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