arch/x86/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  * Copyright 2010 Red Hat, Inc. and/or its affilates.
  10  *
  11  * Authors:
  12  *   Avi Kivity   <avi@qumranet.com>
  13  *   Yaniv Kamay  <yaniv@qumranet.com>
  14  *   Amit Shah    <amit.shah@qumranet.com>
  15  *   Ben-Ami Yassour <benami@il.ibm.com>
  16  *
  17  * This work is licensed under the terms of the GNU GPL, version 2.  See
  18  * the COPYING file in the top-level directory.
  19  *
  20  */
  21
  22 #include <linux/kvm_host.h>
  23 #include "irq.h"
  24 #include "mmu.h"
  25 #include "i8254.h"
  26 #include "tss.h"
  27 #include "kvm_cache_regs.h"
  28 #include "x86.h"
  29
  30 #include <linux/clocksource.h>
  31 #include <linux/interrupt.h>
  32 #include <linux/kvm.h>
  33 #include <linux/fs.h>
  34 #include <linux/vmalloc.h>
  35 #include <linux/module.h>
  36 #include <linux/mman.h>
  37 #include <linux/highmem.h>
  38 #include <linux/iommu.h>
  39 #include <linux/intel-iommu.h>
  40 #include <linux/cpufreq.h>
  41 #include <linux/user-return-notifier.h>
  42 #include <linux/srcu.h>
  43 #include <linux/slab.h>
  44 #include <linux/perf_event.h>
  45 #include <linux/uaccess.h>
  46 #include <trace/events/kvm.h>
  47
  48 #define CREATE_TRACE_POINTS
  49 #include "trace.h"
  50
  51 #include <asm/debugreg.h>
  52 #include <asm/msr.h>
  53 #include <asm/desc.h>
  54 #include <asm/mtrr.h>
  55 #include <asm/mce.h>
  56 #include <asm/i387.h>
  57 #include <asm/xcr.h>
  58
  59 #define MAX_IO_MSRS 256
  60 #define CR0_RESERVED_BITS                                               \
  61         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  62                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  63                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  64 #define CR4_RESERVED_BITS                                               \
  65         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  66                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  67                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  68                           | X86_CR4_OSXSAVE \
  69                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  70
  71 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  72
  73 #define KVM_MAX_MCE_BANKS 32
  74 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
  75
  76 /* EFER defaults:
  77  * - enable syscall per default because its emulated by KVM
  78  * - enable LME and LMA per default on 64 bit KVM
  79  */
  80 #ifdef CONFIG_X86_64
  81 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  82 #else
  83 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  84 #endif
  85
  86 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  87 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  88
  89 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  90 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  91                                     struct kvm_cpuid_entry2 __user *entries);
  92
  93 struct kvm_x86_ops *kvm_x86_ops;
  94 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  95
  96 int ignore_msrs = 0;
  97 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
  98
  99 #define KVM_NR_SHARED_MSRS 16
 100
 101 struct kvm_shared_msrs_global {
 102         int nr;
 103         u32 msrs[KVM_NR_SHARED_MSRS];
 104 };
 105
 106 struct kvm_shared_msrs {
 107         struct user_return_notifier urn;
 108         bool registered;
 109         struct kvm_shared_msr_values {
 110                 u64 host;
 111                 u64 curr;
 112         } values[KVM_NR_SHARED_MSRS];
 113 };
 114
 115 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 116 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
 117
 118 struct kvm_stats_debugfs_item debugfs_entries[] = {
 119         { "pf_fixed", VCPU_STAT(pf_fixed) },
 120         { "pf_guest", VCPU_STAT(pf_guest) },
 121         { "tlb_flush", VCPU_STAT(tlb_flush) },
 122         { "invlpg", VCPU_STAT(invlpg) },
 123         { "exits", VCPU_STAT(exits) },
 124         { "io_exits", VCPU_STAT(io_exits) },
 125         { "mmio_exits", VCPU_STAT(mmio_exits) },
 126         { "signal_exits", VCPU_STAT(signal_exits) },
 127         { "irq_window", VCPU_STAT(irq_window_exits) },
 128         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 129         { "halt_exits", VCPU_STAT(halt_exits) },
 130         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 131         { "hypercalls", VCPU_STAT(hypercalls) },
 132         { "request_irq", VCPU_STAT(request_irq_exits) },
 133         { "irq_exits", VCPU_STAT(irq_exits) },
 134         { "host_state_reload", VCPU_STAT(host_state_reload) },
 135         { "efer_reload", VCPU_STAT(efer_reload) },
 136         { "fpu_reload", VCPU_STAT(fpu_reload) },
 137         { "insn_emulation", VCPU_STAT(insn_emulation) },
 138         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 139         { "irq_injections", VCPU_STAT(irq_injections) },
 140         { "nmi_injections", VCPU_STAT(nmi_injections) },
 141         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 142         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 143         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 144         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 145         { "mmu_flooded", VM_STAT(mmu_flooded) },
 146         { "mmu_recycled", VM_STAT(mmu_recycled) },
 147         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 148         { "mmu_unsync", VM_STAT(mmu_unsync) },
 149         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 150         { "largepages", VM_STAT(lpages) },
 151         { NULL }
 152 };
 153
 154 u64 __read_mostly host_xcr0;
 155
 156 static inline u32 bit(int bitno)
 157 {
 158         return 1 << (bitno & 31);
 159 }
 160
 161 static void kvm_on_user_return(struct user_return_notifier *urn)
 162 {
 163         unsigned slot;
 164         struct kvm_shared_msrs *locals
 165                 = container_of(urn, struct kvm_shared_msrs, urn);
 166         struct kvm_shared_msr_values *values;
 167
 168         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 169                 values = &locals->values[slot];
 170                 if (values->host != values->curr) {
 171                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 172                         values->curr = values->host;
 173                 }
 174         }
 175         locals->registered = false;
 176         user_return_notifier_unregister(urn);
 177 }
 178
 179 static void shared_msr_update(unsigned slot, u32 msr)
 180 {
 181         struct kvm_shared_msrs *smsr;
 182         u64 value;
 183
 184         smsr = &__get_cpu_var(shared_msrs);
 185         /* only read, and nobody should modify it at this time,
 186          * so don't need lock */
 187         if (slot >= shared_msrs_global.nr) {
 188                 printk(KERN_ERR "kvm: invalid MSR slot!");
 189                 return;
 190         }
 191         rdmsrl_safe(msr, &value);
 192         smsr->values[slot].host = value;
 193         smsr->values[slot].curr = value;
 194 }
 195
 196 void kvm_define_shared_msr(unsigned slot, u32 msr)
 197 {
 198         if (slot >= shared_msrs_global.nr)
 199                 shared_msrs_global.nr = slot + 1;
 200         shared_msrs_global.msrs[slot] = msr;
 201         /* we need ensured the shared_msr_global have been updated */
 202         smp_wmb();
 203 }
 204 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 205
 206 static void kvm_shared_msr_cpu_online(void)
 207 {
 208         unsigned i;
 209
 210         for (i = 0; i < shared_msrs_global.nr; ++i)
 211                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 212 }
 213
 214 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 215 {
 216         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 217
 218         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 219                 return;
 220         smsr->values[slot].curr = value;
 221         wrmsrl(shared_msrs_global.msrs[slot], value);
 222         if (!smsr->registered) {
 223                 smsr->urn.on_user_return = kvm_on_user_return;
 224                 user_return_notifier_register(&smsr->urn);
 225                 smsr->registered = true;
 226         }
 227 }
 228 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 229
 230 static void drop_user_return_notifiers(void *ignore)
 231 {
 232         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 233
 234         if (smsr->registered)
 235                 kvm_on_user_return(&smsr->urn);
 236 }
 237
 238 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 239 {
 240         if (irqchip_in_kernel(vcpu->kvm))
 241                 return vcpu->arch.apic_base;
 242         else
 243                 return vcpu->arch.apic_base;
 244 }
 245 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 246
 247 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 248 {
 249         /* TODO: reserve bits check */
 250         if (irqchip_in_kernel(vcpu->kvm))
 251                 kvm_lapic_set_base(vcpu, data);
 252         else
 253                 vcpu->arch.apic_base = data;
 254 }
 255 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 256
 257 #define EXCPT_BENIGN            0
 258 #define EXCPT_CONTRIBUTORY      1
 259 #define EXCPT_PF                2
 260
 261 static int exception_class(int vector)
 262 {
 263         switch (vector) {
 264         case PF_VECTOR:
 265                 return EXCPT_PF;
 266         case DE_VECTOR:
 267         case TS_VECTOR:
 268         case NP_VECTOR:
 269         case SS_VECTOR:
 270         case GP_VECTOR:
 271                 return EXCPT_CONTRIBUTORY;
 272         default:
 273                 break;
 274         }
 275         return EXCPT_BENIGN;
 276 }
 277
 278 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 279                 unsigned nr, bool has_error, u32 error_code,
 280                 bool reinject)
 281 {
 282         u32 prev_nr;
 283         int class1, class2;
 284
 285         if (!vcpu->arch.exception.pending) {
 286         queue:
 287                 vcpu->arch.exception.pending = true;
 288                 vcpu->arch.exception.has_error_code = has_error;
 289                 vcpu->arch.exception.nr = nr;
 290                 vcpu->arch.exception.error_code = error_code;
 291                 vcpu->arch.exception.reinject = reinject;
 292                 return;
 293         }
 294
 295         /* to check exception */
 296         prev_nr = vcpu->arch.exception.nr;
 297         if (prev_nr == DF_VECTOR) {
 298                 /* triple fault -> shutdown */
 299                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 300                 return;
 301         }
 302         class1 = exception_class(prev_nr);
 303         class2 = exception_class(nr);
 304         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 305                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 306                 /* generate double fault per SDM Table 5-5 */
 307                 vcpu->arch.exception.pending = true;
 308                 vcpu->arch.exception.has_error_code = true;
 309                 vcpu->arch.exception.nr = DF_VECTOR;
 310                 vcpu->arch.exception.error_code = 0;
 311         } else
 312                 /* replace previous exception with a new one in a hope
 313                    that instruction re-execution will regenerate lost
 314                    exception */
 315                 goto queue;
 316 }
 317
 318 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 319 {
 320         kvm_multiple_exception(vcpu, nr, false, 0, false);
 321 }
 322 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 323
 324 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 325 {
 326         kvm_multiple_exception(vcpu, nr, false, 0, true);
 327 }
 328 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 329
 330 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 331                            u32 error_code)
 332 {
 333         ++vcpu->stat.pf_guest;
 334         vcpu->arch.cr2 = addr;
 335         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 336 }
 337
 338 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 339 {
 340         vcpu->arch.nmi_pending = 1;
 341 }
 342 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 343
 344 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 345 {
 346         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 347 }
 348 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 349
 350 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 351 {
 352         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 353 }
 354 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 355
 356 /*
 357  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 358  * a #GP and return false.
 359  */
 360 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 361 {
 362         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 363                 return true;
 364         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 365         return false;
 366 }
 367 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 368
 369 /*
 370  * Load the pae pdptrs.  Return true is they are all valid.
 371  */
 372 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 373 {
 374         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 375         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 376         int i;
 377         int ret;
 378         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 379
 380         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 381                                   offset * sizeof(u64), sizeof(pdpte));
 382         if (ret < 0) {
 383                 ret = 0;
 384                 goto out;
 385         }
 386         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 387                 if (is_present_gpte(pdpte[i]) &&
 388                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 389                         ret = 0;
 390                         goto out;
 391                 }
 392         }
 393         ret = 1;
 394
 395         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 396         __set_bit(VCPU_EXREG_PDPTR,
 397                   (unsigned long *)&vcpu->arch.regs_avail);
 398         __set_bit(VCPU_EXREG_PDPTR,
 399                   (unsigned long *)&vcpu->arch.regs_dirty);
 400 out:
 401
 402         return ret;
 403 }
 404 EXPORT_SYMBOL_GPL(load_pdptrs);
 405
 406 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 407 {
 408         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 409         bool changed = true;
 410         int r;
 411
 412         if (is_long_mode(vcpu) || !is_pae(vcpu))
 413                 return false;
 414
 415         if (!test_bit(VCPU_EXREG_PDPTR,
 416                       (unsigned long *)&vcpu->arch.regs_avail))
 417                 return true;
 418
 419         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 420         if (r < 0)
 421                 goto out;
 422         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 423 out:
 424
 425         return changed;
 426 }
 427
 428 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 429 {
 430         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 431         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
 432                                     X86_CR0_CD | X86_CR0_NW;
 433
 434         cr0 |= X86_CR0_ET;
 435
 436 #ifdef CONFIG_X86_64
 437         if (cr0 & 0xffffffff00000000UL)
 438                 return 1;
 439 #endif
 440
 441         cr0 &= ~CR0_RESERVED_BITS;
 442
 443         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 444                 return 1;
 445
 446         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 447                 return 1;
 448
 449         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 450 #ifdef CONFIG_X86_64
 451                 if ((vcpu->arch.efer & EFER_LME)) {
 452                         int cs_db, cs_l;
 453
 454                         if (!is_pae(vcpu))
 455                                 return 1;
 456                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 457                         if (cs_l)
 458                                 return 1;
 459                 } else
 460 #endif
 461                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3))
 462                         return 1;
 463         }
 464
 465         kvm_x86_ops->set_cr0(vcpu, cr0);
 466
 467         if ((cr0 ^ old_cr0) & update_bits)
 468                 kvm_mmu_reset_context(vcpu);
 469         return 0;
 470 }
 471 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 472
 473 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 474 {
 475         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 476 }
 477 EXPORT_SYMBOL_GPL(kvm_lmsw);
 478
 479 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 480 {
 481         u64 xcr0;
 482
 483         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
 484         if (index != XCR_XFEATURE_ENABLED_MASK)
 485                 return 1;
 486         xcr0 = xcr;
 487         if (kvm_x86_ops->get_cpl(vcpu) != 0)
 488                 return 1;
 489         if (!(xcr0 & XSTATE_FP))
 490                 return 1;
 491         if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
 492                 return 1;
 493         if (xcr0 & ~host_xcr0)
 494                 return 1;
 495         vcpu->arch.xcr0 = xcr0;
 496         vcpu->guest_xcr0_loaded = 0;
 497         return 0;
 498 }
 499
 500 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 501 {
 502         if (__kvm_set_xcr(vcpu, index, xcr)) {
 503                 kvm_inject_gp(vcpu, 0);
 504                 return 1;
 505         }
 506         return 0;
 507 }
 508 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 509
 510 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
 511 {
 512         struct kvm_cpuid_entry2 *best;
 513
 514         best = kvm_find_cpuid_entry(vcpu, 1, 0);
 515         return best && (best->ecx & bit(X86_FEATURE_XSAVE));
 516 }
 517
 518 static void update_cpuid(struct kvm_vcpu *vcpu)
 519 {
 520         struct kvm_cpuid_entry2 *best;
 521
 522         best = kvm_find_cpuid_entry(vcpu, 1, 0);
 523         if (!best)
 524                 return;
 525
 526         /* Update OSXSAVE bit */
 527         if (cpu_has_xsave && best->function == 0x1) {
 528                 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
 529                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
 530                         best->ecx |= bit(X86_FEATURE_OSXSAVE);
 531         }
 532 }
 533
 534 int __kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 535 {
 536         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 537         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 538
 539         if (cr4 & CR4_RESERVED_BITS)
 540                 return 1;
 541
 542         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
 543                 return 1;
 544
 545         if (is_long_mode(vcpu)) {
 546                 if (!(cr4 & X86_CR4_PAE))
 547                         return 1;
 548         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 549                    && ((cr4 ^ old_cr4) & pdptr_bits)
 550                    && !load_pdptrs(vcpu, vcpu->arch.cr3))
 551                 return 1;
 552
 553         if (cr4 & X86_CR4_VMXE)
 554                 return 1;
 555
 556         kvm_x86_ops->set_cr4(vcpu, cr4);
 557
 558         if ((cr4 ^ old_cr4) & pdptr_bits)
 559                 kvm_mmu_reset_context(vcpu);
 560
 561         if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
 562                 update_cpuid(vcpu);
 563
 564         return 0;
 565 }
 566
 567 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 568 {
 569         if (__kvm_set_cr4(vcpu, cr4))
 570                 kvm_inject_gp(vcpu, 0);
 571 }
 572 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 573
 574 static int __kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 575 {
 576         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 577                 kvm_mmu_sync_roots(vcpu);
 578                 kvm_mmu_flush_tlb(vcpu);
 579                 return 0;
 580         }
 581
 582         if (is_long_mode(vcpu)) {
 583                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 584                         return 1;
 585         } else {
 586                 if (is_pae(vcpu)) {
 587                         if (cr3 & CR3_PAE_RESERVED_BITS)
 588                                 return 1;
 589                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3))
 590                                 return 1;
 591                 }
 592                 /*
 593                  * We don't check reserved bits in nonpae mode, because
 594                  * this isn't enforced, and VMware depends on this.
 595                  */
 596         }
 597
 598         /*
 599          * Does the new cr3 value map to physical memory? (Note, we
 600          * catch an invalid cr3 even in real-mode, because it would
 601          * cause trouble later on when we turn on paging anyway.)
 602          *
 603          * A real CPU would silently accept an invalid cr3 and would
 604          * attempt to use it - with largely undefined (and often hard
 605          * to debug) behavior on the guest side.
 606          */
 607         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 608                 return 1;
 609         vcpu->arch.cr3 = cr3;
 610         vcpu->arch.mmu.new_cr3(vcpu);
 611         return 0;
 612 }
 613
 614 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 615 {
 616         if (__kvm_set_cr3(vcpu, cr3))
 617                 kvm_inject_gp(vcpu, 0);
 618 }
 619 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 620
 621 int __kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 622 {
 623         if (cr8 & CR8_RESERVED_BITS)
 624                 return 1;
 625         if (irqchip_in_kernel(vcpu->kvm))
 626                 kvm_lapic_set_tpr(vcpu, cr8);
 627         else
 628                 vcpu->arch.cr8 = cr8;
 629         return 0;
 630 }
 631
 632 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 633 {
 634         if (__kvm_set_cr8(vcpu, cr8))
 635                 kvm_inject_gp(vcpu, 0);
 636 }
 637 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 638
 639 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 640 {
 641         if (irqchip_in_kernel(vcpu->kvm))
 642                 return kvm_lapic_get_cr8(vcpu);
 643         else
 644                 return vcpu->arch.cr8;
 645 }
 646 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 647
 648 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 649 {
 650         switch (dr) {
 651         case 0 ... 3:
 652                 vcpu->arch.db[dr] = val;
 653                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 654                         vcpu->arch.eff_db[dr] = val;
 655                 break;
 656         case 4:
 657                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 658                         return 1; /* #UD */
 659                 /* fall through */
 660         case 6:
 661                 if (val & 0xffffffff00000000ULL)
 662                         return -1; /* #GP */
 663                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
 664                 break;
 665         case 5:
 666                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 667                         return 1; /* #UD */
 668                 /* fall through */
 669         default: /* 7 */
 670                 if (val & 0xffffffff00000000ULL)
 671                         return -1; /* #GP */
 672                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 673                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 674                         kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
 675                         vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
 676                 }
 677                 break;
 678         }
 679
 680         return 0;
 681 }
 682
 683 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 684 {
 685         int res;
 686
 687         res = __kvm_set_dr(vcpu, dr, val);
 688         if (res > 0)
 689                 kvm_queue_exception(vcpu, UD_VECTOR);
 690         else if (res < 0)
 691                 kvm_inject_gp(vcpu, 0);
 692
 693         return res;
 694 }
 695 EXPORT_SYMBOL_GPL(kvm_set_dr);
 696
 697 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 698 {
 699         switch (dr) {
 700         case 0 ... 3:
 701                 *val = vcpu->arch.db[dr];
 702                 break;
 703         case 4:
 704                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 705                         return 1;
 706                 /* fall through */
 707         case 6:
 708                 *val = vcpu->arch.dr6;
 709                 break;
 710         case 5:
 711                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 712                         return 1;
 713                 /* fall through */
 714         default: /* 7 */
 715                 *val = vcpu->arch.dr7;
 716                 break;
 717         }
 718
 719         return 0;
 720 }
 721
 722 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 723 {
 724         if (_kvm_get_dr(vcpu, dr, val)) {
 725                 kvm_queue_exception(vcpu, UD_VECTOR);
 726                 return 1;
 727         }
 728         return 0;
 729 }
 730 EXPORT_SYMBOL_GPL(kvm_get_dr);
 731
 732 /*
 733  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 734  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 735  *
 736  * This list is modified at module load time to reflect the
 737  * capabilities of the host cpu. This capabilities test skips MSRs that are
 738  * kvm-specific. Those are put in the beginning of the list.
 739  */
 740
 741 #define KVM_SAVE_MSRS_BEGIN     7
 742 static u32 msrs_to_save[] = {
 743         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 744         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 745         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 746         HV_X64_MSR_APIC_ASSIST_PAGE,
 747         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 748         MSR_K6_STAR,
 749 #ifdef CONFIG_X86_64
 750         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 751 #endif
 752         MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 753 };
 754
 755 static unsigned num_msrs_to_save;
 756
 757 static u32 emulated_msrs[] = {
 758         MSR_IA32_MISC_ENABLE,
 759 };
 760
 761 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
 762 {
 763         u64 old_efer = vcpu->arch.efer;
 764
 765         if (efer & efer_reserved_bits)
 766                 return 1;
 767
 768         if (is_paging(vcpu)
 769             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
 770                 return 1;
 771
 772         if (efer & EFER_FFXSR) {
 773                 struct kvm_cpuid_entry2 *feat;
 774
 775                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 776                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
 777                         return 1;
 778         }
 779
 780         if (efer & EFER_SVME) {
 781                 struct kvm_cpuid_entry2 *feat;
 782
 783                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 784                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
 785                         return 1;
 786         }
 787
 788         efer &= ~EFER_LMA;
 789         efer |= vcpu->arch.efer & EFER_LMA;
 790
 791         kvm_x86_ops->set_efer(vcpu, efer);
 792
 793         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 794         kvm_mmu_reset_context(vcpu);
 795
 796         /* Update reserved bits */
 797         if ((efer ^ old_efer) & EFER_NX)
 798                 kvm_mmu_reset_context(vcpu);
 799
 800         return 0;
 801 }
 802
 803 void kvm_enable_efer_bits(u64 mask)
 804 {
 805        efer_reserved_bits &= ~mask;
 806 }
 807 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 808
 809
 810 /*
 811  * Writes msr value into into the appropriate "register".
 812  * Returns 0 on success, non-0 otherwise.
 813  * Assumes vcpu_load() was already called.
 814  */
 815 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 816 {
 817         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 818 }
 819
 820 /*
 821  * Adapt set_msr() to msr_io()'s calling convention
 822  */
 823 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 824 {
 825         return kvm_set_msr(vcpu, index, *data);
 826 }
 827
 828 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 829 {
 830         int version;
 831         int r;
 832         struct pvclock_wall_clock wc;
 833         struct timespec boot;
 834
 835         if (!wall_clock)
 836                 return;
 837
 838         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
 839         if (r)
 840                 return;
 841
 842         if (version & 1)
 843                 ++version;  /* first time write, random junk */
 844
 845         ++version;
 846
 847         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 848
 849         /*
 850          * The guest calculates current wall clock time by adding
 851          * system time (updated by kvm_write_guest_time below) to the
 852          * wall clock specified here.  guest system time equals host
 853          * system time for us, thus we must fill in host boot time here.
 854          */
 855         getboottime(&boot);
 856
 857         wc.sec = boot.tv_sec;
 858         wc.nsec = boot.tv_nsec;
 859         wc.version = version;
 860
 861         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 862
 863         version++;
 864         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 865 }
 866
 867 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 868 {
 869         uint32_t quotient, remainder;
 870
 871         /* Don't try to replace with do_div(), this one calculates
 872          * "(dividend << 32) / divisor" */
 873         __asm__ ( "divl %4"
 874                   : "=a" (quotient), "=d" (remainder)
 875                   : "0" (0), "1" (dividend), "r" (divisor) );
 876         return quotient;
 877 }
 878
 879 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 880 {
 881         uint64_t nsecs = 1000000000LL;
 882         int32_t  shift = 0;
 883         uint64_t tps64;
 884         uint32_t tps32;
 885
 886         tps64 = tsc_khz * 1000LL;
 887         while (tps64 > nsecs*2) {
 888                 tps64 >>= 1;
 889                 shift--;
 890         }
 891
 892         tps32 = (uint32_t)tps64;
 893         while (tps32 <= (uint32_t)nsecs) {
 894                 tps32 <<= 1;
 895                 shift++;
 896         }
 897
 898         hv_clock->tsc_shift = shift;
 899         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 900
 901         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 902                  __func__, tsc_khz, hv_clock->tsc_shift,
 903                  hv_clock->tsc_to_system_mul);
 904 }
 905
 906 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 907
 908 static void kvm_write_guest_time(struct kvm_vcpu *v)
 909 {
 910         struct timespec ts;
 911         unsigned long flags;
 912         struct kvm_vcpu_arch *vcpu = &v->arch;
 913         void *shared_kaddr;
 914         unsigned long this_tsc_khz;
 915
 916         if ((!vcpu->time_page))
 917                 return;
 918
 919         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 920         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 921                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 922                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 923         }
 924         put_cpu_var(cpu_tsc_khz);
 925
 926         /* Keep irq disabled to prevent changes to the clock */
 927         local_irq_save(flags);
 928         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
 929         ktime_get_ts(&ts);
 930         monotonic_to_bootbased(&ts);
 931         local_irq_restore(flags);
 932
 933         /* With all the info we got, fill in the values */
 934
 935         vcpu->hv_clock.system_time = ts.tv_nsec +
 936                                      (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
 937
 938         vcpu->hv_clock.flags = 0;
 939
 940         /*
 941          * The interface expects us to write an even number signaling that the
 942          * update is finished. Since the guest won't see the intermediate
 943          * state, we just increase by 2 at the end.
 944          */
 945         vcpu->hv_clock.version += 2;
 946
 947         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 948
 949         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 950                sizeof(vcpu->hv_clock));
 951
 952         kunmap_atomic(shared_kaddr, KM_USER0);
 953
 954         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 955 }
 956
 957 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 958 {
 959         struct kvm_vcpu_arch *vcpu = &v->arch;
 960
 961         if (!vcpu->time_page)
 962                 return 0;
 963         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 964         return 1;
 965 }
 966
 967 static bool msr_mtrr_valid(unsigned msr)
 968 {
 969         switch (msr) {
 970         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 971         case MSR_MTRRfix64K_00000:
 972         case MSR_MTRRfix16K_80000:
 973         case MSR_MTRRfix16K_A0000:
 974         case MSR_MTRRfix4K_C0000:
 975         case MSR_MTRRfix4K_C8000:
 976         case MSR_MTRRfix4K_D0000:
 977         case MSR_MTRRfix4K_D8000:
 978         case MSR_MTRRfix4K_E0000:
 979         case MSR_MTRRfix4K_E8000:
 980         case MSR_MTRRfix4K_F0000:
 981         case MSR_MTRRfix4K_F8000:
 982         case MSR_MTRRdefType:
 983         case MSR_IA32_CR_PAT:
 984                 return true;
 985         case 0x2f8:
 986                 return true;
 987         }
 988         return false;
 989 }
 990
 991 static bool valid_pat_type(unsigned t)
 992 {
 993         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 994 }
 995
 996 static bool valid_mtrr_type(unsigned t)
 997 {
 998         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 999 }
1000
1001 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1002 {
1003         int i;
1004
1005         if (!msr_mtrr_valid(msr))
1006                 return false;
1007
1008         if (msr == MSR_IA32_CR_PAT) {
1009                 for (i = 0; i < 8; i++)
1010                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
1011                                 return false;
1012                 return true;
1013         } else if (msr == MSR_MTRRdefType) {
1014                 if (data & ~0xcff)
1015                         return false;
1016                 return valid_mtrr_type(data & 0xff);
1017         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1018                 for (i = 0; i < 8 ; i++)
1019                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1020                                 return false;
1021                 return true;
1022         }
1023
1024         /* variable MTRRs */
1025         return valid_mtrr_type(data & 0xff);
1026 }
1027
1028 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1029 {
1030         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1031
1032         if (!mtrr_valid(vcpu, msr, data))
1033                 return 1;
1034
1035         if (msr == MSR_MTRRdefType) {
1036                 vcpu->arch.mtrr_state.def_type = data;
1037                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1038         } else if (msr == MSR_MTRRfix64K_00000)
1039                 p[0] = data;
1040         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1041                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1042         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1043                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1044         else if (msr == MSR_IA32_CR_PAT)
1045                 vcpu->arch.pat = data;
1046         else {  /* Variable MTRRs */
1047                 int idx, is_mtrr_mask;
1048                 u64 *pt;
1049
1050                 idx = (msr - 0x200) / 2;
1051                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1052                 if (!is_mtrr_mask)
1053                         pt =
1054                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1055                 else
1056                         pt =
1057                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1058                 *pt = data;
1059         }
1060
1061         kvm_mmu_reset_context(vcpu);
1062         return 0;
1063 }
1064
1065 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1066 {
1067         u64 mcg_cap = vcpu->arch.mcg_cap;
1068         unsigned bank_num = mcg_cap & 0xff;
1069
1070         switch (msr) {
1071         case MSR_IA32_MCG_STATUS:
1072                 vcpu->arch.mcg_status = data;
1073                 break;
1074         case MSR_IA32_MCG_CTL:
1075                 if (!(mcg_cap & MCG_CTL_P))
1076                         return 1;
1077                 if (data != 0 && data != ~(u64)0)
1078                         return -1;
1079                 vcpu->arch.mcg_ctl = data;
1080                 break;
1081         default:
1082                 if (msr >= MSR_IA32_MC0_CTL &&
1083                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1084                         u32 offset = msr - MSR_IA32_MC0_CTL;
1085                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1086                          * some Linux kernels though clear bit 10 in bank 4 to
1087                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1088                          * this to avoid an uncatched #GP in the guest
1089                          */
1090                         if ((offset & 0x3) == 0 &&
1091                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1092                                 return -1;
1093                         vcpu->arch.mce_banks[offset] = data;
1094                         break;
1095                 }
1096                 return 1;
1097         }
1098         return 0;
1099 }
1100
1101 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1102 {
1103         struct kvm *kvm = vcpu->kvm;
1104         int lm = is_long_mode(vcpu);
1105         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1106                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1107         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1108                 : kvm->arch.xen_hvm_config.blob_size_32;
1109         u32 page_num = data & ~PAGE_MASK;
1110         u64 page_addr = data & PAGE_MASK;
1111         u8 *page;
1112         int r;
1113
1114         r = -E2BIG;
1115         if (page_num >= blob_size)
1116                 goto out;
1117         r = -ENOMEM;
1118         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1119         if (!page)
1120                 goto out;
1121         r = -EFAULT;
1122         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1123                 goto out_free;
1124         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1125                 goto out_free;
1126         r = 0;
1127 out_free:
1128         kfree(page);
1129 out:
1130         return r;
1131 }
1132
1133 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1134 {
1135         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1136 }
1137
1138 static bool kvm_hv_msr_partition_wide(u32 msr)
1139 {
1140         bool r = false;
1141         switch (msr) {
1142         case HV_X64_MSR_GUEST_OS_ID:
1143         case HV_X64_MSR_HYPERCALL:
1144                 r = true;
1145                 break;
1146         }
1147
1148         return r;
1149 }
1150
1151 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1152 {
1153         struct kvm *kvm = vcpu->kvm;
1154
1155         switch (msr) {
1156         case HV_X64_MSR_GUEST_OS_ID:
1157                 kvm->arch.hv_guest_os_id = data;
1158                 /* setting guest os id to zero disables hypercall page */
1159                 if (!kvm->arch.hv_guest_os_id)
1160                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1161                 break;
1162         case HV_X64_MSR_HYPERCALL: {
1163                 u64 gfn;
1164                 unsigned long addr;
1165                 u8 instructions[4];
1166
1167                 /* if guest os id is not set hypercall should remain disabled */
1168                 if (!kvm->arch.hv_guest_os_id)
1169                         break;
1170                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1171                         kvm->arch.hv_hypercall = data;
1172                         break;
1173                 }
1174                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1175                 addr = gfn_to_hva(kvm, gfn);
1176                 if (kvm_is_error_hva(addr))
1177                         return 1;
1178                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1179                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1180                 if (copy_to_user((void __user *)addr, instructions, 4))
1181                         return 1;
1182                 kvm->arch.hv_hypercall = data;
1183                 break;
1184         }
1185         default:
1186                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1187                           "data 0x%llx\n", msr, data);
1188                 return 1;
1189         }
1190         return 0;
1191 }
1192
1193 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1194 {
1195         switch (msr) {
1196         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1197                 unsigned long addr;
1198
1199                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1200                         vcpu->arch.hv_vapic = data;
1201                         break;
1202                 }
1203                 addr = gfn_to_hva(vcpu->kvm, data >>
1204                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1205                 if (kvm_is_error_hva(addr))
1206                         return 1;
1207                 if (clear_user((void __user *)addr, PAGE_SIZE))
1208                         return 1;
1209                 vcpu->arch.hv_vapic = data;
1210                 break;
1211         }
1212         case HV_X64_MSR_EOI:
1213                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1214         case HV_X64_MSR_ICR:
1215                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1216         case HV_X64_MSR_TPR:
1217                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1218         default:
1219                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1220                           "data 0x%llx\n", msr, data);
1221                 return 1;
1222         }
1223
1224         return 0;
1225 }
1226
1227 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1228 {
1229         switch (msr) {
1230         case MSR_EFER:
1231                 return set_efer(vcpu, data);
1232         case MSR_K7_HWCR:
1233                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1234                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
1235                 if (data != 0) {
1236                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1237                                 data);
1238                         return 1;
1239                 }
1240                 break;
1241         case MSR_FAM10H_MMIO_CONF_BASE:
1242                 if (data != 0) {
1243                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1244                                 "0x%llx\n", data);
1245                         return 1;
1246                 }
1247                 break;
1248         case MSR_AMD64_NB_CFG:
1249                 break;
1250         case MSR_IA32_DEBUGCTLMSR:
1251                 if (!data) {
1252                         /* We support the non-activated case already */
1253                         break;
1254                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1255                         /* Values other than LBR and BTF are vendor-specific,
1256                            thus reserved and should throw a #GP */
1257                         return 1;
1258                 }
1259                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1260                         __func__, data);
1261                 break;
1262         case MSR_IA32_UCODE_REV:
1263         case MSR_IA32_UCODE_WRITE:
1264         case MSR_VM_HSAVE_PA:
1265         case MSR_AMD64_PATCH_LOADER:
1266                 break;
1267         case 0x200 ... 0x2ff:
1268                 return set_msr_mtrr(vcpu, msr, data);
1269         case MSR_IA32_APICBASE:
1270                 kvm_set_apic_base(vcpu, data);
1271                 break;
1272         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1273                 return kvm_x2apic_msr_write(vcpu, msr, data);
1274         case MSR_IA32_MISC_ENABLE:
1275                 vcpu->arch.ia32_misc_enable_msr = data;
1276                 break;
1277         case MSR_KVM_WALL_CLOCK_NEW:
1278         case MSR_KVM_WALL_CLOCK:
1279                 vcpu->kvm->arch.wall_clock = data;
1280                 kvm_write_wall_clock(vcpu->kvm, data);
1281                 break;
1282         case MSR_KVM_SYSTEM_TIME_NEW:
1283         case MSR_KVM_SYSTEM_TIME: {
1284                 if (vcpu->arch.time_page) {
1285                         kvm_release_page_dirty(vcpu->arch.time_page);
1286                         vcpu->arch.time_page = NULL;
1287                 }
1288
1289                 vcpu->arch.time = data;
1290
1291                 /* we verify if the enable bit is set... */
1292                 if (!(data & 1))
1293                         break;
1294
1295                 /* ...but clean it before doing the actual write */
1296                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1297
1298                 vcpu->arch.time_page =
1299                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1300
1301                 if (is_error_page(vcpu->arch.time_page)) {
1302                         kvm_release_page_clean(vcpu->arch.time_page);
1303                         vcpu->arch.time_page = NULL;
1304                 }
1305
1306                 kvm_request_guest_time_update(vcpu);
1307                 break;
1308         }
1309         case MSR_IA32_MCG_CTL:
1310         case MSR_IA32_MCG_STATUS:
1311         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1312                 return set_msr_mce(vcpu, msr, data);
1313
1314         /* Performance counters are not protected by a CPUID bit,
1315          * so we should check all of them in the generic path for the sake of
1316          * cross vendor migration.
1317          * Writing a zero into the event select MSRs disables them,
1318          * which we perfectly emulate ;-). Any other value should be at least
1319          * reported, some guests depend on them.
1320          */
1321         case MSR_P6_EVNTSEL0:
1322         case MSR_P6_EVNTSEL1:
1323         case MSR_K7_EVNTSEL0:
1324         case MSR_K7_EVNTSEL1:
1325         case MSR_K7_EVNTSEL2:
1326         case MSR_K7_EVNTSEL3:
1327                 if (data != 0)
1328                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1329                                 "0x%x data 0x%llx\n", msr, data);
1330                 break;
1331         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1332          * so we ignore writes to make it happy.
1333          */
1334         case MSR_P6_PERFCTR0:
1335         case MSR_P6_PERFCTR1:
1336         case MSR_K7_PERFCTR0:
1337         case MSR_K7_PERFCTR1:
1338         case MSR_K7_PERFCTR2:
1339         case MSR_K7_PERFCTR3:
1340                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1341                         "0x%x data 0x%llx\n", msr, data);
1342                 break;
1343         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1344                 if (kvm_hv_msr_partition_wide(msr)) {
1345                         int r;
1346                         mutex_lock(&vcpu->kvm->lock);
1347                         r = set_msr_hyperv_pw(vcpu, msr, data);
1348                         mutex_unlock(&vcpu->kvm->lock);
1349                         return r;
1350                 } else
1351                         return set_msr_hyperv(vcpu, msr, data);
1352                 break;
1353         default:
1354                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1355                         return xen_hvm_config(vcpu, data);
1356                 if (!ignore_msrs) {
1357                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1358                                 msr, data);
1359                         return 1;
1360                 } else {
1361                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1362                                 msr, data);
1363                         break;
1364                 }
1365         }
1366         return 0;
1367 }
1368 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1369
1370
1371 /*
1372  * Reads an msr value (of 'msr_index') into 'pdata'.
1373  * Returns 0 on success, non-0 otherwise.
1374  * Assumes vcpu_load() was already called.
1375  */
1376 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1377 {
1378         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1379 }
1380
1381 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1382 {
1383         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1384
1385         if (!msr_mtrr_valid(msr))
1386                 return 1;
1387
1388         if (msr == MSR_MTRRdefType)
1389                 *pdata = vcpu->arch.mtrr_state.def_type +
1390                          (vcpu->arch.mtrr_state.enabled << 10);
1391         else if (msr == MSR_MTRRfix64K_00000)
1392                 *pdata = p[0];
1393         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1394                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1395         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1396                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1397         else if (msr == MSR_IA32_CR_PAT)
1398                 *pdata = vcpu->arch.pat;
1399         else {  /* Variable MTRRs */
1400                 int idx, is_mtrr_mask;
1401                 u64 *pt;
1402
1403                 idx = (msr - 0x200) / 2;
1404                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1405                 if (!is_mtrr_mask)
1406                         pt =
1407                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1408                 else
1409                         pt =
1410                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1411                 *pdata = *pt;
1412         }
1413
1414         return 0;
1415 }
1416
1417 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1418 {
1419         u64 data;
1420         u64 mcg_cap = vcpu->arch.mcg_cap;
1421         unsigned bank_num = mcg_cap & 0xff;
1422
1423         switch (msr) {
1424         case MSR_IA32_P5_MC_ADDR:
1425         case MSR_IA32_P5_MC_TYPE:
1426                 data = 0;
1427                 break;
1428         case MSR_IA32_MCG_CAP:
1429                 data = vcpu->arch.mcg_cap;
1430                 break;
1431         case MSR_IA32_MCG_CTL:
1432                 if (!(mcg_cap & MCG_CTL_P))
1433                         return 1;
1434                 data = vcpu->arch.mcg_ctl;
1435                 break;
1436         case MSR_IA32_MCG_STATUS:
1437                 data = vcpu->arch.mcg_status;
1438                 break;
1439         default:
1440                 if (msr >= MSR_IA32_MC0_CTL &&
1441                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1442                         u32 offset = msr - MSR_IA32_MC0_CTL;
1443                         data = vcpu->arch.mce_banks[offset];
1444                         break;
1445                 }
1446                 return 1;
1447         }
1448         *pdata = data;
1449         return 0;
1450 }
1451
1452 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1453 {
1454         u64 data = 0;
1455         struct kvm *kvm = vcpu->kvm;
1456
1457         switch (msr) {
1458         case HV_X64_MSR_GUEST_OS_ID:
1459                 data = kvm->arch.hv_guest_os_id;
1460                 break;
1461         case HV_X64_MSR_HYPERCALL:
1462                 data = kvm->arch.hv_hypercall;
1463                 break;
1464         default:
1465                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1466                 return 1;
1467         }
1468
1469         *pdata = data;
1470         return 0;
1471 }
1472
1473 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1474 {
1475         u64 data = 0;
1476
1477         switch (msr) {
1478         case HV_X64_MSR_VP_INDEX: {
1479                 int r;
1480                 struct kvm_vcpu *v;
1481                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1482                         if (v == vcpu)
1483                                 data = r;
1484                 break;
1485         }
1486         case HV_X64_MSR_EOI:
1487                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1488         case HV_X64_MSR_ICR:
1489                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1490         case HV_X64_MSR_TPR:
1491                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1492         default:
1493                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1494                 return 1;
1495         }
1496         *pdata = data;
1497         return 0;
1498 }
1499
1500 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1501 {
1502         u64 data;
1503
1504         switch (msr) {
1505         case MSR_IA32_PLATFORM_ID:
1506         case MSR_IA32_UCODE_REV:
1507         case MSR_IA32_EBL_CR_POWERON:
1508         case MSR_IA32_DEBUGCTLMSR:
1509         case MSR_IA32_LASTBRANCHFROMIP:
1510         case MSR_IA32_LASTBRANCHTOIP:
1511         case MSR_IA32_LASTINTFROMIP:
1512         case MSR_IA32_LASTINTTOIP:
1513         case MSR_K8_SYSCFG:
1514         case MSR_K7_HWCR:
1515         case MSR_VM_HSAVE_PA:
1516         case MSR_P6_PERFCTR0:
1517         case MSR_P6_PERFCTR1:
1518         case MSR_P6_EVNTSEL0:
1519         case MSR_P6_EVNTSEL1:
1520         case MSR_K7_EVNTSEL0:
1521         case MSR_K7_PERFCTR0:
1522         case MSR_K8_INT_PENDING_MSG:
1523         case MSR_AMD64_NB_CFG:
1524         case MSR_FAM10H_MMIO_CONF_BASE:
1525                 data = 0;
1526                 break;
1527         case MSR_MTRRcap:
1528                 data = 0x500 | KVM_NR_VAR_MTRR;
1529                 break;
1530         case 0x200 ... 0x2ff:
1531                 return get_msr_mtrr(vcpu, msr, pdata);
1532         case 0xcd: /* fsb frequency */
1533                 data = 3;
1534                 break;
1535         case MSR_IA32_APICBASE:
1536                 data = kvm_get_apic_base(vcpu);
1537                 break;
1538         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1539                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1540                 break;
1541         case MSR_IA32_MISC_ENABLE:
1542                 data = vcpu->arch.ia32_misc_enable_msr;
1543                 break;
1544         case MSR_IA32_PERF_STATUS:
1545                 /* TSC increment by tick */
1546                 data = 1000ULL;
1547                 /* CPU multiplier */
1548                 data |= (((uint64_t)4ULL) << 40);
1549                 break;
1550         case MSR_EFER:
1551                 data = vcpu->arch.efer;
1552                 break;
1553         case MSR_KVM_WALL_CLOCK:
1554         case MSR_KVM_WALL_CLOCK_NEW:
1555                 data = vcpu->kvm->arch.wall_clock;
1556                 break;
1557         case MSR_KVM_SYSTEM_TIME:
1558         case MSR_KVM_SYSTEM_TIME_NEW:
1559                 data = vcpu->arch.time;
1560                 break;
1561         case MSR_IA32_P5_MC_ADDR:
1562         case MSR_IA32_P5_MC_TYPE:
1563         case MSR_IA32_MCG_CAP:
1564         case MSR_IA32_MCG_CTL:
1565         case MSR_IA32_MCG_STATUS:
1566         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1567                 return get_msr_mce(vcpu, msr, pdata);
1568         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1569                 if (kvm_hv_msr_partition_wide(msr)) {
1570                         int r;
1571                         mutex_lock(&vcpu->kvm->lock);
1572                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1573                         mutex_unlock(&vcpu->kvm->lock);
1574                         return r;
1575                 } else
1576                         return get_msr_hyperv(vcpu, msr, pdata);
1577                 break;
1578         default:
1579                 if (!ignore_msrs) {
1580                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1581                         return 1;
1582                 } else {
1583                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1584                         data = 0;
1585                 }
1586                 break;
1587         }
1588         *pdata = data;
1589         return 0;
1590 }
1591 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1592
1593 /*
1594  * Read or write a bunch of msrs. All parameters are kernel addresses.
1595  *
1596  * @return number of msrs set successfully.
1597  */
1598 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1599                     struct kvm_msr_entry *entries,
1600                     int (*do_msr)(struct kvm_vcpu *vcpu,
1601                                   unsigned index, u64 *data))
1602 {
1603         int i, idx;
1604
1605         idx = srcu_read_lock(&vcpu->kvm->srcu);
1606         for (i = 0; i < msrs->nmsrs; ++i)
1607                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1608                         break;
1609         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1610
1611         return i;
1612 }
1613
1614 /*
1615  * Read or write a bunch of msrs. Parameters are user addresses.
1616  *
1617  * @return number of msrs set successfully.
1618  */
1619 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1620                   int (*do_msr)(struct kvm_vcpu *vcpu,
1621                                 unsigned index, u64 *data),
1622                   int writeback)
1623 {
1624         struct kvm_msrs msrs;
1625         struct kvm_msr_entry *entries;
1626         int r, n;
1627         unsigned size;
1628
1629         r = -EFAULT;
1630         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1631                 goto out;
1632
1633         r = -E2BIG;
1634         if (msrs.nmsrs >= MAX_IO_MSRS)
1635                 goto out;
1636
1637         r = -ENOMEM;
1638         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1639         entries = kmalloc(size, GFP_KERNEL);
1640         if (!entries)
1641                 goto out;
1642
1643         r = -EFAULT;
1644         if (copy_from_user(entries, user_msrs->entries, size))
1645                 goto out_free;
1646
1647         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1648         if (r < 0)
1649                 goto out_free;
1650
1651         r = -EFAULT;
1652         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1653                 goto out_free;
1654
1655         r = n;
1656
1657 out_free:
1658         kfree(entries);
1659 out:
1660         return r;
1661 }
1662
1663 int kvm_dev_ioctl_check_extension(long ext)
1664 {
1665         int r;
1666
1667         switch (ext) {
1668         case KVM_CAP_IRQCHIP:
1669         case KVM_CAP_HLT:
1670         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1671         case KVM_CAP_SET_TSS_ADDR:
1672         case KVM_CAP_EXT_CPUID:
1673         case KVM_CAP_CLOCKSOURCE:
1674         case KVM_CAP_PIT:
1675         case KVM_CAP_NOP_IO_DELAY:
1676         case KVM_CAP_MP_STATE:
1677         case KVM_CAP_SYNC_MMU:
1678         case KVM_CAP_REINJECT_CONTROL:
1679         case KVM_CAP_IRQ_INJECT_STATUS:
1680         case KVM_CAP_ASSIGN_DEV_IRQ:
1681         case KVM_CAP_IRQFD:
1682         case KVM_CAP_IOEVENTFD:
1683         case KVM_CAP_PIT2:
1684         case KVM_CAP_PIT_STATE2:
1685         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1686         case KVM_CAP_XEN_HVM:
1687         case KVM_CAP_ADJUST_CLOCK:
1688         case KVM_CAP_VCPU_EVENTS:
1689         case KVM_CAP_HYPERV:
1690         case KVM_CAP_HYPERV_VAPIC:
1691         case KVM_CAP_HYPERV_SPIN:
1692         case KVM_CAP_PCI_SEGMENT:
1693         case KVM_CAP_DEBUGREGS:
1694         case KVM_CAP_X86_ROBUST_SINGLESTEP:
1695                 r = 1;
1696                 break;
1697         case KVM_CAP_COALESCED_MMIO:
1698                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1699                 break;
1700         case KVM_CAP_VAPIC:
1701                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1702                 break;
1703         case KVM_CAP_NR_VCPUS:
1704                 r = KVM_MAX_VCPUS;
1705                 break;
1706         case KVM_CAP_NR_MEMSLOTS:
1707                 r = KVM_MEMORY_SLOTS;
1708                 break;
1709         case KVM_CAP_PV_MMU:    /* obsolete */
1710                 r = 0;
1711                 break;
1712         case KVM_CAP_IOMMU:
1713                 r = iommu_found();
1714                 break;
1715         case KVM_CAP_MCE:
1716                 r = KVM_MAX_MCE_BANKS;
1717                 break;
1718         default:
1719                 r = 0;
1720                 break;
1721         }
1722         return r;
1723
1724 }
1725
1726 long kvm_arch_dev_ioctl(struct file *filp,
1727                         unsigned int ioctl, unsigned long arg)
1728 {
1729         void __user *argp = (void __user *)arg;
1730         long r;
1731
1732         switch (ioctl) {
1733         case KVM_GET_MSR_INDEX_LIST: {
1734                 struct kvm_msr_list __user *user_msr_list = argp;
1735                 struct kvm_msr_list msr_list;
1736                 unsigned n;
1737
1738                 r = -EFAULT;
1739                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1740                         goto out;
1741                 n = msr_list.nmsrs;
1742                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1743                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1744                         goto out;
1745                 r = -E2BIG;
1746                 if (n < msr_list.nmsrs)
1747                         goto out;
1748                 r = -EFAULT;
1749                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1750                                  num_msrs_to_save * sizeof(u32)))
1751                         goto out;
1752                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1753                                  &emulated_msrs,
1754                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1755                         goto out;
1756                 r = 0;
1757                 break;
1758         }
1759         case KVM_GET_SUPPORTED_CPUID: {
1760                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1761                 struct kvm_cpuid2 cpuid;
1762
1763                 r = -EFAULT;
1764                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1765                         goto out;
1766                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1767                                                       cpuid_arg->entries);
1768                 if (r)
1769                         goto out;
1770
1771                 r = -EFAULT;
1772                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1773                         goto out;
1774                 r = 0;
1775                 break;
1776         }
1777         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1778                 u64 mce_cap;
1779
1780                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1781                 r = -EFAULT;
1782                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1783                         goto out;
1784                 r = 0;
1785                 break;
1786         }
1787         default:
1788                 r = -EINVAL;
1789         }
1790 out:
1791         return r;
1792 }
1793
1794 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1795 {
1796         kvm_x86_ops->vcpu_load(vcpu, cpu);
1797         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1798                 unsigned long khz = cpufreq_quick_get(cpu);
1799                 if (!khz)
1800                         khz = tsc_khz;
1801                 per_cpu(cpu_tsc_khz, cpu) = khz;
1802         }
1803         kvm_request_guest_time_update(vcpu);
1804 }
1805
1806 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1807 {
1808         kvm_x86_ops->vcpu_put(vcpu);
1809         kvm_put_guest_fpu(vcpu);
1810 }
1811
1812 static int is_efer_nx(void)
1813 {
1814         unsigned long long efer = 0;
1815
1816         rdmsrl_safe(MSR_EFER, &efer);
1817         return efer & EFER_NX;
1818 }
1819
1820 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1821 {
1822         int i;
1823         struct kvm_cpuid_entry2 *e, *entry;
1824
1825         entry = NULL;
1826         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1827                 e = &vcpu->arch.cpuid_entries[i];
1828                 if (e->function == 0x80000001) {
1829                         entry = e;
1830                         break;
1831                 }
1832         }
1833         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1834                 entry->edx &= ~(1 << 20);
1835                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1836         }
1837 }
1838
1839 /* when an old userspace process fills a new kernel module */
1840 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1841                                     struct kvm_cpuid *cpuid,
1842                                     struct kvm_cpuid_entry __user *entries)
1843 {
1844         int r, i;
1845         struct kvm_cpuid_entry *cpuid_entries;
1846
1847         r = -E2BIG;
1848         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1849                 goto out;
1850         r = -ENOMEM;
1851         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1852         if (!cpuid_entries)
1853                 goto out;
1854         r = -EFAULT;
1855         if (copy_from_user(cpuid_entries, entries,
1856                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1857                 goto out_free;
1858         for (i = 0; i < cpuid->nent; i++) {
1859                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1860                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1861                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1862                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1863                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1864                 vcpu->arch.cpuid_entries[i].index = 0;
1865                 vcpu->arch.cpuid_entries[i].flags = 0;
1866                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1867                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1868                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1869         }
1870         vcpu->arch.cpuid_nent = cpuid->nent;
1871         cpuid_fix_nx_cap(vcpu);
1872         r = 0;
1873         kvm_apic_set_version(vcpu);
1874         kvm_x86_ops->cpuid_update(vcpu);
1875         update_cpuid(vcpu);
1876
1877 out_free:
1878         vfree(cpuid_entries);
1879 out:
1880         return r;
1881 }
1882
1883 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1884                                      struct kvm_cpuid2 *cpuid,
1885                                      struct kvm_cpuid_entry2 __user *entries)
1886 {
1887         int r;
1888
1889         r = -E2BIG;
1890         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1891                 goto out;
1892         r = -EFAULT;
1893         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1894                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1895                 goto out;
1896         vcpu->arch.cpuid_nent = cpuid->nent;
1897         kvm_apic_set_version(vcpu);
1898         kvm_x86_ops->cpuid_update(vcpu);
1899         update_cpuid(vcpu);
1900         return 0;
1901
1902 out:
1903         return r;
1904 }
1905
1906 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1907                                      struct kvm_cpuid2 *cpuid,
1908                                      struct kvm_cpuid_entry2 __user *entries)
1909 {
1910         int r;
1911
1912         r = -E2BIG;
1913         if (cpuid->nent < vcpu->arch.cpuid_nent)
1914                 goto out;
1915         r = -EFAULT;
1916         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1917                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1918                 goto out;
1919         return 0;
1920
1921 out:
1922         cpuid->nent = vcpu->arch.cpuid_nent;
1923         return r;
1924 }
1925
1926 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1927                            u32 index)
1928 {
1929         entry->function = function;
1930         entry->index = index;
1931         cpuid_count(entry->function, entry->index,
1932                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1933         entry->flags = 0;
1934 }
1935
1936 #define F(x) bit(X86_FEATURE_##x)
1937
1938 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1939                          u32 index, int *nent, int maxnent)
1940 {
1941         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1942 #ifdef CONFIG_X86_64
1943         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1944                                 ? F(GBPAGES) : 0;
1945         unsigned f_lm = F(LM);
1946 #else
1947         unsigned f_gbpages = 0;
1948         unsigned f_lm = 0;
1949 #endif
1950         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1951
1952         /* cpuid 1.edx */
1953         const u32 kvm_supported_word0_x86_features =
1954                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1955                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1956                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1957                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1958                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1959                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1960                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1961                 0 /* HTT, TM, Reserved, PBE */;
1962         /* cpuid 0x80000001.edx */
1963         const u32 kvm_supported_word1_x86_features =
1964                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1965                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1966                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1967                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1968                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1969                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1970                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1971                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1972         /* cpuid 1.ecx */
1973         const u32 kvm_supported_word4_x86_features =
1974                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1975                 0 /* DS-CPL, VMX, SMX, EST */ |
1976                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1977                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1978                 0 /* Reserved, DCA */ | F(XMM4_1) |
1979                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1980                 0 /* Reserved, AES */ | F(XSAVE) | 0 /* OSXSAVE */;
1981         /* cpuid 0x80000001.ecx */
1982         const u32 kvm_supported_word6_x86_features =
1983                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1984                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1985                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1986                 0 /* SKINIT */ | 0 /* WDT */;
1987
1988         /* all calls to cpuid_count() should be made on the same cpu */
1989         get_cpu();
1990         do_cpuid_1_ent(entry, function, index);
1991         ++*nent;
1992
1993         switch (function) {
1994         case 0:
1995                 entry->eax = min(entry->eax, (u32)0xd);
1996                 break;
1997         case 1:
1998                 entry->edx &= kvm_supported_word0_x86_features;
1999                 entry->ecx &= kvm_supported_word4_x86_features;
2000                 /* we support x2apic emulation even if host does not support
2001                  * it since we emulate x2apic in software */
2002                 entry->ecx |= F(X2APIC);
2003                 break;
2004         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2005          * may return different values. This forces us to get_cpu() before
2006          * issuing the first command, and also to emulate this annoying behavior
2007          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2008         case 2: {
2009                 int t, times = entry->eax & 0xff;
2010
2011                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2012                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2013                 for (t = 1; t < times && *nent < maxnent; ++t) {
2014                         do_cpuid_1_ent(&entry[t], function, 0);
2015                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2016                         ++*nent;
2017                 }
2018                 break;
2019         }
2020         /* function 4 and 0xb have additional index. */
2021         case 4: {
2022                 int i, cache_type;
2023
2024                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2025                 /* read more entries until cache_type is zero */
2026                 for (i = 1; *nent < maxnent; ++i) {
2027                         cache_type = entry[i - 1].eax & 0x1f;
2028                         if (!cache_type)
2029                                 break;
2030                         do_cpuid_1_ent(&entry[i], function, i);
2031                         entry[i].flags |=
2032                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2033                         ++*nent;
2034                 }
2035                 break;
2036         }
2037         case 0xb: {
2038                 int i, level_type;
2039
2040                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2041                 /* read more entries until level_type is zero */
2042                 for (i = 1; *nent < maxnent; ++i) {
2043                         level_type = entry[i - 1].ecx & 0xff00;
2044                         if (!level_type)
2045                                 break;
2046                         do_cpuid_1_ent(&entry[i], function, i);
2047                         entry[i].flags |=
2048                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2049                         ++*nent;
2050                 }
2051                 break;
2052         }
2053         case 0xd: {
2054                 int i;
2055
2056                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2057                 for (i = 1; *nent < maxnent; ++i) {
2058                         if (entry[i - 1].eax == 0 && i != 2)
2059                                 break;
2060                         do_cpuid_1_ent(&entry[i], function, i);
2061                         entry[i].flags |=
2062                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2063                         ++*nent;
2064                 }
2065                 break;
2066         }
2067         case KVM_CPUID_SIGNATURE: {
2068                 char signature[12] = "KVMKVMKVM\0\0";
2069                 u32 *sigptr = (u32 *)signature;
2070                 entry->eax = 0;
2071                 entry->ebx = sigptr[0];
2072                 entry->ecx = sigptr[1];
2073                 entry->edx = sigptr[2];
2074                 break;
2075         }
2076         case KVM_CPUID_FEATURES:
2077                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2078                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
2079                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
2080                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2081                 entry->ebx = 0;
2082                 entry->ecx = 0;
2083                 entry->edx = 0;
2084                 break;
2085         case 0x80000000:
2086                 entry->eax = min(entry->eax, 0x8000001a);
2087                 break;
2088         case 0x80000001:
2089                 entry->edx &= kvm_supported_word1_x86_features;
2090                 entry->ecx &= kvm_supported_word6_x86_features;
2091                 break;
2092         }
2093
2094         kvm_x86_ops->set_supported_cpuid(function, entry);
2095
2096         put_cpu();
2097 }
2098
2099 #undef F
2100
2101 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2102                                      struct kvm_cpuid_entry2 __user *entries)
2103 {
2104         struct kvm_cpuid_entry2 *cpuid_entries;
2105         int limit, nent = 0, r = -E2BIG;
2106         u32 func;
2107
2108         if (cpuid->nent < 1)
2109                 goto out;
2110         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2111                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2112         r = -ENOMEM;
2113         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2114         if (!cpuid_entries)
2115                 goto out;
2116
2117         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2118         limit = cpuid_entries[0].eax;
2119         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2120                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2121                              &nent, cpuid->nent);
2122         r = -E2BIG;
2123         if (nent >= cpuid->nent)
2124                 goto out_free;
2125
2126         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2127         limit = cpuid_entries[nent - 1].eax;
2128         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2129                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2130                              &nent, cpuid->nent);
2131
2132
2133
2134         r = -E2BIG;
2135         if (nent >= cpuid->nent)
2136                 goto out_free;
2137
2138         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2139                      cpuid->nent);
2140
2141         r = -E2BIG;
2142         if (nent >= cpuid->nent)
2143                 goto out_free;
2144
2145         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2146                      cpuid->nent);
2147
2148         r = -E2BIG;
2149         if (nent >= cpuid->nent)
2150                 goto out_free;
2151
2152         r = -EFAULT;
2153         if (copy_to_user(entries, cpuid_entries,
2154                          nent * sizeof(struct kvm_cpuid_entry2)))
2155                 goto out_free;
2156         cpuid->nent = nent;
2157         r = 0;
2158
2159 out_free:
2160         vfree(cpuid_entries);
2161 out:
2162         return r;
2163 }
2164
2165 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2166                                     struct kvm_lapic_state *s)
2167 {
2168         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2169
2170         return 0;
2171 }
2172
2173 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2174                                     struct kvm_lapic_state *s)
2175 {
2176         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2177         kvm_apic_post_state_restore(vcpu);
2178         update_cr8_intercept(vcpu);
2179
2180         return 0;
2181 }
2182
2183 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2184                                     struct kvm_interrupt *irq)
2185 {
2186         if (irq->irq < 0 || irq->irq >= 256)
2187                 return -EINVAL;
2188         if (irqchip_in_kernel(vcpu->kvm))
2189                 return -ENXIO;
2190
2191         kvm_queue_interrupt(vcpu, irq->irq, false);
2192
2193         return 0;
2194 }
2195
2196 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2197 {
2198         kvm_inject_nmi(vcpu);
2199
2200         return 0;
2201 }
2202
2203 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2204                                            struct kvm_tpr_access_ctl *tac)
2205 {
2206         if (tac->flags)
2207                 return -EINVAL;
2208         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2209         return 0;
2210 }
2211
2212 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2213                                         u64 mcg_cap)
2214 {
2215         int r;
2216         unsigned bank_num = mcg_cap & 0xff, bank;
2217
2218         r = -EINVAL;
2219         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2220                 goto out;
2221         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2222                 goto out;
2223         r = 0;
2224         vcpu->arch.mcg_cap = mcg_cap;
2225         /* Init IA32_MCG_CTL to all 1s */
2226         if (mcg_cap & MCG_CTL_P)
2227                 vcpu->arch.mcg_ctl = ~(u64)0;
2228         /* Init IA32_MCi_CTL to all 1s */
2229         for (bank = 0; bank < bank_num; bank++)
2230                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2231 out:
2232         return r;
2233 }
2234
2235 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2236                                       struct kvm_x86_mce *mce)
2237 {
2238         u64 mcg_cap = vcpu->arch.mcg_cap;
2239         unsigned bank_num = mcg_cap & 0xff;
2240         u64 *banks = vcpu->arch.mce_banks;
2241
2242         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2243                 return -EINVAL;
2244         /*
2245          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2246          * reporting is disabled
2247          */
2248         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2249             vcpu->arch.mcg_ctl != ~(u64)0)
2250                 return 0;
2251         banks += 4 * mce->bank;
2252         /*
2253          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2254          * reporting is disabled for the bank
2255          */
2256         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2257                 return 0;
2258         if (mce->status & MCI_STATUS_UC) {
2259                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2260                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2261                         printk(KERN_DEBUG "kvm: set_mce: "
2262                                "injects mce exception while "
2263                                "previous one is in progress!\n");
2264                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2265                         return 0;
2266                 }
2267                 if (banks[1] & MCI_STATUS_VAL)
2268                         mce->status |= MCI_STATUS_OVER;
2269                 banks[2] = mce->addr;
2270                 banks[3] = mce->misc;
2271                 vcpu->arch.mcg_status = mce->mcg_status;
2272                 banks[1] = mce->status;
2273                 kvm_queue_exception(vcpu, MC_VECTOR);
2274         } else if (!(banks[1] & MCI_STATUS_VAL)
2275                    || !(banks[1] & MCI_STATUS_UC)) {
2276                 if (banks[1] & MCI_STATUS_VAL)
2277                         mce->status |= MCI_STATUS_OVER;
2278                 banks[2] = mce->addr;
2279                 banks[3] = mce->misc;
2280                 banks[1] = mce->status;
2281         } else
2282                 banks[1] |= MCI_STATUS_OVER;
2283         return 0;
2284 }
2285
2286 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2287                                                struct kvm_vcpu_events *events)
2288 {
2289         events->exception.injected =
2290                 vcpu->arch.exception.pending &&
2291                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2292         events->exception.nr = vcpu->arch.exception.nr;
2293         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2294         events->exception.error_code = vcpu->arch.exception.error_code;
2295
2296         events->interrupt.injected =
2297                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2298         events->interrupt.nr = vcpu->arch.interrupt.nr;
2299         events->interrupt.soft = 0;
2300         events->interrupt.shadow =
2301                 kvm_x86_ops->get_interrupt_shadow(vcpu,
2302                         KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2303
2304         events->nmi.injected = vcpu->arch.nmi_injected;
2305         events->nmi.pending = vcpu->arch.nmi_pending;
2306         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2307
2308         events->sipi_vector = vcpu->arch.sipi_vector;
2309
2310         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2311                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2312                          | KVM_VCPUEVENT_VALID_SHADOW);
2313 }
2314
2315 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2316                                               struct kvm_vcpu_events *events)
2317 {
2318         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2319                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2320                               | KVM_VCPUEVENT_VALID_SHADOW))
2321                 return -EINVAL;
2322
2323         vcpu->arch.exception.pending = events->exception.injected;
2324         vcpu->arch.exception.nr = events->exception.nr;
2325         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2326         vcpu->arch.exception.error_code = events->exception.error_code;
2327
2328         vcpu->arch.interrupt.pending = events->interrupt.injected;
2329         vcpu->arch.interrupt.nr = events->interrupt.nr;
2330         vcpu->arch.interrupt.soft = events->interrupt.soft;
2331         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2332                 kvm_pic_clear_isr_ack(vcpu->kvm);
2333         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2334                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2335                                                   events->interrupt.shadow);
2336
2337         vcpu->arch.nmi_injected = events->nmi.injected;
2338         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2339                 vcpu->arch.nmi_pending = events->nmi.pending;
2340         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2341
2342         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2343                 vcpu->arch.sipi_vector = events->sipi_vector;
2344
2345         return 0;
2346 }
2347
2348 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2349                                              struct kvm_debugregs *dbgregs)
2350 {
2351         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2352         dbgregs->dr6 = vcpu->arch.dr6;
2353         dbgregs->dr7 = vcpu->arch.dr7;
2354         dbgregs->flags = 0;
2355 }
2356
2357 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2358                                             struct kvm_debugregs *dbgregs)
2359 {
2360         if (dbgregs->flags)
2361                 return -EINVAL;
2362
2363         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2364         vcpu->arch.dr6 = dbgregs->dr6;
2365         vcpu->arch.dr7 = dbgregs->dr7;
2366
2367         return 0;
2368 }
2369
2370 long kvm_arch_vcpu_ioctl(struct file *filp,
2371                          unsigned int ioctl, unsigned long arg)
2372 {
2373         struct kvm_vcpu *vcpu = filp->private_data;
2374         void __user *argp = (void __user *)arg;
2375         int r;
2376         struct kvm_lapic_state *lapic = NULL;
2377
2378         switch (ioctl) {
2379         case KVM_GET_LAPIC: {
2380                 r = -EINVAL;
2381                 if (!vcpu->arch.apic)
2382                         goto out;
2383                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2384
2385                 r = -ENOMEM;
2386                 if (!lapic)
2387                         goto out;
2388                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2389                 if (r)
2390                         goto out;
2391                 r = -EFAULT;
2392                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2393                         goto out;
2394                 r = 0;
2395                 break;
2396         }
2397         case KVM_SET_LAPIC: {
2398                 r = -EINVAL;
2399                 if (!vcpu->arch.apic)
2400                         goto out;
2401                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2402                 r = -ENOMEM;
2403                 if (!lapic)
2404                         goto out;
2405                 r = -EFAULT;
2406                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2407                         goto out;
2408                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2409                 if (r)
2410                         goto out;
2411                 r = 0;
2412                 break;
2413         }
2414         case KVM_INTERRUPT: {
2415                 struct kvm_interrupt irq;
2416
2417                 r = -EFAULT;
2418                 if (copy_from_user(&irq, argp, sizeof irq))
2419                         goto out;
2420                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2421                 if (r)
2422                         goto out;
2423                 r = 0;
2424                 break;
2425         }
2426         case KVM_NMI: {
2427                 r = kvm_vcpu_ioctl_nmi(vcpu);
2428                 if (r)
2429                         goto out;
2430                 r = 0;
2431                 break;
2432         }
2433         case KVM_SET_CPUID: {
2434                 struct kvm_cpuid __user *cpuid_arg = argp;
2435                 struct kvm_cpuid cpuid;
2436
2437                 r = -EFAULT;
2438                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2439                         goto out;
2440                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2441                 if (r)
2442                         goto out;
2443                 break;
2444         }
2445         case KVM_SET_CPUID2: {
2446                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2447                 struct kvm_cpuid2 cpuid;
2448
2449                 r = -EFAULT;
2450                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2451                         goto out;
2452                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2453                                               cpuid_arg->entries);
2454                 if (r)
2455                         goto out;
2456                 break;
2457         }
2458         case KVM_GET_CPUID2: {
2459                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2460                 struct kvm_cpuid2 cpuid;
2461
2462                 r = -EFAULT;
2463                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2464                         goto out;
2465                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2466                                               cpuid_arg->entries);
2467                 if (r)
2468                         goto out;
2469                 r = -EFAULT;
2470                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2471                         goto out;
2472                 r = 0;
2473                 break;
2474         }
2475         case KVM_GET_MSRS:
2476                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2477                 break;
2478         case KVM_SET_MSRS:
2479                 r = msr_io(vcpu, argp, do_set_msr, 0);
2480                 break;
2481         case KVM_TPR_ACCESS_REPORTING: {
2482                 struct kvm_tpr_access_ctl tac;
2483
2484                 r = -EFAULT;
2485                 if (copy_from_user(&tac, argp, sizeof tac))
2486                         goto out;
2487                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2488                 if (r)
2489                         goto out;
2490                 r = -EFAULT;
2491                 if (copy_to_user(argp, &tac, sizeof tac))
2492                         goto out;
2493                 r = 0;
2494                 break;
2495         };
2496         case KVM_SET_VAPIC_ADDR: {
2497                 struct kvm_vapic_addr va;
2498
2499                 r = -EINVAL;
2500                 if (!irqchip_in_kernel(vcpu->kvm))
2501                         goto out;
2502                 r = -EFAULT;
2503                 if (copy_from_user(&va, argp, sizeof va))
2504                         goto out;
2505                 r = 0;
2506                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2507                 break;
2508         }
2509         case KVM_X86_SETUP_MCE: {
2510                 u64 mcg_cap;
2511
2512                 r = -EFAULT;
2513                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2514                         goto out;
2515                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2516                 break;
2517         }
2518         case KVM_X86_SET_MCE: {
2519                 struct kvm_x86_mce mce;
2520
2521                 r = -EFAULT;
2522                 if (copy_from_user(&mce, argp, sizeof mce))
2523                         goto out;
2524                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2525                 break;
2526         }
2527         case KVM_GET_VCPU_EVENTS: {
2528                 struct kvm_vcpu_events events;
2529
2530                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2531
2532                 r = -EFAULT;
2533                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2534                         break;
2535                 r = 0;
2536                 break;
2537         }
2538         case KVM_SET_VCPU_EVENTS: {
2539                 struct kvm_vcpu_events events;
2540
2541                 r = -EFAULT;
2542                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2543                         break;
2544
2545                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2546                 break;
2547         }
2548         case KVM_GET_DEBUGREGS: {
2549                 struct kvm_debugregs dbgregs;
2550
2551                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2552
2553                 r = -EFAULT;
2554                 if (copy_to_user(argp, &dbgregs,
2555                                  sizeof(struct kvm_debugregs)))
2556                         break;
2557                 r = 0;
2558                 break;
2559         }
2560         case KVM_SET_DEBUGREGS: {
2561                 struct kvm_debugregs dbgregs;
2562
2563                 r = -EFAULT;
2564                 if (copy_from_user(&dbgregs, argp,
2565                                    sizeof(struct kvm_debugregs)))
2566                         break;
2567
2568                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2569                 break;
2570         }
2571         default:
2572                 r = -EINVAL;
2573         }
2574 out:
2575         kfree(lapic);
2576         return r;
2577 }
2578
2579 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2580 {
2581         int ret;
2582
2583         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2584                 return -1;
2585         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2586         return ret;
2587 }
2588
2589 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2590                                               u64 ident_addr)
2591 {
2592         kvm->arch.ept_identity_map_addr = ident_addr;
2593         return 0;
2594 }
2595
2596 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2597                                           u32 kvm_nr_mmu_pages)
2598 {
2599         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2600                 return -EINVAL;
2601
2602         mutex_lock(&kvm->slots_lock);
2603         spin_lock(&kvm->mmu_lock);
2604
2605         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2606         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2607
2608         spin_unlock(&kvm->mmu_lock);
2609         mutex_unlock(&kvm->slots_lock);
2610         return 0;
2611 }
2612
2613 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2614 {
2615         return kvm->arch.n_alloc_mmu_pages;
2616 }
2617
2618 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2619 {
2620         int i;
2621         struct kvm_mem_alias *alias;
2622         struct kvm_mem_aliases *aliases;
2623
2624         aliases = kvm_aliases(kvm);
2625
2626         for (i = 0; i < aliases->naliases; ++i) {
2627                 alias = &aliases->aliases[i];
2628                 if (alias->flags & KVM_ALIAS_INVALID)
2629                         continue;
2630                 if (gfn >= alias->base_gfn
2631                     && gfn < alias->base_gfn + alias->npages)
2632                         return alias->target_gfn + gfn - alias->base_gfn;
2633         }
2634         return gfn;
2635 }
2636
2637 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2638 {
2639         int i;
2640         struct kvm_mem_alias *alias;
2641         struct kvm_mem_aliases *aliases;
2642
2643         aliases = kvm_aliases(kvm);
2644
2645         for (i = 0; i < aliases->naliases; ++i) {
2646                 alias = &aliases->aliases[i];
2647                 if (gfn >= alias->base_gfn
2648                     && gfn < alias->base_gfn + alias->npages)
2649                         return alias->target_gfn + gfn - alias->base_gfn;
2650         }
2651         return gfn;
2652 }
2653
2654 /*
2655  * Set a new alias region.  Aliases map a portion of physical memory into
2656  * another portion.  This is useful for memory windows, for example the PC
2657  * VGA region.
2658  */
2659 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2660                                          struct kvm_memory_alias *alias)
2661 {
2662         int r, n;
2663         struct kvm_mem_alias *p;
2664         struct kvm_mem_aliases *aliases, *old_aliases;
2665
2666         r = -EINVAL;
2667         /* General sanity checks */
2668         if (alias->memory_size & (PAGE_SIZE - 1))
2669                 goto out;
2670         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2671                 goto out;
2672         if (alias->slot >= KVM_ALIAS_SLOTS)
2673                 goto out;
2674         if (alias->guest_phys_addr + alias->memory_size
2675             < alias->guest_phys_addr)
2676                 goto out;
2677         if (alias->target_phys_addr + alias->memory_size
2678             < alias->target_phys_addr)
2679                 goto out;
2680
2681         r = -ENOMEM;
2682         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2683         if (!aliases)
2684                 goto out;
2685
2686         mutex_lock(&kvm->slots_lock);
2687
2688         /* invalidate any gfn reference in case of deletion/shrinking */
2689         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2690         aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2691         old_aliases = kvm->arch.aliases;
2692         rcu_assign_pointer(kvm->arch.aliases, aliases);
2693         synchronize_srcu_expedited(&kvm->srcu);
2694         kvm_mmu_zap_all(kvm);
2695         kfree(old_aliases);
2696
2697         r = -ENOMEM;
2698         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2699         if (!aliases)
2700                 goto out_unlock;
2701
2702         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2703
2704         p = &aliases->aliases[alias->slot];
2705         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2706         p->npages = alias->memory_size >> PAGE_SHIFT;
2707         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2708         p->flags &= ~(KVM_ALIAS_INVALID);
2709
2710         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2711                 if (aliases->aliases[n - 1].npages)
2712                         break;
2713         aliases->naliases = n;
2714
2715         old_aliases = kvm->arch.aliases;
2716         rcu_assign_pointer(kvm->arch.aliases, aliases);
2717         synchronize_srcu_expedited(&kvm->srcu);
2718         kfree(old_aliases);
2719         r = 0;
2720
2721 out_unlock:
2722         mutex_unlock(&kvm->slots_lock);
2723 out:
2724         return r;
2725 }
2726
2727 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2728 {
2729         int r;
2730
2731         r = 0;
2732         switch (chip->chip_id) {
2733         case KVM_IRQCHIP_PIC_MASTER:
2734                 memcpy(&chip->chip.pic,
2735                         &pic_irqchip(kvm)->pics[0],
2736                         sizeof(struct kvm_pic_state));
2737                 break;
2738         case KVM_IRQCHIP_PIC_SLAVE:
2739                 memcpy(&chip->chip.pic,
2740                         &pic_irqchip(kvm)->pics[1],
2741                         sizeof(struct kvm_pic_state));
2742                 break;
2743         case KVM_IRQCHIP_IOAPIC:
2744                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2745                 break;
2746         default:
2747                 r = -EINVAL;
2748                 break;
2749         }
2750         return r;
2751 }
2752
2753 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2754 {
2755         int r;
2756
2757         r = 0;
2758         switch (chip->chip_id) {
2759         case KVM_IRQCHIP_PIC_MASTER:
2760                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2761                 memcpy(&pic_irqchip(kvm)->pics[0],
2762                         &chip->chip.pic,
2763                         sizeof(struct kvm_pic_state));
2764                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2765                 break;
2766         case KVM_IRQCHIP_PIC_SLAVE:
2767                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2768                 memcpy(&pic_irqchip(kvm)->pics[1],
2769                         &chip->chip.pic,
2770                         sizeof(struct kvm_pic_state));
2771                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2772                 break;
2773         case KVM_IRQCHIP_IOAPIC:
2774                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2775                 break;
2776         default:
2777                 r = -EINVAL;
2778                 break;
2779         }
2780         kvm_pic_update_irq(pic_irqchip(kvm));
2781         return r;
2782 }
2783
2784 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2785 {
2786         int r = 0;
2787
2788         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2789         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2790         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2791         return r;
2792 }
2793
2794 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2795 {
2796         int r = 0;
2797
2798         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2799         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2800         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2801         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2802         return r;
2803 }
2804
2805 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2806 {
2807         int r = 0;
2808
2809         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2810         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2811                 sizeof(ps->channels));
2812         ps->flags = kvm->arch.vpit->pit_state.flags;
2813         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2814         return r;
2815 }
2816
2817 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2818 {
2819         int r = 0, start = 0;
2820         u32 prev_legacy, cur_legacy;
2821         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2822         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2823         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2824         if (!prev_legacy && cur_legacy)
2825                 start = 1;
2826         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2827                sizeof(kvm->arch.vpit->pit_state.channels));
2828         kvm->arch.vpit->pit_state.flags = ps->flags;
2829         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2830         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2831         return r;
2832 }
2833
2834 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2835                                  struct kvm_reinject_control *control)
2836 {
2837         if (!kvm->arch.vpit)
2838                 return -ENXIO;
2839         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2840         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2841         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2842         return 0;
2843 }
2844
2845 /*
2846  * Get (and clear) the dirty memory log for a memory slot.
2847  */
2848 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2849                                       struct kvm_dirty_log *log)
2850 {
2851         int r, i;
2852         struct kvm_memory_slot *memslot;
2853         unsigned long n;
2854         unsigned long is_dirty = 0;
2855
2856         mutex_lock(&kvm->slots_lock);
2857
2858         r = -EINVAL;
2859         if (log->slot >= KVM_MEMORY_SLOTS)
2860                 goto out;
2861
2862         memslot = &kvm->memslots->memslots[log->slot];
2863         r = -ENOENT;
2864         if (!memslot->dirty_bitmap)
2865                 goto out;
2866
2867         n = kvm_dirty_bitmap_bytes(memslot);
2868
2869         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2870                 is_dirty = memslot->dirty_bitmap[i];
2871
2872         /* If nothing is dirty, don't bother messing with page tables. */
2873         if (is_dirty) {
2874                 struct kvm_memslots *slots, *old_slots;
2875                 unsigned long *dirty_bitmap;
2876
2877                 spin_lock(&kvm->mmu_lock);
2878                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2879                 spin_unlock(&kvm->mmu_lock);
2880
2881                 r = -ENOMEM;
2882                 dirty_bitmap = vmalloc(n);
2883                 if (!dirty_bitmap)
2884                         goto out;
2885                 memset(dirty_bitmap, 0, n);
2886
2887                 r = -ENOMEM;
2888                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2889                 if (!slots) {
2890                         vfree(dirty_bitmap);
2891                         goto out;
2892                 }
2893                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2894                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2895
2896                 old_slots = kvm->memslots;
2897                 rcu_assign_pointer(kvm->memslots, slots);
2898                 synchronize_srcu_expedited(&kvm->srcu);
2899                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2900                 kfree(old_slots);
2901
2902                 r = -EFAULT;
2903                 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n)) {
2904                         vfree(dirty_bitmap);
2905                         goto out;
2906                 }
2907                 vfree(dirty_bitmap);
2908         } else {
2909                 r = -EFAULT;
2910                 if (clear_user(log->dirty_bitmap, n))
2911                         goto out;
2912         }
2913
2914         r = 0;
2915 out:
2916         mutex_unlock(&kvm->slots_lock);
2917         return r;
2918 }
2919
2920 long kvm_arch_vm_ioctl(struct file *filp,
2921                        unsigned int ioctl, unsigned long arg)
2922 {
2923         struct kvm *kvm = filp->private_data;
2924         void __user *argp = (void __user *)arg;
2925         int r = -ENOTTY;
2926         /*
2927          * This union makes it completely explicit to gcc-3.x
2928          * that these two variables' stack usage should be
2929          * combined, not added together.
2930          */
2931         union {
2932                 struct kvm_pit_state ps;
2933                 struct kvm_pit_state2 ps2;
2934                 struct kvm_memory_alias alias;
2935                 struct kvm_pit_config pit_config;
2936         } u;
2937
2938         switch (ioctl) {
2939         case KVM_SET_TSS_ADDR:
2940                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2941                 if (r < 0)
2942                         goto out;
2943                 break;
2944         case KVM_SET_IDENTITY_MAP_ADDR: {
2945                 u64 ident_addr;
2946
2947                 r = -EFAULT;
2948                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2949                         goto out;
2950                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2951                 if (r < 0)
2952                         goto out;
2953                 break;
2954         }
2955         case KVM_SET_MEMORY_REGION: {
2956                 struct kvm_memory_region kvm_mem;
2957                 struct kvm_userspace_memory_region kvm_userspace_mem;
2958
2959                 r = -EFAULT;
2960                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2961                         goto out;
2962                 kvm_userspace_mem.slot = kvm_mem.slot;
2963                 kvm_userspace_mem.flags = kvm_mem.flags;
2964                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2965                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2966                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2967                 if (r)
2968                         goto out;
2969                 break;
2970         }
2971         case KVM_SET_NR_MMU_PAGES:
2972                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2973                 if (r)
2974                         goto out;
2975                 break;
2976         case KVM_GET_NR_MMU_PAGES:
2977                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2978                 break;
2979         case KVM_SET_MEMORY_ALIAS:
2980                 r = -EFAULT;
2981                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2982                         goto out;
2983                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2984                 if (r)
2985                         goto out;
2986                 break;
2987         case KVM_CREATE_IRQCHIP: {
2988                 struct kvm_pic *vpic;
2989
2990                 mutex_lock(&kvm->lock);
2991                 r = -EEXIST;
2992                 if (kvm->arch.vpic)
2993                         goto create_irqchip_unlock;
2994                 r = -ENOMEM;
2995                 vpic = kvm_create_pic(kvm);
2996                 if (vpic) {
2997                         r = kvm_ioapic_init(kvm);
2998                         if (r) {
2999                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3000                                                           &vpic->dev);
3001                                 kfree(vpic);
3002                                 goto create_irqchip_unlock;
3003                         }
3004                 } else
3005                         goto create_irqchip_unlock;
3006                 smp_wmb();
3007                 kvm->arch.vpic = vpic;
3008                 smp_wmb();
3009                 r = kvm_setup_default_irq_routing(kvm);
3010                 if (r) {
3011                         mutex_lock(&kvm->irq_lock);
3012                         kvm_ioapic_destroy(kvm);
3013                         kvm_destroy_pic(kvm);
3014                         mutex_unlock(&kvm->irq_lock);
3015                 }
3016         create_irqchip_unlock:
3017                 mutex_unlock(&kvm->lock);
3018                 break;
3019         }
3020         case KVM_CREATE_PIT:
3021                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3022                 goto create_pit;
3023         case KVM_CREATE_PIT2:
3024                 r = -EFAULT;
3025                 if (copy_from_user(&u.pit_config, argp,
3026                                    sizeof(struct kvm_pit_config)))
3027                         goto out;
3028         create_pit:
3029                 mutex_lock(&kvm->slots_lock);
3030                 r = -EEXIST;
3031                 if (kvm->arch.vpit)
3032                         goto create_pit_unlock;
3033                 r = -ENOMEM;
3034                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3035                 if (kvm->arch.vpit)
3036                         r = 0;
3037         create_pit_unlock:
3038                 mutex_unlock(&kvm->slots_lock);
3039                 break;
3040         case KVM_IRQ_LINE_STATUS:
3041         case KVM_IRQ_LINE: {
3042                 struct kvm_irq_level irq_event;
3043
3044                 r = -EFAULT;
3045                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3046                         goto out;
3047                 r = -ENXIO;
3048                 if (irqchip_in_kernel(kvm)) {
3049                         __s32 status;
3050                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3051                                         irq_event.irq, irq_event.level);
3052                         if (ioctl == KVM_IRQ_LINE_STATUS) {
3053                                 r = -EFAULT;
3054                                 irq_event.status = status;
3055                                 if (copy_to_user(argp, &irq_event,
3056                                                         sizeof irq_event))
3057                                         goto out;
3058                         }
3059                         r = 0;
3060                 }
3061                 break;
3062         }
3063         case KVM_GET_IRQCHIP: {
3064                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3065                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3066
3067                 r = -ENOMEM;
3068                 if (!chip)
3069                         goto out;
3070                 r = -EFAULT;
3071                 if (copy_from_user(chip, argp, sizeof *chip))
3072                         goto get_irqchip_out;
3073                 r = -ENXIO;
3074                 if (!irqchip_in_kernel(kvm))
3075                         goto get_irqchip_out;
3076                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3077                 if (r)
3078                         goto get_irqchip_out;
3079                 r = -EFAULT;
3080                 if (copy_to_user(argp, chip, sizeof *chip))
3081                         goto get_irqchip_out;
3082                 r = 0;
3083         get_irqchip_out:
3084                 kfree(chip);
3085                 if (r)
3086                         goto out;
3087                 break;
3088         }
3089         case KVM_SET_IRQCHIP: {
3090                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3091                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3092
3093                 r = -ENOMEM;
3094                 if (!chip)
3095                         goto out;
3096                 r = -EFAULT;
3097                 if (copy_from_user(chip, argp, sizeof *chip))
3098                         goto set_irqchip_out;
3099                 r = -ENXIO;
3100                 if (!irqchip_in_kernel(kvm))
3101                         goto set_irqchip_out;
3102                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3103                 if (r)
3104                         goto set_irqchip_out;
3105                 r = 0;
3106         set_irqchip_out:
3107                 kfree(chip);
3108                 if (r)
3109                         goto out;
3110                 break;
3111         }
3112         case KVM_GET_PIT: {
3113                 r = -EFAULT;
3114                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3115                         goto out;
3116                 r = -ENXIO;
3117                 if (!kvm->arch.vpit)
3118                         goto out;
3119                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3120                 if (r)
3121                         goto out;
3122                 r = -EFAULT;
3123                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3124                         goto out;
3125                 r = 0;
3126                 break;
3127         }
3128         case KVM_SET_PIT: {
3129                 r = -EFAULT;
3130                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3131                         goto out;
3132                 r = -ENXIO;
3133                 if (!kvm->arch.vpit)
3134                         goto out;
3135                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3136                 if (r)
3137                         goto out;
3138                 r = 0;
3139                 break;
3140         }
3141         case KVM_GET_PIT2: {
3142                 r = -ENXIO;
3143                 if (!kvm->arch.vpit)
3144                         goto out;
3145                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3146                 if (r)
3147                         goto out;
3148                 r = -EFAULT;
3149                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3150                         goto out;
3151                 r = 0;
3152                 break;
3153         }
3154         case KVM_SET_PIT2: {
3155                 r = -EFAULT;
3156                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3157                         goto out;
3158                 r = -ENXIO;
3159                 if (!kvm->arch.vpit)
3160                         goto out;
3161                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3162                 if (r)
3163                         goto out;
3164                 r = 0;
3165                 break;
3166         }
3167         case KVM_REINJECT_CONTROL: {
3168                 struct kvm_reinject_control control;
3169                 r =  -EFAULT;
3170                 if (copy_from_user(&control, argp, sizeof(control)))
3171                         goto out;
3172                 r = kvm_vm_ioctl_reinject(kvm, &control);
3173                 if (r)
3174                         goto out;
3175                 r = 0;
3176                 break;
3177         }
3178         case KVM_XEN_HVM_CONFIG: {
3179                 r = -EFAULT;
3180                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3181                                    sizeof(struct kvm_xen_hvm_config)))
3182                         goto out;
3183                 r = -EINVAL;
3184                 if (kvm->arch.xen_hvm_config.flags)
3185                         goto out;
3186                 r = 0;
3187                 break;
3188         }
3189         case KVM_SET_CLOCK: {
3190                 struct timespec now;
3191                 struct kvm_clock_data user_ns;
3192                 u64 now_ns;
3193                 s64 delta;
3194
3195                 r = -EFAULT;
3196                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3197                         goto out;
3198
3199                 r = -EINVAL;
3200                 if (user_ns.flags)
3201                         goto out;
3202
3203                 r = 0;
3204                 ktime_get_ts(&now);
3205                 now_ns = timespec_to_ns(&now);
3206                 delta = user_ns.clock - now_ns;
3207                 kvm->arch.kvmclock_offset = delta;
3208                 break;
3209         }
3210         case KVM_GET_CLOCK: {
3211                 struct timespec now;
3212                 struct kvm_clock_data user_ns;
3213                 u64 now_ns;
3214
3215                 ktime_get_ts(&now);
3216                 now_ns = timespec_to_ns(&now);
3217                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3218                 user_ns.flags = 0;
3219
3220                 r = -EFAULT;
3221                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3222                         goto out;
3223                 r = 0;
3224                 break;
3225         }
3226
3227         default:
3228                 ;
3229         }
3230 out:
3231         return r;
3232 }
3233
3234 static void kvm_init_msr_list(void)
3235 {
3236         u32 dummy[2];
3237         unsigned i, j;
3238
3239         /* skip the first msrs in the list. KVM-specific */
3240         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3241                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3242                         continue;
3243                 if (j < i)
3244                         msrs_to_save[j] = msrs_to_save[i];
3245                 j++;
3246         }
3247         num_msrs_to_save = j;
3248 }
3249
3250 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3251                            const void *v)
3252 {
3253         if (vcpu->arch.apic &&
3254             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3255                 return 0;
3256
3257         return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3258 }
3259
3260 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3261 {
3262         if (vcpu->arch.apic &&
3263             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3264                 return 0;
3265
3266         return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3267 }
3268
3269 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3270                         struct kvm_segment *var, int seg)
3271 {
3272         kvm_x86_ops->set_segment(vcpu, var, seg);
3273 }
3274
3275 void kvm_get_segment(struct kvm_vcpu *vcpu,
3276                      struct kvm_segment *var, int seg)
3277 {
3278         kvm_x86_ops->get_segment(vcpu, var, seg);
3279 }
3280
3281 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3282 {
3283         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3284         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3285 }
3286
3287  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3288 {
3289         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3290         access |= PFERR_FETCH_MASK;
3291         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3292 }
3293
3294 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3295 {
3296         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3297         access |= PFERR_WRITE_MASK;
3298         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3299 }
3300
3301 /* uses this to access any guest's mapped memory without checking CPL */
3302 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3303 {
3304         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, 0, error);
3305 }
3306
3307 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3308                                       struct kvm_vcpu *vcpu, u32 access,
3309                                       u32 *error)
3310 {
3311         void *data = val;
3312         int r = X86EMUL_CONTINUE;
3313
3314         while (bytes) {
3315                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr, access, error);
3316                 unsigned offset = addr & (PAGE_SIZE-1);
3317                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3318                 int ret;
3319
3320                 if (gpa == UNMAPPED_GVA) {
3321                         r = X86EMUL_PROPAGATE_FAULT;
3322                         goto out;
3323                 }
3324                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3325                 if (ret < 0) {
3326                         r = X86EMUL_IO_NEEDED;
3327                         goto out;
3328                 }
3329
3330                 bytes -= toread;
3331                 data += toread;
3332                 addr += toread;
3333         }
3334 out:
3335         return r;
3336 }
3337
3338 /* used for instruction fetching */
3339 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3340                                 struct kvm_vcpu *vcpu, u32 *error)
3341 {
3342         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3343         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3344                                           access | PFERR_FETCH_MASK, error);
3345 }
3346
3347 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3348                                struct kvm_vcpu *vcpu, u32 *error)
3349 {
3350         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3351         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3352                                           error);
3353 }
3354
3355 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3356                                struct kvm_vcpu *vcpu, u32 *error)
3357 {
3358         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3359 }
3360
3361 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3362                                        unsigned int bytes,
3363                                        struct kvm_vcpu *vcpu,
3364                                        u32 *error)
3365 {
3366         void *data = val;
3367         int r = X86EMUL_CONTINUE;
3368
3369         while (bytes) {
3370                 gpa_t gpa =  vcpu->arch.mmu.gva_to_gpa(vcpu, addr,
3371                                                        PFERR_WRITE_MASK, error);
3372                 unsigned offset = addr & (PAGE_SIZE-1);
3373                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3374                 int ret;
3375
3376                 if (gpa == UNMAPPED_GVA) {
3377                         r = X86EMUL_PROPAGATE_FAULT;
3378                         goto out;
3379                 }
3380                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3381                 if (ret < 0) {
3382                         r = X86EMUL_IO_NEEDED;
3383                         goto out;
3384                 }
3385
3386                 bytes -= towrite;
3387                 data += towrite;
3388                 addr += towrite;
3389         }
3390 out:
3391         return r;
3392 }
3393
3394 static int emulator_read_emulated(unsigned long addr,
3395                                   void *val,
3396                                   unsigned int bytes,
3397                                   unsigned int *error_code,
3398                                   struct kvm_vcpu *vcpu)
3399 {
3400         gpa_t                 gpa;
3401
3402         if (vcpu->mmio_read_completed) {
3403                 memcpy(val, vcpu->mmio_data, bytes);
3404                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3405                                vcpu->mmio_phys_addr, *(u64 *)val);
3406                 vcpu->mmio_read_completed = 0;
3407                 return X86EMUL_CONTINUE;
3408         }
3409
3410         gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, error_code);
3411
3412         if (gpa == UNMAPPED_GVA)
3413                 return X86EMUL_PROPAGATE_FAULT;
3414
3415         /* For APIC access vmexit */
3416         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3417                 goto mmio;
3418
3419         if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3420                                 == X86EMUL_CONTINUE)
3421                 return X86EMUL_CONTINUE;
3422
3423 mmio:
3424         /*
3425          * Is this MMIO handled locally?
3426          */
3427         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3428                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3429                 return X86EMUL_CONTINUE;
3430         }
3431
3432         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3433
3434         vcpu->mmio_needed = 1;
3435         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3436         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3437         vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3438         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3439
3440         return X86EMUL_IO_NEEDED;
3441 }
3442
3443 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3444                           const void *val, int bytes)
3445 {
3446         int ret;
3447
3448         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3449         if (ret < 0)
3450                 return 0;
3451         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3452         return 1;
3453 }
3454
3455 static int emulator_write_emulated_onepage(unsigned long addr,
3456                                            const void *val,
3457                                            unsigned int bytes,
3458                                            unsigned int *error_code,
3459                                            struct kvm_vcpu *vcpu)
3460 {
3461         gpa_t                 gpa;
3462
3463         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error_code);
3464
3465         if (gpa == UNMAPPED_GVA)
3466                 return X86EMUL_PROPAGATE_FAULT;
3467
3468         /* For APIC access vmexit */
3469         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3470                 goto mmio;
3471
3472         if (emulator_write_phys(vcpu, gpa, val, bytes))
3473                 return X86EMUL_CONTINUE;
3474
3475 mmio:
3476         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3477         /*
3478          * Is this MMIO handled locally?
3479          */
3480         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3481                 return X86EMUL_CONTINUE;
3482
3483         vcpu->mmio_needed = 1;
3484         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3485         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3486         vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3487         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3488         memcpy(vcpu->run->mmio.data, val, bytes);
3489
3490         return X86EMUL_CONTINUE;
3491 }
3492
3493 int emulator_write_emulated(unsigned long addr,
3494                             const void *val,
3495                             unsigned int bytes,
3496                             unsigned int *error_code,
3497                             struct kvm_vcpu *vcpu)
3498 {
3499         /* Crossing a page boundary? */
3500         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3501                 int rc, now;
3502
3503                 now = -addr & ~PAGE_MASK;
3504                 rc = emulator_write_emulated_onepage(addr, val, now, error_code,
3505                                                      vcpu);
3506                 if (rc != X86EMUL_CONTINUE)
3507                         return rc;
3508                 addr += now;
3509                 val += now;
3510                 bytes -= now;
3511         }
3512         return emulator_write_emulated_onepage(addr, val, bytes, error_code,
3513                                                vcpu);
3514 }
3515
3516 #define CMPXCHG_TYPE(t, ptr, old, new) \
3517         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3518
3519 #ifdef CONFIG_X86_64
3520 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3521 #else
3522 #  define CMPXCHG64(ptr, old, new) \
3523         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3524 #endif
3525
3526 static int emulator_cmpxchg_emulated(unsigned long addr,
3527                                      const void *old,
3528                                      const void *new,
3529                                      unsigned int bytes,
3530                                      unsigned int *error_code,
3531                                      struct kvm_vcpu *vcpu)
3532 {
3533         gpa_t gpa;
3534         struct page *page;
3535         char *kaddr;
3536         bool exchanged;
3537
3538         /* guests cmpxchg8b have to be emulated atomically */
3539         if (bytes > 8 || (bytes & (bytes - 1)))
3540                 goto emul_write;
3541
3542         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3543
3544         if (gpa == UNMAPPED_GVA ||
3545             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3546                 goto emul_write;
3547
3548         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3549                 goto emul_write;
3550
3551         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3552
3553         kaddr = kmap_atomic(page, KM_USER0);
3554         kaddr += offset_in_page(gpa);
3555         switch (bytes) {
3556         case 1:
3557                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3558                 break;
3559         case 2:
3560                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3561                 break;
3562         case 4:
3563                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3564                 break;
3565         case 8:
3566                 exchanged = CMPXCHG64(kaddr, old, new);
3567                 break;
3568         default:
3569                 BUG();
3570         }
3571         kunmap_atomic(kaddr, KM_USER0);
3572         kvm_release_page_dirty(page);
3573
3574         if (!exchanged)
3575                 return X86EMUL_CMPXCHG_FAILED;
3576
3577         kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3578
3579         return X86EMUL_CONTINUE;
3580
3581 emul_write:
3582         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3583
3584         return emulator_write_emulated(addr, new, bytes, error_code, vcpu);
3585 }
3586
3587 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3588 {
3589         /* TODO: String I/O for in kernel device */
3590         int r;
3591
3592         if (vcpu->arch.pio.in)
3593                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3594                                     vcpu->arch.pio.size, pd);
3595         else
3596                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3597                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
3598                                      pd);
3599         return r;
3600 }
3601
3602
3603 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
3604                              unsigned int count, struct kvm_vcpu *vcpu)
3605 {
3606         if (vcpu->arch.pio.count)
3607                 goto data_avail;
3608
3609         trace_kvm_pio(1, port, size, 1);
3610
3611         vcpu->arch.pio.port = port;
3612         vcpu->arch.pio.in = 1;
3613         vcpu->arch.pio.count  = count;
3614         vcpu->arch.pio.size = size;
3615
3616         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3617         data_avail:
3618                 memcpy(val, vcpu->arch.pio_data, size * count);
3619                 vcpu->arch.pio.count = 0;
3620                 return 1;
3621         }
3622
3623         vcpu->run->exit_reason = KVM_EXIT_IO;
3624         vcpu->run->io.direction = KVM_EXIT_IO_IN;
3625         vcpu->run->io.size = size;
3626         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3627         vcpu->run->io.count = count;
3628         vcpu->run->io.port = port;
3629
3630         return 0;
3631 }
3632
3633 static int emulator_pio_out_emulated(int size, unsigned short port,
3634                               const void *val, unsigned int count,
3635                               struct kvm_vcpu *vcpu)
3636 {
3637         trace_kvm_pio(0, port, size, 1);
3638
3639         vcpu->arch.pio.port = port;
3640         vcpu->arch.pio.in = 0;
3641         vcpu->arch.pio.count = count;
3642         vcpu->arch.pio.size = size;
3643
3644         memcpy(vcpu->arch.pio_data, val, size * count);
3645
3646         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3647                 vcpu->arch.pio.count = 0;
3648                 return 1;
3649         }
3650
3651         vcpu->run->exit_reason = KVM_EXIT_IO;
3652         vcpu->run->io.direction = KVM_EXIT_IO_OUT;
3653         vcpu->run->io.size = size;
3654         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3655         vcpu->run->io.count = count;
3656         vcpu->run->io.port = port;
3657
3658         return 0;
3659 }
3660
3661 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3662 {
3663         return kvm_x86_ops->get_segment_base(vcpu, seg);
3664 }
3665
3666 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3667 {
3668         kvm_mmu_invlpg(vcpu, address);
3669         return X86EMUL_CONTINUE;
3670 }
3671
3672 int emulate_clts(struct kvm_vcpu *vcpu)
3673 {
3674         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3675         kvm_x86_ops->fpu_activate(vcpu);
3676         return X86EMUL_CONTINUE;
3677 }
3678
3679 int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu)
3680 {
3681         return _kvm_get_dr(vcpu, dr, dest);
3682 }
3683
3684 int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu)
3685 {
3686
3687         return __kvm_set_dr(vcpu, dr, value);
3688 }
3689
3690 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3691 {
3692         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3693 }
3694
3695 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
3696 {
3697         unsigned long value;
3698
3699         switch (cr) {
3700         case 0:
3701                 value = kvm_read_cr0(vcpu);
3702                 break;
3703         case 2:
3704                 value = vcpu->arch.cr2;
3705                 break;
3706         case 3:
3707                 value = vcpu->arch.cr3;
3708                 break;
3709         case 4:
3710                 value = kvm_read_cr4(vcpu);
3711                 break;
3712         case 8:
3713                 value = kvm_get_cr8(vcpu);
3714                 break;
3715         default:
3716                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3717                 return 0;
3718         }
3719
3720         return value;
3721 }
3722
3723 static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
3724 {
3725         int res = 0;
3726
3727         switch (cr) {
3728         case 0:
3729                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
3730                 break;
3731         case 2:
3732                 vcpu->arch.cr2 = val;
3733                 break;
3734         case 3:
3735                 res = __kvm_set_cr3(vcpu, val);
3736                 break;
3737         case 4:
3738                 res = __kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
3739                 break;
3740         case 8:
3741                 res = __kvm_set_cr8(vcpu, val & 0xfUL);
3742                 break;
3743         default:
3744                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3745                 res = -1;
3746         }
3747
3748         return res;
3749 }
3750
3751 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
3752 {
3753         return kvm_x86_ops->get_cpl(vcpu);
3754 }
3755
3756 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
3757 {
3758         kvm_x86_ops->get_gdt(vcpu, dt);
3759 }
3760
3761 static unsigned long emulator_get_cached_segment_base(int seg,
3762                                                       struct kvm_vcpu *vcpu)
3763 {
3764         return get_segment_base(vcpu, seg);
3765 }
3766
3767 static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg,
3768                                            struct kvm_vcpu *vcpu)
3769 {
3770         struct kvm_segment var;
3771
3772         kvm_get_segment(vcpu, &var, seg);
3773
3774         if (var.unusable)
3775                 return false;
3776
3777         if (var.g)
3778                 var.limit >>= 12;
3779         set_desc_limit(desc, var.limit);
3780         set_desc_base(desc, (unsigned long)var.base);
3781         desc->type = var.type;
3782         desc->s = var.s;
3783         desc->dpl = var.dpl;
3784         desc->p = var.present;
3785         desc->avl = var.avl;
3786         desc->l = var.l;
3787         desc->d = var.db;
3788         desc->g = var.g;
3789
3790         return true;
3791 }
3792
3793 static void emulator_set_cached_descriptor(struct desc_struct *desc, int seg,
3794                                            struct kvm_vcpu *vcpu)
3795 {
3796         struct kvm_segment var;
3797
3798         /* needed to preserve selector */
3799         kvm_get_segment(vcpu, &var, seg);
3800
3801         var.base = get_desc_base(desc);
3802         var.limit = get_desc_limit(desc);
3803         if (desc->g)
3804                 var.limit = (var.limit << 12) | 0xfff;
3805         var.type = desc->type;
3806         var.present = desc->p;
3807         var.dpl = desc->dpl;
3808         var.db = desc->d;
3809         var.s = desc->s;
3810         var.l = desc->l;
3811         var.g = desc->g;
3812         var.avl = desc->avl;
3813         var.present = desc->p;
3814         var.unusable = !var.present;
3815         var.padding = 0;
3816
3817         kvm_set_segment(vcpu, &var, seg);
3818         return;
3819 }
3820
3821 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
3822 {
3823         struct kvm_segment kvm_seg;
3824
3825         kvm_get_segment(vcpu, &kvm_seg, seg);
3826         return kvm_seg.selector;
3827 }
3828
3829 static void emulator_set_segment_selector(u16 sel, int seg,
3830                                           struct kvm_vcpu *vcpu)
3831 {
3832         struct kvm_segment kvm_seg;
3833
3834         kvm_get_segment(vcpu, &kvm_seg, seg);
3835         kvm_seg.selector = sel;
3836         kvm_set_segment(vcpu, &kvm_seg, seg);
3837 }
3838
3839 static struct x86_emulate_ops emulate_ops = {
3840         .read_std            = kvm_read_guest_virt_system,
3841         .write_std           = kvm_write_guest_virt_system,
3842         .fetch               = kvm_fetch_guest_virt,
3843         .read_emulated       = emulator_read_emulated,
3844         .write_emulated      = emulator_write_emulated,
3845         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3846         .pio_in_emulated     = emulator_pio_in_emulated,
3847         .pio_out_emulated    = emulator_pio_out_emulated,
3848         .get_cached_descriptor = emulator_get_cached_descriptor,
3849         .set_cached_descriptor = emulator_set_cached_descriptor,
3850         .get_segment_selector = emulator_get_segment_selector,
3851         .set_segment_selector = emulator_set_segment_selector,
3852         .get_cached_segment_base = emulator_get_cached_segment_base,
3853         .get_gdt             = emulator_get_gdt,
3854         .get_cr              = emulator_get_cr,
3855         .set_cr              = emulator_set_cr,
3856         .cpl                 = emulator_get_cpl,
3857         .get_dr              = emulator_get_dr,
3858         .set_dr              = emulator_set_dr,
3859         .set_msr             = kvm_set_msr,
3860         .get_msr             = kvm_get_msr,
3861 };
3862
3863 static void cache_all_regs(struct kvm_vcpu *vcpu)
3864 {
3865         kvm_register_read(vcpu, VCPU_REGS_RAX);
3866         kvm_register_read(vcpu, VCPU_REGS_RSP);
3867         kvm_register_read(vcpu, VCPU_REGS_RIP);
3868         vcpu->arch.regs_dirty = ~0;
3869 }
3870
3871 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
3872 {
3873         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
3874         /*
3875          * an sti; sti; sequence only disable interrupts for the first
3876          * instruction. So, if the last instruction, be it emulated or
3877          * not, left the system with the INT_STI flag enabled, it
3878          * means that the last instruction is an sti. We should not
3879          * leave the flag on in this case. The same goes for mov ss
3880          */
3881         if (!(int_shadow & mask))
3882                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
3883 }
3884
3885 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
3886 {
3887         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
3888         if (ctxt->exception == PF_VECTOR)
3889                 kvm_inject_page_fault(vcpu, ctxt->cr2, ctxt->error_code);
3890         else if (ctxt->error_code_valid)
3891                 kvm_queue_exception_e(vcpu, ctxt->exception, ctxt->error_code);
3892         else
3893                 kvm_queue_exception(vcpu, ctxt->exception);
3894 }
3895
3896 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
3897 {
3898         ++vcpu->stat.insn_emulation_fail;
3899         trace_kvm_emulate_insn_failed(vcpu);
3900         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3901         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3902         vcpu->run->internal.ndata = 0;
3903         kvm_queue_exception(vcpu, UD_VECTOR);
3904         return EMULATE_FAIL;
3905 }
3906
3907 int emulate_instruction(struct kvm_vcpu *vcpu,
3908                         unsigned long cr2,
3909                         u16 error_code,
3910                         int emulation_type)
3911 {
3912         int r;
3913         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
3914
3915         kvm_clear_exception_queue(vcpu);
3916         vcpu->arch.mmio_fault_cr2 = cr2;
3917         /*
3918          * TODO: fix emulate.c to use guest_read/write_register
3919          * instead of direct ->regs accesses, can save hundred cycles
3920          * on Intel for instructions that don't read/change RSP, for
3921          * for example.
3922          */
3923         cache_all_regs(vcpu);
3924
3925         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3926                 int cs_db, cs_l;
3927                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3928
3929                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3930                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
3931                 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
3932                 vcpu->arch.emulate_ctxt.mode =
3933                         (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
3934                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3935                         ? X86EMUL_MODE_VM86 : cs_l
3936                         ? X86EMUL_MODE_PROT64 : cs_db
3937                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3938                 memset(c, 0, sizeof(struct decode_cache));
3939                 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
3940                 vcpu->arch.emulate_ctxt.interruptibility = 0;
3941                 vcpu->arch.emulate_ctxt.exception = -1;
3942
3943                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3944                 trace_kvm_emulate_insn_start(vcpu);
3945
3946                 /* Only allow emulation of specific instructions on #UD
3947                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3948                 if (emulation_type & EMULTYPE_TRAP_UD) {
3949                         if (!c->twobyte)
3950                                 return EMULATE_FAIL;
3951                         switch (c->b) {
3952                         case 0x01: /* VMMCALL */
3953                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3954                                         return EMULATE_FAIL;
3955                                 break;
3956                         case 0x34: /* sysenter */
3957                         case 0x35: /* sysexit */
3958                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3959                                         return EMULATE_FAIL;
3960                                 break;
3961                         case 0x05: /* syscall */
3962                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3963                                         return EMULATE_FAIL;
3964                                 break;
3965                         default:
3966                                 return EMULATE_FAIL;
3967                         }
3968
3969                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3970                                 return EMULATE_FAIL;
3971                 }
3972
3973                 ++vcpu->stat.insn_emulation;
3974                 if (r)  {
3975                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3976                                 return EMULATE_DONE;
3977                         if (emulation_type & EMULTYPE_SKIP)
3978                                 return EMULATE_FAIL;
3979                         return handle_emulation_failure(vcpu);
3980                 }
3981         }
3982
3983         if (emulation_type & EMULTYPE_SKIP) {
3984                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3985                 return EMULATE_DONE;
3986         }
3987
3988         /* this is needed for vmware backdor interface to work since it
3989            changes registers values  during IO operation */
3990         memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
3991
3992 restart:
3993         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3994
3995         if (r) { /* emulation failed */
3996                 /*
3997                  * if emulation was due to access to shadowed page table
3998                  * and it failed try to unshadow page and re-entetr the
3999                  * guest to let CPU execute the instruction.
4000                  */
4001                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
4002                         return EMULATE_DONE;
4003
4004                 return handle_emulation_failure(vcpu);
4005         }
4006
4007         toggle_interruptibility(vcpu, vcpu->arch.emulate_ctxt.interruptibility);
4008         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4009         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4010         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4011
4012         if (vcpu->arch.emulate_ctxt.exception >= 0) {
4013                 inject_emulated_exception(vcpu);
4014                 return EMULATE_DONE;
4015         }
4016
4017         if (vcpu->arch.pio.count) {
4018                 if (!vcpu->arch.pio.in)
4019                         vcpu->arch.pio.count = 0;
4020                 return EMULATE_DO_MMIO;
4021         }
4022
4023         if (vcpu->mmio_needed) {
4024                 if (vcpu->mmio_is_write)
4025                         vcpu->mmio_needed = 0;
4026                 return EMULATE_DO_MMIO;
4027         }
4028
4029         if (vcpu->arch.emulate_ctxt.restart)
4030                 goto restart;
4031
4032         return EMULATE_DONE;
4033 }
4034 EXPORT_SYMBOL_GPL(emulate_instruction);
4035
4036 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4037 {
4038         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4039         int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
4040         /* do not return to emulator after return from userspace */
4041         vcpu->arch.pio.count = 0;
4042         return ret;
4043 }
4044 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4045
4046 static void bounce_off(void *info)
4047 {
4048         /* nothing */
4049 }
4050
4051 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4052                                      void *data)
4053 {
4054         struct cpufreq_freqs *freq = data;
4055         struct kvm *kvm;
4056         struct kvm_vcpu *vcpu;
4057         int i, send_ipi = 0;
4058
4059         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4060                 return 0;
4061         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4062                 return 0;
4063         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
4064
4065         spin_lock(&kvm_lock);
4066         list_for_each_entry(kvm, &vm_list, vm_list) {
4067                 kvm_for_each_vcpu(i, vcpu, kvm) {
4068                         if (vcpu->cpu != freq->cpu)
4069                                 continue;
4070                         if (!kvm_request_guest_time_update(vcpu))
4071                                 continue;
4072                         if (vcpu->cpu != smp_processor_id())
4073                                 send_ipi++;
4074                 }
4075         }
4076         spin_unlock(&kvm_lock);
4077
4078         if (freq->old < freq->new && send_ipi) {
4079                 /*
4080                  * We upscale the frequency.  Must make the guest
4081                  * doesn't see old kvmclock values while running with
4082                  * the new frequency, otherwise we risk the guest sees
4083                  * time go backwards.
4084                  *
4085                  * In case we update the frequency for another cpu
4086                  * (which might be in guest context) send an interrupt
4087                  * to kick the cpu out of guest context.  Next time
4088                  * guest context is entered kvmclock will be updated,
4089                  * so the guest will not see stale values.
4090                  */
4091                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
4092         }
4093         return 0;
4094 }
4095
4096 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4097         .notifier_call  = kvmclock_cpufreq_notifier
4098 };
4099
4100 static void kvm_timer_init(void)
4101 {
4102         int cpu;
4103
4104         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4105                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4106                                           CPUFREQ_TRANSITION_NOTIFIER);
4107                 for_each_online_cpu(cpu) {
4108                         unsigned long khz = cpufreq_get(cpu);
4109                         if (!khz)
4110                                 khz = tsc_khz;
4111                         per_cpu(cpu_tsc_khz, cpu) = khz;
4112                 }
4113         } else {
4114                 for_each_possible_cpu(cpu)
4115                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
4116         }
4117 }
4118
4119 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4120
4121 static int kvm_is_in_guest(void)
4122 {
4123         return percpu_read(current_vcpu) != NULL;
4124 }
4125
4126 static int kvm_is_user_mode(void)
4127 {
4128         int user_mode = 3;
4129
4130         if (percpu_read(current_vcpu))
4131                 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4132
4133         return user_mode != 0;
4134 }
4135
4136 static unsigned long kvm_get_guest_ip(void)
4137 {
4138         unsigned long ip = 0;
4139
4140         if (percpu_read(current_vcpu))
4141                 ip = kvm_rip_read(percpu_read(current_vcpu));
4142
4143         return ip;
4144 }
4145
4146 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4147         .is_in_guest            = kvm_is_in_guest,
4148         .is_user_mode           = kvm_is_user_mode,
4149         .get_guest_ip           = kvm_get_guest_ip,
4150 };
4151
4152 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4153 {
4154         percpu_write(current_vcpu, vcpu);
4155 }
4156 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4157
4158 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4159 {
4160         percpu_write(current_vcpu, NULL);
4161 }
4162 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4163
4164 int kvm_arch_init(void *opaque)
4165 {
4166         int r;
4167         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4168
4169         if (kvm_x86_ops) {
4170                 printk(KERN_ERR "kvm: already loaded the other module\n");
4171                 r = -EEXIST;
4172                 goto out;
4173         }
4174
4175         if (!ops->cpu_has_kvm_support()) {
4176                 printk(KERN_ERR "kvm: no hardware support\n");
4177                 r = -EOPNOTSUPP;
4178                 goto out;
4179         }
4180         if (ops->disabled_by_bios()) {
4181                 printk(KERN_ERR "kvm: disabled by bios\n");
4182                 r = -EOPNOTSUPP;
4183                 goto out;
4184         }
4185
4186         r = kvm_mmu_module_init();
4187         if (r)
4188                 goto out;
4189
4190         kvm_init_msr_list();
4191
4192         kvm_x86_ops = ops;
4193         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4194         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
4195         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4196                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
4197
4198         kvm_timer_init();
4199
4200         perf_register_guest_info_callbacks(&kvm_guest_cbs);
4201
4202         if (cpu_has_xsave)
4203                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4204
4205         return 0;
4206
4207 out:
4208         return r;
4209 }
4210
4211 void kvm_arch_exit(void)
4212 {
4213         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4214
4215         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4216                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4217                                             CPUFREQ_TRANSITION_NOTIFIER);
4218         kvm_x86_ops = NULL;
4219         kvm_mmu_module_exit();
4220 }
4221
4222 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4223 {
4224         ++vcpu->stat.halt_exits;
4225         if (irqchip_in_kernel(vcpu->kvm)) {
4226                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4227                 return 1;
4228         } else {
4229                 vcpu->run->exit_reason = KVM_EXIT_HLT;
4230                 return 0;
4231         }
4232 }
4233 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4234
4235 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4236                            unsigned long a1)
4237 {
4238         if (is_long_mode(vcpu))
4239                 return a0;
4240         else
4241                 return a0 | ((gpa_t)a1 << 32);
4242 }
4243
4244 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4245 {
4246         u64 param, ingpa, outgpa, ret;
4247         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4248         bool fast, longmode;
4249         int cs_db, cs_l;
4250
4251         /*
4252          * hypercall generates UD from non zero cpl and real mode
4253          * per HYPER-V spec
4254          */
4255         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4256                 kvm_queue_exception(vcpu, UD_VECTOR);
4257                 return 0;
4258         }
4259
4260         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4261         longmode = is_long_mode(vcpu) && cs_l == 1;
4262
4263         if (!longmode) {
4264                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4265                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4266                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4267                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4268                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4269                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4270         }
4271 #ifdef CONFIG_X86_64
4272         else {
4273                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4274                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4275                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4276         }
4277 #endif
4278
4279         code = param & 0xffff;
4280         fast = (param >> 16) & 0x1;
4281         rep_cnt = (param >> 32) & 0xfff;
4282         rep_idx = (param >> 48) & 0xfff;
4283
4284         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4285
4286         switch (code) {
4287         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4288                 kvm_vcpu_on_spin(vcpu);
4289                 break;
4290         default:
4291                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4292                 break;
4293         }
4294
4295         ret = res | (((u64)rep_done & 0xfff) << 32);
4296         if (longmode) {
4297                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4298         } else {
4299                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4300                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4301         }
4302
4303         return 1;
4304 }
4305
4306 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4307 {
4308         unsigned long nr, a0, a1, a2, a3, ret;
4309         int r = 1;
4310
4311         if (kvm_hv_hypercall_enabled(vcpu->kvm))
4312                 return kvm_hv_hypercall(vcpu);
4313
4314         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4315         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4316         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4317         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4318         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4319
4320         trace_kvm_hypercall(nr, a0, a1, a2, a3);
4321
4322         if (!is_long_mode(vcpu)) {
4323                 nr &= 0xFFFFFFFF;
4324                 a0 &= 0xFFFFFFFF;
4325                 a1 &= 0xFFFFFFFF;
4326                 a2 &= 0xFFFFFFFF;
4327                 a3 &= 0xFFFFFFFF;
4328         }
4329
4330         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4331                 ret = -KVM_EPERM;
4332                 goto out;
4333         }
4334
4335         switch (nr) {
4336         case KVM_HC_VAPIC_POLL_IRQ:
4337                 ret = 0;
4338                 break;
4339         case KVM_HC_MMU_OP:
4340                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4341                 break;
4342         default:
4343                 ret = -KVM_ENOSYS;
4344                 break;
4345         }
4346 out:
4347         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4348         ++vcpu->stat.hypercalls;
4349         return r;
4350 }
4351 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4352
4353 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4354 {
4355         char instruction[3];
4356         unsigned long rip = kvm_rip_read(vcpu);
4357
4358         /*
4359          * Blow out the MMU to ensure that no other VCPU has an active mapping
4360          * to ensure that the updated hypercall appears atomically across all
4361          * VCPUs.
4362          */
4363         kvm_mmu_zap_all(vcpu->kvm);
4364
4365         kvm_x86_ops->patch_hypercall(vcpu, instruction);
4366
4367         return emulator_write_emulated(rip, instruction, 3, NULL, vcpu);
4368 }
4369
4370 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4371 {
4372         struct desc_ptr dt = { limit, base };
4373
4374         kvm_x86_ops->set_gdt(vcpu, &dt);
4375 }
4376
4377 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4378 {
4379         struct desc_ptr dt = { limit, base };
4380
4381         kvm_x86_ops->set_idt(vcpu, &dt);
4382 }
4383
4384 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4385 {
4386         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4387         int j, nent = vcpu->arch.cpuid_nent;
4388
4389         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4390         /* when no next entry is found, the current entry[i] is reselected */
4391         for (j = i + 1; ; j = (j + 1) % nent) {
4392                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4393                 if (ej->function == e->function) {
4394                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4395                         return j;
4396                 }
4397         }
4398         return 0; /* silence gcc, even though control never reaches here */
4399 }
4400
4401 /* find an entry with matching function, matching index (if needed), and that
4402  * should be read next (if it's stateful) */
4403 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4404         u32 function, u32 index)
4405 {
4406         if (e->function != function)
4407                 return 0;
4408         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4409                 return 0;
4410         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4411             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4412                 return 0;
4413         return 1;
4414 }
4415
4416 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4417                                               u32 function, u32 index)
4418 {
4419         int i;
4420         struct kvm_cpuid_entry2 *best = NULL;
4421
4422         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4423                 struct kvm_cpuid_entry2 *e;
4424
4425                 e = &vcpu->arch.cpuid_entries[i];
4426                 if (is_matching_cpuid_entry(e, function, index)) {
4427                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4428                                 move_to_next_stateful_cpuid_entry(vcpu, i);
4429                         best = e;
4430                         break;
4431                 }
4432                 /*
4433                  * Both basic or both extended?
4434                  */
4435                 if (((e->function ^ function) & 0x80000000) == 0)
4436                         if (!best || e->function > best->function)
4437                                 best = e;
4438         }
4439         return best;
4440 }
4441 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4442
4443 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4444 {
4445         struct kvm_cpuid_entry2 *best;
4446
4447         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
4448         if (!best || best->eax < 0x80000008)
4449                 goto not_found;
4450         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4451         if (best)
4452                 return best->eax & 0xff;
4453 not_found:
4454         return 36;
4455 }
4456
4457 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4458 {
4459         u32 function, index;
4460         struct kvm_cpuid_entry2 *best;
4461
4462         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4463         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4464         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4465         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4466         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4467         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4468         best = kvm_find_cpuid_entry(vcpu, function, index);
4469         if (best) {
4470                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4471                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4472                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4473                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4474         }
4475         kvm_x86_ops->skip_emulated_instruction(vcpu);
4476         trace_kvm_cpuid(function,
4477                         kvm_register_read(vcpu, VCPU_REGS_RAX),
4478                         kvm_register_read(vcpu, VCPU_REGS_RBX),
4479                         kvm_register_read(vcpu, VCPU_REGS_RCX),
4480                         kvm_register_read(vcpu, VCPU_REGS_RDX));
4481 }
4482 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4483
4484 /*
4485  * Check if userspace requested an interrupt window, and that the
4486  * interrupt window is open.
4487  *
4488  * No need to exit to userspace if we already have an interrupt queued.
4489  */
4490 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4491 {
4492         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4493                 vcpu->run->request_interrupt_window &&
4494                 kvm_arch_interrupt_allowed(vcpu));
4495 }
4496
4497 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4498 {
4499         struct kvm_run *kvm_run = vcpu->run;
4500
4501         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4502         kvm_run->cr8 = kvm_get_cr8(vcpu);
4503         kvm_run->apic_base = kvm_get_apic_base(vcpu);
4504         if (irqchip_in_kernel(vcpu->kvm))
4505                 kvm_run->ready_for_interrupt_injection = 1;
4506         else
4507                 kvm_run->ready_for_interrupt_injection =
4508                         kvm_arch_interrupt_allowed(vcpu) &&
4509                         !kvm_cpu_has_interrupt(vcpu) &&
4510                         !kvm_event_needs_reinjection(vcpu);
4511 }
4512
4513 static void vapic_enter(struct kvm_vcpu *vcpu)
4514 {
4515         struct kvm_lapic *apic = vcpu->arch.apic;
4516         struct page *page;
4517
4518         if (!apic || !apic->vapic_addr)
4519                 return;
4520
4521         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4522
4523         vcpu->arch.apic->vapic_page = page;
4524 }
4525
4526 static void vapic_exit(struct kvm_vcpu *vcpu)
4527 {
4528         struct kvm_lapic *apic = vcpu->arch.apic;
4529         int idx;
4530
4531         if (!apic || !apic->vapic_addr)
4532                 return;
4533
4534         idx = srcu_read_lock(&vcpu->kvm->srcu);
4535         kvm_release_page_dirty(apic->vapic_page);
4536         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4537         srcu_read_unlock(&vcpu->kvm->srcu, idx);
4538 }
4539
4540 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4541 {
4542         int max_irr, tpr;
4543
4544         if (!kvm_x86_ops->update_cr8_intercept)
4545                 return;
4546
4547         if (!vcpu->arch.apic)
4548                 return;
4549
4550         if (!vcpu->arch.apic->vapic_addr)
4551                 max_irr = kvm_lapic_find_highest_irr(vcpu);
4552         else
4553                 max_irr = -1;
4554
4555         if (max_irr != -1)
4556                 max_irr >>= 4;
4557
4558         tpr = kvm_lapic_get_cr8(vcpu);
4559
4560         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4561 }
4562
4563 static void inject_pending_event(struct kvm_vcpu *vcpu)
4564 {
4565         /* try to reinject previous events if any */
4566         if (vcpu->arch.exception.pending) {
4567                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
4568                                         vcpu->arch.exception.has_error_code,
4569                                         vcpu->arch.exception.error_code);
4570                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4571                                           vcpu->arch.exception.has_error_code,
4572                                           vcpu->arch.exception.error_code,
4573                                           vcpu->arch.exception.reinject);
4574                 return;
4575         }
4576
4577         if (vcpu->arch.nmi_injected) {
4578                 kvm_x86_ops->set_nmi(vcpu);
4579                 return;
4580         }
4581
4582         if (vcpu->arch.interrupt.pending) {
4583                 kvm_x86_ops->set_irq(vcpu);
4584                 return;
4585         }
4586
4587         /* try to inject new event if pending */
4588         if (vcpu->arch.nmi_pending) {
4589                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4590                         vcpu->arch.nmi_pending = false;
4591                         vcpu->arch.nmi_injected = true;
4592                         kvm_x86_ops->set_nmi(vcpu);
4593                 }
4594         } else if (kvm_cpu_has_interrupt(vcpu)) {
4595                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4596                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4597                                             false);
4598                         kvm_x86_ops->set_irq(vcpu);
4599                 }
4600         }
4601 }
4602
4603 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
4604 {
4605         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
4606                         !vcpu->guest_xcr0_loaded) {
4607                 /* kvm_set_xcr() also depends on this */
4608                 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
4609                 vcpu->guest_xcr0_loaded = 1;
4610         }
4611 }
4612
4613 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
4614 {
4615         if (vcpu->guest_xcr0_loaded) {
4616                 if (vcpu->arch.xcr0 != host_xcr0)
4617                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
4618                 vcpu->guest_xcr0_loaded = 0;
4619         }
4620 }
4621
4622 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4623 {
4624         int r;
4625         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4626                 vcpu->run->request_interrupt_window;
4627
4628         if (vcpu->requests)
4629                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4630                         kvm_mmu_unload(vcpu);
4631
4632         r = kvm_mmu_reload(vcpu);
4633         if (unlikely(r))
4634                 goto out;
4635
4636         if (vcpu->requests) {
4637                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
4638                         __kvm_migrate_timers(vcpu);
4639                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4640                         kvm_write_guest_time(vcpu);
4641                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4642                         kvm_mmu_sync_roots(vcpu);
4643                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4644                         kvm_x86_ops->tlb_flush(vcpu);
4645                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4646                                        &vcpu->requests)) {
4647                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4648                         r = 0;
4649                         goto out;
4650                 }
4651                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4652                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4653                         r = 0;
4654                         goto out;
4655                 }
4656                 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4657                         vcpu->fpu_active = 0;
4658                         kvm_x86_ops->fpu_deactivate(vcpu);
4659                 }
4660         }
4661
4662         preempt_disable();
4663
4664         kvm_x86_ops->prepare_guest_switch(vcpu);
4665         if (vcpu->fpu_active)
4666                 kvm_load_guest_fpu(vcpu);
4667         kvm_load_guest_xcr0(vcpu);
4668
4669         atomic_set(&vcpu->guest_mode, 1);
4670         smp_wmb();
4671
4672         local_irq_disable();
4673
4674         if (!atomic_read(&vcpu->guest_mode) || vcpu->requests
4675             || need_resched() || signal_pending(current)) {
4676                 atomic_set(&vcpu->guest_mode, 0);
4677                 smp_wmb();
4678                 local_irq_enable();
4679                 preempt_enable();
4680                 r = 1;
4681                 goto out;
4682         }
4683
4684         inject_pending_event(vcpu);
4685
4686         /* enable NMI/IRQ window open exits if needed */
4687         if (vcpu->arch.nmi_pending)
4688                 kvm_x86_ops->enable_nmi_window(vcpu);
4689         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4690                 kvm_x86_ops->enable_irq_window(vcpu);
4691
4692         if (kvm_lapic_enabled(vcpu)) {
4693                 update_cr8_intercept(vcpu);
4694                 kvm_lapic_sync_to_vapic(vcpu);
4695         }
4696
4697         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4698
4699         kvm_guest_enter();
4700
4701         if (unlikely(vcpu->arch.switch_db_regs)) {
4702                 set_debugreg(0, 7);
4703                 set_debugreg(vcpu->arch.eff_db[0], 0);
4704                 set_debugreg(vcpu->arch.eff_db[1], 1);
4705                 set_debugreg(vcpu->arch.eff_db[2], 2);
4706                 set_debugreg(vcpu->arch.eff_db[3], 3);
4707         }
4708
4709         trace_kvm_entry(vcpu->vcpu_id);
4710         kvm_x86_ops->run(vcpu);
4711
4712         /*
4713          * If the guest has used debug registers, at least dr7
4714          * will be disabled while returning to the host.
4715          * If we don't have active breakpoints in the host, we don't
4716          * care about the messed up debug address registers. But if
4717          * we have some of them active, restore the old state.
4718          */
4719         if (hw_breakpoint_active())
4720                 hw_breakpoint_restore();
4721
4722         atomic_set(&vcpu->guest_mode, 0);
4723         smp_wmb();
4724         local_irq_enable();
4725
4726         ++vcpu->stat.exits;
4727
4728         /*
4729          * We must have an instruction between local_irq_enable() and
4730          * kvm_guest_exit(), so the timer interrupt isn't delayed by
4731          * the interrupt shadow.  The stat.exits increment will do nicely.
4732          * But we need to prevent reordering, hence this barrier():
4733          */
4734         barrier();
4735
4736         kvm_guest_exit();
4737
4738         preempt_enable();
4739
4740         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4741
4742         /*
4743          * Profile KVM exit RIPs:
4744          */
4745         if (unlikely(prof_on == KVM_PROFILING)) {
4746                 unsigned long rip = kvm_rip_read(vcpu);
4747                 profile_hit(KVM_PROFILING, (void *)rip);
4748         }
4749
4750
4751         kvm_lapic_sync_from_vapic(vcpu);
4752
4753         r = kvm_x86_ops->handle_exit(vcpu);
4754 out:
4755         return r;
4756 }
4757
4758
4759 static int __vcpu_run(struct kvm_vcpu *vcpu)
4760 {
4761         int r;
4762         struct kvm *kvm = vcpu->kvm;
4763
4764         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4765                 pr_debug("vcpu %d received sipi with vector # %x\n",
4766                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
4767                 kvm_lapic_reset(vcpu);
4768                 r = kvm_arch_vcpu_reset(vcpu);
4769                 if (r)
4770                         return r;
4771                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4772         }
4773
4774         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4775         vapic_enter(vcpu);
4776
4777         r = 1;
4778         while (r > 0) {
4779                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4780                         r = vcpu_enter_guest(vcpu);
4781                 else {
4782                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4783                         kvm_vcpu_block(vcpu);
4784                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4785                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4786                         {
4787                                 switch(vcpu->arch.mp_state) {
4788                                 case KVM_MP_STATE_HALTED:
4789                                         vcpu->arch.mp_state =
4790                                                 KVM_MP_STATE_RUNNABLE;
4791                                 case KVM_MP_STATE_RUNNABLE:
4792                                         break;
4793                                 case KVM_MP_STATE_SIPI_RECEIVED:
4794                                 default:
4795                                         r = -EINTR;
4796                                         break;
4797                                 }
4798                         }
4799                 }
4800
4801                 if (r <= 0)
4802                         break;
4803
4804                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4805                 if (kvm_cpu_has_pending_timer(vcpu))
4806                         kvm_inject_pending_timer_irqs(vcpu);
4807
4808                 if (dm_request_for_irq_injection(vcpu)) {
4809                         r = -EINTR;
4810                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4811                         ++vcpu->stat.request_irq_exits;
4812                 }
4813                 if (signal_pending(current)) {
4814                         r = -EINTR;
4815                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4816                         ++vcpu->stat.signal_exits;
4817                 }
4818                 if (need_resched()) {
4819                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4820                         kvm_resched(vcpu);
4821                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4822                 }
4823         }
4824
4825         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4826
4827         vapic_exit(vcpu);
4828
4829         return r;
4830 }
4831
4832 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4833 {
4834         int r;
4835         sigset_t sigsaved;
4836
4837         if (vcpu->sigset_active)
4838                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4839
4840         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4841                 kvm_vcpu_block(vcpu);
4842                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4843                 r = -EAGAIN;
4844                 goto out;
4845         }
4846
4847         /* re-sync apic's tpr */
4848         if (!irqchip_in_kernel(vcpu->kvm))
4849                 kvm_set_cr8(vcpu, kvm_run->cr8);
4850
4851         if (vcpu->arch.pio.count || vcpu->mmio_needed ||
4852             vcpu->arch.emulate_ctxt.restart) {
4853                 if (vcpu->mmio_needed) {
4854                         memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4855                         vcpu->mmio_read_completed = 1;
4856                         vcpu->mmio_needed = 0;
4857                 }
4858                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4859                 r = emulate_instruction(vcpu, 0, 0, EMULTYPE_NO_DECODE);
4860                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4861                 if (r != EMULATE_DONE) {
4862                         r = 0;
4863                         goto out;
4864                 }
4865         }
4866         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4867                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4868                                      kvm_run->hypercall.ret);
4869
4870         r = __vcpu_run(vcpu);
4871
4872 out:
4873         post_kvm_run_save(vcpu);
4874         if (vcpu->sigset_active)
4875                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4876
4877         return r;
4878 }
4879
4880 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4881 {
4882         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4883         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4884         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4885         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4886         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4887         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4888         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4889         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4890 #ifdef CONFIG_X86_64
4891         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4892         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4893         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4894         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4895         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4896         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4897         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4898         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4899 #endif
4900
4901         regs->rip = kvm_rip_read(vcpu);
4902         regs->rflags = kvm_get_rflags(vcpu);
4903
4904         return 0;
4905 }
4906
4907 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4908 {
4909         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4910         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4911         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4912         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4913         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4914         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4915         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4916         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4917 #ifdef CONFIG_X86_64
4918         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4919         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4920         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4921         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4922         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4923         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4924         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4925         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4926 #endif
4927
4928         kvm_rip_write(vcpu, regs->rip);
4929         kvm_set_rflags(vcpu, regs->rflags);
4930
4931         vcpu->arch.exception.pending = false;
4932
4933         return 0;
4934 }
4935
4936 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4937 {
4938         struct kvm_segment cs;
4939
4940         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4941         *db = cs.db;
4942         *l = cs.l;
4943 }
4944 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4945
4946 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4947                                   struct kvm_sregs *sregs)
4948 {
4949         struct desc_ptr dt;
4950
4951         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4952         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4953         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4954         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4955         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4956         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4957
4958         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4959         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4960
4961         kvm_x86_ops->get_idt(vcpu, &dt);
4962         sregs->idt.limit = dt.size;
4963         sregs->idt.base = dt.address;
4964         kvm_x86_ops->get_gdt(vcpu, &dt);
4965         sregs->gdt.limit = dt.size;
4966         sregs->gdt.base = dt.address;
4967
4968         sregs->cr0 = kvm_read_cr0(vcpu);
4969         sregs->cr2 = vcpu->arch.cr2;
4970         sregs->cr3 = vcpu->arch.cr3;
4971         sregs->cr4 = kvm_read_cr4(vcpu);
4972         sregs->cr8 = kvm_get_cr8(vcpu);
4973         sregs->efer = vcpu->arch.efer;
4974         sregs->apic_base = kvm_get_apic_base(vcpu);
4975
4976         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4977
4978         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4979                 set_bit(vcpu->arch.interrupt.nr,
4980                         (unsigned long *)sregs->interrupt_bitmap);
4981
4982         return 0;
4983 }
4984
4985 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4986                                     struct kvm_mp_state *mp_state)
4987 {
4988         mp_state->mp_state = vcpu->arch.mp_state;
4989         return 0;
4990 }
4991
4992 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4993                                     struct kvm_mp_state *mp_state)
4994 {
4995         vcpu->arch.mp_state = mp_state->mp_state;
4996         return 0;
4997 }
4998
4999 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5000                     bool has_error_code, u32 error_code)
5001 {
5002         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5003         int cs_db, cs_l, ret;
5004         cache_all_regs(vcpu);
5005
5006         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5007
5008         vcpu->arch.emulate_ctxt.vcpu = vcpu;
5009         vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
5010         vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
5011         vcpu->arch.emulate_ctxt.mode =
5012                 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5013                 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
5014                 ? X86EMUL_MODE_VM86 : cs_l
5015                 ? X86EMUL_MODE_PROT64 : cs_db
5016                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
5017         memset(c, 0, sizeof(struct decode_cache));
5018         memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
5019
5020         ret = emulator_task_switch(&vcpu->arch.emulate_ctxt, &emulate_ops,
5021                                    tss_selector, reason, has_error_code,
5022                                    error_code);
5023
5024         if (ret)
5025                 return EMULATE_FAIL;
5026
5027         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5028         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5029         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5030         return EMULATE_DONE;
5031 }
5032 EXPORT_SYMBOL_GPL(kvm_task_switch);
5033
5034 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5035                                   struct kvm_sregs *sregs)
5036 {
5037         int mmu_reset_needed = 0;
5038         int pending_vec, max_bits;
5039         struct desc_ptr dt;
5040
5041         dt.size = sregs->idt.limit;
5042         dt.address = sregs->idt.base;
5043         kvm_x86_ops->set_idt(vcpu, &dt);
5044         dt.size = sregs->gdt.limit;
5045         dt.address = sregs->gdt.base;
5046         kvm_x86_ops->set_gdt(vcpu, &dt);
5047
5048         vcpu->arch.cr2 = sregs->cr2;
5049         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5050         vcpu->arch.cr3 = sregs->cr3;
5051
5052         kvm_set_cr8(vcpu, sregs->cr8);
5053
5054         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5055         kvm_x86_ops->set_efer(vcpu, sregs->efer);
5056         kvm_set_apic_base(vcpu, sregs->apic_base);
5057
5058         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5059         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5060         vcpu->arch.cr0 = sregs->cr0;
5061
5062         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5063         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5064         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5065                 load_pdptrs(vcpu, vcpu->arch.cr3);
5066                 mmu_reset_needed = 1;
5067         }
5068
5069         if (mmu_reset_needed)
5070                 kvm_mmu_reset_context(vcpu);
5071
5072         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5073         pending_vec = find_first_bit(
5074                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5075         if (pending_vec < max_bits) {
5076                 kvm_queue_interrupt(vcpu, pending_vec, false);
5077                 pr_debug("Set back pending irq %d\n", pending_vec);
5078                 if (irqchip_in_kernel(vcpu->kvm))
5079                         kvm_pic_clear_isr_ack(vcpu->kvm);
5080         }
5081
5082         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5083         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5084         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5085         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5086         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5087         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5088
5089         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5090         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5091
5092         update_cr8_intercept(vcpu);
5093
5094         /* Older userspace won't unhalt the vcpu on reset. */
5095         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5096             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5097             !is_protmode(vcpu))
5098                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5099
5100         return 0;
5101 }
5102
5103 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5104                                         struct kvm_guest_debug *dbg)
5105 {
5106         unsigned long rflags;
5107         int i, r;
5108
5109         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5110                 r = -EBUSY;
5111                 if (vcpu->arch.exception.pending)
5112                         goto out;
5113                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5114                         kvm_queue_exception(vcpu, DB_VECTOR);
5115                 else
5116                         kvm_queue_exception(vcpu, BP_VECTOR);
5117         }
5118
5119         /*
5120          * Read rflags as long as potentially injected trace flags are still
5121          * filtered out.
5122          */
5123         rflags = kvm_get_rflags(vcpu);
5124
5125         vcpu->guest_debug = dbg->control;
5126         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5127                 vcpu->guest_debug = 0;
5128
5129         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5130                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5131                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5132                 vcpu->arch.switch_db_regs =
5133                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5134         } else {
5135                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5136                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5137                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5138         }
5139
5140         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5141                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5142                         get_segment_base(vcpu, VCPU_SREG_CS);
5143
5144         /*
5145          * Trigger an rflags update that will inject or remove the trace
5146          * flags.
5147          */
5148         kvm_set_rflags(vcpu, rflags);
5149
5150         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5151
5152         r = 0;
5153
5154 out:
5155
5156         return r;
5157 }
5158
5159 /*
5160  * Translate a guest virtual address to a guest physical address.
5161  */
5162 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5163                                     struct kvm_translation *tr)
5164 {
5165         unsigned long vaddr = tr->linear_address;
5166         gpa_t gpa;
5167         int idx;
5168
5169         idx = srcu_read_lock(&vcpu->kvm->srcu);
5170         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5171         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5172         tr->physical_address = gpa;
5173         tr->valid = gpa != UNMAPPED_GVA;
5174         tr->writeable = 1;
5175         tr->usermode = 0;
5176
5177         return 0;
5178 }
5179
5180 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5181 {
5182         struct i387_fxsave_struct *fxsave =
5183                         &vcpu->arch.guest_fpu.state->fxsave;
5184
5185         memcpy(fpu->fpr, fxsave->st_space, 128);
5186         fpu->fcw = fxsave->cwd;
5187         fpu->fsw = fxsave->swd;
5188         fpu->ftwx = fxsave->twd;
5189         fpu->last_opcode = fxsave->fop;
5190         fpu->last_ip = fxsave->rip;
5191         fpu->last_dp = fxsave->rdp;
5192         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5193
5194         return 0;
5195 }
5196
5197 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5198 {
5199         struct i387_fxsave_struct *fxsave =
5200                         &vcpu->arch.guest_fpu.state->fxsave;
5201
5202         memcpy(fxsave->st_space, fpu->fpr, 128);
5203         fxsave->cwd = fpu->fcw;
5204         fxsave->swd = fpu->fsw;
5205         fxsave->twd = fpu->ftwx;
5206         fxsave->fop = fpu->last_opcode;
5207         fxsave->rip = fpu->last_ip;
5208         fxsave->rdp = fpu->last_dp;
5209         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5210
5211         return 0;
5212 }
5213
5214 int fx_init(struct kvm_vcpu *vcpu)
5215 {
5216         int err;
5217
5218         err = fpu_alloc(&vcpu->arch.guest_fpu);
5219         if (err)
5220                 return err;
5221
5222         fpu_finit(&vcpu->arch.guest_fpu);
5223
5224         /*
5225          * Ensure guest xcr0 is valid for loading
5226          */
5227         vcpu->arch.xcr0 = XSTATE_FP;
5228
5229         vcpu->arch.cr0 |= X86_CR0_ET;
5230
5231         return 0;
5232 }
5233 EXPORT_SYMBOL_GPL(fx_init);
5234
5235 static void fx_free(struct kvm_vcpu *vcpu)
5236 {
5237         fpu_free(&vcpu->arch.guest_fpu);
5238 }
5239
5240 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5241 {
5242         if (vcpu->guest_fpu_loaded)
5243                 return;
5244
5245         /*
5246          * Restore all possible states in the guest,
5247          * and assume host would use all available bits.
5248          * Guest xcr0 would be loaded later.
5249          */
5250         kvm_put_guest_xcr0(vcpu);
5251         vcpu->guest_fpu_loaded = 1;
5252         unlazy_fpu(current);
5253         fpu_restore_checking(&vcpu->arch.guest_fpu);
5254         trace_kvm_fpu(1);
5255 }
5256
5257 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5258 {
5259         kvm_put_guest_xcr0(vcpu);
5260
5261         if (!vcpu->guest_fpu_loaded)
5262                 return;
5263
5264         vcpu->guest_fpu_loaded = 0;
5265         fpu_save_init(&vcpu->arch.guest_fpu);
5266         ++vcpu->stat.fpu_reload;
5267         set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5268         trace_kvm_fpu(0);
5269 }
5270
5271 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5272 {
5273         if (vcpu->arch.time_page) {
5274                 kvm_release_page_dirty(vcpu->arch.time_page);
5275                 vcpu->arch.time_page = NULL;
5276         }
5277
5278         fx_free(vcpu);
5279         kvm_x86_ops->vcpu_free(vcpu);
5280 }
5281
5282 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5283                                                 unsigned int id)
5284 {
5285         return kvm_x86_ops->vcpu_create(kvm, id);
5286 }
5287
5288 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5289 {
5290         int r;
5291
5292         vcpu->arch.mtrr_state.have_fixed = 1;
5293         vcpu_load(vcpu);
5294         r = kvm_arch_vcpu_reset(vcpu);
5295         if (r == 0)
5296                 r = kvm_mmu_setup(vcpu);
5297         vcpu_put(vcpu);
5298         if (r < 0)
5299                 goto free_vcpu;
5300
5301         return 0;
5302 free_vcpu:
5303         kvm_x86_ops->vcpu_free(vcpu);
5304         return r;
5305 }
5306
5307 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5308 {
5309         vcpu_load(vcpu);
5310         kvm_mmu_unload(vcpu);
5311         vcpu_put(vcpu);
5312
5313         fx_free(vcpu);
5314         kvm_x86_ops->vcpu_free(vcpu);
5315 }
5316
5317 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5318 {
5319         vcpu->arch.nmi_pending = false;
5320         vcpu->arch.nmi_injected = false;
5321
5322         vcpu->arch.switch_db_regs = 0;
5323         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5324         vcpu->arch.dr6 = DR6_FIXED_1;
5325         vcpu->arch.dr7 = DR7_FIXED_1;
5326
5327         return kvm_x86_ops->vcpu_reset(vcpu);
5328 }
5329
5330 int kvm_arch_hardware_enable(void *garbage)
5331 {
5332         /*
5333          * Since this may be called from a hotplug notifcation,
5334          * we can't get the CPU frequency directly.
5335          */
5336         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5337                 int cpu = raw_smp_processor_id();
5338                 per_cpu(cpu_tsc_khz, cpu) = 0;
5339         }
5340
5341         kvm_shared_msr_cpu_online();
5342
5343         return kvm_x86_ops->hardware_enable(garbage);
5344 }
5345
5346 void kvm_arch_hardware_disable(void *garbage)
5347 {
5348         kvm_x86_ops->hardware_disable(garbage);
5349         drop_user_return_notifiers(garbage);
5350 }
5351
5352 int kvm_arch_hardware_setup(void)
5353 {
5354         return kvm_x86_ops->hardware_setup();
5355 }
5356
5357 void kvm_arch_hardware_unsetup(void)
5358 {
5359         kvm_x86_ops->hardware_unsetup();
5360 }
5361
5362 void kvm_arch_check_processor_compat(void *rtn)
5363 {
5364         kvm_x86_ops->check_processor_compatibility(rtn);
5365 }
5366
5367 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5368 {
5369         struct page *page;
5370         struct kvm *kvm;
5371         int r;
5372
5373         BUG_ON(vcpu->kvm == NULL);
5374         kvm = vcpu->kvm;
5375
5376         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5377         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5378                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5379         else
5380                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5381
5382         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5383         if (!page) {
5384                 r = -ENOMEM;
5385                 goto fail;
5386         }
5387         vcpu->arch.pio_data = page_address(page);
5388
5389         r = kvm_mmu_create(vcpu);
5390         if (r < 0)
5391                 goto fail_free_pio_data;
5392
5393         if (irqchip_in_kernel(kvm)) {
5394                 r = kvm_create_lapic(vcpu);
5395                 if (r < 0)
5396                         goto fail_mmu_destroy;
5397         }
5398
5399         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5400                                        GFP_KERNEL);
5401         if (!vcpu->arch.mce_banks) {
5402                 r = -ENOMEM;
5403                 goto fail_free_lapic;
5404         }
5405         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5406
5407         return 0;
5408 fail_free_lapic:
5409         kvm_free_lapic(vcpu);
5410 fail_mmu_destroy:
5411         kvm_mmu_destroy(vcpu);
5412 fail_free_pio_data:
5413         free_page((unsigned long)vcpu->arch.pio_data);
5414 fail:
5415         return r;
5416 }
5417
5418 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5419 {
5420         int idx;
5421
5422         kfree(vcpu->arch.mce_banks);
5423         kvm_free_lapic(vcpu);
5424         idx = srcu_read_lock(&vcpu->kvm->srcu);
5425         kvm_mmu_destroy(vcpu);
5426         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5427         free_page((unsigned long)vcpu->arch.pio_data);
5428 }
5429
5430 struct  kvm *kvm_arch_create_vm(void)
5431 {
5432         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5433
5434         if (!kvm)
5435                 return ERR_PTR(-ENOMEM);
5436
5437         kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5438         if (!kvm->arch.aliases) {
5439                 kfree(kvm);
5440                 return ERR_PTR(-ENOMEM);
5441         }
5442
5443         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5444         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5445
5446         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5447         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5448
5449         rdtscll(kvm->arch.vm_init_tsc);
5450
5451         return kvm;
5452 }
5453
5454 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5455 {
5456         vcpu_load(vcpu);
5457         kvm_mmu_unload(vcpu);
5458         vcpu_put(vcpu);
5459 }
5460
5461 static void kvm_free_vcpus(struct kvm *kvm)
5462 {
5463         unsigned int i;
5464         struct kvm_vcpu *vcpu;
5465
5466         /*
5467          * Unpin any mmu pages first.
5468          */
5469         kvm_for_each_vcpu(i, vcpu, kvm)
5470                 kvm_unload_vcpu_mmu(vcpu);
5471         kvm_for_each_vcpu(i, vcpu, kvm)
5472                 kvm_arch_vcpu_free(vcpu);
5473
5474         mutex_lock(&kvm->lock);
5475         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5476                 kvm->vcpus[i] = NULL;
5477
5478         atomic_set(&kvm->online_vcpus, 0);
5479         mutex_unlock(&kvm->lock);
5480 }
5481
5482 void kvm_arch_sync_events(struct kvm *kvm)
5483 {
5484         kvm_free_all_assigned_devices(kvm);
5485 }
5486
5487 void kvm_arch_destroy_vm(struct kvm *kvm)
5488 {
5489         kvm_iommu_unmap_guest(kvm);
5490         kvm_free_pit(kvm);
5491         kfree(kvm->arch.vpic);
5492         kfree(kvm->arch.vioapic);
5493         kvm_free_vcpus(kvm);
5494         kvm_free_physmem(kvm);
5495         if (kvm->arch.apic_access_page)
5496                 put_page(kvm->arch.apic_access_page);
5497         if (kvm->arch.ept_identity_pagetable)
5498                 put_page(kvm->arch.ept_identity_pagetable);
5499         cleanup_srcu_struct(&kvm->srcu);
5500         kfree(kvm->arch.aliases);
5501         kfree(kvm);
5502 }
5503
5504 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5505                                 struct kvm_memory_slot *memslot,
5506                                 struct kvm_memory_slot old,
5507                                 struct kvm_userspace_memory_region *mem,
5508                                 int user_alloc)
5509 {
5510         int npages = memslot->npages;
5511
5512         /*To keep backward compatibility with older userspace,
5513          *x86 needs to hanlde !user_alloc case.
5514          */
5515         if (!user_alloc) {
5516                 if (npages && !old.rmap) {
5517                         unsigned long userspace_addr;
5518
5519                         down_write(&current->mm->mmap_sem);
5520                         userspace_addr = do_mmap(NULL, 0,
5521                                                  npages * PAGE_SIZE,
5522                                                  PROT_READ | PROT_WRITE,
5523                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5524                                                  0);
5525                         up_write(&current->mm->mmap_sem);
5526
5527                         if (IS_ERR((void *)userspace_addr))
5528                                 return PTR_ERR((void *)userspace_addr);
5529
5530                         memslot->userspace_addr = userspace_addr;
5531                 }
5532         }
5533
5534
5535         return 0;
5536 }
5537
5538 void kvm_arch_commit_memory_region(struct kvm *kvm,
5539                                 struct kvm_userspace_memory_region *mem,
5540                                 struct kvm_memory_slot old,
5541                                 int user_alloc)
5542 {
5543
5544         int npages = mem->memory_size >> PAGE_SHIFT;
5545
5546         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5547                 int ret;
5548
5549                 down_write(&current->mm->mmap_sem);
5550                 ret = do_munmap(current->mm, old.userspace_addr,
5551                                 old.npages * PAGE_SIZE);
5552                 up_write(&current->mm->mmap_sem);
5553                 if (ret < 0)
5554                         printk(KERN_WARNING
5555                                "kvm_vm_ioctl_set_memory_region: "
5556                                "failed to munmap memory\n");
5557         }
5558
5559         spin_lock(&kvm->mmu_lock);
5560         if (!kvm->arch.n_requested_mmu_pages) {
5561                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5562                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5563         }
5564
5565         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5566         spin_unlock(&kvm->mmu_lock);
5567 }
5568
5569 void kvm_arch_flush_shadow(struct kvm *kvm)
5570 {
5571         kvm_mmu_zap_all(kvm);
5572         kvm_reload_remote_mmus(kvm);
5573 }
5574
5575 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5576 {
5577         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5578                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5579                 || vcpu->arch.nmi_pending ||
5580                 (kvm_arch_interrupt_allowed(vcpu) &&
5581                  kvm_cpu_has_interrupt(vcpu));
5582 }
5583
5584 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5585 {
5586         int me;
5587         int cpu = vcpu->cpu;
5588
5589         if (waitqueue_active(&vcpu->wq)) {
5590                 wake_up_interruptible(&vcpu->wq);
5591                 ++vcpu->stat.halt_wakeup;
5592         }
5593
5594         me = get_cpu();
5595         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5596                 if (atomic_xchg(&vcpu->guest_mode, 0))
5597                         smp_send_reschedule(cpu);
5598         put_cpu();
5599 }
5600
5601 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5602 {
5603         return kvm_x86_ops->interrupt_allowed(vcpu);
5604 }
5605
5606 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
5607 {
5608         unsigned long current_rip = kvm_rip_read(vcpu) +
5609                 get_segment_base(vcpu, VCPU_SREG_CS);
5610
5611         return current_rip == linear_rip;
5612 }
5613 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
5614
5615 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5616 {
5617         unsigned long rflags;
5618
5619         rflags = kvm_x86_ops->get_rflags(vcpu);
5620         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5621                 rflags &= ~X86_EFLAGS_TF;
5622         return rflags;
5623 }
5624 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5625
5626 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5627 {
5628         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5629             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
5630                 rflags |= X86_EFLAGS_TF;
5631         kvm_x86_ops->set_rflags(vcpu, rflags);
5632 }
5633 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5634
5635 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5636 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5637 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5638 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5639 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5640 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5641 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5642 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5643 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5644 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5645 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
5646 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);