAVE);
 598         }
 599 }
 600
 601 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 602 {
 603         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 604         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 605
 606         if (cr4 & CR4_RESERVED_BITS)
 607                 return 1;
 608
 609         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
 610                 return 1;
 611
 612         if (is_long_mode(vcpu)) {
 613                 if (!(cr4 & X86_CR4_PAE))
 614                         return 1;
 615         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 616                    && ((cr4 ^ old_cr4) & pdptr_bits)
 617                    && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 618                                    kvm_read_cr3(vcpu)))
 619                 return 1;
 620
 621         if (cr4 & X86_CR4_VMXE)
 622                 return 1;
 623
 624         kvm_x86_ops->set_cr4(vcpu, cr4);
 625
 626         if ((cr4 ^ old_cr4) & pdptr_bits)
 627                 kvm_mmu_reset_context(vcpu);
 628
 629         if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
 630                 update_cpuid(vcpu);
 631
 632         return 0;
 633 }
 634 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 635
 636 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 637 {
 638         if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
 639                 kvm_mmu_sync_roots(vcpu);
 640                 kvm_mmu_flush_tlb(vcpu);
 641                 return 0;
 642         }
 643
 644         if (is_long_mode(vcpu)) {
 645                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 646                         return 1;
 647         } else {
 648                 if (is_pae(vcpu)) {
 649                         if (cr3 & CR3_PAE_RESERVED_BITS)
 650                                 return 1;
 651                         if (is_paging(vcpu) &&
 652                             !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 653                                 return 1;
 654                 }
 655                 /*
 656                  * We don't check reserved bits in nonpae mode, because
 657                  * this isn't enforced, and VMware depends on this.
 658                  */
 659         }
 660
 661         /*
 662          * Does the new cr3 value map to physical memory? (Note, we
 663          * catch an invalid cr3 even in real-mode, because it would
 664          * cause trouble later on when we turn on paging anyway.)
 665          *
 666          * A real CPU would silently accept an invalid cr3 and would
 667          * attempt to use it - with largely undefined (and often hard
 668          * to debug) behavior on the guest side.
 669          */
 670         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 671                 return 1;
 672         vcpu->arch.cr3 = cr3;
 673         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
 674         vcpu->arch.mmu.new_cr3(vcpu);
 675         return 0;
 676 }
 677 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 678
 679 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 680 {
 681         if (cr8 & CR8_RESERVED_BITS)
 682                 return 1;
 683         if (irqchip_in_kernel(vcpu->kvm))
 684                 kvm_lapic_set_tpr(vcpu, cr8);
 685         else
 686                 vcpu->arch.cr8 = cr8;
 687         return 0;
 688 }
 689 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 690
 691 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 692 {
 693         if (irqchip_in_kernel(vcpu->kvm))
 694                 return kvm_lapic_get_cr8(vcpu);
 695         else
 696                 return vcpu->arch.cr8;
 697 }
 698 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 699
 700 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 701 {
 702         switch (dr) {
 703         case 0 ... 3:
 704                 vcpu->arch.db[dr] = val;
 705                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 706                         vcpu->arch.eff_db[dr] = val;
 707                 break;
 708         case 4:
 709                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 710                         return 1; /* #UD */
 711                 /* fall through */
 712         case 6:
 713                 if (val & 0xffffffff00000000ULL)
 714                         return -1; /* #GP */
 715                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
 716                 break;
 717         case 5:
 718                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 719                         return 1; /* #UD */
 720                 /* fall through */
 721         default: /* 7 */
 722                 if (val & 0xffffffff00000000ULL)
 723                         return -1; /* #GP */
 724                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 725                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 726                         kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
 727                         vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
 728                 }
 729                 break;
 730         }
 731
 732         return 0;
 733 }
 734
 735 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 736 {
 737         int res;
 738
 739         res = __kvm_set_dr(vcpu, dr, val);
 740         if (res > 0)
 741                 kvm_queue_exception(vcpu, UD_VECTOR);
 742         else if (res < 0)
 743                 kvm_inject_gp(vcpu, 0);
 744
 745         return res;
 746 }
 747 EXPORT_SYMBOL_GPL(kvm_set_dr);
 748
 749 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 750 {
 751         switch (dr) {
 752         case 0 ... 3:
 753                 *val = vcpu->arch.db[dr];
 754                 break;
 755         case 4:
 756                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 757                         return 1;
 758                 /* fall through */
 759         case 6:
 760                 *val = vcpu->arch.dr6;
 761                 break;
 762         case 5:
 763                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 764                         return 1;
 765                 /* fall through */
 766         default: /* 7 */
 767                 *val = vcpu->arch.dr7;
 768                 break;
 769         }
 770
 771         return 0;
 772 }
 773
 774 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 775 {
 776         if (_kvm_get_dr(vcpu, dr, val)) {
 777                 kvm_queue_exception(vcpu, UD_VECTOR);
 778                 return 1;
 779         }
 780         return 0;
 781 }
 782 EXPORT_SYMBOL_GPL(kvm_get_dr);
 783
 784 /*
 785  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 786  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 787  *
 788  * This list is modified at module load time to reflect the
 789  * capabilities of the host cpu. This capabilities test skips MSRs that are
 790  * kvm-specific. Those are put in the beginning of the list.
 791  */
 792
 793 #define KVM_SAVE_MSRS_BEGIN     8
 794 static u32 msrs_to_save[] = {
 795         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 796         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 797         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 798         HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN,
 799         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 800         MSR_STAR,
 801 #ifdef CONFIG_X86_64
 802         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 803 #endif
 804         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 805 };
 806
 807 static unsigned num_msrs_to_save;
 808
 809 static u32 emulated_msrs[] = {
 810         MSR_IA32_MISC_ENABLE,
 811         MSR_IA32_MCG_STATUS,
 812         MSR_IA32_MCG_CTL,
 813 };
 814
 815 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
 816 {
 817         u64 old_efer = vcpu->arch.efer;
 818
 819         if (efer & efer_reserved_bits)
 820                 return 1;
 821
 822         if (is_paging(vcpu)
 823             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
 824                 return 1;
 825
 826         if (efer & EFER_FFXSR) {
 827                 struct kvm_cpuid_entry2 *feat;
 828
 829                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 830                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
 831                         return 1;
 832         }
 833
 834         if (efer & EFER_SVME) {
 835                 struct kvm_cpuid_entry2 *feat;
 836
 837                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 838                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
 839                         return 1;
 840         }
 841
 842         efer &= ~EFER_LMA;
 843         efer |= vcpu->arch.efer & EFER_LMA;
 844
 845         kvm_x86_ops->set_efer(vcpu, efer);
 846
 847         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 848
 849         /* Update reserved bits */
 850         if ((efer ^ old_efer) & EFER_NX)
 851                 kvm_mmu_reset_context(vcpu);
 852
 853         return 0;
 854 }
 855
 856 void kvm_enable_efer_bits(u64 mask)
 857 {
 858        efer_reserved_bits &= ~mask;
 859 }
 860 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 861
 862
 863 /*
 864  * Writes msr value into into the appropriate "register".
 865  * Returns 0 on success, non-0 otherwise.
 866  * Assumes vcpu_load() was already called.
 867  */
 868 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 869 {
 870         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 871 }
 872
 873 /*
 874  * Adapt set_msr() to msr_io()'s calling convention
 875  */
 876 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 877 {
 878         return kvm_set_msr(vcpu, index, *data);
 879 }
 880
 881 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 882 {
 883         int version;
 884         int r;
 885         struct pvclock_wall_clock wc;
 886         struct timespec boot;
 887
 888         if (!wall_clock)
 889                 return;
 890
 891         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
 892         if (r)
 893                 return;
 894
 895         if (version & 1)
 896                 ++version;  /* first time write, random junk */
 897
 898         ++version;
 899
 900         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 901
 902         /*
 903          * The guest calculates current wall clock time by adding
 904          * system time (updated by kvm_guest_time_update below) to the
 905          * wall clock specified here.  guest system time equals host
 906          * system time for us, thus we must fill in host boot time here.
 907          */
 908         getboottime(&boot);
 909
 910         wc.sec = boot.tv_sec;
 911         wc.nsec = boot.tv_nsec;
 912         wc.version = version;
 913
 914         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 915
 916         version++;
 917         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 918 }
 919
 920 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 921 {
 922         uint32_t quotient, remainder;
 923
 924         /* Don't try to replace with do_div(), this one calculates
 925          * "(dividend << 32) / divisor" */
 926         __asm__ ( "divl %4"
 927                   : "=a" (quotient), "=d" (remainder)
 928                   : "0" (0), "1" (dividend), "r" (divisor) );
 929         return quotient;
 930 }
 931
 932 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
 933                                s8 *pshift, u32 *pmultiplier)
 934 {
 935         uint64_t scaled64;
 936         int32_t  shift = 0;
 937         uint64_t tps64;
 938         uint32_t tps32;
 939
 940         tps64 = base_khz * 1000LL;
 941         scaled64 = scaled_khz * 1000LL;
 942         while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
 943                 tps64 >>= 1;
 944                 shift--;
 945         }
 946
 947         tps32 = (uint32_t)tps64;
 948         while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
 949                 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
 950                         scaled64 >>= 1;
 951                 else
 952                         tps32 <<= 1;
 953                 shift++;
 954         }
 955
 956         *pshift = shift;
 957         *pmultiplier = div_frac(scaled64, tps32);
 958
 959         pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
 960                  __func__, base_khz, scaled_khz, shift, *pmultiplier);
 961 }
 962
 963 static inline u64 get_kernel_ns(void)
 964 {
 965         struct timespec ts;
 966
 967         WARN_ON(preemptible());
 968         ktime_get_ts(&ts);
 969         monotonic_to_bootbased(&ts);
 970         return timespec_to_ns(&ts);
 971 }
 972
 973 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 974 unsigned long max_tsc_khz;
 975
 976 static inline int kvm_tsc_changes_freq(void)
 977 {
 978         int cpu = get_cpu();
 979         int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
 980                   cpufreq_quick_get(cpu) != 0;
 981         put_cpu();
 982         return ret;
 983 }
 984
 985 static inline u64 nsec_to_cycles(u64 nsec)
 986 {
 987         u64 ret;
 988
 989         WARN_ON(preemptible());
 990         if (kvm_tsc_changes_freq())
 991                 printk_once(KERN_WARNING
 992                  "kvm: unreliable cycle conversion on adjustable rate TSC\n");
 993         ret = nsec * __this_cpu_read(cpu_tsc_khz);
 994         do_div(ret, USEC_PER_SEC);
 995         return ret;
 996 }
 997
 998 static void kvm_arch_set_tsc_khz(struct kvm *kvm, u32 this_tsc_khz)
 999 {
1000         /* Compute a scale to convert nanoseconds in TSC cycles */
1001         kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1002                            &kvm->arch.virtual_tsc_shift,
1003                            &kvm->arch.virtual_tsc_mult);
1004         kvm->arch.virtual_tsc_khz = this_tsc_khz;
1005 }
1006
1007 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1008 {
1009         u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1010                                       vcpu->kvm->arch.virtual_tsc_mult,
1011                                       vcpu->kvm->arch.virtual_tsc_shift);
1012         tsc += vcpu->arch.last_tsc_write;
1013         return tsc;
1014 }
1015
1016 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1017 {
1018         struct kvm *kvm = vcpu->kvm;
1019         u64 offset, ns, elapsed;
1020         unsigned long flags;
1021         s64 sdiff;
1022
1023         raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1024         offset = data - native_read_tsc();
1025         ns = get_kernel_ns();
1026         elapsed = ns - kvm->arch.last_tsc_nsec;
1027         sdiff = data - kvm->arch.last_tsc_write;
1028         if (sdiff < 0)
1029                 sdiff = -sdiff;
1030
1031         /*
1032          * Special case: close write to TSC within 5 seconds of
1033          * another CPU is interpreted as an attempt to synchronize
1034          * The 5 seconds is to accommodate host load / swapping as
1035          * well as any reset of TSC during the boot process.
1036          *
1037          * In that case, for a reliable TSC, we can match TSC offsets,
1038          * or make a best guest using elapsed value.
1039          */
1040         if (sdiff < nsec_to_cycles(5ULL * NSEC_PER_SEC) &&
1041             elapsed < 5ULL * NSEC_PER_SEC) {
1042                 if (!check_tsc_unstable()) {
1043                         offset = kvm->arch.last_tsc_offset;
1044                         pr_debug("kvm: matched tsc offset for %llu\n", data);
1045                 } else {
1046                         u64 delta = nsec_to_cycles(elapsed);
1047                         offset += delta;
1048                         pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1049                 }
1050                 ns = kvm->arch.last_tsc_nsec;
1051         }
1052         kvm->arch.last_tsc_nsec = ns;
1053         kvm->arch.last_tsc_write = data;
1054         kvm->arch.last_tsc_offset = offset;
1055         kvm_x86_ops->write_tsc_offset(vcpu, offset);
1056         raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1057
1058         /* Reset of TSC must disable overshoot protection below */
1059         vcpu->arch.hv_clock.tsc_timestamp = 0;
1060         vcpu->arch.last_tsc_write = data;
1061         vcpu->arch.last_tsc_nsec = ns;
1062 }
1063 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1064
1065 static int kvm_guest_time_update(struct kvm_vcpu *v)
1066 {
1067         unsigned long flags;
1068         struct kvm_vcpu_arch *vcpu = &v->arch;
1069         void *shared_kaddr;
1070         unsigned long this_tsc_khz;
1071         s64 kernel_ns, max_kernel_ns;
1072         u64 tsc_timestamp;
1073
1074         /* Keep irq disabled to prevent changes to the clock */
1075         local_irq_save(flags);
1076         kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1077         kernel_ns = get_kernel_ns();
1078         this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1079
1080         if (unlikely(this_tsc_khz == 0)) {
1081                 local_irq_restore(flags);
1082                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1083                 return 1;
1084         }
1085
1086         /*
1087          * We may have to catch up the TSC to match elapsed wall clock
1088          * time for two reasons, even if kvmclock is used.
1089          *   1) CPU could have been running below the maximum TSC rate
1090          *   2) Broken TSC compensation resets the base at each VCPU
1091          *      entry to avoid unknown leaps of TSC even when running
1092          *      again on the same CPU.  This may cause apparent elapsed
1093          *      time to disappear, and the guest to stand still or run
1094          *      very slowly.
1095          */
1096         if (vcpu->tsc_catchup) {
1097                 u64 tsc = compute_guest_tsc(v, kernel_ns);
1098                 if (tsc > tsc_timestamp) {
1099                         kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1100                         tsc_timestamp = tsc;
1101                 }
1102         }
1103
1104         local_irq_restore(flags);
1105
1106         if (!vcpu->time_page)
1107                 return 0;
1108
1109         /*
1110          * Time as measured by the TSC may go backwards when resetting the base
1111          * tsc_timestamp.  The reason for this is that the TSC resolution is
1112          * higher than the resolution of the other clock scales.  Thus, many
1113          * possible measurments of the TSC correspond to one measurement of any
1114          * other clock, and so a spread of values is possible.  This is not a
1115          * problem for the computation of the nanosecond clock; with TSC rates
1116          * around 1GHZ, there can only be a few cycles which correspond to one
1117          * nanosecond value, and any path through this code will inevitably
1118          * take longer than that.  However, with the kernel_ns value itself,
1119          * the precision may be much lower, down to HZ granularity.  If the
1120          * first sampling of TSC against kernel_ns ends in the low part of the
1121          * range, and the second in the high end of the range, we can get:
1122          *
1123          * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1124          *
1125          * As the sampling errors potentially range in the thousands of cycles,
1126          * it is possible such a time value has already been observed by the
1127          * guest.  To protect against this, we must compute the system time as
1128          * observed by the guest and ensure the new system time is greater.
1129          */
1130         max_kernel_ns = 0;
1131         if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1132                 max_kernel_ns = vcpu->last_guest_tsc -
1133                                 vcpu->hv_clock.tsc_timestamp;
1134                 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1135                                     vcpu->hv_clock.tsc_to_system_mul,
1136                                     vcpu->hv_clock.tsc_shift);
1137                 max_kernel_ns += vcpu->last_kernel_ns;
1138         }
1139
1140         if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1141                 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1142                                    &vcpu->hv_clock.tsc_shift,
1143                                    &vcpu->hv_clock.tsc_to_system_mul);
1144                 vcpu->hw_tsc_khz = this_tsc_khz;
1145         }
1146
1147         if (max_kernel_ns > kernel_ns)
1148                 kernel_ns = max_kernel_ns;
1149
1150         /* With all the info we got, fill in the values */
1151         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1152         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1153         vcpu->last_kernel_ns = kernel_ns;
1154         vcpu->last_guest_tsc = tsc_timestamp;
1155         vcpu->hv_clock.flags = 0;
1156
1157         /*
1158          * The interface expects us to write an even number signaling that the
1159          * update is finished. Since the guest won't see the intermediate
1160          * state, we just increase by 2 at the end.
1161          */
1162         vcpu->hv_clock.version += 2;
1163
1164         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1165
1166         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1167                sizeof(vcpu->hv_clock));
1168
1169         kunmap_atomic(shared_kaddr, KM_USER0);
1170
1171         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1172         return 0;
1173 }
1174
1175 static bool msr_mtrr_valid(unsigned msr)
1176 {
1177         switch (msr) {
1178         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1179         case MSR_MTRRfix64K_00000:
1180         case MSR_MTRRfix16K_80000:
1181         case MSR_MTRRfix16K_A0000:
1182         case MSR_MTRRfix4K_C0000:
1183         case MSR_MTRRfix4K_C8000:
1184         case MSR_MTRRfix4K_D0000:
1185         case MSR_MTRRfix4K_D8000:
1186         case MSR_MTRRfix4K_E0000:
1187         case MSR_MTRRfix4K_E8000:
1188         case MSR_MTRRfix4K_F0000:
1189         case MSR_MTRRfix4K_F8000:
1190         case MSR_MTRRdefType:
1191         case MSR_IA32_CR_PAT:
1192                 return true;
1193         case 0x2f8:
1194                 return true;
1195         }
1196         return false;
1197 }
1198
1199 static bool valid_pat_type(unsigned t)
1200 {
1201         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1202 }
1203
1204 static bool valid_mtrr_type(unsigned t)
1205 {
1206         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1207 }
1208
1209 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1210 {
1211         int i;
1212
1213         if (!msr_mtrr_valid(msr))
1214                 return false;
1215
1216         if (msr == MSR_IA32_CR_PAT) {
1217                 for (i = 0; i < 8; i++)
1218                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
1219                                 return false;
1220                 return true;
1221         } else if (msr == MSR_MTRRdefType) {
1222                 if (data & ~0xcff)
1223                         return false;
1224                 return valid_mtrr_type(data & 0xff);
1225         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1226                 for (i = 0; i < 8 ; i++)
1227                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1228                                 return false;
1229                 return true;
1230         }
1231
1232         /* variable MTRRs */
1233         return valid_mtrr_type(data & 0xff);
1234 }
1235
1236 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1237 {
1238         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1239
1240         if (!mtrr_valid(vcpu, msr, data))
1241                 return 1;
1242
1243         if (msr == MSR_MTRRdefType) {
1244                 vcpu->arch.mtrr_state.def_type = data;
1245                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1246         } else if (msr == MSR_MTRRfix64K_00000)
1247                 p[0] = data;
1248         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1249                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1250         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1251                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1252         else if (msr == MSR_IA32_CR_PAT)
1253                 vcpu->arch.pat = data;
1254         else {  /* Variable MTRRs */
1255                 int idx, is_mtrr_mask;
1256                 u64 *pt;
1257
1258                 idx = (msr - 0x200) / 2;
1259                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1260                 if (!is_mtrr_mask)
1261                         pt =
1262                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1263                 else
1264                         pt =
1265                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1266                 *pt = data;
1267         }
1268
1269         kvm_mmu_reset_context(vcpu);
1270         return 0;
1271 }
1272
1273 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1274 {
1275         u64 mcg_cap = vcpu->arch.mcg_cap;
1276         unsigned bank_num = mcg_cap & 0xff;
1277
1278         switch (msr) {
1279         case MSR_IA32_MCG_STATUS:
1280                 vcpu->arch.mcg_status = data;
1281                 break;
1282         case MSR_IA32_MCG_CTL:
1283                 if (!(mcg_cap & MCG_CTL_P))
1284                         return 1;
1285                 if (data != 0 && data != ~(u64)0)
1286                         return -1;
1287                 vcpu->arch.mcg_ctl = data;
1288                 break;
1289         default:
1290                 if (msr >= MSR_IA32_MC0_CTL &&
1291                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1292                         u32 offset = msr - MSR_IA32_MC0_CTL;
1293                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1294                          * some Linux kernels though clear bit 10 in bank 4 to
1295                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1296                          * this to avoid an uncatched #GP in the guest
1297                          */
1298                         if ((offset & 0x3) == 0 &&
1299                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1300                                 return -1;
1301                         vcpu->arch.mce_banks[offset] = data;
1302                         break;
1303                 }
1304                 return 1;
1305         }
1306         return 0;
1307 }
1308
1309 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1310 {
1311         struct kvm *kvm = vcpu->kvm;
1312         int lm = is_long_mode(vcpu);
1313         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1314                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1315         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1316                 : kvm->arch.xen_hvm_config.blob_size_32;
1317         u32 page_num = data & ~PAGE_MASK;
1318         u64 page_addr = data & PAGE_MASK;
1319         u8 *page;
1320         int r;
1321
1322         r = -E2BIG;
1323         if (page_num >= blob_size)
1324                 goto out;
1325         r = -ENOMEM;
1326         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1327         if (!page)
1328                 goto out;
1329         r = -EFAULT;
1330         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1331                 goto out_free;
1332         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1333                 goto out_free;
1334         r = 0;
1335 out_free:
1336         kfree(page);
1337 out:
1338         return r;
1339 }
1340
1341 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1342 {
1343         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1344 }
1345
1346 static bool kvm_hv_msr_partition_wide(u32 msr)
1347 {
1348         bool r = false;
1349         switch (msr) {
1350         case HV_X64_MSR_GUEST_OS_ID:
1351         case HV_X64_MSR_HYPERCALL:
1352                 r = true;
1353                 break;
1354         }
1355
1356         return r;
1357 }
1358
1359 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1360 {
1361         struct kvm *kvm = vcpu->kvm;
1362
1363         switch (msr) {
1364         case HV_X64_MSR_GUEST_OS_ID:
1365                 kvm->arch.hv_guest_os_id = data;
1366                 /* setting guest os id to zero disables hypercall page */
1367                 if (!kvm->arch.hv_guest_os_id)
1368                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1369                 break;
1370         case HV_X64_MSR_HYPERCALL: {
1371                 u64 gfn;
1372                 unsigned long addr;
1373                 u8 instructions[4];
1374
1375                 /* if guest os id is not set hypercall should remain disabled */
1376                 if (!kvm->arch.hv_guest_os_id)
1377                         break;
1378                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1379                         kvm->arch.hv_hypercall = data;
1380                         break;
1381                 }
1382                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1383                 addr = gfn_to_hva(kvm, gfn);
1384                 if (kvm_is_error_hva(addr))
1385                         return 1;
1386                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1387                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1388                 if (copy_to_user((void __user *)addr, instructions, 4))
1389                         return 1;
1390                 kvm->arch.hv_hypercall = data;
1391                 break;
1392         }
1393         default:
1394                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1395                           "data 0x%llx\n", msr, data);
1396                 return 1;
1397         }
1398         return 0;
1399 }
1400
1401 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1402 {
1403         switch (msr) {
1404         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1405                 unsigned long addr;
1406
1407                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1408                         vcpu->arch.hv_vapic = data;
1409                         break;
1410                 }
1411                 addr = gfn_to_hva(vcpu->kvm, data >>
1412                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1413                 if (kvm_is_error_hva(addr))
1414                         return 1;
1415                 if (clear_user((void __user *)addr, PAGE_SIZE))
1416                         return 1;
1417                 vcpu->arch.hv_vapic = data;
1418                 break;
1419         }
1420         case HV_X64_MSR_EOI:
1421                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1422         case HV_X64_MSR_ICR:
1423                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1424         case HV_X64_MSR_TPR:
1425                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1426         default:
1427                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1428                           "data 0x%llx\n", msr, data);
1429                 return 1;
1430         }
1431
1432         return 0;
1433 }
1434
1435 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1436 {
1437         gpa_t gpa = data & ~0x3f;
1438
1439         /* Bits 2:5 are resrved, Should be zero */
1440         if (data & 0x3c)
1441                 return 1;
1442
1443         vcpu->arch.apf.msr_val = data;
1444
1445         if (!(data & KVM_ASYNC_PF_ENABLED)) {
1446                 kvm_clear_async_pf_completion_queue(vcpu);
1447                 kvm_async_pf_hash_reset(vcpu);
1448                 return 0;
1449         }
1450
1451         if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1452                 return 1;
1453
1454         vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1455         kvm_async_pf_wakeup_all(vcpu);
1456         return 0;
1457 }
1458
1459 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1460 {
1461         if (vcpu->arch.time_page) {
1462                 kvm_release_page_dirty(vcpu->arch.time_page);
1463                 vcpu->arch.time_page = NULL;
1464         }
1465 }
1466
1467 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1468 {
1469         switch (msr) {
1470         case MSR_EFER:
1471                 return set_efer(vcpu, data);
1472         case MSR_K7_HWCR:
1473                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1474                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
1475                 if (data != 0) {
1476                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1477                                 data);
1478                         return 1;
1479                 }
1480                 break;
1481         case MSR_FAM10H_MMIO_CONF_BASE:
1482                 if (data != 0) {
1483                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1484                                 "0x%llx\n", data);
1485                         return 1;
1486                 }
1487                 break;
1488         case MSR_AMD64_NB_CFG:
1489                 break;
1490         case MSR_IA32_DEBUGCTLMSR:
1491                 if (!data) {
1492                         /* We support the non-activated case already */
1493                         break;
1494                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1495                         /* Values other than LBR and BTF are vendor-specific,
1496                            thus reserved and should throw a #GP */
1497                         return 1;
1498                 }
1499                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1500                         __func__, data);
1501                 break;
1502         case MSR_IA32_UCODE_REV:
1503         case MSR_IA32_UCODE_WRITE:
1504         case MSR_VM_HSAVE_PA:
1505         case MSR_AMD64_PATCH_LOADER:
1506                 break;
1507         case 0x200 ... 0x2ff:
1508                 return set_msr_mtrr(vcpu, msr, data);
1509         case MSR_IA32_APICBASE:
1510                 kvm_set_apic_base(vcpu, data);
1511                 break;
1512         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1513                 return kvm_x2apic_msr_write(vcpu, msr, data);
1514         case MSR_IA32_MISC_ENABLE:
1515                 vcpu->arch.ia32_misc_enable_msr = data;
1516                 break;
1517         case MSR_KVM_WALL_CLOCK_NEW:
1518         case MSR_KVM_WALL_CLOCK:
1519                 vcpu->kvm->arch.wall_clock = data;
1520                 kvm_write_wall_clock(vcpu->kvm, data);
1521                 break;
1522         case MSR_KVM_SYSTEM_TIME_NEW:
1523         case MSR_KVM_SYSTEM_TIME: {
1524                 kvmclock_reset(vcpu);
1525
1526                 vcpu->arch.time = data;
1527                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1528
1529                 /* we verify if the enable bit is set... */
1530                 if (!(data & 1))
1531                         break;
1532
1533                 /* ...but clean it before doing the actual write */
1534                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1535
1536                 vcpu->arch.time_page =
1537                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1538
1539                 if (is_error_page(vcpu->arch.time_page)) {
1540                         kvm_release_page_clean(vcpu->arch.time_page);
1541                         vcpu->arch.time_page = NULL;
1542                 }
1543                 break;
1544         }
1545         case MSR_KVM_ASYNC_PF_EN:
1546                 if (kvm_pv_enable_async_pf(vcpu, data))
1547                         return 1;
1548                 break;
1549         case MSR_IA32_MCG_CTL:
1550         case MSR_IA32_MCG_STATUS:
1551         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1552                 return set_msr_mce(vcpu, msr, data);
1553
1554         /* Performance counters are not protected by a CPUID bit,
1555          * so we should check all of them in the generic path for the sake of
1556          * cross vendor migration.
1557          * Writing a zero into the event select MSRs disables them,
1558          * which we perfectly emulate ;-). Any other value should be at least
1559          * reported, some guests depend on them.
1560          */
1561         case MSR_P6_EVNTSEL0:
1562         case MSR_P6_EVNTSEL1:
1563         case MSR_K7_EVNTSEL0:
1564         case MSR_K7_EVNTSEL1:
1565         case MSR_K7_EVNTSEL2:
1566         case MSR_K7_EVNTSEL3:
1567                 if (data != 0)
1568                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1569                                 "0x%x data 0x%llx\n", msr, data);
1570                 break;
1571         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1572          * so we ignore writes to make it happy.
1573          */
1574         case MSR_P6_PERFCTR0:
1575         case MSR_P6_PERFCTR1:
1576         case MSR_K7_PERFCTR0:
1577         case MSR_K7_PERFCTR1:
1578         case MSR_K7_PERFCTR2:
1579         case MSR_K7_PERFCTR3:
1580                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1581                         "0x%x data 0x%llx\n", msr, data);
1582                 break;
1583         case MSR_K7_CLK_CTL:
1584                 /*
1585                  * Ignore all writes to this no longer documented MSR.
1586                  * Writes are only relevant for old K7 processors,
1587                  * all pre-dating SVM, but a recommended workaround from
1588                  * AMD for these chips. It is possible to speicify the
1589                  * affected processor models on the command line, hence
1590                  * the need to ignore the workaround.
1591                  */
1592                 break;
1593         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1594                 if (kvm_hv_msr_partition_wide(msr)) {
1595                         int r;
1596                         mutex_lock(&vcpu->kvm->lock);
1597                         r = set_msr_hyperv_pw(vcpu, msr, data);
1598                         mutex_unlock(&vcpu->kvm->lock);
1599                         return r;
1600                 } else
1601                         return set_msr_hyperv(vcpu, msr, data);
1602                 break;
1603         case MSR_IA32_BBL_CR_CTL3:
1604                 /* Drop writes to this legacy MSR -- see rdmsr
1605                  * counterpart for further detail.
1606                  */
1607                 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1608                 break;
1609         default:
1610                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1611                         return xen_hvm_config(vcpu, data);
1612                 if (!ignore_msrs) {
1613                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1614                                 msr, data);
1615                         return 1;
1616                 } else {
1617                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1618                                 msr, data);
1619                         break;
1620                 }
1621         }
1622         return 0;
1623 }
1624 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1625
1626
1627 /*
1628  * Reads an msr value (of 'msr_index') into 'pdata'.
1629  * Returns 0 on success, non-0 otherwise.
1630  * Assumes vcpu_load() was already called.
1631  */
1632 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1633 {
1634         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1635 }
1636
1637 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1638 {
1639         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1640
1641         if (!msr_mtrr_valid(msr))
1642                 return 1;
1643
1644         if (msr == MSR_MTRRdefType)
1645                 *pdata = vcpu->arch.mtrr_state.def_type +
1646                          (vcpu->arch.mtrr_state.enabled << 10);
1647         else if (msr == MSR_MTRRfix64K_00000)
1648                 *pdata = p[0];
1649         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1650                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1651         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1652                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1653         else if (msr == MSR_IA32_CR_PAT)
1654                 *pdata = vcpu->arch.pat;
1655         else {  /* Variable MTRRs */
1656                 int idx, is_mtrr_mask;
1657                 u64 *pt;
1658
1659                 idx = (msr - 0x200) / 2;
1660                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1661                 if (!is_mtrr_mask)
1662                         pt =
1663                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1664                 else
1665                         pt =
1666                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1667                 *pdata = *pt;
1668         }
1669
1670         return 0;
1671 }
1672
1673 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1674 {
1675         u64 data;
1676         u64 mcg_cap = vcpu->arch.mcg_cap;
1677         unsigned bank_num = mcg_cap & 0xff;
1678
1679         switch (msr) {
1680         case MSR_IA32_P5_MC_ADDR:
1681         case MSR_IA32_P5_MC_TYPE:
1682                 data = 0;
1683                 break;
1684         case MSR_IA32_MCG_CAP:
1685                 data = vcpu->arch.mcg_cap;
1686                 break;
1687         case MSR_IA32_MCG_CTL:
1688                 if (!(mcg_cap & MCG_CTL_P))
1689                         return 1;
1690                 data = vcpu->arch.mcg_ctl;
1691                 break;
1692         case MSR_IA32_MCG_STATUS:
1693                 data = vcpu->arch.mcg_status;
1694                 break;
1695         default:
1696                 if (msr >= MSR_IA32_MC0_CTL &&
1697                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1698                         u32 offset = msr - MSR_IA32_MC0_CTL;
1699                         data = vcpu->arch.mce_banks[offset];
1700                         break;
1701                 }
1702                 return 1;
1703         }
1704         *pdata = data;
1705         return 0;
1706 }
1707
1708 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1709 {
1710         u64 data = 0;
1711         struct kvm *kvm = vcpu->kvm;
1712
1713         switch (msr) {
1714         case HV_X64_MSR_GUEST_OS_ID:
1715                 data = kvm->arch.hv_guest_os_id;
1716                 break;
1717         case HV_X64_MSR_HYPERCALL:
1718                 data = kvm->arch.hv_hypercall;
1719                 break;
1720         default:
1721                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1722                 return 1;
1723         }
1724
1725         *pdata = data;
1726         return 0;
1727 }
1728
1729 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1730 {
1731         u64 data = 0;
1732
1733         switch (msr) {
1734         case HV_X64_MSR_VP_INDEX: {
1735                 int r;
1736                 struct kvm_vcpu *v;
1737                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1738                         if (v == vcpu)
1739                                 data = r;
1740                 break;
1741         }
1742         case HV_X64_MSR_EOI:
1743                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1744         case HV_X64_MSR_ICR:
1745                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1746         case HV_X64_MSR_TPR:
1747                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1748         default:
1749                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1750                 return 1;
1751         }
1752         *pdata = data;
1753         return 0;
1754 }
1755
1756 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1757 {
1758         u64 data;
1759
1760         switch (msr) {
1761         case MSR_IA32_PLATFORM_ID:
1762         case MSR_IA32_UCODE_REV:
1763         case MSR_IA32_EBL_CR_POWERON:
1764         case MSR_IA32_DEBUGCTLMSR:
1765         case MSR_IA32_LASTBRANCHFROMIP:
1766         case MSR_IA32_LASTBRANCHTOIP:
1767         case MSR_IA32_LASTINTFROMIP:
1768         case MSR_IA32_LASTINTTOIP:
1769         case MSR_K8_SYSCFG:
1770         case MSR_K7_HWCR:
1771         case MSR_VM_HSAVE_PA:
1772         case MSR_P6_PERFCTR0:
1773         case MSR_P6_PERFCTR1:
1774         case MSR_P6_EVNTSEL0:
1775         case MSR_P6_EVNTSEL1:
1776         case MSR_K7_EVNTSEL0:
1777         case MSR_K7_PERFCTR0:
1778         case MSR_K8_INT_PENDING_MSG:
1779         case MSR_AMD64_NB_CFG:
1780         case MSR_FAM10H_MMIO_CONF_BASE:
1781                 data = 0;
1782                 break;
1783         case MSR_MTRRcap:
1784                 data = 0x500 | KVM_NR_VAR_MTRR;
1785                 break;
1786         case 0x200 ... 0x2ff:
1787                 return get_msr_mtrr(vcpu, msr, pdata);
1788         case 0xcd: /* fsb frequency */
1789                 data = 3;
1790                 break;
1791                 /*
1792                  * MSR_EBC_FREQUENCY_ID
1793                  * Conservative value valid for even the basic CPU models.
1794                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1795                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1796                  * and 266MHz for model 3, or 4. Set Core Clock
1797                  * Frequency to System Bus Frequency Ratio to 1 (bits
1798                  * 31:24) even though these are only valid for CPU
1799                  * models > 2, however guests may end up dividing or
1800                  * multiplying by zero otherwise.
1801                  */
1802         case MSR_EBC_FREQUENCY_ID:
1803                 data = 1 << 24;
1804                 break;
1805         case MSR_IA32_APICBASE:
1806                 data = kvm_get_apic_base(vcpu);
1807                 break;
1808         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1809                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1810                 break;
1811         case MSR_IA32_MISC_ENABLE:
1812                 data = vcpu->arch.ia32_misc_enable_msr;
1813                 break;
1814         case MSR_IA32_PERF_STATUS:
1815                 /* TSC increment by tick */
1816                 data = 1000ULL;
1817                 /* CPU multiplier */
1818                 data |= (((uint64_t)4ULL) << 40);
1819                 break;
1820         case MSR_EFER:
1821                 data = vcpu->arch.efer;
1822                 break;
1823         case MSR_KVM_WALL_CLOCK:
1824         case MSR_KVM_WALL_CLOCK_NEW:
1825                 data = vcpu->kvm->arch.wall_clock;
1826                 break;
1827         case MSR_KVM_SYSTEM_TIME:
1828         case MSR_KVM_SYSTEM_TIME_NEW:
1829                 data = vcpu->arch.time;
1830                 break;
1831         case MSR_KVM_ASYNC_PF_EN:
1832                 data = vcpu->arch.apf.msr_val;
1833                 break;
1834         case MSR_IA32_P5_MC_ADDR:
1835         case MSR_IA32_P5_MC_TYPE:
1836         case MSR_IA32_MCG_CAP:
1837         case MSR_IA32_MCG_CTL:
1838         case MSR_IA32_MCG_STATUS:
1839         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1840                 return get_msr_mce(vcpu, msr, pdata);
1841         case MSR_K7_CLK_CTL:
1842                 /*
1843                  * Provide expected ramp-up count for K7. All other
1844                  * are set to zero, indicating minimum divisors for
1845                  * every field.
1846                  *
1847                  * This prevents guest kernels on AMD host with CPU
1848                  * type 6, model 8 and higher from exploding due to
1849                  * the rdmsr failing.
1850                  */
1851                 data = 0x20000000;
1852                 break;
1853         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1854                 if (kvm_hv_msr_partition_wide(msr)) {
1855                         int r;
1856                         mutex_lock(&vcpu->kvm->lock);
1857                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1858                         mutex_unlock(&vcpu->kvm->lock);
1859                         return r;
1860                 } else
1861                         return get_msr_hyperv(vcpu, msr, pdata);
1862                 break;
1863         case MSR_IA32_BBL_CR_CTL3:
1864                 /* This legacy MSR exists but isn't fully documented in current
1865                  * silicon.  It is however accessed by winxp in very narrow
1866                  * scenarios where it sets bit #19, itself documented as
1867                  * a "reserved" bit.  Best effort attempt to source coherent
1868                  * read data here should the balance of the register be
1869                  * interpreted by the guest:
1870                  *
1871                  * L2 cache control register 3: 64GB range, 256KB size,
1872                  * enabled, latency 0x1, configured
1873                  */
1874                 data = 0xbe702111;
1875                 break;
1876         default:
1877                 if (!ignore_msrs) {
1878                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1879                         return 1;
1880                 } else {
1881                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1882                         data = 0;
1883                 }
1884                 break;
1885         }
1886         *pdata = data;
1887         return 0;
1888 }
1889 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1890
1891 /*
1892  * Read or write a bunch of msrs. All parameters are kernel addresses.
1893  *
1894  * @return number of msrs set successfully.
1895  */
1896 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1897                     struct kvm_msr_entry *entries,
1898                     int (*do_msr)(struct kvm_vcpu *vcpu,
1899                                   unsigned index, u64 *data))
1900 {
1901         int i, idx;
1902
1903         idx = srcu_read_lock(&vcpu->kvm->srcu);
1904         for (i = 0; i < msrs->nmsrs; ++i)
1905                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1906                         break;
1907         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1908
1909         return i;
1910 }
1911
1912 /*
1913  * Read or write a bunch of msrs. Parameters are user addresses.
1914  *
1915  * @return number of msrs set successfully.
1916  */
1917 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1918                   int (*do_msr)(struct kvm_vcpu *vcpu,
1919                                 unsigned index, u64 *data),
1920                   int writeback)
1921 {
1922         struct kvm_msrs msrs;
1923         struct kvm_msr_entry *entries;
1924         int r, n;
1925         unsigned size;
1926
1927         r = -EFAULT;
1928         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1929                 goto out;
1930
1931         r = -E2BIG;
1932         if (msrs.nmsrs >= MAX_IO_MSRS)
1933                 goto out;
1934
1935         r = -ENOMEM;
1936         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1937         entries = kmalloc(size, GFP_KERNEL);
1938         if (!entries)
1939                 goto out;
1940
1941         r = -EFAULT;
1942         if (copy_from_user(entries, user_msrs->entries, size))
1943                 goto out_free;
1944
1945         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1946         if (r < 0)
1947                 goto out_free;
1948
1949         r = -EFAULT;
1950         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1951                 goto out_free;
1952
1953         r = n;
1954
1955 out_free:
1956         kfree(entries);
1957 out:
1958         return r;
1959 }
1960
1961 int kvm_dev_ioctl_check_extension(long ext)
1962 {
1963         int r;
1964
1965         switch (ext) {
1966         case KVM_CAP_IRQCHIP:
1967         case KVM_CAP_HLT:
1968         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1969         case KVM_CAP_SET_TSS_ADDR:
1970         case KVM_CAP_EXT_CPUID:
1971         case KVM_CAP_CLOCKSOURCE:
1972         case KVM_CAP_PIT:
1973         case KVM_CAP_NOP_IO_DELAY:
1974         case KVM_CAP_MP_STATE:
1975         case KVM_CAP_SYNC_MMU:
1976         case KVM_CAP_USER_NMI:
1977         case KVM_CAP_REINJECT_CONTROL:
1978         case KVM_CAP_IRQ_INJECT_STATUS:
1979         case KVM_CAP_ASSIGN_DEV_IRQ:
1980         case KVM_CAP_IRQFD:
1981         case KVM_CAP_IOEVENTFD:
1982         case KVM_CAP_PIT2:
1983         case KVM_CAP_PIT_STATE2:
1984         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1985         case KVM_CAP_XEN_HVM:
1986         case KVM_CAP_ADJUST_CLOCK:
1987         case KVM_CAP_VCPU_EVENTS:
1988         case KVM_CAP_HYPERV:
1989         case KVM_CAP_HYPERV_VAPIC:
1990         case KVM_CAP_HYPERV_SPIN:
1991         case KVM_CAP_PCI_SEGMENT:
1992         case KVM_CAP_DEBUGREGS:
1993         case KVM_CAP_X86_ROBUST_SINGLESTEP:
1994         case KVM_CAP_XSAVE:
1995         case KVM_CAP_ASYNC_PF:
1996                 r = 1;
1997                 break;
1998         case KVM_CAP_COALESCED_MMIO:
1999                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2000                 break;
2001         case KVM_CAP_VAPIC:
2002                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2003                 break;
2004         case KVM_CAP_NR_VCPUS:
2005                 r = KVM_MAX_VCPUS;
2006                 break;
2007         case KVM_CAP_NR_MEMSLOTS:
2008                 r = KVM_MEMORY_SLOTS;
2009                 break;
2010         case KVM_CAP_PV_MMU:    /* obsolete */
2011                 r = 0;
2012                 break;
2013         case KVM_CAP_IOMMU:
2014                 r = iommu_found();
2015                 break;
2016         case KVM_CAP_MCE:
2017                 r = KVM_MAX_MCE_BANKS;
2018                 break;
2019         case KVM_CAP_XCRS:
2020                 r = cpu_has_xsave;
2021                 break;
2022         default:
2023                 r = 0;
2024                 break;
2025         }
2026         return r;
2027
2028 }
2029
2030 long kvm_arch_dev_ioctl(struct file *filp,
2031                         unsigned int ioctl, unsigned long arg)
2032 {
2033         void __user *argp = (void __user *)arg;
2034         long r;
2035
2036         switch (ioctl) {
2037         case KVM_GET_MSR_INDEX_LIST: {
2038                 struct kvm_msr_list __user *user_msr_list = argp;
2039                 struct kvm_msr_list msr_list;
2040                 unsigned n;
2041
2042                 r = -EFAULT;
2043                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2044                         goto out;
2045                 n = msr_list.nmsrs;
2046                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2047                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2048                         goto out;
2049                 r = -E2BIG;
2050                 if (n < msr_list.nmsrs)
2051                         goto out;
2052                 r = -EFAULT;
2053                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2054                                  num_msrs_to_save * sizeof(u32)))
2055                         goto out;
2056                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2057                                  &emulated_msrs,
2058                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2059                         goto out;
2060                 r = 0;
2061                 break;
2062         }
2063         case KVM_GET_SUPPORTED_CPUID: {
2064                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2065                 struct kvm_cpuid2 cpuid;
2066
2067                 r = -EFAULT;
2068                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2069                         goto out;
2070                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2071                                                       cpuid_arg->entries);
2072                 if (r)
2073                         goto out;
2074
2075                 r = -EFAULT;
2076                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2077                         goto out;
2078                 r = 0;
2079                 break;
2080         }
2081         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2082                 u64 mce_cap;
2083
2084                 mce_cap = KVM_MCE_CAP_SUPPORTED;
2085                 r = -EFAULT;
2086                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2087                         goto out;
2088                 r = 0;
2089                 break;
2090         }
2091         default:
2092                 r = -EINVAL;
2093         }
2094 out:
2095         return r;
2096 }
2097
2098 static void wbinvd_ipi(void *garbage)
2099 {
2100         wbinvd();
2101 }
2102
2103 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2104 {
2105         return vcpu->kvm->arch.iommu_domain &&
2106                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2107 }
2108
2109 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2110 {
2111         /* Address WBINVD may be executed by guest */
2112         if (need_emulate_wbinvd(vcpu)) {
2113                 if (kvm_x86_ops->has_wbinvd_exit())
2114                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2115                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2116                         smp_call_function_single(vcpu->cpu,
2117                                         wbinvd_ipi, NULL, 1);
2118         }
2119
2120         kvm_x86_ops->vcpu_load(vcpu, cpu);
2121         if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2122                 /* Make sure TSC doesn't go backwards */
2123                 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2124                                 native_read_tsc() - vcpu->arch.last_host_tsc;
2125                 if (tsc_delta < 0)
2126                         mark_tsc_unstable("KVM discovered backwards TSC");
2127                 if (check_tsc_unstable()) {
2128                         kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2129                         vcpu->arch.tsc_catchup = 1;
2130                 }
2131                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2132                 if (vcpu->cpu != cpu)
2133                         kvm_migrate_timers(vcpu);
2134                 vcpu->cpu = cpu;
2135         }
2136 }
2137
2138 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2139 {
2140         kvm_x86_ops->vcpu_put(vcpu);
2141         kvm_put_guest_fpu(vcpu);
2142         vcpu->arch.last_host_tsc = native_read_tsc();
2143 }
2144
2145 static int is_efer_nx(void)
2146 {
2147         unsigned long long efer = 0;
2148
2149         rdmsrl_safe(MSR_EFER, &efer);
2150         return efer & EFER_NX;
2151 }
2152
2153 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2154 {
2155         int i;
2156         struct kvm_cpuid_entry2 *e, *entry;
2157
2158         entry = NULL;
2159         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2160                 e = &vcpu->arch.cpuid_entries[i];
2161                 if (e->function == 0x80000001) {
2162                         entry = e;
2163                         break;
2164                 }
2165         }
2166         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2167                 entry->edx &= ~(1 << 20);
2168                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2169         }
2170 }
2171
2172 /* when an old userspace process fills a new kernel module */
2173 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2174                                     struct kvm_cpuid *cpuid,
2175                                     struct kvm_cpuid_entry __user *entries)
2176 {
2177         int r, i;
2178         struct kvm_cpuid_entry *cpuid_entries;
2179
2180         r = -E2BIG;
2181         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2182                 goto out;
2183         r = -ENOMEM;
2184         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2185         if (!cpuid_entries)
2186                 goto out;
2187         r = -EFAULT;
2188         if (copy_from_user(cpuid_entries, entries,
2189                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2190                 goto out_free;
2191         for (i = 0; i < cpuid->nent; i++) {
2192                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2193                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2194                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2195                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2196                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2197                 vcpu->arch.cpuid_entries[i].index = 0;
2198                 vcpu->arch.cpuid_entries[i].flags = 0;
2199                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2200                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2201                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2202         }
2203         vcpu->arch.cpuid_nent = cpuid->nent;
2204         cpuid_fix_nx_cap(vcpu);
2205         r = 0;
2206         kvm_apic_set_version(vcpu);
2207         kvm_x86_ops->cpuid_update(vcpu);
2208         update_cpuid(vcpu);
2209
2210 out_free:
2211         vfree(cpuid_entries);
2212 out:
2213         return r;
2214 }
2215
2216 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2217                                      struct kvm_cpuid2 *cpuid,
2218                                      struct kvm_cpuid_entry2 __user *entries)
2219 {
2220         int r;
2221
2222         r = -E2BIG;
2223         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2224                 goto out;
2225         r = -EFAULT;
2226         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2227                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2228                 goto out;
2229         vcpu->arch.cpuid_nent = cpuid->nent;
2230         kvm_apic_set_version(vcpu);
2231         kvm_x86_ops->cpuid_update(vcpu);
2232         update_cpuid(vcpu);
2233         return 0;
2234
2235 out:
2236         return r;
2237 }
2238
2239 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2240                                      struct kvm_cpuid2 *cpuid,
2241                                      struct kvm_cpuid_entry2 __user *entries)
2242 {
2243         int r;
2244
2245         r = -E2BIG;
2246         if (cpuid->nent < vcpu->arch.cpuid_nent)
2247                 goto out;
2248         r = -EFAULT;
2249         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2250                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2251                 goto out;
2252         return 0;
2253
2254 out:
2255         cpuid->nent = vcpu->arch.cpuid_nent;
2256         return r;
2257 }
2258
2259 static void cpuid_mask(u32 *word, int wordnum)
2260 {
2261         *word &= boot_cpu_data.x86_capability[wordnum];
2262 }
2263
2264 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2265                            u32 index)
2266 {
2267         entry->function = function;
2268         entry->index = index;
2269         cpuid_count(entry->function, entry->index,
2270                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2271         entry->flags = 0;
2272 }
2273
2274 #define F(x) bit(X86_FEATURE_##x)
2275
2276 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2277                          u32 index, int *nent, int maxnent)
2278 {
2279         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2280 #ifdef CONFIG_X86_64
2281         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2282                                 ? F(GBPAGES) : 0;
2283         unsigned f_lm = F(LM);
2284 #else
2285         unsigned f_gbpages = 0;
2286         unsigned f_lm = 0;
2287 #endif
2288         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2289
2290         /* cpuid 1.edx */
2291         const u32 kvm_supported_word0_x86_features =
2292                 F(FPU) | F(VME) | F(DE) | F(PSE) |
2293                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2294                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2295                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2296                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2297                 0 /* Reserved, DS, ACPI */ | F(MMX) |
2298                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2299                 0 /* HTT, TM, Reserved, PBE */;
2300         /* cpuid 0x80000001.edx */
2301         const u32 kvm_supported_word1_x86_features =
2302                 F(FPU) | F(VME) | F(DE) | F(PSE) |
2303                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2304                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2305                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2306                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2307                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2308                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2309                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2310         /* cpuid 1.ecx */
2311         const u32 kvm_supported_word4_x86_features =
2312                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2313                 0 /* DS-CPL, VMX, SMX, EST */ |
2314                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2315                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2316                 0 /* Reserved, DCA */ | F(XMM4_1) |
2317                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2318                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2319                 F(F16C);
2320         /* cpuid 0x80000001.ecx */
2321         const u32 kvm_supported_word6_x86_features =
2322                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2323                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2324                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2325                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2326
2327         /* all calls to cpuid_count() should be made on the same cpu */
2328         get_cpu();
2329         do_cpuid_1_ent(entry, function, index);
2330         ++*nent;
2331
2332         switch (function) {
2333         case 0:
2334                 entry->eax = min(entry->eax, (u32)0xd);
2335                 break;
2336         case 1:
2337                 entry->edx &= kvm_supported_word0_x86_features;
2338                 cpuid_mask(&entry->edx, 0);
2339                 entry->ecx &= kvm_supported_word4_x86_features;
2340                 cpuid_mask(&entry->ecx, 4);
2341                 /* we support x2apic emulation even if host does not support
2342                  * it since we emulate x2apic in software */
2343                 entry->ecx |= F(X2APIC);
2344                 break;
2345         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2346          * may return different values. This forces us to get_cpu() before
2347          * issuing the first command, and also to emulate this annoying behavior
2348          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2349         case 2: {
2350                 int t, times = entry->eax & 0xff;
2351
2352                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2353                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2354                 for (t = 1; t < times && *nent < maxnent; ++t) {
2355                         do_cpuid_1_ent(&entry[t], function, 0);
2356                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2357                         ++*nent;
2358                 }
2359                 break;
2360         }
2361         /* function 4 and 0xb have additional index. */
2362         case 4: {
2363                 int i, cache_type;
2364
2365                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2366                 /* read more entries until cache_type is zero */
2367                 for (i = 1; *nent < maxnent; ++i) {
2368                         cache_type = entry[i - 1].eax & 0x1f;
2369                         if (!cache_type)
2370                                 break;
2371                         do_cpuid_1_ent(&entry[i], function, i);
2372                         entry[i].flags |=
2373                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2374                         ++*nent;
2375                 }
2376                 break;
2377         }
2378         case 0xb: {
2379                 int i, level_type;
2380
2381                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2382                 /* read more entries until level_type is zero */
2383                 for (i = 1; *nent < maxnent; ++i) {
2384                         level_type = entry[i - 1].ecx & 0xff00;
2385                         if (!level_type)
2386                                 break;
2387                         do_cpuid_1_ent(&entry[i], function, i);
2388                         entry[i].flags |=
2389                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2390                         ++*nent;
2391                 }
2392                 break;
2393         }
2394         case 0xd: {
2395                 int i;
2396
2397                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2398                 for (i = 1; *nent < maxnent && i < 64; ++i) {
2399                         if (entry[i].eax == 0)
2400                                 continue;
2401                         do_cpuid_1_ent(&entry[i], function, i);
2402                         entry[i].flags |=
2403                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2404                         ++*nent;
2405                 }
2406                 break;
2407         }
2408         case KVM_CPUID_SIGNATURE: {
2409                 char signature[12] = "KVMKVMKVM\0\0";
2410                 u32 *sigptr = (u32 *)signature;
2411                 entry->eax = 0;
2412                 entry->ebx = sigptr[0];
2413                 entry->ecx = sigptr[1];
2414                 entry->edx = sigptr[2];
2415                 break;
2416         }
2417         case KVM_CPUID_FEATURES:
2418                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2419                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
2420                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
2421                              (1 << KVM_FEATURE_ASYNC_PF) |
2422                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2423                 entry->ebx = 0;
2424                 entry->ecx = 0;
2425                 entry->edx = 0;
2426                 break;
2427         case 0x80000000:
2428                 entry->eax = min(entry->eax, 0x8000001a);
2429                 break;
2430         case 0x80000001:
2431                 entry->edx &= kvm_supported_word1_x86_features;
2432                 cpuid_mask(&entry->edx, 1);
2433                 entry->ecx &= kvm_supported_word6_x86_features;
2434                 cpuid_mask(&entry->ecx, 6);
2435                 break;
2436         }
2437
2438         kvm_x86_ops->set_supported_cpuid(function, entry);
2439
2440         put_cpu();
2441 }
2442
2443 #undef F
2444
2445 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2446                                      struct kvm_cpuid_entry2 __user *entries)
2447 {
2448         struct kvm_cpuid_entry2 *cpuid_entries;
2449         int limit, nent = 0, r = -E2BIG;
2450         u32 func;
2451
2452         if (cpuid->nent < 1)
2453                 goto out;
2454         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2455                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2456         r = -ENOMEM;
2457         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2458         if (!cpuid_entries)
2459                 goto out;
2460
2461         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2462         limit = cpuid_entries[0].eax;
2463         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2464                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2465                              &nent, cpuid->nent);
2466         r = -E2BIG;
2467         if (nent >= cpuid->nent)
2468                 goto out_free;
2469
2470         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2471         limit = cpuid_entries[nent - 1].eax;
2472         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2473                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2474                              &nent, cpuid->nent);
2475
2476
2477
2478         r = -E2BIG;
2479         if (nent >= cpuid->nent)
2480                 goto out_free;
2481
2482         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2483                      cpuid->nent);
2484
2485         r = -E2BIG;
2486         if (nent >= cpuid->nent)
2487                 goto out_free;
2488
2489         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2490                      cpuid->nent);
2491
2492         r = -E2BIG;
2493         if (nent >= cpuid->nent)
2494                 goto out_free;
2495
2496         r = -EFAULT;
2497         if (copy_to_user(entries, cpuid_entries,
2498                          nent * sizeof(struct kvm_cpuid_entry2)))
2499                 goto out_free;
2500         cpuid->nent = nent;
2501         r = 0;
2502
2503 out_free:
2504         vfree(cpuid_entries);
2505 out:
2506         return r;
2507 }
2508
2509 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2510                                     struct kvm_lapic_state *s)
2511 {
2512         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2513
2514         return 0;
2515 }
2516
2517 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2518                                     struct kvm_lapic_state *s)
2519 {
2520         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2521         kvm_apic_post_state_restore(vcpu);
2522         update_cr8_intercept(vcpu);
2523
2524         return 0;
2525 }
2526
2527 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2528                                     struct kvm_interrupt *irq)
2529 {
2530         if (irq->irq < 0 || irq->irq >= 256)
2531                 return -EINVAL;
2532         if (irqchip_in_kernel(vcpu->kvm))
2533                 return -ENXIO;
2534
2535         kvm_queue_interrupt(vcpu, irq->irq, false);
2536         kvm_make_request(KVM_REQ_EVENT, vcpu);
2537
2538         return 0;
2539 }
2540
2541 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2542 {
2543         kvm_inject_nmi(vcpu);
2544
2545         return 0;
2546 }
2547
2548 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2549                                            struct kvm_tpr_access_ctl *tac)
2550 {
2551         if (tac->flags)
2552                 return -EINVAL;
2553         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2554         return 0;
2555 }
2556
2557 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2558                                         u64 mcg_cap)
2559 {
2560         int r;
2561         unsigned bank_num = mcg_cap & 0xff, bank;
2562
2563         r = -EINVAL;
2564         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2565                 goto out;
2566         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2567                 goto out;
2568         r = 0;
2569         vcpu->arch.mcg_cap = mcg_cap;
2570         /* Init IA32_MCG_CTL to all 1s */
2571         if (mcg_cap & MCG_CTL_P)
2572                 vcpu->arch.mcg_ctl = ~(u64)0;
2573         /* Init IA32_MCi_CTL to all 1s */
2574         for (bank = 0; bank < bank_num; bank++)
2575                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2576 out:
2577         return r;
2578 }
2579
2580 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2581                                       struct kvm_x86_mce *mce)
2582 {
2583         u64 mcg_cap = vcpu->arch.mcg_cap;
2584         unsigned bank_num = mcg_cap & 0xff;
2585         u64 *banks = vcpu->arch.mce_banks;
2586
2587         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2588                 return -EINVAL;
2589         /*
2590          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2591          * reporting is disabled
2592          */
2593         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2594             vcpu->arch.mcg_ctl != ~(u64)0)
2595                 return 0;
2596         banks += 4 * mce->bank;
2597         /*
2598          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2599          * reporting is disabled for the bank
2600          */
2601         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2602                 return 0;
2603         if (mce->status & MCI_STATUS_UC) {
2604                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2605                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2606                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2607                         return 0;
2608                 }
2609                 if (banks[1] & MCI_STATUS_VAL)
2610                         mce->status |= MCI_STATUS_OVER;
2611                 banks[2] = mce->addr;
2612                 banks[3] = mce->misc;
2613                 vcpu->arch.mcg_status = mce->mcg_status;
2614                 banks[1] = mce->status;
2615                 kvm_queue_exception(vcpu, MC_VECTOR);
2616         } else if (!(banks[1] & MCI_STATUS_VAL)
2617                    || !(banks[1] & MCI_STATUS_UC)) {
2618                 if (banks[1] & MCI_STATUS_VAL)
2619                         mce->status |= MCI_STATUS_OVER;
2620                 banks[2] = mce->addr;
2621                 banks[3] = mce->misc;
2622                 banks[1] = mce->status;
2623         } else
2624                 banks[1] |= MCI_STATUS_OVER;
2625         return 0;
2626 }
2627
2628 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2629                                                struct kvm_vcpu_events *events)
2630 {
2631         events->exception.injected =
2632                 vcpu->arch.exception.pending &&
2633                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2634         events->exception.nr = vcpu->arch.exception.nr;
2635         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2636         events->exception.pad = 0;
2637         events->exception.error_code = vcpu->arch.exception.error_code;
2638
2639         events->interrupt.injected =
2640                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2641         events->interrupt.nr = vcpu->arch.interrupt.nr;
2642         events->interrupt.soft = 0;
2643         events->interrupt.shadow =
2644                 kvm_x86_ops->get_interrupt_shadow(vcpu,
2645                         KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2646
2647         events->nmi.injected = vcpu->arch.nmi_injected;
2648         events->nmi.pending = vcpu->arch.nmi_pending;
2649         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2650         events->nmi.pad = 0;
2651
2652         events->sipi_vector = vcpu->arch.sipi_vector;
2653
2654         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2655                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2656                          | KVM_VCPUEVENT_VALID_SHADOW);
2657         memset(&events->reserved, 0, sizeof(events->reserved));
2658 }
2659
2660 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2661                                               struct kvm_vcpu_events *events)
2662 {
2663         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2664                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2665                               | KVM_VCPUEVENT_VALID_SHADOW))
2666                 return -EINVAL;
2667
2668         vcpu->arch.exception.pending = events->exception.injected;
2669         vcpu->arch.exception.nr = events->exception.nr;
2670         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2671         vcpu->arch.exception.error_code = events->exception.error_code;
2672
2673         vcpu->arch.interrupt.pending = events->interrupt.injected;
2674         vcpu->arch.interrupt.nr = events->interrupt.nr;
2675         vcpu->arch.interrupt.soft = events->interrupt.soft;
2676         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2677                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2678                                                   events->interrupt.shadow);
2679
2680         vcpu->arch.nmi_injected = events->nmi.injected;
2681         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2682                 vcpu->arch.nmi_pending = events->nmi.pending;
2683         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2684
2685         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2686                 vcpu->arch.sipi_vector = events->sipi_vector;
2687
2688         kvm_make_request(KVM_REQ_EVENT, vcpu);
2689
2690         return 0;
2691 }
2692
2693 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2694                                              struct kvm_debugregs *dbgregs)
2695 {
2696         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2697         dbgregs->dr6 = vcpu->arch.dr6;
2698         dbgregs->dr7 = vcpu->arch.dr7;
2699         dbgregs->flags = 0;
2700         memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2701 }
2702
2703 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2704                                             struct kvm_debugregs *dbgregs)
2705 {
2706         if (dbgregs->flags)
2707                 return -EINVAL;
2708
2709         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2710         vcpu->arch.dr6 = dbgregs->dr6;
2711         vcpu->arch.dr7 = dbgregs->dr7;
2712
2713         return 0;
2714 }
2715
2716 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2717                                          struct kvm_xsave *guest_xsave)
2718 {
2719         if (cpu_has_xsave)
2720                 memcpy(guest_xsave->region,
2721                         &vcpu->arch.guest_fpu.state->xsave,
2722                         xstate_size);
2723         else {
2724                 memcpy(guest_xsave->region,
2725                         &vcpu->arch.guest_fpu.state->fxsave,
2726                         sizeof(struct i387_fxsave_struct));
2727                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2728                         XSTATE_FPSSE;
2729         }
2730 }
2731
2732 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2733                                         struct kvm_xsave *guest_xsave)
2734 {
2735         u64 xstate_bv =
2736                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2737
2738         if (cpu_has_xsave)
2739                 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2740                         guest_xsave->region, xstate_size);
2741         else {
2742                 if (xstate_bv & ~XSTATE_FPSSE)
2743                         return -EINVAL;
2744                 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2745                         guest_xsave->region, sizeof(struct i387_fxsave_struct));
2746         }
2747         return 0;
2748 }
2749
2750 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2751                                         struct kvm_xcrs *guest_xcrs)
2752 {
2753         if (!cpu_has_xsave) {
2754                 guest_xcrs->nr_xcrs = 0;
2755                 return;
2756         }
2757
2758         guest_xcrs->nr_xcrs = 1;
2759         guest_xcrs->flags = 0;
2760         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2761         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2762 }
2763
2764 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2765                                        struct kvm_xcrs *guest_xcrs)
2766 {
2767         int i, r = 0;
2768
2769         if (!cpu_has_xsave)
2770                 return -EINVAL;
2771
2772         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2773                 return -EINVAL;
2774
2775         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2776                 /* Only support XCR0 currently */
2777                 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2778                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2779                                 guest_xcrs->xcrs[0].value);
2780                         break;
2781                 }
2782         if (r)
2783                 r = -EINVAL;
2784         return r;
2785 }
2786
2787 long kvm_arch_vcpu_ioctl(struct file *filp,
2788                          unsigned int ioctl, unsigned long arg)
2789 {
2790         struct kvm_vcpu *vcpu = filp->private_data;
2791         void __user *argp = (void __user *)arg;
2792         int r;
2793         union {
2794                 struct kvm_lapic_state *lapic;
2795                 struct kvm_xsave *xsave;
2796                 struct kvm_xcrs *xcrs;
2797                 void *buffer;
2798         } u;
2799
2800         u.buffer = NULL;
2801         switch (ioctl) {
2802         case KVM_GET_LAPIC: {
2803                 r = -EINVAL;
2804                 if (!vcpu->arch.apic)
2805                         goto out;
2806                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2807
2808                 r = -ENOMEM;
2809                 if (!u.lapic)
2810                         goto out;
2811                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2812                 if (r)
2813                         goto out;
2814                 r = -EFAULT;
2815                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2816                         goto out;
2817                 r = 0;
2818                 break;
2819         }
2820         case KVM_SET_LAPIC: {
2821                 r = -EINVAL;
2822                 if (!vcpu->arch.apic)
2823                         goto out;
2824                 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2825                 r = -ENOMEM;
2826                 if (!u.lapic)
2827                         goto out;
2828                 r = -EFAULT;
2829                 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2830                         goto out;
2831                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2832                 if (r)
2833                         goto out;
2834                 r = 0;
2835                 break;
2836         }
2837         case KVM_INTERRUPT: {
2838                 struct kvm_interrupt irq;
2839
2840                 r = -EFAULT;
2841                 if (copy_from_user(&irq, argp, sizeof irq))
2842                         goto out;
2843                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2844                 if (r)
2845                         goto out;
2846                 r = 0;
2847                 break;
2848         }
2849         case KVM_NMI: {
2850                 r = kvm_vcpu_ioctl_nmi(vcpu);
2851                 if (r)
2852                         goto out;
2853                 r = 0;
2854                 break;
2855         }
2856         case KVM_SET_CPUID: {
2857                 struct kvm_cpuid __user *cpuid_arg = argp;
2858                 struct kvm_cpuid cpuid;
2859
2860                 r = -EFAULT;
2861                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2862                         goto out;
2863                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2864                 if (r)
2865                         goto out;
2866                 break;
2867         }
2868         case KVM_SET_CPUID2: {
2869                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2870                 struct kvm_cpuid2 cpuid;
2871
2872                 r = -EFAULT;
2873                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2874                         goto out;
2875                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2876                                               cpuid_arg->entries);
2877                 if (r)
2878                         goto out;
2879                 break;
2880         }
2881         case KVM_GET_CPUID2: {
2882                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2883                 struct kvm_cpuid2 cpuid;
2884
2885                 r = -EFAULT;
2886                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2887                         goto out;
2888                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2889                                               cpuid_arg->entries);
2890                 if (r)
2891                         goto out;
2892                 r = -EFAULT;
2893                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2894                         goto out;
2895                 r = 0;
2896                 break;
2897         }
2898         case KVM_GET_MSRS:
2899                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2900                 break;
2901         case KVM_SET_MSRS:
2902                 r = msr_io(vcpu, argp, do_set_msr, 0);
2903                 break;
2904         case KVM_TPR_ACCESS_REPORTING: {
2905                 struct kvm_tpr_access_ctl tac;
2906
2907                 r = -EFAULT;
2908                 if (copy_from_user(&tac, argp, sizeof tac))
2909                         goto out;
2910                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2911                 if (r)
2912                         goto out;
2913                 r = -EFAULT;
2914                 if (copy_to_user(argp, &tac, sizeof tac))
2915                         goto out;
2916                 r = 0;
2917                 break;
2918         };
2919         case KVM_SET_VAPIC_ADDR: {
2920                 struct kvm_vapic_addr va;
2921
2922                 r = -EINVAL;
2923                 if (!irqchip_in_kernel(vcpu->kvm))
2924                         goto out;
2925                 r = -EFAULT;
2926                 if (copy_from_user(&va, argp, sizeof va))
2927                         goto out;
2928                 r = 0;
2929                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2930                 break;
2931         }
2932         case KVM_X86_SETUP_MCE: {
2933                 u64 mcg_cap;
2934
2935                 r = -EFAULT;
2936                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2937                         goto out;
2938                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2939                 break;
2940         }
2941         case KVM_X86_SET_MCE: {
2942                 struct kvm_x86_mce mce;
2943
2944                 r = -EFAULT;
2945                 if (copy_from_user(&mce, argp, sizeof mce))
2946                         goto out;
2947                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2948                 break;
2949         }
2950         case KVM_GET_VCPU_EVENTS: {
2951                 struct kvm_vcpu_events events;
2952
2953                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2954
2955                 r = -EFAULT;
2956                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2957                         break;
2958                 r = 0;
2959                 break;
2960         }
2961         case KVM_SET_VCPU_EVENTS: {
2962                 struct kvm_vcpu_events events;
2963
2964                 r = -EFAULT;
2965                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2966                         break;
2967
2968                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2969                 break;
2970         }
2971         case KVM_GET_DEBUGREGS: {
2972                 struct kvm_debugregs dbgregs;
2973
2974                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2975
2976                 r = -EFAULT;
2977                 if (copy_to_user(argp, &dbgregs,
2978                                  sizeof(struct kvm_debugregs)))
2979                         break;
2980                 r = 0;
2981                 break;
2982         }
2983         case KVM_SET_DEBUGREGS: {
2984                 struct kvm_debugregs dbgregs;
2985
2986                 r = -EFAULT;
2987                 if (copy_from_user(&dbgregs, argp,
2988                                    sizeof(struct kvm_debugregs)))
2989                         break;
2990
2991                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2992                 break;
2993         }
2994         case KVM_GET_XSAVE: {
2995                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2996                 r = -ENOMEM;
2997                 if (!u.xsave)
2998                         break;
2999
3000                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3001
3002                 r = -EFAULT;
3003                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3004                         break;
3005                 r = 0;
3006                 break;
3007         }
3008         case KVM_SET_XSAVE: {
3009                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3010                 r = -ENOMEM;
3011                 if (!u.xsave)
3012                         break;
3013
3014                 r = -EFAULT;
3015                 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3016                         break;
3017
3018                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3019                 break;
3020         }
3021         case KVM_GET_XCRS: {
3022                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3023                 r = -ENOMEM;
3024                 if (!u.xcrs)
3025                         break;
3026
3027                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3028
3029                 r = -EFAULT;
3030                 if (copy_to_user(argp, u.xcrs,
3031                                  sizeof(struct kvm_xcrs)))
3032                         break;
3033                 r = 0;
3034                 break;
3035         }
3036         case KVM_SET_XCRS: {
3037                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3038                 r = -ENOMEM;
3039                 if (!u.xcrs)
3040                         break;
3041
3042                 r = -EFAULT;
3043                 if (copy_from_user(u.xcrs, argp,
3044                                    sizeof(struct kvm_xcrs)))
3045                         break;
3046
3047                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3048                 break;
3049         }
3050         default:
3051                 r = -EINVAL;
3052         }
3053 out:
3054         kfree(u.buffer);
3055         return r;
3056 }
3057
3058 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3059 {
3060         int ret;
3061
3062         if (addr > (unsigned int)(-3 * PAGE_SIZE))
3063                 return -1;
3064         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3065         return ret;
3066 }
3067
3068 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3069                                               u64 ident_addr)
3070 {
3071         kvm->arch.ept_identity_map_addr = ident_addr;
3072         return 0;
3073 }
3074
3075 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3076                                           u32 kvm_nr_mmu_pages)
3077 {
3078         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3079                 return -EINVAL;
3080
3081         mutex_lock(&kvm->slots_lock);
3082         spin_lock(&kvm->mmu_lock);
3083
3084         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3085         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3086
3087         spin_unlock(&kvm->mmu_lock);
3088         mutex_unlock(&kvm->slots_lock);
3089         return 0;
3090 }
3091
3092 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3093 {
3094         return kvm->arch.n_max_mmu_pages;
3095 }
3096
3097 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3098 {
3099         int r;
3100
3101         r = 0;
3102         switch (chip->chip_id) {
3103         case KVM_IRQCHIP_PIC_MASTER:
3104                 memcpy(&chip->chip.pic,
3105                         &pic_irqchip(kvm)->pics[0],
3106                         sizeof(struct kvm_pic_state));
3107                 break;
3108         case KVM_IRQCHIP_PIC_SLAVE:
3109                 memcpy(&chip->chip.pic,
3110                         &pic_irqchip(kvm)->pics[1],
3111                         sizeof(struct kvm_pic_state));
3112                 break;
3113         case KVM_IRQCHIP_IOAPIC:
3114                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3115                 break;
3116         default:
3117                 r = -EINVAL;
3118                 break;
3119         }
3120         return r;
3121 }
3122
3123 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3124 {
3125         int r;
3126
3127         r = 0;
3128         switch (chip->chip_id) {
3129         case KVM_IRQCHIP_PIC_MASTER:
3130                 spin_lock(&pic_irqchip(kvm)->lock);
3131                 memcpy(&pic_irqchip(kvm)->pics[0],
3132                         &chip->chip.pic,
3133                         sizeof(struct kvm_pic_state));
3134                 spin_unlock(&pic_irqchip(kvm)->lock);
3135                 break;
3136         case KVM_IRQCHIP_PIC_SLAVE:
3137                 spin_lock(&pic_irqchip(kvm)->lock);
3138                 memcpy(&pic_irqchip(kvm)->pics[1],
3139                         &chip->chip.pic,
3140                         sizeof(struct kvm_pic_state));
3141                 spin_unlock(&pic_irqchip(kvm)->lock);
3142                 break;
3143         case KVM_IRQCHIP_IOAPIC:
3144                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3145                 break;
3146         default:
3147                 r = -EINVAL;
3148                 break;
3149         }
3150         kvm_pic_update_irq(pic_irqchip(kvm));
3151         return r;
3152 }
3153
3154 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3155 {
3156         int r = 0;
3157
3158         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3159         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3160         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3161         return r;
3162 }
3163
3164 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3165 {
3166         int r = 0;
3167
3168         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3169         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3170         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3171         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3172         return r;
3173 }
3174
3175 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3176 {
3177         int r = 0;
3178
3179         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3180         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3181                 sizeof(ps->channels));
3182         ps->flags = kvm->arch.vpit->pit_state.flags;
3183         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3184         memset(&ps->reserved, 0, sizeof(ps->reserved));
3185         return r;
3186 }
3187
3188 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3189 {
3190         int r = 0, start = 0;
3191         u32 prev_legacy, cur_legacy;
3192         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3193         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3194         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3195         if (!prev_legacy && cur_legacy)
3196                 start = 1;
3197         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3198                sizeof(kvm->arch.vpit->pit_state.channels));
3199         kvm->arch.vpit->pit_state.flags = ps->flags;
3200         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3201         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3202         return r;
3203 }
3204
3205 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3206                                  struct kvm_reinject_control *control)
3207 {
3208         if (!kvm->arch.vpit)
3209                 return -ENXIO;
3210         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3211         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3212         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3213         return 0;
3214 }
3215
3216 /*
3217  * Get (and clear) the dirty memory log for a memory slot.
3218  */
3219 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3220                                       struct kvm_dirty_log *log)
3221 {
3222         int r, i;
3223         struct kvm_memory_slot *memslot;
3224         unsigned long n;
3225         unsigned long is_dirty = 0;
3226
3227         mutex_lock(&kvm->slots_lock);
3228
3229         r = -EINVAL;
3230         if (log->slot >= KVM_MEMORY_SLOTS)
3231                 goto out;
3232
3233         memslot = &kvm->memslots->memslots[log->slot];
3234         r = -ENOENT;
3235         if (!memslot->dirty_bitmap)
3236                 goto out;
3237
3238         n = kvm_dirty_bitmap_bytes(memslot);
3239
3240         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3241                 is_dirty = memslot->dirty_bitmap[i];
3242
3243         /* If nothing is dirty, don't bother messing with page tables. */
3244         if (is_dirty) {
3245                 struct kvm_memslots *slots, *old_slots;
3246                 unsigned long *dirty_bitmap;
3247
3248                 dirty_bitmap = memslot->dirty_bitmap_head;
3249                 if (memslot->dirty_bitmap == dirty_bitmap)
3250                         dirty_bitmap += n / sizeof(long);
3251                 memset(dirty_bitmap, 0, n);
3252
3253                 r = -ENOMEM;
3254                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3255                 if (!slots)
3256                         goto out;
3257                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3258                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3259                 slots->generation++;
3260
3261                 old_slots = kvm->memslots;
3262                 rcu_assign_pointer(kvm->memslots, slots);
3263                 synchronize_srcu_expedited(&kvm->srcu);
3264                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3265                 kfree(old_slots);
3266
3267                 spin_lock(&kvm->mmu_lock);
3268                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3269                 spin_unlock(&kvm->mmu_lock);
3270
3271                 r = -EFAULT;
3272                 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3273                         goto out;
3274         } else {
3275                 r = -EFAULT;
3276                 if (clear_user(log->dirty_bitmap, n))
3277                         goto out;
3278         }
3279
3280         r = 0;
3281 out:
3282         mutex_unlock(&kvm->slots_lock);
3283         return r;
3284 }
3285
3286 long kvm_arch_vm_ioctl(struct file *filp,
3287                        unsigned int ioctl, unsigned long arg)
3288 {
3289         struct kvm *kvm = filp->private_data;
3290         void __user *argp = (void __user *)arg;
3291         int r = -ENOTTY;
3292         /*
3293          * This union makes it completely explicit to gcc-3.x
3294          * that these two variables' stack usage should be
3295          * combined, not added together.
3296          */
3297         union {
3298                 struct kvm_pit_state ps;
3299                 struct kvm_pit_state2 ps2;
3300                 struct kvm_pit_config pit_config;
3301         } u;
3302
3303         switch (ioctl) {
3304         case KVM_SET_TSS_ADDR:
3305                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3306                 if (r < 0)
3307                         goto out;
3308                 break;
3309         case KVM_SET_IDENTITY_MAP_ADDR: {
3310                 u64 ident_addr;
3311
3312                 r = -EFAULT;
3313                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3314                         goto out;
3315                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3316                 if (r < 0)
3317                         goto out;
3318                 break;
3319         }
3320         case KVM_SET_NR_MMU_PAGES:
3321                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3322                 if (r)
3323                         goto out;
3324                 break;
3325         case KVM_GET_NR_MMU_PAGES:
3326                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3327                 break;
3328         case KVM_CREATE_IRQCHIP: {
3329                 struct kvm_pic *vpic;
3330
3331                 mutex_lock(&kvm->lock);
3332                 r = -EEXIST;
3333                 if (kvm->arch.vpic)
3334                         goto create_irqchip_unlock;
3335                 r = -ENOMEM;
3336                 vpic = kvm_create_pic(kvm);
3337                 if (vpic) {
3338                         r = kvm_ioapic_init(kvm);
3339                         if (r) {
3340                                 mutex_lock(&kvm->slots_lock);
3341                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3342                                                           &vpic->dev);
3343                                 mutex_unlock(&kvm->slots_lock);
3344                                 kfree(vpic);
3345                                 goto create_irqchip_unlock;
3346                         }
3347                 } else
3348                         goto create_irqchip_unlock;
3349                 smp_wmb();
3350                 kvm->arch.vpic = vpic;
3351                 smp_wmb();
3352                 r = kvm_setup_default_irq_routing(kvm);
3353                 if (r) {
3354                         mutex_lock(&kvm->slots_lock);
3355                         mutex_lock(&kvm->irq_lock);
3356                         kvm_ioapic_destroy(kvm);
3357                         kvm_destroy_pic(kvm);
3358                         mutex_unlock(&kvm->irq_lock);
3359                         mutex_unlock(&kvm->slots_lock);
3360                 }
3361         create_irqchip_unlock:
3362                 mutex_unlock(&kvm->lock);
3363                 break;
3364         }
3365         case KVM_CREATE_PIT:
3366                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3367                 goto create_pit;
3368         case KVM_CREATE_PIT2:
3369                 r = -EFAULT;
3370                 if (copy_from_user(&u.pit_config, argp,
3371                                    sizeof(struct kvm_pit_config)))
3372                         goto out;
3373         create_pit:
3374                 mutex_lock(&kvm->slots_lock);
3375                 r = -EEXIST;
3376                 if (kvm->arch.vpit)
3377                         goto create_pit_unlock;
3378                 r = -ENOMEM;
3379                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3380                 if (kvm->arch.vpit)
3381                         r = 0;
3382         create_pit_unlock:
3383                 mutex_unlock(&kvm->slots_lock);
3384                 break;
3385         case KVM_IRQ_LINE_STATUS:
3386         case KVM_IRQ_LINE: {
3387                 struct kvm_irq_level irq_event;
3388
3389                 r = -EFAULT;
3390                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3391                         goto out;
3392                 r = -ENXIO;
3393                 if (irqchip_in_kernel(kvm)) {
3394                         __s32 status;
3395                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3396                                         irq_event.irq, irq_event.level);
3397                         if (ioctl == KVM_IRQ_LINE_STATUS) {
3398                                 r = -EFAULT;
3399                                 irq_event.status = status;
3400                                 if (copy_to_user(argp, &irq_event,
3401                                                         sizeof irq_event))
3402                                         goto out;
3403                         }
3404                         r = 0;
3405                 }
3406                 break;
3407         }
3408         case KVM_GET_IRQCHIP: {
3409                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3410                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3411
3412                 r = -ENOMEM;
3413                 if (!chip)
3414                         goto out;
3415                 r = -EFAULT;
3416                 if (copy_from_user(chip, argp, sizeof *chip))
3417                         goto get_irqchip_out;
3418                 r = -ENXIO;
3419                 if (!irqchip_in_kernel(kvm))
3420                         goto get_irqchip_out;
3421                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3422                 if (r)
3423                         goto get_irqchip_out;
3424                 r = -EFAULT;
3425                 if (copy_to_user(argp, chip, sizeof *chip))
3426                         goto get_irqchip_out;
3427                 r = 0;
3428         get_irqchip_out:
3429                 kfree(chip);
3430                 if (r)
3431                         goto out;
3432                 break;
3433         }
3434         case KVM_SET_IRQCHIP: {
3435                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3436                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3437
3438                 r = -ENOMEM;
3439                 if (!chip)
3440                         goto out;
3441                 r = -EFAULT;
3442                 if (copy_from_user(chip, argp, sizeof *chip))
3443                         goto set_irqchip_out;
3444                 r = -ENXIO;
3445                 if (!irqchip_in_kernel(kvm))
3446                         goto set_irqchip_out;
3447                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3448                 if (r)
3449                         goto set_irqchip_out;
3450                 r = 0;
3451         set_irqchip_out:
3452                 kfree(chip);
3453                 if (r)
3454                         goto out;
3455                 break;
3456         }
3457         case KVM_GET_PIT: {
3458                 r = -EFAULT;
3459                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3460                         goto out;
3461                 r = -ENXIO;
3462                 if (!kvm->arch.vpit)
3463                         goto out;
3464                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3465                 if (r)
3466                         goto out;
3467                 r = -EFAULT;
3468                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3469                         goto out;
3470                 r = 0;
3471                 break;
3472         }
3473         case KVM_SET_PIT: {
3474                 r = -EFAULT;
3475                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3476                         goto out;
3477                 r = -ENXIO;
3478                 if (!kvm->arch.vpit)
3479                         goto out;
3480                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3481                 if (r)
3482                         goto out;
3483                 r = 0;
3484                 break;
3485         }
3486         case KVM_GET_PIT2: {
3487                 r = -ENXIO;
3488                 if (!kvm->arch.vpit)
3489                         goto out;
3490                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3491                 if (r)
3492                         goto out;
3493                 r = -EFAULT;
3494                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3495                         goto out;
3496                 r = 0;
3497                 break;
3498         }
3499         case KVM_SET_PIT2: {
3500                 r = -EFAULT;
3501                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3502                         goto out;
3503                 r = -ENXIO;
3504                 if (!kvm->arch.vpit)
3505                         goto out;
3506                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3507                 if (r)
3508                         goto out;
3509                 r = 0;
3510                 break;
3511         }
3512         case KVM_REINJECT_CONTROL: {
3513                 struct kvm_reinject_control control;
3514                 r =  -EFAULT;
3515                 if (copy_from_user(&control, argp, sizeof(control)))
3516                         goto out;
3517                 r = kvm_vm_ioctl_reinject(kvm, &control);
3518                 if (r)
3519                         goto out;
3520                 r = 0;
3521                 break;
3522         }
3523         case KVM_XEN_HVM_CONFIG: {
3524                 r = -EFAULT;
3525                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3526                                    sizeof(struct kvm_xen_hvm_config)))
3527                         goto out;
3528                 r = -EINVAL;
3529                 if (kvm->arch.xen_hvm_config.flags)
3530                         goto out;
3531                 r = 0;
3532                 break;
3533         }
3534         case KVM_SET_CLOCK: {
3535                 struct kvm_clock_data user_ns;
3536                 u64 now_ns;
3537                 s64 delta;
3538
3539                 r = -EFAULT;
3540                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3541                         goto out;
3542
3543                 r = -EINVAL;
3544                 if (user_ns.flags)
3545                         goto out;
3546
3547                 r = 0;
3548                 local_irq_disable();
3549                 now_ns = get_kernel_ns();
3550                 delta = user_ns.clock - now_ns;
3551                 local_irq_enable();
3552                 kvm->arch.kvmclock_offset = delta;
3553                 break;
3554         }
3555         case KVM_GET_CLOCK: {
3556                 struct kvm_clock_data user_ns;
3557                 u64 now_ns;
3558
3559                 local_irq_disable();
3560                 now_ns = get_kernel_ns();
3561                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3562                 local_irq_enable();
3563                 user_ns.flags = 0;
3564                 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3565
3566                 r = -EFAULT;
3567                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3568                         goto out;
3569                 r = 0;
3570                 break;
3571         }
3572
3573         default:
3574                 ;
3575         }
3576 out:
3577         return r;
3578 }
3579
3580 static void kvm_init_msr_list(void)
3581 {
3582         u32 dummy[2];
3583         unsigned i, j;
3584
3585         /* skip the first msrs in the list. KVM-specific */
3586         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3587                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3588                         continue;
3589                 if (j < i)
3590                         msrs_to_save[j] = msrs_to_save[i];
3591                 j++;
3592         }
3593         num_msrs_to_save = j;
3594 }
3595
3596 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3597                            const void *v)
3598 {
3599         int handled = 0;
3600         int n;
3601
3602         do {
3603                 n = min(len, 8);
3604                 if (!(vcpu->arch.apic &&
3605                       !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3606                     && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3607                         break;
3608                 handled += n;
3609                 addr += n;
3610                 len -= n;
3611                 v += n;
3612         } while (len);
3613
3614         return handled;
3615 }
3616
3617 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3618 {
3619         int handled = 0;
3620         int n;
3621
3622         do {
3623                 n = min(len, 8);
3624                 if (!(vcpu->arch.apic &&
3625                       !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3626                     && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3627                         break;
3628                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3629                 handled += n;
3630                 addr += n;
3631                 len -= n;
3632                 v += n;
3633         } while (len);
3634
3635         return handled;
3636 }
3637
3638 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3639                         struct kvm_segment *var, int seg)
3640 {
3641         kvm_x86_ops->set_segment(vcpu, var, seg);
3642 }
3643
3644 void kvm_get_segment(struct kvm_vcpu *vcpu,
3645                      struct kvm_segment *var, int seg)
3646 {
3647         kvm_x86_ops->get_segment(vcpu, var, seg);
3648 }
3649
3650 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3651 {
3652         return gpa;
3653 }
3654
3655 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3656 {
3657         gpa_t t_gpa;
3658         struct x86_exception exception;
3659
3660         BUG_ON(!mmu_is_nested(vcpu));
3661
3662         /* NPT walks are always user-walks */
3663         access |= PFERR_USER_MASK;
3664         t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3665
3666         return t_gpa;
3667 }
3668
3669 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3670                               struct x86_exception *exception)
3671 {
3672         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3673         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3674 }
3675
3676  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3677                                 struct x86_exception *exception)
3678 {
3679         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3680         access |= PFERR_FETCH_MASK;
3681         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3682 }
3683
3684 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3685                                struct x86_exception *exception)
3686 {
3687         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3688         access |= PFERR_WRITE_MASK;
3689         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3690 }
3691
3692 /* uses this to access any guest's mapped memory without checking CPL */
3693 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3694                                 struct x86_exception *exception)
3695 {
3696         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3697 }
3698
3699 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3700                                       struct kvm_vcpu *vcpu, u32 access,
3701                                       struct x86_exception *exception)
3702 {
3703         void *data = val;
3704         int r = X86EMUL_CONTINUE;
3705
3706         while (bytes) {
3707                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3708                                                             exception);
3709                 unsigned offset = addr & (PAGE_SIZE-1);
3710                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3711                 int ret;
3712
3713                 if (gpa == UNMAPPED_GVA)
3714                         return X86EMUL_PROPAGATE_FAULT;
3715                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3716                 if (ret < 0) {
3717                         r = X86EMUL_IO_NEEDED;
3718                         goto out;
3719                 }
3720
3721                 bytes -= toread;
3722                 data += toread;
3723                 addr += toread;
3724         }
3725 out:
3726         return r;
3727 }
3728
3729 /* used for instruction fetching */
3730 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3731                                 struct kvm_vcpu *vcpu,
3732                                 struct x86_exception *exception)
3733 {
3734         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3735         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3736                                           access | PFERR_FETCH_MASK,
3737                                           exception);
3738 }
3739
3740 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3741                                struct kvm_vcpu *vcpu,
3742                                struct x86_exception *exception)
3743 {
3744         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3745         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3746                                           exception);
3747 }
3748
3749 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3750                                       struct kvm_vcpu *vcpu,
3751                                       struct x86_exception *exception)
3752 {
3753         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3754 }
3755
3756 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3757                                        unsigned int bytes,
3758                                        struct kvm_vcpu *vcpu,
3759                                        struct x86_exception *exception)
3760 {
3761         void *data = val;
3762         int r = X86EMUL_CONTINUE;
3763
3764         while (bytes) {
3765                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3766                                                              PFERR_WRITE_MASK,
3767                                                              exception);
3768                 unsigned offset = addr & (PAGE_SIZE-1);
3769                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3770                 int ret;
3771
3772                 if (gpa == UNMAPPED_GVA)
3773                         return X86EMUL_PROPAGATE_FAULT;
3774                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3775                 if (ret < 0) {
3776                         r = X86EMUL_IO_NEEDED;
3777                         goto out;
3778                 }
3779
3780                 bytes -= towrite;
3781                 data += towrite;
3782                 addr += towrite;
3783         }
3784 out:
3785         return r;
3786 }
3787
3788 static int emulator_read_emulated(unsigned long addr,
3789                                   void *val,
3790                                   unsigned int bytes,
3791                                   struct x86_exception *exception,
3792                                   struct kvm_vcpu *vcpu)
3793 {
3794         gpa_t                 gpa;
3795         int handled;
3796
3797         if (vcpu->mmio_read_completed) {
3798                 memcpy(val, vcpu->mmio_data, bytes);
3799                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3800                                vcpu->mmio_phys_addr, *(u64 *)val);
3801                 vcpu->mmio_read_completed = 0;
3802                 return X86EMUL_CONTINUE;
3803         }
3804
3805         gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, exception);
3806
3807         if (gpa == UNMAPPED_GVA)
3808                 return X86EMUL_PROPAGATE_FAULT;
3809
3810         /* For APIC access vmexit */
3811         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3812                 goto mmio;
3813
3814         if (kvm_read_guest_virt(addr, val, bytes, vcpu, exception)
3815             == X86EMUL_CONTINUE)
3816                 return X86EMUL_CONTINUE;
3817
3818 mmio:
3819         /*
3820          * Is this MMIO handled locally?
3821          */
3822         handled = vcpu_mmio_read(vcpu, gpa, bytes, val);
3823
3824         if (handled == bytes)
3825                 return X86EMUL_CONTINUE;
3826
3827         gpa += handled;
3828         bytes -= handled;
3829         val += handled;
3830
3831         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3832
3833         vcpu->mmio_needed = 1;
3834         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3835         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3836         vcpu->mmio_size = bytes;
3837         vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3838         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3839         vcpu->mmio_index = 0;
3840
3841         return X86EMUL_IO_NEEDED;
3842 }
3843
3844 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3845                         const void *val, int bytes)
3846 {
3847         int ret;
3848
3849         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3850         if (ret < 0)
3851                 return 0;
3852         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3853         return 1;
3854 }
3855
3856 static int emulator_write_emulated_onepage(unsigned long addr,
3857                                            const void *val,
3858                                            unsigned int bytes,
3859                                            struct x86_exception *exception,
3860                                            struct kvm_vcpu *vcpu)
3861 {
3862         gpa_t                 gpa;
3863         int handled;
3864
3865         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, exception);
3866
3867         if (gpa == UNMAPPED_GVA)
3868                 return X86EMUL_PROPAGATE_FAULT;
3869
3870         /* For APIC access vmexit */
3871         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3872                 goto mmio;
3873
3874         if (emulator_write_phys(vcpu, gpa, val, bytes))
3875                 return X86EMUL_CONTINUE;
3876
3877 mmio:
3878         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3879         /*
3880          * Is this MMIO handled locally?
3881          */
3882         handled = vcpu_mmio_write(vcpu, gpa, bytes, val);
3883         if (handled == bytes)
3884                 return X86EMUL_CONTINUE;
3885
3886         gpa += handled;
3887         bytes -= handled;
3888         val += handled;
3889
3890         vcpu->mmio_needed = 1;
3891         memcpy(vcpu->mmio_data, val, bytes);
3892         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3893         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3894         vcpu->mmio_size = bytes;
3895         vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3896         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3897         memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
3898         vcpu->mmio_index = 0;
3899
3900         return X86EMUL_CONTINUE;
3901 }
3902
3903 int emulator_write_emulated(unsigned long addr,
3904                             const void *val,
3905                             unsigned int bytes,
3906                             struct x86_exception *exception,
3907                             struct kvm_vcpu *vcpu)
3908 {
3909         /* Crossing a page boundary? */
3910         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3911                 int rc, now;
3912
3913                 now = -addr & ~PAGE_MASK;
3914                 rc = emulator_write_emulated_onepage(addr, val, now, exception,
3915                                                      vcpu);
3916                 if (rc != X86EMUL_CONTINUE)
3917                         return rc;
3918                 addr += now;
3919                 val += now;
3920                 bytes -= now;
3921         }
3922         return emulator_write_emulated_onepage(addr, val, bytes, exception,
3923                                                vcpu);
3924 }
3925
3926 #define CMPXCHG_TYPE(t, ptr, old, new) \
3927         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3928
3929 #ifdef CONFIG_X86_64
3930 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3931 #else
3932 #  define CMPXCHG64(ptr, old, new) \
3933         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3934 #endif
3935
3936 static int emulator_cmpxchg_emulated(unsigned long addr,
3937                                      const void *old,
3938                                      const void *new,
3939                                      unsigned int bytes,
3940                                      struct x86_exception *exception,
3941                                      struct kvm_vcpu *vcpu)
3942 {
3943         gpa_t gpa;
3944         struct page *page;
3945         char *kaddr;
3946         bool exchanged;
3947
3948         /* guests cmpxchg8b have to be emulated atomically */
3949         if (bytes > 8 || (bytes & (bytes - 1)))
3950                 goto emul_write;
3951
3952         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3953
3954         if (gpa == UNMAPPED_GVA ||
3955             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3956                 goto emul_write;
3957
3958         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3959                 goto emul_write;
3960
3961         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3962         if (is_error_page(page)) {
3963                 kvm_release_page_clean(page);
3964                 goto emul_write;
3965         }
3966
3967         kaddr = kmap_atomic(page, KM_USER0);
3968         kaddr += offset_in_page(gpa);
3969         switch (bytes) {
3970         case 1:
3971                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3972                 break;
3973         case 2:
3974                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3975                 break;
3976         case 4:
3977                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3978                 break;
3979         case 8:
3980                 exchanged = CMPXCHG64(kaddr, old, new);
3981                 break;
3982         default:
3983                 BUG();
3984         }
3985         kunmap_atomic(kaddr, KM_USER0);
3986         kvm_release_page_dirty(page);
3987
3988         if (!exchanged)
3989                 return X86EMUL_CMPXCHG_FAILED;
3990
3991         kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3992
3993         return X86EMUL_CONTINUE;
3994
3995 emul_write:
3996         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3997
3998         return emulator_write_emulated(addr, new, bytes, exception, vcpu);
3999 }
4000
4001 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4002 {
4003         /* TODO: String I/O for in kernel device */
4004         int r;
4005
4006         if (vcpu->arch.pio.in)
4007                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4008                                     vcpu->arch.pio.size, pd);
4009         else
4010                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4011                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
4012                                      pd);
4013         return r;
4014 }
4015
4016
4017 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
4018                              unsigned int count, struct kvm_vcpu *vcpu)
4019 {
4020         if (vcpu->arch.pio.count)
4021                 goto data_avail;
4022
4023         trace_kvm_pio(0, port, size, count);
4024
4025         vcpu->arch.pio.port = port;
4026         vcpu->arch.pio.in = 1;
4027         vcpu->arch.pio.count  = count;
4028         vcpu->arch.pio.size = size;
4029
4030         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4031         data_avail:
4032                 memcpy(val, vcpu->arch.pio_data, size * count);
4033                 vcpu->arch.pio.count = 0;
4034                 return 1;
4035         }
4036
4037         vcpu->run->exit_reason = KVM_EXIT_IO;
4038         vcpu->run->io.direction = KVM_EXIT_IO_IN;
4039         vcpu->run->io.size = size;
4040         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4041         vcpu->run->io.count = count;
4042         vcpu->run->io.port = port;
4043
4044         return 0;
4045 }
4046
4047 static int emulator_pio_out_emulated(int size, unsigned short port,
4048                               const void *val, unsigned int count,
4049                               struct kvm_vcpu *vcpu)
4050 {
4051         trace_kvm_pio(1, port, size, count);
4052
4053         vcpu->arch.pio.port = port;
4054         vcpu->arch.pio.in = 0;
4055         vcpu->arch.pio.count = count;
4056         vcpu->arch.pio.size = size;
4057
4058         memcpy(vcpu->arch.pio_data, val, size * count);
4059
4060         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4061                 vcpu->arch.pio.count = 0;
4062                 return 1;
4063         }
4064
4065         vcpu->run->exit_reason = KVM_EXIT_IO;
4066         vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4067         vcpu->run->io.size = size;
4068         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4069         vcpu->run->io.count = count;
4070         vcpu->run->io.port = port;
4071
4072         return 0;
4073 }
4074
4075 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4076 {
4077         return kvm_x86_ops->get_segment_base(vcpu, seg);
4078 }
4079
4080 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
4081 {
4082         kvm_mmu_invlpg(vcpu, address);
4083         return X86EMUL_CONTINUE;
4084 }
4085
4086 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4087 {
4088         if (!need_emulate_wbinvd(vcpu))
4089                 return X86EMUL_CONTINUE;
4090
4091         if (kvm_x86_ops->has_wbinvd_exit()) {
4092                 int cpu = get_cpu();
4093
4094                 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4095                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4096                                 wbinvd_ipi, NULL, 1);
4097                 put_cpu();
4098                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4099         } else
4100                 wbinvd();
4101         return X86EMUL_CONTINUE;
4102 }
4103 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4104
4105 int emulate_clts(struct kvm_vcpu *vcpu)
4106 {
4107         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4108         kvm_x86_ops->fpu_activate(vcpu);
4109         return X86EMUL_CONTINUE;
4110 }
4111
4112 int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu)
4113 {
4114         return _kvm_get_dr(vcpu, dr, dest);
4115 }
4116
4117 int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu)
4118 {
4119
4120         return __kvm_set_dr(vcpu, dr, value);
4121 }
4122
4123 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4124 {
4125         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4126 }
4127
4128 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
4129 {
4130         unsigned long value;
4131
4132         switch (cr) {
4133         case 0:
4134                 value = kvm_read_cr0(vcpu);
4135                 break;
4136         case 2:
4137                 value = vcpu->arch.cr2;
4138                 break;
4139         case 3:
4140                 value = kvm_read_cr3(vcpu);
4141                 break;
4142         case 4:
4143                 value = kvm_read_cr4(vcpu);
4144                 break;
4145         case 8:
4146                 value = kvm_get_cr8(vcpu);
4147                 break;
4148         default:
4149                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4150                 return 0;
4151         }
4152
4153         return value;
4154 }
4155
4156 static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
4157 {
4158         int res = 0;
4159
4160         switch (cr) {
4161         case 0:
4162                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4163                 break;
4164         case 2:
4165                 vcpu->arch.cr2 = val;
4166                 break;
4167         case 3:
4168                 res = kvm_set_cr3(vcpu, val);
4169                 break;
4170         case 4:
4171                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4172                 break;
4173         case 8:
4174                 res = kvm_set_cr8(vcpu, val);
4175                 break;
4176         default:
4177                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4178                 res = -1;
4179         }
4180
4181         return res;
4182 }
4183
4184 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
4185 {
4186         return kvm_x86_ops->get_cpl(vcpu);
4187 }
4188
4189 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4190 {
4191         kvm_x86_ops->get_gdt(vcpu, dt);
4192 }
4193
4194 static void emulator_get_idt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4195 {
4196         kvm_x86_ops->get_idt(vcpu, dt);
4197 }
4198
4199 static unsigned long emulator_get_cached_segment_base(int seg,
4200                                                       struct kvm_vcpu *vcpu)
4201 {
4202         return get_segment_base(vcpu, seg);
4203 }
4204
4205 static bool emulator_get_cached_descriptor(struct desc_struct *desc, u32 *base3,
4206                                            int seg, struct kvm_vcpu *vcpu)
4207 {
4208         struct kvm_segment var;
4209
4210         kvm_get_segment(vcpu, &var, seg);
4211
4212         if (var.unusable)
4213                 return false;
4214
4215         if (var.g)
4216                 var.limit >>= 12;
4217         set_desc_limit(desc, var.limit);
4218         set_desc_base(desc, (unsigned long)var.base);
4219 #ifdef CONFIG_X86_64
4220         if (base3)
4221                 *base3 = var.base >> 32;
4222 #endif
4223         desc->type = var.type;
4224         desc->s = var.s;
4225         desc->dpl = var.dpl;
4226         desc->p = var.present;
4227         desc->avl = var.avl;
4228         desc->l = var.l;
4229         desc->d = var.db;
4230         desc->g = var.g;
4231
4232         return true;
4233 }
4234
4235 static void emulator_set_cached_descriptor(struct desc_struct *desc, u32 base3,
4236                                            int seg, struct kvm_vcpu *vcpu)
4237 {
4238         struct kvm_segment var;
4239
4240         /* needed to preserve selector */
4241         kvm_get_segment(vcpu, &var, seg);
4242
4243         var.base = get_desc_base(desc);
4244 #ifdef CONFIG_X86_64
4245         var.base |= ((u64)base3) << 32;
4246 #endif
4247         var.limit = get_desc_limit(desc);
4248         if (desc->g)
4249                 var.limit = (var.limit << 12) | 0xfff;
4250         var.type = desc->type;
4251         var.present = desc->p;
4252         var.dpl = desc->dpl;
4253         var.db = desc->d;
4254         var.s = desc->s;
4255         var.l = desc->l;
4256         var.g = desc->g;
4257         var.avl = desc->avl;
4258         var.present = desc->p;
4259         var.unusable = !var.present;
4260         var.padding = 0;
4261
4262         kvm_set_segment(vcpu, &var, seg);
4263         return;
4264 }
4265
4266 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
4267 {
4268         struct kvm_segment kvm_seg;
4269
4270         kvm_get_segment(vcpu, &kvm_seg, seg);
4271         return kvm_seg.selector;
4272 }
4273
4274 static void emulator_set_segment_selector(u16 sel, int seg,
4275                                           struct kvm_vcpu *vcpu)
4276 {
4277         struct kvm_segment kvm_seg;
4278
4279         kvm_get_segment(vcpu, &kvm_seg, seg);
4280         kvm_seg.selector = sel;
4281         kvm_set_segment(vcpu, &kvm_seg, seg);
4282 }
4283
4284 static struct x86_emulate_ops emulate_ops = {
4285         .read_std            = kvm_read_guest_virt_system,
4286         .write_std           = kvm_write_guest_virt_system,
4287         .fetch               = kvm_fetch_guest_virt,
4288         .read_emulated       = emulator_read_emulated,
4289         .write_emulated      = emulator_write_emulated,
4290         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
4291         .pio_in_emulated     = emulator_pio_in_emulated,
4292         .pio_out_emulated    = emulator_pio_out_emulated,
4293         .get_cached_descriptor = emulator_get_cached_descriptor,
4294         .set_cached_descriptor = emulator_set_cached_descriptor,
4295         .get_segment_selector = emulator_get_segment_selector,
4296         .set_segment_selector = emulator_set_segment_selector,
4297         .get_cached_segment_base = emulator_get_cached_segment_base,
4298         .get_gdt             = emulator_get_gdt,
4299         .get_idt             = emulator_get_idt,
4300         .get_cr              = emulator_get_cr,
4301         .set_cr              = emulator_set_cr,
4302         .cpl                 = emulator_get_cpl,
4303         .get_dr              = emulator_get_dr,
4304         .set_dr              = emulator_set_dr,
4305         .set_msr             = kvm_set_msr,
4306         .get_msr             = kvm_get_msr,
4307 };
4308
4309 static void cache_all_regs(struct kvm_vcpu *vcpu)
4310 {
4311         kvm_register_read(vcpu, VCPU_REGS_RAX);
4312         kvm_register_read(vcpu, VCPU_REGS_RSP);
4313         kvm_register_read(vcpu, VCPU_REGS_RIP);
4314         vcpu->arch.regs_dirty = ~0;
4315 }
4316
4317 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4318 {
4319         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4320         /*
4321          * an sti; sti; sequence only disable interrupts for the first
4322          * instruction. So, if the last instruction, be it emulated or
4323          * not, left the system with the INT_STI flag enabled, it
4324          * means that the last instruction is an sti. We should not
4325          * leave the flag on in this case. The same goes for mov ss
4326          */
4327         if (!(int_shadow & mask))
4328                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4329 }
4330
4331 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4332 {
4333         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4334         if (ctxt->exception.vector == PF_VECTOR)
4335                 kvm_propagate_fault(vcpu, &ctxt->exception);
4336         else if (ctxt->exception.error_code_valid)
4337                 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4338                                       ctxt->exception.error_code);
4339         else
4340                 kvm_queue_exception(vcpu, ctxt->exception.vector);
4341 }
4342
4343 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4344 {
4345         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4346         int cs_db, cs_l;
4347
4348         cache_all_regs(vcpu);
4349
4350         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4351
4352         vcpu->arch.emulate_ctxt.vcpu = vcpu;
4353         vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
4354         vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4355         vcpu->arch.emulate_ctxt.mode =
4356                 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4357                 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4358                 ? X86EMUL_MODE_VM86 : cs_l
4359                 ? X86EMUL_MODE_PROT64 : cs_db
4360                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4361         memset(c, 0, sizeof(struct decode_cache));
4362         memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4363 }
4364
4365 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq)
4366 {
4367         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4368         int ret;
4369
4370         init_emulate_ctxt(vcpu);
4371
4372         vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4373         vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4374         vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip;
4375         ret = emulate_int_real(&vcpu->arch.emulate_ctxt, &emulate_ops, irq);
4376
4377         if (ret != X86EMUL_CONTINUE)
4378                 return EMULATE_FAIL;
4379
4380         vcpu->arch.emulate_ctxt.eip = c->eip;
4381         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4382         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4383         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4384
4385         if (irq == NMI_VECTOR)
4386                 vcpu->arch.nmi_pending = false;
4387         else
4388                 vcpu->arch.interrupt.pending = false;
4389
4390         return EMULATE_DONE;
4391 }
4392 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4393
4394 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4395 {
4396         int r = EMULATE_DONE;
4397
4398         ++vcpu->stat.insn_emulation_fail;
4399         trace_kvm_emulate_insn_failed(vcpu);
4400         if (!is_guest_mode(vcpu)) {
4401                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4402                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4403                 vcpu->run->internal.ndata = 0;
4404                 r = EMULATE_FAIL;
4405         }
4406         kvm_queue_exception(vcpu, UD_VECTOR);
4407
4408         return r;
4409 }
4410
4411 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4412 {
4413         gpa_t gpa;
4414
4415         if (tdp_enabled)
4416                 return false;
4417
4418         /*
4419          * if emulation was due to access to shadowed page table
4420          * and it failed try to unshadow page and re-entetr the
4421          * guest to let CPU execute the instruction.
4422          */
4423         if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4424                 return true;
4425
4426         gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4427
4428         if (gpa == UNMAPPED_GVA)
4429                 return true; /* let cpu generate fault */
4430
4431         if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4432                 return true;
4433
4434         return false;
4435 }
4436
4437 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4438                             unsigned long cr2,
4439                             int emulation_type,
4440                             void *insn,
4441                             int insn_len)
4442 {
4443         int r;
4444         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4445
4446         kvm_clear_exception_queue(vcpu);
4447         vcpu->arch.mmio_fault_cr2 = cr2;
4448         /*
4449          * TODO: fix emulate.c to use guest_read/write_register
4450          * instead of direct ->regs accesses, can save hundred cycles
4451          * on Intel for instructions that don't read/change RSP, for
4452          * for example.
4453          */
4454         cache_all_regs(vcpu);
4455
4456         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4457                 init_emulate_ctxt(vcpu);
4458                 vcpu->arch.emulate_ctxt.interruptibility = 0;
4459                 vcpu->arch.emulate_ctxt.have_exception = false;
4460                 vcpu->arch.emulate_ctxt.perm_ok = false;
4461
4462                 vcpu->arch.emulate_ctxt.only_vendor_specific_insn
4463                         = emulation_type & EMULTYPE_TRAP_UD;
4464
4465                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, insn, insn_len);
4466
4467                 trace_kvm_emulate_insn_start(vcpu);
4468                 ++vcpu->stat.insn_emulation;
4469                 if (r)  {
4470                         if (emulation_type & EMULTYPE_TRAP_UD)
4471                                 return EMULATE_FAIL;
4472                         if (reexecute_instruction(vcpu, cr2))
4473                                 return EMULATE_DONE;
4474                         if (emulation_type & EMULTYPE_SKIP)
4475                                 return EMULATE_FAIL;
4476                         return handle_emulation_failure(vcpu);
4477                 }
4478         }
4479
4480         if (emulation_type & EMULTYPE_SKIP) {
4481                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4482                 return EMULATE_DONE;
4483         }
4484
4485         /* this is needed for vmware backdor interface to work since it
4486            changes registers values  during IO operation */
4487         memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4488
4489 restart:
4490         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4491
4492         if (r == EMULATION_FAILED) {
4493                 if (reexecute_instruction(vcpu, cr2))
4494                         return EMULATE_DONE;
4495
4496                 return handle_emulation_failure(vcpu);
4497         }
4498
4499         if (vcpu->arch.emulate_ctxt.have_exception) {
4500                 inject_emulated_exception(vcpu);
4501                 r = EMULATE_DONE;
4502         } else if (vcpu->arch.pio.count) {
4503                 if (!vcpu->arch.pio.in)
4504                         vcpu->arch.pio.count = 0;
4505                 r = EMULATE_DO_MMIO;
4506         } else if (vcpu->mmio_needed)
4507                 r = EMULATE_DO_MMIO;
4508         else if (r == EMULATION_RESTART)
4509                 goto restart;
4510         else
4511                 r = EMULATE_DONE;
4512
4513         toggle_interruptibility(vcpu, vcpu->arch.emulate_ctxt.interruptibility);
4514         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4515         kvm_make_request(KVM_REQ_EVENT, vcpu);
4516         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4517         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4518
4519         return r;
4520 }
4521 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4522
4523 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4524 {
4525         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4526         int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
4527         /* do not return to emulator after return from userspace */
4528         vcpu->arch.pio.count = 0;
4529         return ret;
4530 }
4531 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4532
4533 static void tsc_bad(void *info)
4534 {
4535         __this_cpu_write(cpu_tsc_khz, 0);
4536 }
4537
4538 static void tsc_khz_changed(void *data)
4539 {
4540         struct cpufreq_freqs *freq = data;
4541         unsigned long khz = 0;
4542
4543         if (data)
4544                 khz = freq->new;
4545         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4546                 khz = cpufreq_quick_get(raw_smp_processor_id());
4547         if (!khz)
4548                 khz = tsc_khz;
4549         __this_cpu_write(cpu_tsc_khz, khz);
4550 }
4551
4552 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4553                                      void *data)
4554 {
4555         struct cpufreq_freqs *freq = data;
4556         struct kvm *kvm;
4557         struct kvm_vcpu *vcpu;
4558         int i, send_ipi = 0;
4559
4560         /*
4561          * We allow guests to temporarily run on slowing clocks,
4562          * provided we notify them after, or to run on accelerating
4563          * clocks, provided we notify them before.  Thus time never
4564          * goes backwards.
4565          *
4566          * However, we have a problem.  We can't atomically update
4567          * the frequency of a given CPU from this function; it is
4568          * merely a notifier, which can be called from any CPU.
4569          * Changing the TSC frequency at arbitrary points in time
4570          * requires a recomputation of local variables related to
4571          * the TSC for each VCPU.  We must flag these local variables
4572          * to be updated and be sure the update takes place with the
4573          * new frequency before any guests proceed.
4574          *
4575          * Unfortunately, the combination of hotplug CPU and frequency
4576          * change creates an intractable locking scenario; the order
4577          * of when these callouts happen is undefined with respect to
4578          * CPU hotplug, and they can race with each other.  As such,
4579          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4580          * undefined; you can actually have a CPU frequency change take
4581          * place in between the computation of X and the setting of the
4582          * variable.  To protect against this problem, all updates of
4583          * the per_cpu tsc_khz variable are done in an interrupt
4584          * protected IPI, and all callers wishing to update the value
4585          * must wait for a synchronous IPI to complete (which is trivial
4586          * if the caller is on the CPU already).  This establishes the
4587          * necessary total order on variable updates.
4588          *
4589          * Note that because a guest time update may take place
4590          * anytime after the setting of the VCPU's request bit, the
4591          * correct TSC value must be set before the request.  However,
4592          * to ensure the update actually makes it to any guest which
4593          * starts running in hardware virtualization between the set
4594          * and the acquisition of the spinlock, we must also ping the
4595          * CPU after setting the request bit.
4596          *
4597          */
4598
4599         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4600                 return 0;
4601         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4602                 return 0;
4603
4604         smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4605
4606         raw_spin_lock(&kvm_lock);
4607         list_for_each_entry(kvm, &vm_list, vm_list) {
4608                 kvm_for_each_vcpu(i, vcpu, kvm) {
4609                         if (vcpu->cpu != freq->cpu)
4610                                 continue;
4611                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4612                         if (vcpu->cpu != smp_processor_id())
4613                                 send_ipi = 1;
4614                 }
4615         }
4616         raw_spin_unlock(&kvm_lock);
4617
4618         if (freq->old < freq->new && send_ipi) {
4619                 /*
4620                  * We upscale the frequency.  Must make the guest
4621                  * doesn't see old kvmclock values while running with
4622                  * the new frequency, otherwise we risk the guest sees
4623                  * time go backwards.
4624                  *
4625                  * In case we update the frequency for another cpu
4626                  * (which might be in guest context) send an interrupt
4627                  * to kick the cpu out of guest context.  Next time
4628                  * guest context is entered kvmclock will be updated,
4629                  * so the guest will not see stale values.
4630                  */
4631                 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4632         }
4633         return 0;
4634 }
4635
4636 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4637         .notifier_call  = kvmclock_cpufreq_notifier
4638 };
4639
4640 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4641                                         unsigned long action, void *hcpu)
4642 {
4643         unsigned int cpu = (unsigned long)hcpu;
4644
4645         switch (action) {
4646                 case CPU_ONLINE:
4647                 case CPU_DOWN_FAILED:
4648                         smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4649                         break;
4650                 case CPU_DOWN_PREPARE:
4651                         smp_call_function_single(cpu, tsc_bad, NULL, 1);
4652                         break;
4653         }
4654         return NOTIFY_OK;
4655 }
4656
4657 static struct notifier_block kvmclock_cpu_notifier_block = {
4658         .notifier_call  = kvmclock_cpu_notifier,
4659         .priority = -INT_MAX
4660 };
4661
4662 static void kvm_timer_init(void)
4663 {
4664         int cpu;
4665
4666         max_tsc_khz = tsc_khz;
4667         register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4668         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4669 #ifdef CONFIG_CPU_FREQ
4670                 struct cpufreq_policy policy;
4671                 memset(&policy, 0, sizeof(policy));
4672                 cpu = get_cpu();
4673                 cpufreq_get_policy(&policy, cpu);
4674                 if (policy.cpuinfo.max_freq)
4675                         max_tsc_khz = policy.cpuinfo.max_freq;
4676                 put_cpu();
4677 #endif
4678                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4679                                           CPUFREQ_TRANSITION_NOTIFIER);
4680         }
4681         pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4682         for_each_online_cpu(cpu)
4683                 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4684 }
4685
4686 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4687
4688 static int kvm_is_in_guest(void)
4689 {
4690         return percpu_read(current_vcpu) != NULL;
4691 }
4692
4693 static int kvm_is_user_mode(void)
4694 {
4695         int user_mode = 3;
4696
4697         if (percpu_read(current_vcpu))
4698                 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4699
4700         return user_mode != 0;
4701 }
4702
4703 static unsigned long kvm_get_guest_ip(void)
4704 {
4705         unsigned long ip = 0;
4706
4707         if (percpu_read(current_vcpu))
4708                 ip = kvm_rip_read(percpu_read(current_vcpu));
4709
4710         return ip;
4711 }
4712
4713 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4714         .is_in_guest            = kvm_is_in_guest,
4715         .is_user_mode           = kvm_is_user_mode,
4716         .get_guest_ip           = kvm_get_guest_ip,
4717 };
4718
4719 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4720 {
4721         percpu_write(current_vcpu, vcpu);
4722 }
4723 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4724
4725 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4726 {
4727         percpu_write(current_vcpu, NULL);
4728 }
4729 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4730
4731 int kvm_arch_init(void *opaque)
4732 {
4733         int r;
4734         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4735
4736         if (kvm_x86_ops) {
4737                 printk(KERN_ERR "kvm: already loaded the other module\n");
4738                 r = -EEXIST;
4739                 goto out;
4740         }
4741
4742         if (!ops->cpu_has_kvm_support()) {
4743                 printk(KERN_ERR "kvm: no hardware support\n");
4744                 r = -EOPNOTSUPP;
4745                 goto out;
4746         }
4747         if (ops->disabled_by_bios()) {
4748                 printk(KERN_ERR "kvm: disabled by bios\n");
4749                 r = -EOPNOTSUPP;
4750                 goto out;
4751         }
4752
4753         r = kvm_mmu_module_init();
4754         if (r)
4755                 goto out;
4756
4757         kvm_init_msr_list();
4758
4759         kvm_x86_ops = ops;
4760         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4761         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4762                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
4763
4764         kvm_timer_init();
4765
4766         perf_register_guest_info_callbacks(&kvm_guest_cbs);
4767
4768         if (cpu_has_xsave)
4769                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4770
4771         return 0;
4772
4773 out:
4774         return r;
4775 }
4776
4777 void kvm_arch_exit(void)
4778 {
4779         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4780
4781         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4782                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4783                                             CPUFREQ_TRANSITION_NOTIFIER);
4784         unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4785         kvm_x86_ops = NULL;
4786         kvm_mmu_module_exit();
4787 }
4788
4789 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4790 {
4791         ++vcpu->stat.halt_exits;
4792         if (irqchip_in_kernel(vcpu->kvm)) {
4793                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4794                 return 1;
4795         } else {
4796                 vcpu->run->exit_reason = KVM_EXIT_HLT;
4797                 return 0;
4798         }
4799 }
4800 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4801
4802 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4803                            unsigned long a1)
4804 {
4805         if (is_long_mode(vcpu))
4806                 return a0;
4807         else
4808                 return a0 | ((gpa_t)a1 << 32);
4809 }
4810
4811 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4812 {
4813         u64 param, ingpa, outgpa, ret;
4814         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4815         bool fast, longmode;
4816         int cs_db, cs_l;
4817
4818         /*
4819          * hypercall generates UD from non zero cpl and real mode
4820          * per HYPER-V spec
4821          */
4822         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4823                 kvm_queue_exception(vcpu, UD_VECTOR);
4824                 return 0;
4825         }
4826
4827         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4828         longmode = is_long_mode(vcpu) && cs_l == 1;
4829
4830         if (!longmode) {
4831                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4832                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4833                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4834                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4835                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4836                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4837         }
4838 #ifdef CONFIG_X86_64
4839         else {
4840                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4841                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4842                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4843         }
4844 #endif
4845
4846         code = param & 0xffff;
4847         fast = (param >> 16) & 0x1;
4848         rep_cnt = (param >> 32) & 0xfff;
4849         rep_idx = (param >> 48) & 0xfff;
4850
4851         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4852
4853         switch (code) {
4854         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4855                 kvm_vcpu_on_spin(vcpu);
4856                 break;
4857         default:
4858                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4859                 break;
4860         }
4861
4862         ret = res | (((u64)rep_done & 0xfff) << 32);
4863         if (longmode) {
4864                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4865         } else {
4866                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4867                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4868         }
4869
4870         return 1;
4871 }
4872
4873 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4874 {
4875         unsigned long nr, a0, a1, a2, a3, ret;
4876         int r = 1;
4877
4878         if (kvm_hv_hypercall_enabled(vcpu->kvm))
4879                 return kvm_hv_hypercall(vcpu);
4880
4881         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4882         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4883         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4884         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4885         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4886
4887         trace_kvm_hypercall(nr, a0, a1, a2, a3);
4888
4889         if (!is_long_mode(vcpu)) {
4890                 nr &= 0xFFFFFFFF;
4891                 a0 &= 0xFFFFFFFF;
4892                 a1 &= 0xFFFFFFFF;
4893                 a2 &= 0xFFFFFFFF;
4894                 a3 &= 0xFFFFFFFF;
4895         }
4896
4897         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4898                 ret = -KVM_EPERM;
4899                 goto out;
4900         }
4901
4902         switch (nr) {
4903         case KVM_HC_VAPIC_POLL_IRQ:
4904                 ret = 0;
4905                 break;
4906         case KVM_HC_MMU_OP:
4907                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4908                 break;
4909         default:
4910                 ret = -KVM_ENOSYS;
4911                 break;
4912         }
4913 out:
4914         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4915         ++vcpu->stat.hypercalls;
4916         return r;
4917 }
4918 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4919
4920 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4921 {
4922         char instruction[3];
4923         unsigned long rip = kvm_rip_read(vcpu);
4924
4925         /*
4926          * Blow out the MMU to ensure that no other VCPU has an active mapping
4927          * to ensure that the updated hypercall appears atomically across all
4928          * VCPUs.
4929          */
4930         kvm_mmu_zap_all(vcpu->kvm);
4931
4932         kvm_x86_ops->patch_hypercall(vcpu, instruction);
4933
4934         return emulator_write_emulated(rip, instruction, 3, NULL, vcpu);
4935 }
4936
4937 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4938 {
4939         struct desc_ptr dt = { limit, base };
4940
4941         kvm_x86_ops->set_gdt(vcpu, &dt);
4942 }
4943
4944 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4945 {
4946         struct desc_ptr dt = { limit, base };
4947
4948         kvm_x86_ops->set_idt(vcpu, &dt);
4949 }
4950
4951 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4952 {
4953         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4954         int j, nent = vcpu->arch.cpuid_nent;
4955
4956         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4957         /* when no next entry is found, the current entry[i] is reselected */
4958         for (j = i + 1; ; j = (j + 1) % nent) {
4959                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4960                 if (ej->function == e->function) {
4961                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4962                         return j;
4963                 }
4964         }
4965         return 0; /* silence gcc, even though control never reaches here */
4966 }
4967
4968 /* find an entry with matching function, matching index (if needed), and that
4969  * should be read next (if it's stateful) */
4970 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4971         u32 function, u32 index)
4972 {
4973         if (e->function != function)
4974                 return 0;
4975         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4976                 return 0;
4977         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4978             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4979                 return 0;
4980         return 1;
4981 }
4982
4983 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4984                                               u32 function, u32 index)
4985 {
4986         int i;
4987         struct kvm_cpuid_entry2 *best = NULL;
4988
4989         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4990                 struct kvm_cpuid_entry2 *e;
4991
4992                 e = &vcpu->arch.cpuid_entries[i];
4993                 if (is_matching_cpuid_entry(e, function, index)) {
4994                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4995                                 move_to_next_stateful_cpuid_entry(vcpu, i);
4996                         best = e;
4997                         break;
4998                 }
4999         }
5000         return best;
5001 }
5002 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5003
5004 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5005 {
5006         struct kvm_cpuid_entry2 *best;
5007
5008         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5009         if (!best || best->eax < 0x80000008)
5010                 goto not_found;
5011         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5012         if (best)
5013                 return best->eax & 0xff;
5014 not_found:
5015         return 36;
5016 }
5017
5018 /*
5019  * If no match is found, check whether we exceed the vCPU's limit
5020  * and return the content of the highest valid _standard_ leaf instead.
5021  * This is to satisfy the CPUID specification.
5022  */
5023 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5024                                                   u32 function, u32 index)
5025 {
5026         struct kvm_cpuid_entry2 *maxlevel;
5027
5028         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5029         if (!maxlevel || maxlevel->eax >= function)
5030                 return NULL;
5031         if (function & 0x80000000) {
5032                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5033                 if (!maxlevel)
5034                         return NULL;
5035         }
5036         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5037 }
5038
5039 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5040 {
5041         u32 function, index;
5042         struct kvm_cpuid_entry2 *best;
5043
5044         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5045         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5046         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5047         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5048         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5049         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5050         best = kvm_find_cpuid_entry(vcpu, function, index);
5051
5052         if (!best)
5053                 best = check_cpuid_limit(vcpu, function, index);
5054
5055         if (best) {
5056                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5057                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5058                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5059                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5060         }
5061         kvm_x86_ops->skip_emulated_instruction(vcpu);
5062         trace_kvm_cpuid(function,
5063                         kvm_register_read(vcpu, VCPU_REGS_RAX),
5064                         kvm_register_read(vcpu, VCPU_REGS_RBX),
5065                         kvm_register_read(vcpu, VCPU_REGS_RCX),
5066                         kvm_register_read(vcpu, VCPU_REGS_RDX));
5067 }
5068 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5069
5070 /*
5071  * Check if userspace requested an interrupt window, and that the
5072  * interrupt window is open.
5073  *
5074  * No need to exit to userspace if we already have an interrupt queued.
5075  */
5076 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5077 {
5078         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5079                 vcpu->run->request_interrupt_window &&
5080                 kvm_arch_interrupt_allowed(vcpu));
5081 }
5082
5083 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5084 {
5085         struct kvm_run *kvm_run = vcpu->run;
5086
5087         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5088         kvm_run->cr8 = kvm_get_cr8(vcpu);
5089         kvm_run->apic_base = kvm_get_apic_base(vcpu);
5090         if (irqchip_in_kernel(vcpu->kvm))
5091                 kvm_run->ready_for_interrupt_injection = 1;
5092         else
5093                 kvm_run->ready_for_interrupt_injection =
5094                         kvm_arch_interrupt_allowed(vcpu) &&
5095                         !kvm_cpu_has_interrupt(vcpu) &&
5096                         !kvm_event_needs_reinjection(vcpu);
5097 }
5098
5099 static void vapic_enter(struct kvm_vcpu *vcpu)
5100 {
5101         struct kvm_lapic *apic = vcpu->arch.apic;
5102         struct page *page;
5103
5104         if (!apic || !apic->vapic_addr)
5105                 return;
5106
5107         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5108
5109         vcpu->arch.apic->vapic_page = page;
5110 }
5111
5112 static void vapic_exit(struct kvm_vcpu *vcpu)
5113 {
5114         struct kvm_lapic *apic = vcpu->arch.apic;
5115         int idx;
5116
5117         if (!apic || !apic->vapic_addr)
5118                 return;
5119
5120         idx = srcu_read_lock(&vcpu->kvm->srcu);
5121         kvm_release_page_dirty(apic->vapic_page);
5122         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5123         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5124 }
5125
5126 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5127 {
5128         int max_irr, tpr;
5129
5130         if (!kvm_x86_ops->update_cr8_intercept)
5131                 return;
5132
5133         if (!vcpu->arch.apic)
5134                 return;
5135
5136         if (!vcpu->arch.apic->vapic_addr)
5137                 max_irr = kvm_lapic_find_highest_irr(vcpu);
5138         else
5139                 max_irr = -1;
5140
5141         if (max_irr != -1)
5142                 max_irr >>= 4;
5143
5144         tpr = kvm_lapic_get_cr8(vcpu);
5145
5146         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5147 }
5148
5149 static void inject_pending_event(struct kvm_vcpu *vcpu)
5150 {
5151         /* try to reinject previous events if any */
5152         if (vcpu->arch.exception.pending) {
5153                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5154                                         vcpu->arch.exception.has_error_code,
5155                                         vcpu->arch.exception.error_code);
5156                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5157                                           vcpu->arch.exception.has_error_code,
5158                                           vcpu->arch.exception.error_code,
5159                                           vcpu->arch.exception.reinject);
5160                 return;
5161         }
5162
5163         if (vcpu->arch.nmi_injected) {
5164                 kvm_x86_ops->set_nmi(vcpu);
5165                 return;
5166         }
5167
5168         if (vcpu->arch.interrupt.pending) {
5169                 kvm_x86_ops->set_irq(vcpu);
5170                 return;
5171         }
5172
5173         /* try to inject new event if pending */
5174         if (vcpu->arch.nmi_pending) {
5175                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5176                         vcpu->arch.nmi_pending = false;
5177                         vcpu->arch.nmi_injected = true;
5178                         kvm_x86_ops->set_nmi(vcpu);
5179                 }
5180         } else if (kvm_cpu_has_interrupt(vcpu)) {
5181                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5182                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5183                                             false);
5184                         kvm_x86_ops->set_irq(vcpu);
5185                 }
5186         }
5187 }
5188
5189 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5190 {
5191         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5192                         !vcpu->guest_xcr0_loaded) {
5193                 /* kvm_set_xcr() also depends on this */
5194                 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5195                 vcpu->guest_xcr0_loaded = 1;
5196         }
5197 }
5198
5199 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5200 {
5201         if (vcpu->guest_xcr0_loaded) {
5202                 if (vcpu->arch.xcr0 != host_xcr0)
5203                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5204                 vcpu->guest_xcr0_loaded = 0;
5205         }
5206 }
5207
5208 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5209 {
5210         int r;
5211         bool nmi_pending;
5212         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5213                 vcpu->run->request_interrupt_window;
5214
5215         if (vcpu->requests) {
5216                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5217                         kvm_mmu_unload(vcpu);
5218                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5219                         __kvm_migrate_timers(vcpu);
5220                 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5221                         r = kvm_guest_time_update(vcpu);
5222                         if (unlikely(r))
5223                                 goto out;
5224                 }
5225                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5226                         kvm_mmu_sync_roots(vcpu);
5227                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5228                         kvm_x86_ops->tlb_flush(vcpu);
5229                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5230                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5231                         r = 0;
5232                         goto out;
5233                 }
5234                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5235                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5236                         r = 0;
5237                         goto out;
5238                 }
5239                 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5240                         vcpu->fpu_active = 0;
5241                         kvm_x86_ops->fpu_deactivate(vcpu);
5242                 }
5243                 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5244                         /* Page is swapped out. Do synthetic halt */
5245                         vcpu->arch.apf.halted = true;
5246                         r = 1;
5247                         goto out;
5248                 }
5249         }
5250
5251         r = kvm_mmu_reload(vcpu);
5252         if (unlikely(r))
5253                 goto out;
5254
5255         /*
5256          * An NMI can be injected between local nmi_pending read and
5257          * vcpu->arch.nmi_pending read inside inject_pending_event().
5258          * But in that case, KVM_REQ_EVENT will be set, which makes
5259          * the race described above benign.
5260          */
5261         nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5262
5263         if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5264                 inject_pending_event(vcpu);
5265
5266                 /* enable NMI/IRQ window open exits if needed */
5267                 if (nmi_pending)
5268                         kvm_x86_ops->enable_nmi_window(vcpu);
5269                 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5270                         kvm_x86_ops->enable_irq_window(vcpu);
5271
5272                 if (kvm_lapic_enabled(vcpu)) {
5273                         update_cr8_intercept(vcpu);
5274                         kvm_lapic_sync_to_vapic(vcpu);
5275                 }
5276         }
5277
5278         preempt_disable();
5279
5280         kvm_x86_ops->prepare_guest_switch(vcpu);
5281         if (vcpu->fpu_active)
5282                 kvm_load_guest_fpu(vcpu);
5283         kvm_load_guest_xcr0(vcpu);
5284
5285         vcpu->mode = IN_GUEST_MODE;
5286
5287         /* We should set ->mode before check ->requests,
5288          * see the comment in make_all_cpus_request.
5289          */
5290         smp_mb();
5291
5292         local_irq_disable();
5293
5294         if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5295             || need_resched() || signal_pending(current)) {
5296                 vcpu->mode = OUTSIDE_GUEST_MODE;
5297                 smp_wmb();
5298                 local_irq_enable();
5299                 preempt_enable();
5300                 kvm_x86_ops->cancel_injection(vcpu);
5301                 r = 1;
5302                 goto out;
5303         }
5304
5305         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5306
5307         kvm_guest_enter();
5308
5309         if (unlikely(vcpu->arch.switch_db_regs)) {
5310                 set_debugreg(0, 7);
5311                 set_debugreg(vcpu->arch.eff_db[0], 0);
5312                 set_debugreg(vcpu->arch.eff_db[1], 1);
5313                 set_debugreg(vcpu->arch.eff_db[2], 2);
5314                 set_debugreg(vcpu->arch.eff_db[3], 3);
5315         }
5316
5317         trace_kvm_entry(vcpu->vcpu_id);
5318         kvm_x86_ops->run(vcpu);
5319
5320         /*
5321          * If the guest has used debug registers, at least dr7
5322          * will be disabled while returning to the host.
5323          * If we don't have active breakpoints in the host, we don't
5324          * care about the messed up debug address registers. But if
5325          * we have some of them active, restore the old state.
5326          */
5327         if (hw_breakpoint_active())
5328                 hw_breakpoint_restore();
5329
5330         kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5331
5332         vcpu->mode = OUTSIDE_GUEST_MODE;
5333         smp_wmb();
5334         local_irq_enable();
5335
5336         ++vcpu->stat.exits;
5337
5338         /*
5339          * We must have an instruction between local_irq_enable() and
5340          * kvm_guest_exit(), so the timer interrupt isn't delayed by
5341          * the interrupt shadow.  The stat.exits increment will do nicely.
5342          * But we need to prevent reordering, hence this barrier():
5343          */
5344         barrier();
5345
5346         kvm_guest_exit();
5347
5348         preempt_enable();
5349
5350         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5351
5352         /*
5353          * Profile KVM exit RIPs:
5354          */
5355         if (unlikely(prof_on == KVM_PROFILING)) {
5356                 unsigned long rip = kvm_rip_read(vcpu);
5357                 profile_hit(KVM_PROFILING, (void *)rip);
5358         }
5359
5360
5361         kvm_lapic_sync_from_vapic(vcpu);
5362
5363         r = kvm_x86_ops->handle_exit(vcpu);
5364 out:
5365         return r;
5366 }
5367
5368
5369 static int __vcpu_run(struct kvm_vcpu *vcpu)
5370 {
5371         int r;
5372         struct kvm *kvm = vcpu->kvm;
5373
5374         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5375                 pr_debug("vcpu %d received sipi with vector # %x\n",
5376                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
5377                 kvm_lapic_reset(vcpu);
5378                 r = kvm_arch_vcpu_reset(vcpu);
5379                 if (r)
5380                         return r;
5381                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5382         }
5383
5384         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5385         vapic_enter(vcpu);
5386
5387         r = 1;
5388         while (r > 0) {
5389                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5390                     !vcpu->arch.apf.halted)
5391                         r = vcpu_enter_guest(vcpu);
5392                 else {
5393                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5394                         kvm_vcpu_block(vcpu);
5395                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5396                         if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5397                         {
5398                                 switch(vcpu->arch.mp_state) {
5399                                 case KVM_MP_STATE_HALTED:
5400                                         vcpu->arch.mp_state =
5401                                                 KVM_MP_STATE_RUNNABLE;
5402                                 case KVM_MP_STATE_RUNNABLE:
5403                                         vcpu->arch.apf.halted = false;
5404                                         break;
5405                                 case KVM_MP_STATE_SIPI_RECEIVED:
5406                                 default:
5407                                         r = -EINTR;
5408                                         break;
5409                                 }
5410                         }
5411                 }
5412
5413                 if (r <= 0)
5414                         break;
5415
5416                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5417                 if (kvm_cpu_has_pending_timer(vcpu))
5418                         kvm_inject_pending_timer_irqs(vcpu);
5419
5420                 if (dm_request_for_irq_injection(vcpu)) {
5421                         r = -EINTR;
5422                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5423                         ++vcpu->stat.request_irq_exits;
5424                 }
5425
5426                 kvm_check_async_pf_completion(vcpu);
5427
5428                 if (signal_pending(current)) {
5429                         r = -EINTR;
5430                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5431                         ++vcpu->stat.signal_exits;
5432                 }
5433                 if (need_resched()) {
5434                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5435                         kvm_resched(vcpu);
5436                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5437                 }
5438         }
5439
5440         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5441
5442         vapic_exit(vcpu);
5443
5444         return r;
5445 }
5446
5447 static int complete_mmio(struct kvm_vcpu *vcpu)
5448 {
5449         struct kvm_run *run = vcpu->run;
5450         int r;
5451
5452         if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5453                 return 1;
5454
5455         if (vcpu->mmio_needed) {
5456                 vcpu->mmio_needed = 0;
5457                 if (!vcpu->mmio_is_write)
5458                         memcpy(vcpu->mmio_data, run->mmio.data, 8);
5459                 vcpu->mmio_index += 8;
5460                 if (vcpu->mmio_index < vcpu->mmio_size) {
5461                         run->exit_reason = KVM_EXIT_MMIO;
5462                         run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5463                         memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5464                         run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5465                         run->mmio.is_write = vcpu->mmio_is_write;
5466                         vcpu->mmio_needed = 1;
5467                         return 0;
5468                 }
5469                 if (vcpu->mmio_is_write)
5470                         return 1;
5471                 vcpu->mmio_read_completed = 1;
5472         }
5473         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5474         r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5475         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5476         if (r != EMULATE_DONE)
5477                 return 0;
5478         return 1;
5479 }
5480
5481 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5482 {
5483         int r;
5484         sigset_t sigsaved;
5485
5486         if (!tsk_used_math(current) && init_fpu(current))
5487                 return -ENOMEM;
5488
5489         if (vcpu->sigset_active)
5490                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5491
5492         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5493                 kvm_vcpu_block(vcpu);
5494                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5495                 r = -EAGAIN;
5496                 goto out;
5497         }
5498
5499         /* re-sync apic's tpr */
5500         if (!irqchip_in_kernel(vcpu->kvm)) {
5501                 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5502                         r = -EINVAL;
5503                         goto out;
5504                 }
5505         }
5506
5507         r = complete_mmio(vcpu);
5508         if (r <= 0)
5509                 goto out;
5510
5511         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5512                 kvm_register_write(vcpu, VCPU_REGS_RAX,
5513                                      kvm_run->hypercall.ret);
5514
5515         r = __vcpu_run(vcpu);
5516
5517 out:
5518         post_kvm_run_save(vcpu);
5519         if (vcpu->sigset_active)
5520                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5521
5522         return r;
5523 }
5524
5525 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5526 {
5527         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5528         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5529         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5530         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5531         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5532         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5533         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5534         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5535 #ifdef CONFIG_X86_64
5536         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5537         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5538         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5539         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5540         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5541         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5542         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5543         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5544 #endif
5545
5546         regs->rip = kvm_rip_read(vcpu);
5547         regs->rflags = kvm_get_rflags(vcpu);
5548
5549         return 0;
5550 }
5551
5552 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5553 {
5554         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5555         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5556         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5557         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5558         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5559         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5560         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5561         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5562 #ifdef CONFIG_X86_64
5563         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5564         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5565         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5566         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5567         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5568         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5569         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5570         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5571 #endif
5572
5573         kvm_rip_write(vcpu, regs->rip);
5574         kvm_set_rflags(vcpu, regs->rflags);
5575
5576         vcpu->arch.exception.pending = false;
5577
5578         kvm_make_request(KVM_REQ_EVENT, vcpu);
5579
5580         return 0;
5581 }
5582
5583 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5584 {
5585         struct kvm_segment cs;
5586
5587         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5588         *db = cs.db;
5589         *l = cs.l;
5590 }
5591 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5592
5593 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5594                                   struct kvm_sregs *sregs)
5595 {
5596         struct desc_ptr dt;
5597
5598         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5599         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5600         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5601         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5602         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5603         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5604
5605         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5606         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5607
5608         kvm_x86_ops->get_idt(vcpu, &dt);
5609         sregs->idt.limit = dt.size;
5610         sregs->idt.base = dt.address;
5611         kvm_x86_ops->get_gdt(vcpu, &dt);
5612         sregs->gdt.limit = dt.size;
5613         sregs->gdt.base = dt.address;
5614
5615         sregs->cr0 = kvm_read_cr0(vcpu);
5616         sregs->cr2 = vcpu->arch.cr2;
5617         sregs->cr3 = kvm_read_cr3(vcpu);
5618         sregs->cr4 = kvm_read_cr4(vcpu);
5619         sregs->cr8 = kvm_get_cr8(vcpu);
5620         sregs->efer = vcpu->arch.efer;
5621         sregs->apic_base = kvm_get_apic_base(vcpu);
5622
5623         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5624
5625         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5626                 set_bit(vcpu->arch.interrupt.nr,
5627                         (unsigned long *)sregs->interrupt_bitmap);
5628
5629         return 0;
5630 }
5631
5632 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5633                                     struct kvm_mp_state *mp_state)
5634 {
5635         mp_state->mp_state = vcpu->arch.mp_state;
5636         return 0;
5637 }
5638
5639 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5640                                     struct kvm_mp_state *mp_state)
5641 {
5642         vcpu->arch.mp_state = mp_state->mp_state;
5643         kvm_make_request(KVM_REQ_EVENT, vcpu);
5644         return 0;
5645 }
5646
5647 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5648                     bool has_error_code, u32 error_code)
5649 {
5650         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5651         int ret;
5652
5653         init_emulate_ctxt(vcpu);
5654
5655         ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5656                                    tss_selector, reason, has_error_code,
5657                                    error_code);
5658
5659         if (ret)
5660                 return EMULATE_FAIL;
5661
5662         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5663         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5664         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5665         kvm_make_request(KVM_REQ_EVENT, vcpu);
5666         return EMULATE_DONE;
5667 }
5668 EXPORT_SYMBOL_GPL(kvm_task_switch);
5669
5670 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5671                                   struct kvm_sregs *sregs)
5672 {
5673         int mmu_reset_needed = 0;
5674         int pending_vec, max_bits, idx;
5675         struct desc_ptr dt;
5676
5677         dt.size = sregs->idt.limit;
5678         dt.address = sregs->idt.base;
5679         kvm_x86_ops->set_idt(vcpu, &dt);
5680         dt.size = sregs->gdt.limit;
5681         dt.address = sregs->gdt.base;
5682         kvm_x86_ops->set_gdt(vcpu, &dt);
5683
5684         vcpu->arch.cr2 = sregs->cr2;
5685         mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
5686         vcpu->arch.cr3 = sregs->cr3;
5687         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
5688
5689         kvm_set_cr8(vcpu, sregs->cr8);
5690
5691         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5692         kvm_x86_ops->set_efer(vcpu, sregs->efer);
5693         kvm_set_apic_base(vcpu, sregs->apic_base);
5694
5695         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5696         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5697         vcpu->arch.cr0 = sregs->cr0;
5698
5699         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5700         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5701         if (sregs->cr4 & X86_CR4_OSXSAVE)
5702                 update_cpuid(vcpu);
5703
5704         idx = srcu_read_lock(&vcpu->kvm->srcu);
5705         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5706                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
5707                 mmu_reset_needed = 1;
5708         }
5709         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5710
5711         if (mmu_reset_needed)
5712                 kvm_mmu_reset_context(vcpu);
5713
5714         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5715         pending_vec = find_first_bit(
5716                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5717         if (pending_vec < max_bits) {
5718                 kvm_queue_interrupt(vcpu, pending_vec, false);
5719                 pr_debug("Set back pending irq %d\n", pending_vec);
5720         }
5721
5722         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5723         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5724         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5725         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5726         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5727         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5728
5729         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5730         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5731
5732         update_cr8_intercept(vcpu);
5733
5734         /* Older userspace won't unhalt the vcpu on reset. */
5735         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5736             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5737             !is_protmode(vcpu))
5738                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5739
5740         kvm_make_request(KVM_REQ_EVENT, vcpu);
5741
5742         return 0;
5743 }
5744
5745 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5746                                         struct kvm_guest_debug *dbg)
5747 {
5748         unsigned long rflags;
5749         int i, r;
5750
5751         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5752                 r = -EBUSY;
5753                 if (vcpu->arch.exception.pending)
5754                         goto out;
5755                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5756                         kvm_queue_exception(vcpu, DB_VECTOR);
5757                 else
5758                         kvm_queue_exception(vcpu, BP_VECTOR);
5759         }
5760
5761         /*
5762          * Read rflags as long as potentially injected trace flags are still
5763          * filtered out.
5764          */
5765         rflags = kvm_get_rflags(vcpu);
5766
5767         vcpu->guest_debug = dbg->control;
5768         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5769                 vcpu->guest_debug = 0;
5770
5771         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5772                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5773                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5774                 vcpu->arch.switch_db_regs =
5775                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5776         } else {
5777                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5778                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5779                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5780         }
5781
5782         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5783                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5784                         get_segment_base(vcpu, VCPU_SREG_CS);
5785
5786         /*
5787          * Trigger an rflags update that will inject or remove the trace
5788          * flags.
5789          */
5790         kvm_set_rflags(vcpu, rflags);
5791
5792         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5793
5794         r = 0;
5795
5796 out:
5797
5798         return r;
5799 }
5800
5801 /*
5802  * Translate a guest virtual address to a guest physical address.
5803  */
5804 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5805                                     struct kvm_translation *tr)
5806 {
5807         unsigned long vaddr = tr->linear_address;
5808         gpa_t gpa;
5809         int idx;
5810
5811         idx = srcu_read_lock(&vcpu->kvm->srcu);
5812         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5813         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5814         tr->physical_address = gpa;
5815         tr->valid = gpa != UNMAPPED_GVA;
5816         tr->writeable = 1;
5817         tr->usermode = 0;
5818
5819         return 0;
5820 }
5821
5822 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5823 {
5824         struct i387_fxsave_struct *fxsave =
5825                         &vcpu->arch.guest_fpu.state->fxsave;
5826
5827         memcpy(fpu->fpr, fxsave->st_space, 128);
5828         fpu->fcw = fxsave->cwd;
5829         fpu->fsw = fxsave->swd;
5830         fpu->ftwx = fxsave->twd;
5831         fpu->last_opcode = fxsave->fop;
5832         fpu->last_ip = fxsave->rip;
5833         fpu->last_dp = fxsave->rdp;
5834         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5835
5836         return 0;
5837 }
5838
5839 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5840 {
5841         struct i387_fxsave_struct *fxsave =
5842                         &vcpu->arch.guest_fpu.state->fxsave;
5843
5844         memcpy(fxsave->st_space, fpu->fpr, 128);
5845         fxsave->cwd = fpu->fcw;
5846         fxsave->swd = fpu->fsw;
5847         fxsave->twd = fpu->ftwx;
5848         fxsave->fop = fpu->last_opcode;
5849         fxsave->rip = fpu->last_ip;
5850         fxsave->rdp = fpu->last_dp;
5851         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5852
5853         return 0;
5854 }
5855
5856 int fx_init(struct kvm_vcpu *vcpu)
5857 {
5858         int err;
5859
5860         err = fpu_alloc(&vcpu->arch.guest_fpu);
5861         if (err)
5862                 return err;
5863
5864         fpu_finit(&vcpu->arch.guest_fpu);
5865
5866         /*
5867          * Ensure guest xcr0 is valid for loading
5868          */
5869         vcpu->arch.xcr0 = XSTATE_FP;
5870
5871         vcpu->arch.cr0 |= X86_CR0_ET;
5872
5873         return 0;
5874 }
5875 EXPORT_SYMBOL_GPL(fx_init);
5876
5877 static void fx_free(struct kvm_vcpu *vcpu)
5878 {
5879         fpu_free(&vcpu->arch.guest_fpu);
5880 }
5881
5882 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5883 {
5884         if (vcpu->guest_fpu_loaded)
5885                 return;
5886
5887         /*
5888          * Restore all possible states in the guest,
5889          * and assume host would use all available bits.
5890          * Guest xcr0 would be loaded later.
5891          */
5892         kvm_put_guest_xcr0(vcpu);
5893         vcpu->guest_fpu_loaded = 1;
5894         unlazy_fpu(current);
5895         fpu_restore_checking(&vcpu->arch.guest_fpu);
5896         trace_kvm_fpu(1);
5897 }
5898
5899 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5900 {
5901         kvm_put_guest_xcr0(vcpu);
5902
5903         if (!vcpu->guest_fpu_loaded)
5904                 return;
5905
5906         vcpu->guest_fpu_loaded = 0;
5907         fpu_save_init(&vcpu->arch.guest_fpu);
5908         ++vcpu->stat.fpu_reload;
5909         kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
5910         trace_kvm_fpu(0);
5911 }
5912
5913 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5914 {
5915         kvmclock_reset(vcpu);
5916
5917         free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5918         fx_free(vcpu);
5919         kvm_x86_ops->vcpu_free(vcpu);
5920 }
5921
5922 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5923                                                 unsigned int id)
5924 {
5925         if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5926                 printk_once(KERN_WARNING
5927                 "kvm: SMP vm created on host with unstable TSC; "
5928                 "guest TSC will not be reliable\n");
5929         return kvm_x86_ops->vcpu_create(kvm, id);
5930 }
5931
5932 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5933 {
5934         int r;
5935
5936         vcpu->arch.mtrr_state.have_fixed = 1;
5937         vcpu_load(vcpu);
5938         r = kvm_arch_vcpu_reset(vcpu);
5939         if (r == 0)
5940                 r = kvm_mmu_setup(vcpu);
5941         vcpu_put(vcpu);
5942         if (r < 0)
5943                 goto free_vcpu;
5944
5945         return 0;
5946 free_vcpu:
5947         kvm_x86_ops->vcpu_free(vcpu);
5948         return r;
5949 }
5950
5951 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5952 {
5953         vcpu->arch.apf.msr_val = 0;
5954
5955         vcpu_load(vcpu);
5956         kvm_mmu_unload(vcpu);
5957         vcpu_put(vcpu);
5958
5959         fx_free(vcpu);
5960         kvm_x86_ops->vcpu_free(vcpu);
5961 }
5962
5963 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5964 {
5965         vcpu->arch.nmi_pending = false;
5966         vcpu->arch.nmi_injected = false;
5967
5968         vcpu->arch.switch_db_regs = 0;
5969         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5970         vcpu->arch.dr6 = DR6_FIXED_1;
5971         vcpu->arch.dr7 = DR7_FIXED_1;
5972
5973         kvm_make_request(KVM_REQ_EVENT, vcpu);
5974         vcpu->arch.apf.msr_val = 0;
5975
5976         kvmclock_reset(vcpu);
5977
5978         kvm_clear_async_pf_completion_queue(vcpu);
5979         kvm_async_pf_hash_reset(vcpu);
5980         vcpu->arch.apf.halted = false;
5981
5982         return kvm_x86_ops->vcpu_reset(vcpu);
5983 }
5984
5985 int kvm_arch_hardware_enable(void *garbage)
5986 {
5987         struct kvm *kvm;
5988         struct kvm_vcpu *vcpu;
5989         int i;
5990
5991         kvm_shared_msr_cpu_online();
5992         list_for_each_entry(kvm, &vm_list, vm_list)
5993                 kvm_for_each_vcpu(i, vcpu, kvm)
5994                         if (vcpu->cpu == smp_processor_id())
5995                                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5996         return kvm_x86_ops->hardware_enable(garbage);
5997 }
5998
5999 void kvm_arch_hardware_disable(void *garbage)
6000 {
6001         kvm_x86_ops->hardware_disable(garbage);
6002         drop_user_return_notifiers(garbage);
6003 }
6004
6005 int kvm_arch_hardware_setup(void)
6006 {
6007         return kvm_x86_ops->hardware_setup();
6008 }
6009
6010 void kvm_arch_hardware_unsetup(void)
6011 {
6012         kvm_x86_ops->hardware_unsetup();
6013 }
6014
6015 void kvm_arch_check_processor_compat(void *rtn)
6016 {
6017         kvm_x86_ops->check_processor_compatibility(rtn);
6018 }
6019
6020 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6021 {
6022         struct page *page;
6023         struct kvm *kvm;
6024         int r;
6025
6026         BUG_ON(vcpu->kvm == NULL);
6027         kvm = vcpu->kvm;
6028
6029         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6030         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6031         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6032         vcpu->arch.mmu.translate_gpa = translate_gpa;
6033         vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6034         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6035                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6036         else
6037                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6038
6039         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6040         if (!page) {
6041                 r = -ENOMEM;
6042                 goto fail;
6043         }
6044         vcpu->arch.pio_data = page_address(page);
6045
6046         if (!kvm->arch.virtual_tsc_khz)
6047                 kvm_arch_set_tsc_khz(kvm, max_tsc_khz);
6048
6049         r = kvm_mmu_create(vcpu);
6050         if (r < 0)
6051                 goto fail_free_pio_data;
6052
6053         if (irqchip_in_kernel(kvm)) {
6054                 r = kvm_create_lapic(vcpu);
6055                 if (r < 0)
6056                         goto fail_mmu_destroy;
6057         }
6058
6059         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6060                                        GFP_KERNEL);
6061         if (!vcpu->arch.mce_banks) {
6062                 r = -ENOMEM;
6063                 goto fail_free_lapic;
6064         }
6065         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6066
6067         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6068                 goto fail_free_mce_banks;
6069
6070         kvm_async_pf_hash_reset(vcpu);
6071
6072         return 0;
6073 fail_free_mce_banks:
6074         kfree(vcpu->arch.mce_banks);
6075 fail_free_lapic:
6076         kvm_free_lapic(vcpu);
6077 fail_mmu_destroy:
6078         kvm_mmu_destroy(vcpu);
6079 fail_free_pio_data:
6080         free_page((unsigned long)vcpu->arch.pio_data);
6081 fail:
6082         return r;
6083 }
6084
6085 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6086 {
6087         int idx;
6088
6089         kfree(vcpu->arch.mce_banks);
6090         kvm_free_lapic(vcpu);
6091         idx = srcu_read_lock(&vcpu->kvm->srcu);
6092         kvm_mmu_destroy(vcpu);
6093         srcu_read_unlock(&vcpu->kvm->srcu, idx);
6094         free_page((unsigned long)vcpu->arch.pio_data);
6095 }
6096
6097 int kvm_arch_init_vm(struct kvm *kvm)
6098 {
6099         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6100         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6101
6102         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6103         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6104
6105         raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6106
6107         return 0;
6108 }
6109
6110 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6111 {
6112         vcpu_load(vcpu);
6113         kvm_mmu_unload(vcpu);
6114         vcpu_put(vcpu);
6115 }
6116
6117 static void kvm_free_vcpus(struct kvm *kvm)
6118 {
6119         unsigned int i;
6120         struct kvm_vcpu *vcpu;
6121
6122         /*
6123          * Unpin any mmu pages first.
6124          */
6125         kvm_for_each_vcpu(i, vcpu, kvm) {
6126                 kvm_clear_async_pf_completion_queue(vcpu);
6127                 kvm_unload_vcpu_mmu(vcpu);
6128         }
6129         kvm_for_each_vcpu(i, vcpu, kvm)
6130                 kvm_arch_vcpu_free(vcpu);
6131
6132         mutex_lock(&kvm->lock);
6133         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6134                 kvm->vcpus[i] = NULL;
6135
6136         atomic_set(&kvm->online_vcpus, 0);
6137         mutex_unlock(&kvm->lock);
6138 }
6139
6140 void kvm_arch_sync_events(struct kvm *kvm)
6141 {
6142         kvm_free_all_assigned_devices(kvm);
6143         kvm_free_pit(kvm);
6144 }
6145
6146 void kvm_arch_destroy_vm(struct kvm *kvm)
6147 {
6148         kvm_iommu_unmap_guest(kvm);
6149         kfree(kvm->arch.vpic);
6150         kfree(kvm->arch.vioapic);
6151         kvm_free_vcpus(kvm);
6152         if (kvm->arch.apic_access_page)
6153                 put_page(kvm->arch.apic_access_page);
6154         if (kvm->arch.ept_identity_pagetable)
6155                 put_page(kvm->arch.ept_identity_pagetable);
6156 }
6157
6158 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6159                                 struct kvm_memory_slot *memslot,
6160                                 struct kvm_memory_slot old,
6161                                 struct kvm_userspace_memory_region *mem,
6162                                 int user_alloc)
6163 {
6164         int npages = memslot->npages;
6165         int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6166
6167         /* Prevent internal slot pages from being moved by fork()/COW. */
6168         if (memslot->id >= KVM_MEMORY_SLOTS)
6169                 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6170
6171         /*To keep backward compatibility with older userspace,
6172          *x86 needs to hanlde !user_alloc case.
6173          */
6174         if (!user_alloc) {
6175                 if (npages && !old.rmap) {
6176                         unsigned long userspace_addr;
6177
6178                         down_write(&current->mm->mmap_sem);
6179                         userspace_addr = do_mmap(NULL, 0,
6180                                                  npages * PAGE_SIZE,
6181                                                  PROT_READ | PROT_WRITE,
6182                                                  map_flags,
6183                                                  0);
6184                         up_write(&current->mm->mmap_sem);
6185
6186                         if (IS_ERR((void *)userspace_addr))
6187                                 return PTR_ERR((void *)userspace_addr);
6188
6189                         memslot->userspace_addr = userspace_addr;
6190                 }
6191         }
6192
6193
6194         return 0;
6195 }
6196
6197 void kvm_arch_commit_memory_region(struct kvm *kvm,
6198                                 struct kvm_userspace_memory_region *mem,
6199                                 struct kvm_memory_slot old,
6200                                 int user_alloc)
6201 {
6202
6203         int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6204
6205         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6206                 int ret;
6207
6208                 down_write(&current->mm->mmap_sem);
6209                 ret = do_munmap(current->mm, old.userspace_addr,
6210                                 old.npages * PAGE_SIZE);
6211                 up_write(&current->mm->mmap_sem);
6212                 if (ret < 0)
6213                         printk(KERN_WARNING
6214                                "kvm_vm_ioctl_set_memory_region: "
6215                                "failed to munmap memory\n");
6216         }
6217
6218         if (!kvm->arch.n_requested_mmu_pages)
6219                 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6220
6221         spin_lock(&kvm->mmu_lock);
6222         if (nr_mmu_pages)
6223                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6224         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6225         spin_unlock(&kvm->mmu_lock);
6226 }
6227
6228 void kvm_arch_flush_shadow(struct kvm *kvm)
6229 {
6230         kvm_mmu_zap_all(kvm);
6231         kvm_reload_remote_mmus(kvm);
6232 }
6233
6234 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6235 {
6236         return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6237                 !vcpu->arch.apf.halted)
6238                 || !list_empty_careful(&vcpu->async_pf.done)
6239                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6240                 || vcpu->arch.nmi_pending ||
6241                 (kvm_arch_interrupt_allowed(vcpu) &&
6242                  kvm_cpu_has_interrupt(vcpu));
6243 }
6244
6245 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6246 {
6247         int me;
6248         int cpu = vcpu->cpu;
6249
6250         if (waitqueue_active(&vcpu->wq)) {
6251                 wake_up_interruptible(&vcpu->wq);
6252                 ++vcpu->stat.halt_wakeup;
6253         }
6254
6255         me = get_cpu();
6256         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6257                 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6258                         smp_send_reschedule(cpu);
6259         put_cpu();
6260 }
6261
6262 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6263 {
6264         return kvm_x86_ops->interrupt_allowed(vcpu);
6265 }
6266
6267 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6268 {
6269         unsigned long current_rip = kvm_rip_read(vcpu) +
6270                 get_segment_base(vcpu, VCPU_SREG_CS);
6271
6272         return current_rip == linear_rip;
6273 }
6274 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6275
6276 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6277 {
6278         unsigned long rflags;
6279
6280         rflags = kvm_x86_ops->get_rflags(vcpu);
6281         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6282                 rflags &= ~X86_EFLAGS_TF;
6283         return rflags;
6284 }
6285 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6286
6287 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6288 {
6289         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6290             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6291                 rflags |= X86_EFLAGS_TF;
6292         kvm_x86_ops->set_rflags(vcpu, rflags);
6293         kvm_make_request(KVM_REQ_EVENT, vcpu);
6294 }
6295 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6296
6297 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6298 {
6299         int r;
6300
6301         if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6302               is_error_page(work->page))
6303                 return;
6304
6305         r = kvm_mmu_reload(vcpu);
6306         if (unlikely(r))
6307                 return;
6308
6309         if (!vcpu->arch.mmu.direct_map &&
6310               work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6311                 return;
6312
6313         vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6314 }
6315
6316 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6317 {
6318         return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6319 }
6320
6321 static inline u32 kvm_async_pf_next_probe(u32 key)
6322 {
6323         return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6324 }
6325
6326 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6327 {
6328         u32 key = kvm_async_pf_hash_fn(gfn);
6329
6330         while (vcpu->arch.apf.gfns[key] != ~0)
6331                 key = kvm_async_pf_next_probe(key);
6332
6333         vcpu->arch.apf.gfns[key] = gfn;
6334 }
6335
6336 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6337 {
6338         int i;
6339         u32 key = kvm_async_pf_hash_fn(gfn);
6340
6341         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6342                      (vcpu->arch.apf.gfns[key] != gfn &&
6343                       vcpu->arch.apf.gfns[key] != ~0); i++)
6344                 key = kvm_async_pf_next_probe(key);
6345
6346         return key;
6347 }
6348
6349 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6350 {
6351         return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6352 }
6353
6354 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6355 {
6356         u32 i, j, k;
6357
6358         i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6359         while (true) {
6360                 vcpu->arch.apf.gfns[i] = ~0;
6361                 do {
6362                         j = kvm_async_pf_next_probe(j);
6363                         if (vcpu->arch.apf.gfns[j] == ~0)
6364                                 return;
6365                         k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6366                         /*
6367                          * k lies cyclically in ]i,j]
6368                          * |    i.k.j |
6369                          * |....j i.k.| or  |.k..j i...|
6370                          */
6371                 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6372                 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6373                 i = j;
6374         }
6375 }
6376
6377 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6378 {
6379
6380         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6381                                       sizeof(val));
6382 }
6383
6384 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6385                                      struct kvm_async_pf *work)
6386 {
6387         struct x86_exception fault;
6388
6389         trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6390         kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6391
6392         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6393             (vcpu->arch.apf.send_user_only &&
6394              kvm_x86_ops->get_cpl(vcpu) == 0))
6395                 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6396         else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6397                 fault.vector = PF_VECTOR;
6398                 fault.error_code_valid = true;
6399                 fault.error_code = 0;
6400                 fault.nested_page_fault = false;
6401                 fault.address = work->arch.token;
6402                 kvm_inject_page_fault(vcpu, &fault);
6403         }
6404 }
6405
6406 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6407                                  struct kvm_async_pf *work)
6408 {
6409         struct x86_exception fault;
6410
6411         trace_kvm_async_pf_ready(work->arch.token, work->gva);
6412         if (is_error_page(work->page))
6413                 work->arch.token = ~0; /* broadcast wakeup */
6414         else
6415                 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6416
6417         if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6418             !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6419                 fault.vector = PF_VECTOR;
6420                 fault.error_code_valid = true;
6421                 fault.error_code = 0;
6422                 fault.nested_page_fault = false;
6423                 fault.address = work->arch.token;
6424                 kvm_inject_page_fault(vcpu, &fault);
6425         }
6426         vcpu->arch.apf.halted = false;
6427 }
6428
6429 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6430 {
6431         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6432                 return true;
6433         else
6434                 return !kvm_event_needs_reinjection(vcpu) &&
6435                         kvm_x86_ops->interrupt_allowed(vcpu);
6436 }
6437
6438 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6439 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6440 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6441 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6442 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6443 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6444 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6445 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6446 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6447 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6448 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6449 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);