arch/x86/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  10  *
  11  * Authors:
  12  *   Avi Kivity   <avi@qumranet.com>
  13  *   Yaniv Kamay  <yaniv@qumranet.com>
  14  *   Amit Shah    <amit.shah@qumranet.com>
  15  *   Ben-Ami Yassour <benami@il.ibm.com>
  16  *
  17  * This work is licensed under the terms of the GNU GPL, version 2.  See
  18  * the COPYING file in the top-level directory.
  19  *
  20  */
  21
  22 #include <linux/kvm_host.h>
  23 #include "irq.h"
  24 #include "mmu.h"
  25 #include "i8254.h"
  26 #include "tss.h"
  27 #include "kvm_cache_regs.h"
  28 #include "x86.h"
  29
  30 #include <linux/clocksource.h>
  31 #include <linux/interrupt.h>
  32 #include <linux/kvm.h>
  33 #include <linux/fs.h>
  34 #include <linux/vmalloc.h>
  35 #include <linux/module.h>
  36 #include <linux/mman.h>
  37 #include <linux/highmem.h>
  38 #include <linux/iommu.h>
  39 #include <linux/intel-iommu.h>
  40 #include <linux/cpufreq.h>
  41 #include <linux/user-return-notifier.h>
  42 #include <linux/srcu.h>
  43 #include <linux/slab.h>
  44 #include <linux/perf_event.h>
  45 #include <linux/uaccess.h>
  46 #include <linux/hash.h>
  47 #include <trace/events/kvm.h>
  48
  49 #define CREATE_TRACE_POINTS
  50 #include "trace.h"
  51
  52 #include <asm/debugreg.h>
  53 #include <asm/msr.h>
  54 #include <asm/desc.h>
  55 #include <asm/mtrr.h>
  56 #include <asm/mce.h>
  57 #include <asm/i387.h>
  58 #include <asm/xcr.h>
  59 #include <asm/pvclock.h>
  60 #include <asm/div64.h>
  61
  62 #define MAX_IO_MSRS 256
  63 #define KVM_MAX_MCE_BANKS 32
  64 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
  65
  66 #define emul_to_vcpu(ctxt) \
  67         container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
  68
  69 /* EFER defaults:
  70  * - enable syscall per default because its emulated by KVM
  71  * - enable LME and LMA per default on 64 bit KVM
  72  */
  73 #ifdef CONFIG_X86_64
  74 static
  75 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
  76 #else
  77 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
  78 #endif
  79
  80 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  81 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  82
  83 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  84 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  85                                     struct kvm_cpuid_entry2 __user *entries);
  86
  87 struct kvm_x86_ops *kvm_x86_ops;
  88 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  89
  90 int ignore_msrs = 0;
  91 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
  92
  93 bool kvm_has_tsc_control;
  94 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
  95 u32  kvm_max_guest_tsc_khz;
  96 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
  97
  98 #define KVM_NR_SHARED_MSRS 16
  99
 100 struct kvm_shared_msrs_global {
 101         int nr;
 102         u32 msrs[KVM_NR_SHARED_MSRS];
 103 };
 104
 105 struct kvm_shared_msrs {
 106         struct user_return_notifier urn;
 107         bool registered;
 108         struct kvm_shared_msr_values {
 109                 u64 host;
 110                 u64 curr;
 111         } values[KVM_NR_SHARED_MSRS];
 112 };
 113
 114 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 115 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
 116
 117 struct kvm_stats_debugfs_item debugfs_entries[] = {
 118         { "pf_fixed", VCPU_STAT(pf_fixed) },
 119         { "pf_guest", VCPU_STAT(pf_guest) },
 120         { "tlb_flush", VCPU_STAT(tlb_flush) },
 121         { "invlpg", VCPU_STAT(invlpg) },
 122         { "exits", VCPU_STAT(exits) },
 123         { "io_exits", VCPU_STAT(io_exits) },
 124         { "mmio_exits", VCPU_STAT(mmio_exits) },
 125         { "signal_exits", VCPU_STAT(signal_exits) },
 126         { "irq_window", VCPU_STAT(irq_window_exits) },
 127         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 128         { "halt_exits", VCPU_STAT(halt_exits) },
 129         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 130         { "hypercalls", VCPU_STAT(hypercalls) },
 131         { "request_irq", VCPU_STAT(request_irq_exits) },
 132         { "irq_exits", VCPU_STAT(irq_exits) },
 133         { "host_state_reload", VCPU_STAT(host_state_reload) },
 134         { "efer_reload", VCPU_STAT(efer_reload) },
 135         { "fpu_reload", VCPU_STAT(fpu_reload) },
 136         { "insn_emulation", VCPU_STAT(insn_emulation) },
 137         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 138         { "irq_injections", VCPU_STAT(irq_injections) },
 139         { "nmi_injections", VCPU_STAT(nmi_injections) },
 140         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 141         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 142         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 143         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 144         { "mmu_flooded", VM_STAT(mmu_flooded) },
 145         { "mmu_recycled", VM_STAT(mmu_recycled) },
 146         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 147         { "mmu_unsync", VM_STAT(mmu_unsync) },
 148         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 149         { "largepages", VM_STAT(lpages) },
 150         { NULL }
 151 };
 152
 153 u64 __read_mostly host_xcr0;
 154
 155 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
 156
 157 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
 158 {
 159         int i;
 160         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
 161                 vcpu->arch.apf.gfns[i] = ~0;
 162 }
 163
 164 static void kvm_on_user_return(struct user_return_notifier *urn)
 165 {
 166         unsigned slot;
 167         struct kvm_shared_msrs *locals
 168                 = container_of(urn, struct kvm_shared_msrs, urn);
 169         struct kvm_shared_msr_values *values;
 170
 171         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 172                 values = &locals->values[slot];
 173                 if (values->host != values->curr) {
 174                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 175                         values->curr = values->host;
 176                 }
 177         }
 178         locals->registered = false;
 179         user_return_notifier_unregister(urn);
 180 }
 181
 182 static void shared_msr_update(unsigned slot, u32 msr)
 183 {
 184         struct kvm_shared_msrs *smsr;
 185         u64 value;
 186
 187         smsr = &__get_cpu_var(shared_msrs);
 188         /* only read, and nobody should modify it at this time,
 189          * so don't need lock */
 190         if (slot >= shared_msrs_global.nr) {
 191                 printk(KERN_ERR "kvm: invalid MSR slot!");
 192                 return;
 193         }
 194         rdmsrl_safe(msr, &value);
 195         smsr->values[slot].host = value;
 196         smsr->values[slot].curr = value;
 197 }
 198
 199 void kvm_define_shared_msr(unsigned slot, u32 msr)
 200 {
 201         if (slot >= shared_msrs_global.nr)
 202                 shared_msrs_global.nr = slot + 1;
 203         shared_msrs_global.msrs[slot] = msr;
 204         /* we need ensured the shared_msr_global have been updated */
 205         smp_wmb();
 206 }
 207 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 208
 209 static void kvm_shared_msr_cpu_online(void)
 210 {
 211         unsigned i;
 212
 213         for (i = 0; i < shared_msrs_global.nr; ++i)
 214                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 215 }
 216
 217 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 218 {
 219         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 220
 221         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 222                 return;
 223         smsr->values[slot].curr = value;
 224         wrmsrl(shared_msrs_global.msrs[slot], value);
 225         if (!smsr->registered) {
 226                 smsr->urn.on_user_return = kvm_on_user_return;
 227                 user_return_notifier_register(&smsr->urn);
 228                 smsr->registered = true;
 229         }
 230 }
 231 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 232
 233 static void drop_user_return_notifiers(void *ignore)
 234 {
 235         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 236
 237         if (smsr->registered)
 238                 kvm_on_user_return(&smsr->urn);
 239 }
 240
 241 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 242 {
 243         if (irqchip_in_kernel(vcpu->kvm))
 244                 return vcpu->arch.apic_base;
 245         else
 246                 return vcpu->arch.apic_base;
 247 }
 248 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 249
 250 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 251 {
 252         /* TODO: reserve bits check */
 253         if (irqchip_in_kernel(vcpu->kvm))
 254                 kvm_lapic_set_base(vcpu, data);
 255         else
 256                 vcpu->arch.apic_base = data;
 257 }
 258 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 259
 260 #define EXCPT_BENIGN            0
 261 #define EXCPT_CONTRIBUTORY      1
 262 #define EXCPT_PF                2
 263
 264 static int exception_class(int vector)
 265 {
 266         switch (vector) {
 267         case PF_VECTOR:
 268                 return EXCPT_PF;
 269         case DE_VECTOR:
 270         case TS_VECTOR:
 271         case NP_VECTOR:
 272         case SS_VECTOR:
 273         case GP_VECTOR:
 274                 return EXCPT_CONTRIBUTORY;
 275         default:
 276                 break;
 277         }
 278         return EXCPT_BENIGN;
 279 }
 280
 281 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 282                 unsigned nr, bool has_error, u32 error_code,
 283                 bool reinject)
 284 {
 285         u32 prev_nr;
 286         int class1, class2;
 287
 288         kvm_make_request(KVM_REQ_EVENT, vcpu);
 289
 290         if (!vcpu->arch.exception.pending) {
 291         queue:
 292                 vcpu->arch.exception.pending = true;
 293                 vcpu->arch.exception.has_error_code = has_error;
 294                 vcpu->arch.exception.nr = nr;
 295                 vcpu->arch.exception.error_code = error_code;
 296                 vcpu->arch.exception.reinject = reinject;
 297                 return;
 298         }
 299
 300         /* to check exception */
 301         prev_nr = vcpu->arch.exception.nr;
 302         if (prev_nr == DF_VECTOR) {
 303                 /* triple fault -> shutdown */
 304                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
 305                 return;
 306         }
 307         class1 = exception_class(prev_nr);
 308         class2 = exception_class(nr);
 309         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 310                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 311                 /* generate double fault per SDM Table 5-5 */
 312                 vcpu->arch.exception.pending = true;
 313                 vcpu->arch.exception.has_error_code = true;
 314                 vcpu->arch.exception.nr = DF_VECTOR;
 315                 vcpu->arch.exception.error_code = 0;
 316         } else
 317                 /* replace previous exception with a new one in a hope
 318                    that instruction re-execution will regenerate lost
 319                    exception */
 320                 goto queue;
 321 }
 322
 323 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 324 {
 325         kvm_multiple_exception(vcpu, nr, false, 0, false);
 326 }
 327 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 328
 329 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 330 {
 331         kvm_multiple_exception(vcpu, nr, false, 0, true);
 332 }
 333 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 334
 335 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
 336 {
 337         if (err)
 338                 kvm_inject_gp(vcpu, 0);
 339         else
 340                 kvm_x86_ops->skip_emulated_instruction(vcpu);
 341 }
 342 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
 343
 344 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 345 {
 346         ++vcpu->stat.pf_guest;
 347         vcpu->arch.cr2 = fault->address;
 348         kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
 349 }
 350 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
 351
 352 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 353 {
 354         if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
 355                 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
 356         else
 357                 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
 358 }
 359
 360 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 361 {
 362         kvm_make_request(KVM_REQ_EVENT, vcpu);
 363         vcpu->arch.nmi_pending = 1;
 364 }
 365 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 366
 367 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 368 {
 369         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 370 }
 371 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 372
 373 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 374 {
 375         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 376 }
 377 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 378
 379 /*
 380  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 381  * a #GP and return false.
 382  */
 383 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 384 {
 385         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 386                 return true;
 387         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 388         return false;
 389 }
 390 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 391
 392 /*
 393  * This function will be used to read from the physical memory of the currently
 394  * running guest. The difference to kvm_read_guest_page is that this function
 395  * can read from guest physical or from the guest's guest physical memory.
 396  */
 397 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 398                             gfn_t ngfn, void *data, int offset, int len,
 399                             u32 access)
 400 {
 401         gfn_t real_gfn;
 402         gpa_t ngpa;
 403
 404         ngpa     = gfn_to_gpa(ngfn);
 405         real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
 406         if (real_gfn == UNMAPPED_GVA)
 407                 return -EFAULT;
 408
 409         real_gfn = gpa_to_gfn(real_gfn);
 410
 411         return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
 412 }
 413 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
 414
 415 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
 416                                void *data, int offset, int len, u32 access)
 417 {
 418         return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
 419                                        data, offset, len, access);
 420 }
 421
 422 /*
 423  * Load the pae pdptrs.  Return true is they are all valid.
 424  */
 425 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
 426 {
 427         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 428         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 429         int i;
 430         int ret;
 431         u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
 432
 433         ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
 434                                       offset * sizeof(u64), sizeof(pdpte),
 435                                       PFERR_USER_MASK|PFERR_WRITE_MASK);
 436         if (ret < 0) {
 437                 ret = 0;
 438                 goto out;
 439         }
 440         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 441                 if (is_present_gpte(pdpte[i]) &&
 442                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 443                         ret = 0;
 444                         goto out;
 445                 }
 446         }
 447         ret = 1;
 448
 449         memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
 450         __set_bit(VCPU_EXREG_PDPTR,
 451                   (unsigned long *)&vcpu->arch.regs_avail);
 452         __set_bit(VCPU_EXREG_PDPTR,
 453                   (unsigned long *)&vcpu->arch.regs_dirty);
 454 out:
 455
 456         return ret;
 457 }
 458 EXPORT_SYMBOL_GPL(load_pdptrs);
 459
 460 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 461 {
 462         u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
 463         bool changed = true;
 464         int offset;
 465         gfn_t gfn;
 466         int r;
 467
 468         if (is_long_mode(vcpu) || !is_pae(vcpu))
 469                 return false;
 470
 471         if (!test_bit(VCPU_EXREG_PDPTR,
 472                       (unsigned long *)&vcpu->arch.regs_avail))
 473                 return true;
 474
 475         gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
 476         offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
 477         r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
 478                                        PFERR_USER_MASK | PFERR_WRITE_MASK);
 479         if (r < 0)
 480                 goto out;
 481         changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
 482 out:
 483
 484         return changed;
 485 }
 486
 487 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 488 {
 489         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 490         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
 491                                     X86_CR0_CD | X86_CR0_NW;
 492
 493         cr0 |= X86_CR0_ET;
 494
 495 #ifdef CONFIG_X86_64
 496         if (cr0 & 0xffffffff00000000UL)
 497                 return 1;
 498 #endif
 499
 500         cr0 &= ~CR0_RESERVED_BITS;
 501
 502         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 503                 return 1;
 504
 505         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 506                 return 1;
 507
 508         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 509 #ifdef CONFIG_X86_64
 510                 if ((vcpu->arch.efer & EFER_LME)) {
 511                         int cs_db, cs_l;
 512
 513                         if (!is_pae(vcpu))
 514                                 return 1;
 515                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 516                         if (cs_l)
 517                                 return 1;
 518                 } else
 519 #endif
 520                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 521                                                  kvm_read_cr3(vcpu)))
 522                         return 1;
 523         }
 524
 525         kvm_x86_ops->set_cr0(vcpu, cr0);
 526
 527         if ((cr0 ^ old_cr0) & X86_CR0_PG) {
 528                 kvm_clear_async_pf_completion_queue(vcpu);
 529                 kvm_async_pf_hash_reset(vcpu);
 530         }
 531
 532         if ((cr0 ^ old_cr0) & update_bits)
 533                 kvm_mmu_reset_context(vcpu);
 534         return 0;
 535 }
 536 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 537
 538 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 539 {
 540         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 541 }
 542 EXPORT_SYMBOL_GPL(kvm_lmsw);
 543
 544 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 545 {
 546         u64 xcr0;
 547
 548         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
 549         if (index != XCR_XFEATURE_ENABLED_MASK)
 550                 return 1;
 551         xcr0 = xcr;
 552         if (kvm_x86_ops->get_cpl(vcpu) != 0)
 553                 return 1;
 554         if (!(xcr0 & XSTATE_FP))
 555                 return 1;
 556         if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
 557                 return 1;
 558         if (xcr0 & ~host_xcr0)
 559                 return 1;
 560         vcpu->arch.xcr0 = xcr0;
 561         vcpu->guest_xcr0_loaded = 0;
 562         return 0;
 563 }
 564
 565 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 566 {
 567         if (__kvm_set_xcr(vcpu, index, xcr)) {
 568                 kvm_inject_gp(vcpu, 0);
 569                 return 1;
 570         }
 571         return 0;
 572 }
 573 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 574
 575 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
 576 {
 577         struct kvm_cpuid_entry2 *best;
 578
 579         best = kvm_find_cpuid_entry(vcpu, 1, 0);
 580         return best && (best->ecx & bit(X86_FEATURE_XSAVE));
 581 }
 582
 583 static bool guest_cpuid_has_smep(struct kvm_vcpu *vcpu)
 584 {
 585         struct kvm_cpuid_entry2 *best;
 586
 587         best = kvm_find_cpuid_entry(vcpu, 7, 0);
 588         return best && (best->ebx & bit(X86_FEATURE_SMEP));
 589 }
 590
 591 static bool guest_cpuid_has_fsgsbase(struct kvm_vcpu *vcpu)
 592 {
 593         struct kvm_cpuid_entry2 *best;
 594
 595         best = kvm_find_cpuid_entry(vcpu, 7, 0);
 596         return best && (best->ebx & bit(X86_FEATURE_FSGSBASE));
 597 }
 598
 599 static void update_cpuid(struct kvm_vcpu *vcpu)
 600 {
 601         struct kvm_cpuid_entry2 *best;
 602
 603         best = kvm_find_cpuid_entry(vcpu, 1, 0);
 604         if (!best)
 605                 return;
 606
 607         /* Update OSXSAVE bit */
 608         if (cpu_has_xsave && best->function == 0x1) {
 609                 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
 610                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
 611                         best->ecx |= bit(X86_FEATURE_OSXSAVE);
 612         }
 613 }
 614
 615 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 616 {
 617         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 618         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
 619                                    X86_CR4_PAE | X86_CR4_SMEP;
 620         if (cr4 & CR4_RESERVED_BITS)
 621                 return 1;
 622
 623         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
 624                 return 1;
 625
 626         if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
 627                 return 1;
 628
 629         if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
 630                 return 1;
 631
 632         if (is_long_mode(vcpu)) {
 633                 if (!(cr4 & X86_CR4_PAE))
 634                         return 1;
 635         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 636                    && ((cr4 ^ old_cr4) & pdptr_bits)
 637                    && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 638                                    kvm_read_cr3(vcpu)))
 639                 return 1;
 640
 641         if (kvm_x86_ops->set_cr4(vcpu, cr4))
 642                 return 1;
 643
 644         if ((cr4 ^ old_cr4) & pdptr_bits)
 645                 kvm_mmu_reset_context(vcpu);
 646
 647         if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
 648                 update_cpuid(vcpu);
 649
 650         return 0;
 651 }
 652 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 653
 654 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 655 {
 656         if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
 657                 kvm_mmu_sync_roots(vcpu);
 658                 kvm_mmu_flush_tlb(vcpu);
 659                 return 0;
 660         }
 661
 662         if (is_long_mode(vcpu)) {
 663                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 664                         return 1;
 665         } else {
 666                 if (is_pae(vcpu)) {
 667                         if (cr3 & CR3_PAE_RESERVED_BITS)
 668                                 return 1;
 669                         if (is_paging(vcpu) &&
 670                             !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 671                                 return 1;
 672                 }
 673                 /*
 674                  * We don't check reserved bits in nonpae mode, because
 675                  * this isn't enforced, and VMware depends on this.
 676                  */
 677         }
 678
 679         /*
 680          * Does the new cr3 value map to physical memory? (Note, we
 681          * catch an invalid cr3 even in real-mode, because it would
 682          * cause trouble later on when we turn on paging anyway.)
 683          *
 684          * A real CPU would silently accept an invalid cr3 and would
 685          * attempt to use it - with largely undefined (and often hard
 686          * to debug) behavior on the guest side.
 687          */
 688         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 689                 return 1;
 690         vcpu->arch.cr3 = cr3;
 691         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
 692         vcpu->arch.mmu.new_cr3(vcpu);
 693         return 0;
 694 }
 695 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 696
 697 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 698 {
 699         if (cr8 & CR8_RESERVED_BITS)
 700                 return 1;
 701         if (irqchip_in_kernel(vcpu->kvm))
 702                 kvm_lapic_set_tpr(vcpu, cr8);
 703         else
 704                 vcpu->arch.cr8 = cr8;
 705         return 0;
 706 }
 707 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 708
 709 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 710 {
 711         if (irqchip_in_kernel(vcpu->kvm))
 712                 return kvm_lapic_get_cr8(vcpu);
 713         else
 714                 return vcpu->arch.cr8;
 715 }
 716 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 717
 718 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 719 {
 720         switch (dr) {
 721         case 0 ... 3:
 722                 vcpu->arch.db[dr] = val;
 723                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 724                         vcpu->arch.eff_db[dr] = val;
 725                 break;
 726         case 4:
 727                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 728                         return 1; /* #UD */
 729                 /* fall through */
 730         case 6:
 731                 if (val & 0xffffffff00000000ULL)
 732                         return -1; /* #GP */
 733                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
 734                 break;
 735         case 5:
 736                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 737                         return 1; /* #UD */
 738                 /* fall through */
 739         default: /* 7 */
 740                 if (val & 0xffffffff00000000ULL)
 741                         return -1; /* #GP */
 742                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 743                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 744                         kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
 745                         vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
 746                 }
 747                 break;
 748         }
 749
 750         return 0;
 751 }
 752
 753 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 754 {
 755         int res;
 756
 757         res = __kvm_set_dr(vcpu, dr, val);
 758         if (res > 0)
 759                 kvm_queue_exception(vcpu, UD_VECTOR);
 760         else if (res < 0)
 761                 kvm_inject_gp(vcpu, 0);
 762
 763         return res;
 764 }
 765 EXPORT_SYMBOL_GPL(kvm_set_dr);
 766
 767 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 768 {
 769         switch (dr) {
 770         case 0 ... 3:
 771                 *val = vcpu->arch.db[dr];
 772                 break;
 773         case 4:
 774                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 775                         return 1;
 776                 /* fall through */
 777         case 6:
 778                 *val = vcpu->arch.dr6;
 779                 break;
 780         case 5:
 781                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 782                         return 1;
 783                 /* fall through */
 784         default: /* 7 */
 785                 *val = vcpu->arch.dr7;
 786                 break;
 787         }
 788
 789         return 0;
 790 }
 791
 792 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 793 {
 794         if (_kvm_get_dr(vcpu, dr, val)) {
 795                 kvm_queue_exception(vcpu, UD_VECTOR);
 796                 return 1;
 797         }
 798         return 0;
 799 }
 800 EXPORT_SYMBOL_GPL(kvm_get_dr);
 801
 802 /*
 803  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 804  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 805  *
 806  * This list is modified at module load time to reflect the
 807  * capabilities of the host cpu. This capabilities test skips MSRs that are
 808  * kvm-specific. Those are put in the beginning of the list.
 809  */
 810
 811 #define KVM_SAVE_MSRS_BEGIN     9
 812 static u32 msrs_to_save[] = {
 813         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 814         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 815         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 816         HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
 817         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 818         MSR_STAR,
 819 #ifdef CONFIG_X86_64
 820         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 821 #endif
 822         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 823 };
 824
 825 static unsigned num_msrs_to_save;
 826
 827 static u32 emulated_msrs[] = {
 828         MSR_IA32_MISC_ENABLE,
 829         MSR_IA32_MCG_STATUS,
 830         MSR_IA32_MCG_CTL,
 831 };
 832
 833 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
 834 {
 835         u64 old_efer = vcpu->arch.efer;
 836
 837         if (efer & efer_reserved_bits)
 838                 return 1;
 839
 840         if (is_paging(vcpu)
 841             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
 842                 return 1;
 843
 844         if (efer & EFER_FFXSR) {
 845                 struct kvm_cpuid_entry2 *feat;
 846
 847                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 848                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
 849                         return 1;
 850         }
 851
 852         if (efer & EFER_SVME) {
 853                 struct kvm_cpuid_entry2 *feat;
 854
 855                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 856                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
 857                         return 1;
 858         }
 859
 860         efer &= ~EFER_LMA;
 861         efer |= vcpu->arch.efer & EFER_LMA;
 862
 863         kvm_x86_ops->set_efer(vcpu, efer);
 864
 865         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 866
 867         /* Update reserved bits */
 868         if ((efer ^ old_efer) & EFER_NX)
 869                 kvm_mmu_reset_context(vcpu);
 870
 871         return 0;
 872 }
 873
 874 void kvm_enable_efer_bits(u64 mask)
 875 {
 876        efer_reserved_bits &= ~mask;
 877 }
 878 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 879
 880
 881 /*
 882  * Writes msr value into into the appropriate "register".
 883  * Returns 0 on success, non-0 otherwise.
 884  * Assumes vcpu_load() was already called.
 885  */
 886 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 887 {
 888         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 889 }
 890
 891 /*
 892  * Adapt set_msr() to msr_io()'s calling convention
 893  */
 894 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 895 {
 896         return kvm_set_msr(vcpu, index, *data);
 897 }
 898
 899 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 900 {
 901         int version;
 902         int r;
 903         struct pvclock_wall_clock wc;
 904         struct timespec boot;
 905
 906         if (!wall_clock)
 907                 return;
 908
 909         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
 910         if (r)
 911                 return;
 912
 913         if (version & 1)
 914                 ++version;  /* first time write, random junk */
 915
 916         ++version;
 917
 918         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 919
 920         /*
 921          * The guest calculates current wall clock time by adding
 922          * system time (updated by kvm_guest_time_update below) to the
 923          * wall clock specified here.  guest system time equals host
 924          * system time for us, thus we must fill in host boot time here.
 925          */
 926         getboottime(&boot);
 927
 928         wc.sec = boot.tv_sec;
 929         wc.nsec = boot.tv_nsec;
 930         wc.version = version;
 931
 932         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 933
 934         version++;
 935         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 936 }
 937
 938 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 939 {
 940         uint32_t quotient, remainder;
 941
 942         /* Don't try to replace with do_div(), this one calculates
 943          * "(dividend << 32) / divisor" */
 944         __asm__ ( "divl %4"
 945                   : "=a" (quotient), "=d" (remainder)
 946                   : "0" (0), "1" (dividend), "r" (divisor) );
 947         return quotient;
 948 }
 949
 950 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
 951                                s8 *pshift, u32 *pmultiplier)
 952 {
 953         uint64_t scaled64;
 954         int32_t  shift = 0;
 955         uint64_t tps64;
 956         uint32_t tps32;
 957
 958         tps64 = base_khz * 1000LL;
 959         scaled64 = scaled_khz * 1000LL;
 960         while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
 961                 tps64 >>= 1;
 962                 shift--;
 963         }
 964
 965         tps32 = (uint32_t)tps64;
 966         while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
 967                 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
 968                         scaled64 >>= 1;
 969                 else
 970                         tps32 <<= 1;
 971                 shift++;
 972         }
 973
 974         *pshift = shift;
 975         *pmultiplier = div_frac(scaled64, tps32);
 976
 977         pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
 978                  __func__, base_khz, scaled_khz, shift, *pmultiplier);
 979 }
 980
 981 static inline u64 get_kernel_ns(void)
 982 {
 983         struct timespec ts;
 984
 985         WARN_ON(preemptible());
 986         ktime_get_ts(&ts);
 987         monotonic_to_bootbased(&ts);
 988         return timespec_to_ns(&ts);
 989 }
 990
 991 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 992 unsigned long max_tsc_khz;
 993
 994 static inline int kvm_tsc_changes_freq(void)
 995 {
 996         int cpu = get_cpu();
 997         int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
 998                   cpufreq_quick_get(cpu) != 0;
 999         put_cpu();
1000         return ret;
1001 }
1002
1003 static u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
1004 {
1005         if (vcpu->arch.virtual_tsc_khz)
1006                 return vcpu->arch.virtual_tsc_khz;
1007         else
1008                 return __this_cpu_read(cpu_tsc_khz);
1009 }
1010
1011 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1012 {
1013         u64 ret;
1014
1015         WARN_ON(preemptible());
1016         if (kvm_tsc_changes_freq())
1017                 printk_once(KERN_WARNING
1018                  "kvm: unreliable cycle conversion on adjustable rate TSC\n");
1019         ret = nsec * vcpu_tsc_khz(vcpu);
1020         do_div(ret, USEC_PER_SEC);
1021         return ret;
1022 }
1023
1024 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1025 {
1026         /* Compute a scale to convert nanoseconds in TSC cycles */
1027         kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1028                            &vcpu->arch.tsc_catchup_shift,
1029                            &vcpu->arch.tsc_catchup_mult);
1030 }
1031
1032 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1033 {
1034         u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1035                                       vcpu->arch.tsc_catchup_mult,
1036                                       vcpu->arch.tsc_catchup_shift);
1037         tsc += vcpu->arch.last_tsc_write;
1038         return tsc;
1039 }
1040
1041 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1042 {
1043         struct kvm *kvm = vcpu->kvm;
1044         u64 offset, ns, elapsed;
1045         unsigned long flags;
1046         s64 sdiff;
1047
1048         raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1049         offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1050         ns = get_kernel_ns();
1051         elapsed = ns - kvm->arch.last_tsc_nsec;
1052         sdiff = data - kvm->arch.last_tsc_write;
1053         if (sdiff < 0)
1054                 sdiff = -sdiff;
1055
1056         /*
1057          * Special case: close write to TSC within 5 seconds of
1058          * another CPU is interpreted as an attempt to synchronize
1059          * The 5 seconds is to accommodate host load / swapping as
1060          * well as any reset of TSC during the boot process.
1061          *
1062          * In that case, for a reliable TSC, we can match TSC offsets,
1063          * or make a best guest using elapsed value.
1064          */
1065         if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1066             elapsed < 5ULL * NSEC_PER_SEC) {
1067                 if (!check_tsc_unstable()) {
1068                         offset = kvm->arch.last_tsc_offset;
1069                         pr_debug("kvm: matched tsc offset for %llu\n", data);
1070                 } else {
1071                         u64 delta = nsec_to_cycles(vcpu, elapsed);
1072                         offset += delta;
1073                         pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1074                 }
1075                 ns = kvm->arch.last_tsc_nsec;
1076         }
1077         kvm->arch.last_tsc_nsec = ns;
1078         kvm->arch.last_tsc_write = data;
1079         kvm->arch.last_tsc_offset = offset;
1080         kvm_x86_ops->write_tsc_offset(vcpu, offset);
1081         raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1082
1083         /* Reset of TSC must disable overshoot protection below */
1084         vcpu->arch.hv_clock.tsc_timestamp = 0;
1085         vcpu->arch.last_tsc_write = data;
1086         vcpu->arch.last_tsc_nsec = ns;
1087 }
1088 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1089
1090 static int kvm_guest_time_update(struct kvm_vcpu *v)
1091 {
1092         unsigned long flags;
1093         struct kvm_vcpu_arch *vcpu = &v->arch;
1094         void *shared_kaddr;
1095         unsigned long this_tsc_khz;
1096         s64 kernel_ns, max_kernel_ns;
1097         u64 tsc_timestamp;
1098
1099         /* Keep irq disabled to prevent changes to the clock */
1100         local_irq_save(flags);
1101         kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1102         kernel_ns = get_kernel_ns();
1103         this_tsc_khz = vcpu_tsc_khz(v);
1104         if (unlikely(this_tsc_khz == 0)) {
1105                 local_irq_restore(flags);
1106                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1107                 return 1;
1108         }
1109
1110         /*
1111          * We may have to catch up the TSC to match elapsed wall clock
1112          * time for two reasons, even if kvmclock is used.
1113          *   1) CPU could have been running below the maximum TSC rate
1114          *   2) Broken TSC compensation resets the base at each VCPU
1115          *      entry to avoid unknown leaps of TSC even when running
1116          *      again on the same CPU.  This may cause apparent elapsed
1117          *      time to disappear, and the guest to stand still or run
1118          *      very slowly.
1119          */
1120         if (vcpu->tsc_catchup) {
1121                 u64 tsc = compute_guest_tsc(v, kernel_ns);
1122                 if (tsc > tsc_timestamp) {
1123                         kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1124                         tsc_timestamp = tsc;
1125                 }
1126         }
1127
1128         local_irq_restore(flags);
1129
1130         if (!vcpu->time_page)
1131                 return 0;
1132
1133         /*
1134          * Time as measured by the TSC may go backwards when resetting the base
1135          * tsc_timestamp.  The reason for this is that the TSC resolution is
1136          * higher than the resolution of the other clock scales.  Thus, many
1137          * possible measurments of the TSC correspond to one measurement of any
1138          * other clock, and so a spread of values is possible.  This is not a
1139          * problem for the computation of the nanosecond clock; with TSC rates
1140          * around 1GHZ, there can only be a few cycles which correspond to one
1141          * nanosecond value, and any path through this code will inevitably
1142          * take longer than that.  However, with the kernel_ns value itself,
1143          * the precision may be much lower, down to HZ granularity.  If the
1144          * first sampling of TSC against kernel_ns ends in the low part of the
1145          * range, and the second in the high end of the range, we can get:
1146          *
1147          * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1148          *
1149          * As the sampling errors potentially range in the thousands of cycles,
1150          * it is possible such a time value has already been observed by the
1151          * guest.  To protect against this, we must compute the system time as
1152          * observed by the guest and ensure the new system time is greater.
1153          */
1154         max_kernel_ns = 0;
1155         if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1156                 max_kernel_ns = vcpu->last_guest_tsc -
1157                                 vcpu->hv_clock.tsc_timestamp;
1158                 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1159                                     vcpu->hv_clock.tsc_to_system_mul,
1160                                     vcpu->hv_clock.tsc_shift);
1161                 max_kernel_ns += vcpu->last_kernel_ns;
1162         }
1163
1164         if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1165                 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1166                                    &vcpu->hv_clock.tsc_shift,
1167                                    &vcpu->hv_clock.tsc_to_system_mul);
1168                 vcpu->hw_tsc_khz = this_tsc_khz;
1169         }
1170
1171         if (max_kernel_ns > kernel_ns)
1172                 kernel_ns = max_kernel_ns;
1173
1174         /* With all the info we got, fill in the values */
1175         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1176         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1177         vcpu->last_kernel_ns = kernel_ns;
1178         vcpu->last_guest_tsc = tsc_timestamp;
1179         vcpu->hv_clock.flags = 0;
1180
1181         /*
1182          * The interface expects us to write an even number signaling that the
1183          * update is finished. Since the guest won't see the intermediate
1184          * state, we just increase by 2 at the end.
1185          */
1186         vcpu->hv_clock.version += 2;
1187
1188         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1189
1190         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1191                sizeof(vcpu->hv_clock));
1192
1193         kunmap_atomic(shared_kaddr, KM_USER0);
1194
1195         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1196         return 0;
1197 }
1198
1199 static bool msr_mtrr_valid(unsigned msr)
1200 {
1201         switch (msr) {
1202         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1203         case MSR_MTRRfix64K_00000:
1204         case MSR_MTRRfix16K_80000:
1205         case MSR_MTRRfix16K_A0000:
1206         case MSR_MTRRfix4K_C0000:
1207         case MSR_MTRRfix4K_C8000:
1208         case MSR_MTRRfix4K_D0000:
1209         case MSR_MTRRfix4K_D8000:
1210         case MSR_MTRRfix4K_E0000:
1211         case MSR_MTRRfix4K_E8000:
1212         case MSR_MTRRfix4K_F0000:
1213         case MSR_MTRRfix4K_F8000:
1214         case MSR_MTRRdefType:
1215         case MSR_IA32_CR_PAT:
1216                 return true;
1217         case 0x2f8:
1218                 return true;
1219         }
1220         return false;
1221 }
1222
1223 static bool valid_pat_type(unsigned t)
1224 {
1225         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1226 }
1227
1228 static bool valid_mtrr_type(unsigned t)
1229 {
1230         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1231 }
1232
1233 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1234 {
1235         int i;
1236
1237         if (!msr_mtrr_valid(msr))
1238                 return false;
1239
1240         if (msr == MSR_IA32_CR_PAT) {
1241                 for (i = 0; i < 8; i++)
1242                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
1243                                 return false;
1244                 return true;
1245         } else if (msr == MSR_MTRRdefType) {
1246                 if (data & ~0xcff)
1247                         return false;
1248                 return valid_mtrr_type(data & 0xff);
1249         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1250                 for (i = 0; i < 8 ; i++)
1251                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1252                                 return false;
1253                 return true;
1254         }
1255
1256         /* variable MTRRs */
1257         return valid_mtrr_type(data & 0xff);
1258 }
1259
1260 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1261 {
1262         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1263
1264         if (!mtrr_valid(vcpu, msr, data))
1265                 return 1;
1266
1267         if (msr == MSR_MTRRdefType) {
1268                 vcpu->arch.mtrr_state.def_type = data;
1269                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1270         } else if (msr == MSR_MTRRfix64K_00000)
1271                 p[0] = data;
1272         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1273                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1274         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1275                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1276         else if (msr == MSR_IA32_CR_PAT)
1277                 vcpu->arch.pat = data;
1278         else {  /* Variable MTRRs */
1279                 int idx, is_mtrr_mask;
1280                 u64 *pt;
1281
1282                 idx = (msr - 0x200) / 2;
1283                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1284                 if (!is_mtrr_mask)
1285                         pt =
1286                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1287                 else
1288                         pt =
1289                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1290                 *pt = data;
1291         }
1292
1293         kvm_mmu_reset_context(vcpu);
1294         return 0;
1295 }
1296
1297 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1298 {
1299         u64 mcg_cap = vcpu->arch.mcg_cap;
1300         unsigned bank_num = mcg_cap & 0xff;
1301
1302         switch (msr) {
1303         case MSR_IA32_MCG_STATUS:
1304                 vcpu->arch.mcg_status = data;
1305                 break;
1306         case MSR_IA32_MCG_CTL:
1307                 if (!(mcg_cap & MCG_CTL_P))
1308                         return 1;
1309                 if (data != 0 && data != ~(u64)0)
1310                         return -1;
1311                 vcpu->arch.mcg_ctl = data;
1312                 break;
1313         default:
1314                 if (msr >= MSR_IA32_MC0_CTL &&
1315                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1316                         u32 offset = msr - MSR_IA32_MC0_CTL;
1317                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1318                          * some Linux kernels though clear bit 10 in bank 4 to
1319                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1320                          * this to avoid an uncatched #GP in the guest
1321                          */
1322                         if ((offset & 0x3) == 0 &&
1323                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1324                                 return -1;
1325                         vcpu->arch.mce_banks[offset] = data;
1326                         break;
1327                 }
1328                 return 1;
1329         }
1330         return 0;
1331 }
1332
1333 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1334 {
1335         struct kvm *kvm = vcpu->kvm;
1336         int lm = is_long_mode(vcpu);
1337         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1338                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1339         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1340                 : kvm->arch.xen_hvm_config.blob_size_32;
1341         u32 page_num = data & ~PAGE_MASK;
1342         u64 page_addr = data & PAGE_MASK;
1343         u8 *page;
1344         int r;
1345
1346         r = -E2BIG;
1347         if (page_num >= blob_size)
1348                 goto out;
1349         r = -ENOMEM;
1350         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1351         if (!page)
1352                 goto out;
1353         r = -EFAULT;
1354         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1355                 goto out_free;
1356         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1357                 goto out_free;
1358         r = 0;
1359 out_free:
1360         kfree(page);
1361 out:
1362         return r;
1363 }
1364
1365 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1366 {
1367         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1368 }
1369
1370 static bool kvm_hv_msr_partition_wide(u32 msr)
1371 {
1372         bool r = false;
1373         switch (msr) {
1374         case HV_X64_MSR_GUEST_OS_ID:
1375         case HV_X64_MSR_HYPERCALL:
1376                 r = true;
1377                 break;
1378         }
1379
1380         return r;
1381 }
1382
1383 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1384 {
1385         struct kvm *kvm = vcpu->kvm;
1386
1387         switch (msr) {
1388         case HV_X64_MSR_GUEST_OS_ID:
1389                 kvm->arch.hv_guest_os_id = data;
1390                 /* setting guest os id to zero disables hypercall page */
1391                 if (!kvm->arch.hv_guest_os_id)
1392                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1393                 break;
1394         case HV_X64_MSR_HYPERCALL: {
1395                 u64 gfn;
1396                 unsigned long addr;
1397                 u8 instructions[4];
1398
1399                 /* if guest os id is not set hypercall should remain disabled */
1400                 if (!kvm->arch.hv_guest_os_id)
1401                         break;
1402                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1403                         kvm->arch.hv_hypercall = data;
1404                         break;
1405                 }
1406                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1407                 addr = gfn_to_hva(kvm, gfn);
1408                 if (kvm_is_error_hva(addr))
1409                         return 1;
1410                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1411                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1412                 if (__copy_to_user((void __user *)addr, instructions, 4))
1413                         return 1;
1414                 kvm->arch.hv_hypercall = data;
1415                 break;
1416         }
1417         default:
1418                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1419                           "data 0x%llx\n", msr, data);
1420                 return 1;
1421         }
1422         return 0;
1423 }
1424
1425 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1426 {
1427         switch (msr) {
1428         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1429                 unsigned long addr;
1430
1431                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1432                         vcpu->arch.hv_vapic = data;
1433                         break;
1434                 }
1435                 addr = gfn_to_hva(vcpu->kvm, data >>
1436                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1437                 if (kvm_is_error_hva(addr))
1438                         return 1;
1439                 if (__clear_user((void __user *)addr, PAGE_SIZE))
1440                         return 1;
1441                 vcpu->arch.hv_vapic = data;
1442                 break;
1443         }
1444         case HV_X64_MSR_EOI:
1445                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1446         case HV_X64_MSR_ICR:
1447                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1448         case HV_X64_MSR_TPR:
1449                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1450         default:
1451                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1452                           "data 0x%llx\n", msr, data);
1453                 return 1;
1454         }
1455
1456         return 0;
1457 }
1458
1459 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1460 {
1461         gpa_t gpa = data & ~0x3f;
1462
1463         /* Bits 2:5 are resrved, Should be zero */
1464         if (data & 0x3c)
1465                 return 1;
1466
1467         vcpu->arch.apf.msr_val = data;
1468
1469         if (!(data & KVM_ASYNC_PF_ENABLED)) {
1470                 kvm_clear_async_pf_completion_queue(vcpu);
1471                 kvm_async_pf_hash_reset(vcpu);
1472                 return 0;
1473         }
1474
1475         if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1476                 return 1;
1477
1478         vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1479         kvm_async_pf_wakeup_all(vcpu);
1480         return 0;
1481 }
1482
1483 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1484 {
1485         if (vcpu->arch.time_page) {
1486                 kvm_release_page_dirty(vcpu->arch.time_page);
1487                 vcpu->arch.time_page = NULL;
1488         }
1489 }
1490
1491 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1492 {
1493         u64 delta;
1494
1495         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1496                 return;
1497
1498         delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1499         vcpu->arch.st.last_steal = current->sched_info.run_delay;
1500         vcpu->arch.st.accum_steal = delta;
1501 }
1502
1503 static void record_steal_time(struct kvm_vcpu *vcpu)
1504 {
1505         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1506                 return;
1507
1508         if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1509                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1510                 return;
1511
1512         vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1513         vcpu->arch.st.steal.version += 2;
1514         vcpu->arch.st.accum_steal = 0;
1515
1516         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1517                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1518 }
1519
1520 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1521 {
1522         switch (msr) {
1523         case MSR_EFER:
1524                 return set_efer(vcpu, data);
1525         case MSR_K7_HWCR:
1526                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1527                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
1528                 if (data != 0) {
1529                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1530                                 data);
1531                         return 1;
1532                 }
1533                 break;
1534         case MSR_FAM10H_MMIO_CONF_BASE:
1535                 if (data != 0) {
1536                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1537                                 "0x%llx\n", data);
1538                         return 1;
1539                 }
1540                 break;
1541         case MSR_AMD64_NB_CFG:
1542                 break;
1543         case MSR_IA32_DEBUGCTLMSR:
1544                 if (!data) {
1545                         /* We support the non-activated case already */
1546                         break;
1547                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1548                         /* Values other than LBR and BTF are vendor-specific,
1549                            thus reserved and should throw a #GP */
1550                         return 1;
1551                 }
1552                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1553                         __func__, data);
1554                 break;
1555         case MSR_IA32_UCODE_REV:
1556         case MSR_IA32_UCODE_WRITE:
1557         case MSR_VM_HSAVE_PA:
1558         case MSR_AMD64_PATCH_LOADER:
1559                 break;
1560         case 0x200 ... 0x2ff:
1561                 return set_msr_mtrr(vcpu, msr, data);
1562         case MSR_IA32_APICBASE:
1563                 kvm_set_apic_base(vcpu, data);
1564                 break;
1565         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1566                 return kvm_x2apic_msr_write(vcpu, msr, data);
1567         case MSR_IA32_MISC_ENABLE:
1568                 vcpu->arch.ia32_misc_enable_msr = data;
1569                 break;
1570         case MSR_KVM_WALL_CLOCK_NEW:
1571         case MSR_KVM_WALL_CLOCK:
1572                 vcpu->kvm->arch.wall_clock = data;
1573                 kvm_write_wall_clock(vcpu->kvm, data);
1574                 break;
1575         case MSR_KVM_SYSTEM_TIME_NEW:
1576         case MSR_KVM_SYSTEM_TIME: {
1577                 kvmclock_reset(vcpu);
1578
1579                 vcpu->arch.time = data;
1580                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1581
1582                 /* we verify if the enable bit is set... */
1583                 if (!(data & 1))
1584                         break;
1585
1586                 /* ...but clean it before doing the actual write */
1587                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1588
1589                 vcpu->arch.time_page =
1590                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1591
1592                 if (is_error_page(vcpu->arch.time_page)) {
1593                         kvm_release_page_clean(vcpu->arch.time_page);
1594                         vcpu->arch.time_page = NULL;
1595                 }
1596                 break;
1597         }
1598         case MSR_KVM_ASYNC_PF_EN:
1599                 if (kvm_pv_enable_async_pf(vcpu, data))
1600                         return 1;
1601                 break;
1602         case MSR_KVM_STEAL_TIME:
1603
1604                 if (unlikely(!sched_info_on()))
1605                         return 1;
1606
1607                 if (data & KVM_STEAL_RESERVED_MASK)
1608                         return 1;
1609
1610                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1611                                                         data & KVM_STEAL_VALID_BITS))
1612                         return 1;
1613
1614                 vcpu->arch.st.msr_val = data;
1615
1616                 if (!(data & KVM_MSR_ENABLED))
1617                         break;
1618
1619                 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1620
1621                 preempt_disable();
1622                 accumulate_steal_time(vcpu);
1623                 preempt_enable();
1624
1625                 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1626
1627                 break;
1628
1629         case MSR_IA32_MCG_CTL:
1630         case MSR_IA32_MCG_STATUS:
1631         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1632                 return set_msr_mce(vcpu, msr, data);
1633
1634         /* Performance counters are not protected by a CPUID bit,
1635          * so we should check all of them in the generic path for the sake of
1636          * cross vendor migration.
1637          * Writing a zero into the event select MSRs disables them,
1638          * which we perfectly emulate ;-). Any other value should be at least
1639          * reported, some guests depend on them.
1640          */
1641         case MSR_P6_EVNTSEL0:
1642         case MSR_P6_EVNTSEL1:
1643         case MSR_K7_EVNTSEL0:
1644         case MSR_K7_EVNTSEL1:
1645         case MSR_K7_EVNTSEL2:
1646         case MSR_K7_EVNTSEL3:
1647                 if (data != 0)
1648                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1649                                 "0x%x data 0x%llx\n", msr, data);
1650                 break;
1651         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1652          * so we ignore writes to make it happy.
1653          */
1654         case MSR_P6_PERFCTR0:
1655         case MSR_P6_PERFCTR1:
1656         case MSR_K7_PERFCTR0:
1657         case MSR_K7_PERFCTR1:
1658         case MSR_K7_PERFCTR2:
1659         case MSR_K7_PERFCTR3:
1660                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1661                         "0x%x data 0x%llx\n", msr, data);
1662                 break;
1663         case MSR_K7_CLK_CTL:
1664                 /*
1665                  * Ignore all writes to this no longer documented MSR.
1666                  * Writes are only relevant for old K7 processors,
1667                  * all pre-dating SVM, but a recommended workaround from
1668                  * AMD for these chips. It is possible to speicify the
1669                  * affected processor models on the command line, hence
1670                  * the need to ignore the workaround.
1671                  */
1672                 break;
1673         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1674                 if (kvm_hv_msr_partition_wide(msr)) {
1675                         int r;
1676                         mutex_lock(&vcpu->kvm->lock);
1677                         r = set_msr_hyperv_pw(vcpu, msr, data);
1678                         mutex_unlock(&vcpu->kvm->lock);
1679                         return r;
1680                 } else
1681                         return set_msr_hyperv(vcpu, msr, data);
1682                 break;
1683         case MSR_IA32_BBL_CR_CTL3:
1684                 /* Drop writes to this legacy MSR -- see rdmsr
1685                  * counterpart for further detail.
1686                  */
1687                 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1688                 break;
1689         default:
1690                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1691                         return xen_hvm_config(vcpu, data);
1692                 if (!ignore_msrs) {
1693                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1694                                 msr, data);
1695                         return 1;
1696                 } else {
1697                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1698                                 msr, data);
1699                         break;
1700                 }
1701         }
1702         return 0;
1703 }
1704 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1705
1706
1707 /*
1708  * Reads an msr value (of 'msr_index') into 'pdata'.
1709  * Returns 0 on success, non-0 otherwise.
1710  * Assumes vcpu_load() was already called.
1711  */
1712 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1713 {
1714         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1715 }
1716
1717 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1718 {
1719         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1720
1721         if (!msr_mtrr_valid(msr))
1722                 return 1;
1723
1724         if (msr == MSR_MTRRdefType)
1725                 *pdata = vcpu->arch.mtrr_state.def_type +
1726                          (vcpu->arch.mtrr_state.enabled << 10);
1727         else if (msr == MSR_MTRRfix64K_00000)
1728                 *pdata = p[0];
1729         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1730                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1731         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1732                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1733         else if (msr == MSR_IA32_CR_PAT)
1734                 *pdata = vcpu->arch.pat;
1735         else {  /* Variable MTRRs */
1736                 int idx, is_mtrr_mask;
1737                 u64 *pt;
1738
1739                 idx = (msr - 0x200) / 2;
1740                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1741                 if (!is_mtrr_mask)
1742                         pt =
1743                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1744                 else
1745                         pt =
1746                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1747                 *pdata = *pt;
1748         }
1749
1750         return 0;
1751 }
1752
1753 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1754 {
1755         u64 data;
1756         u64 mcg_cap = vcpu->arch.mcg_cap;
1757         unsigned bank_num = mcg_cap & 0xff;
1758
1759         switch (msr) {
1760         case MSR_IA32_P5_MC_ADDR:
1761         case MSR_IA32_P5_MC_TYPE:
1762                 data = 0;
1763                 break;
1764         case MSR_IA32_MCG_CAP:
1765                 data = vcpu->arch.mcg_cap;
1766                 break;
1767         case MSR_IA32_MCG_CTL:
1768                 if (!(mcg_cap & MCG_CTL_P))
1769                         return 1;
1770                 data = vcpu->arch.mcg_ctl;
1771                 break;
1772         case MSR_IA32_MCG_STATUS:
1773                 data = vcpu->arch.mcg_status;
1774                 break;
1775         default:
1776                 if (msr >= MSR_IA32_MC0_CTL &&
1777                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1778                         u32 offset = msr - MSR_IA32_MC0_CTL;
1779                         data = vcpu->arch.mce_banks[offset];
1780                         break;
1781                 }
1782                 return 1;
1783         }
1784         *pdata = data;
1785         return 0;
1786 }
1787
1788 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1789 {
1790         u64 data = 0;
1791         struct kvm *kvm = vcpu->kvm;
1792
1793         switch (msr) {
1794         case HV_X64_MSR_GUEST_OS_ID:
1795                 data = kvm->arch.hv_guest_os_id;
1796                 break;
1797         case HV_X64_MSR_HYPERCALL:
1798                 data = kvm->arch.hv_hypercall;
1799                 break;
1800         default:
1801                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1802                 return 1;
1803         }
1804
1805         *pdata = data;
1806         return 0;
1807 }
1808
1809 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1810 {
1811         u64 data = 0;
1812
1813         switch (msr) {
1814         case HV_X64_MSR_VP_INDEX: {
1815                 int r;
1816                 struct kvm_vcpu *v;
1817                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1818                         if (v == vcpu)
1819                                 data = r;
1820                 break;
1821         }
1822         case HV_X64_MSR_EOI:
1823                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1824         case HV_X64_MSR_ICR:
1825                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1826         case HV_X64_MSR_TPR:
1827                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1828         default:
1829                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1830                 return 1;
1831         }
1832         *pdata = data;
1833         return 0;
1834 }
1835
1836 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1837 {
1838         u64 data;
1839
1840         switch (msr) {
1841         case MSR_IA32_PLATFORM_ID:
1842         case MSR_IA32_UCODE_REV:
1843         case MSR_IA32_EBL_CR_POWERON:
1844         case MSR_IA32_DEBUGCTLMSR:
1845         case MSR_IA32_LASTBRANCHFROMIP:
1846         case MSR_IA32_LASTBRANCHTOIP:
1847         case MSR_IA32_LASTINTFROMIP:
1848         case MSR_IA32_LASTINTTOIP:
1849         case MSR_K8_SYSCFG:
1850         case MSR_K7_HWCR:
1851         case MSR_VM_HSAVE_PA:
1852         case MSR_P6_PERFCTR0:
1853         case MSR_P6_PERFCTR1:
1854         case MSR_P6_EVNTSEL0:
1855         case MSR_P6_EVNTSEL1:
1856         case MSR_K7_EVNTSEL0:
1857         case MSR_K7_PERFCTR0:
1858         case MSR_K8_INT_PENDING_MSG:
1859         case MSR_AMD64_NB_CFG:
1860         case MSR_FAM10H_MMIO_CONF_BASE:
1861                 data = 0;
1862                 break;
1863         case MSR_MTRRcap:
1864                 data = 0x500 | KVM_NR_VAR_MTRR;
1865                 break;
1866         case 0x200 ... 0x2ff:
1867                 return get_msr_mtrr(vcpu, msr, pdata);
1868         case 0xcd: /* fsb frequency */
1869                 data = 3;
1870                 break;
1871                 /*
1872                  * MSR_EBC_FREQUENCY_ID
1873                  * Conservative value valid for even the basic CPU models.
1874                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1875                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1876                  * and 266MHz for model 3, or 4. Set Core Clock
1877                  * Frequency to System Bus Frequency Ratio to 1 (bits
1878                  * 31:24) even though these are only valid for CPU
1879                  * models > 2, however guests may end up dividing or
1880                  * multiplying by zero otherwise.
1881                  */
1882         case MSR_EBC_FREQUENCY_ID:
1883                 data = 1 << 24;
1884                 break;
1885         case MSR_IA32_APICBASE:
1886                 data = kvm_get_apic_base(vcpu);
1887                 break;
1888         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1889                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1890                 break;
1891         case MSR_IA32_MISC_ENABLE:
1892                 data = vcpu->arch.ia32_misc_enable_msr;
1893                 break;
1894         case MSR_IA32_PERF_STATUS:
1895                 /* TSC increment by tick */
1896                 data = 1000ULL;
1897                 /* CPU multiplier */
1898                 data |= (((uint64_t)4ULL) << 40);
1899                 break;
1900         case MSR_EFER:
1901                 data = vcpu->arch.efer;
1902                 break;
1903         case MSR_KVM_WALL_CLOCK:
1904         case MSR_KVM_WALL_CLOCK_NEW:
1905                 data = vcpu->kvm->arch.wall_clock;
1906                 break;
1907         case MSR_KVM_SYSTEM_TIME:
1908         case MSR_KVM_SYSTEM_TIME_NEW:
1909                 data = vcpu->arch.time;
1910                 break;
1911         case MSR_KVM_ASYNC_PF_EN:
1912                 data = vcpu->arch.apf.msr_val;
1913                 break;
1914         case MSR_KVM_STEAL_TIME:
1915                 data = vcpu->arch.st.msr_val;
1916                 break;
1917         case MSR_IA32_P5_MC_ADDR:
1918         case MSR_IA32_P5_MC_TYPE:
1919         case MSR_IA32_MCG_CAP:
1920         case MSR_IA32_MCG_CTL:
1921         case MSR_IA32_MCG_STATUS:
1922         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1923                 return get_msr_mce(vcpu, msr, pdata);
1924         case MSR_K7_CLK_CTL:
1925                 /*
1926                  * Provide expected ramp-up count for K7. All other
1927                  * are set to zero, indicating minimum divisors for
1928                  * every field.
1929                  *
1930                  * This prevents guest kernels on AMD host with CPU
1931                  * type 6, model 8 and higher from exploding due to
1932                  * the rdmsr failing.
1933                  */
1934                 data = 0x20000000;
1935                 break;
1936         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1937                 if (kvm_hv_msr_partition_wide(msr)) {
1938                         int r;
1939                         mutex_lock(&vcpu->kvm->lock);
1940                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1941                         mutex_unlock(&vcpu->kvm->lock);
1942                         return r;
1943                 } else
1944                         return get_msr_hyperv(vcpu, msr, pdata);
1945                 break;
1946         case MSR_IA32_BBL_CR_CTL3:
1947                 /* This legacy MSR exists but isn't fully documented in current
1948                  * silicon.  It is however accessed by winxp in very narrow
1949                  * scenarios where it sets bit #19, itself documented as
1950                  * a "reserved" bit.  Best effort attempt to source coherent
1951                  * read data here should the balance of the register be
1952                  * interpreted by the guest:
1953                  *
1954                  * L2 cache control register 3: 64GB range, 256KB size,
1955                  * enabled, latency 0x1, configured
1956                  */
1957                 data = 0xbe702111;
1958                 break;
1959         default:
1960                 if (!ignore_msrs) {
1961                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1962                         return 1;
1963                 } else {
1964                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1965                         data = 0;
1966                 }
1967                 break;
1968         }
1969         *pdata = data;
1970         return 0;
1971 }
1972 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1973
1974 /*
1975  * Read or write a bunch of msrs. All parameters are kernel addresses.
1976  *
1977  * @return number of msrs set successfully.
1978  */
1979 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1980                     struct kvm_msr_entry *entries,
1981                     int (*do_msr)(struct kvm_vcpu *vcpu,
1982                                   unsigned index, u64 *data))
1983 {
1984         int i, idx;
1985
1986         idx = srcu_read_lock(&vcpu->kvm->srcu);
1987         for (i = 0; i < msrs->nmsrs; ++i)
1988                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1989                         break;
1990         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1991
1992         return i;
1993 }
1994
1995 /*
1996  * Read or write a bunch of msrs. Parameters are user addresses.
1997  *
1998  * @return number of msrs set successfully.
1999  */
2000 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2001                   int (*do_msr)(struct kvm_vcpu *vcpu,
2002                                 unsigned index, u64 *data),
2003                   int writeback)
2004 {
2005         struct kvm_msrs msrs;
2006         struct kvm_msr_entry *entries;
2007         int r, n;
2008         unsigned size;
2009
2010         r = -EFAULT;
2011         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2012                 goto out;
2013
2014         r = -E2BIG;
2015         if (msrs.nmsrs >= MAX_IO_MSRS)
2016                 goto out;
2017
2018         r = -ENOMEM;
2019         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2020         entries = kmalloc(size, GFP_KERNEL);
2021         if (!entries)
2022                 goto out;
2023
2024         r = -EFAULT;
2025         if (copy_from_user(entries, user_msrs->entries, size))
2026                 goto out_free;
2027
2028         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2029         if (r < 0)
2030                 goto out_free;
2031
2032         r = -EFAULT;
2033         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2034                 goto out_free;
2035
2036         r = n;
2037
2038 out_free:
2039         kfree(entries);
2040 out:
2041         return r;
2042 }
2043
2044 int kvm_dev_ioctl_check_extension(long ext)
2045 {
2046         int r;
2047
2048         switch (ext) {
2049         case KVM_CAP_IRQCHIP:
2050         case KVM_CAP_HLT:
2051         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2052         case KVM_CAP_SET_TSS_ADDR:
2053         case KVM_CAP_EXT_CPUID:
2054         case KVM_CAP_CLOCKSOURCE:
2055         case KVM_CAP_PIT:
2056         case KVM_CAP_NOP_IO_DELAY:
2057         case KVM_CAP_MP_STATE:
2058         case KVM_CAP_SYNC_MMU:
2059         case KVM_CAP_USER_NMI:
2060         case KVM_CAP_REINJECT_CONTROL:
2061         case KVM_CAP_IRQ_INJECT_STATUS:
2062         case KVM_CAP_ASSIGN_DEV_IRQ:
2063         case KVM_CAP_IRQFD:
2064         case KVM_CAP_IOEVENTFD:
2065         case KVM_CAP_PIT2:
2066         case KVM_CAP_PIT_STATE2:
2067         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2068         case KVM_CAP_XEN_HVM:
2069         case KVM_CAP_ADJUST_CLOCK:
2070         case KVM_CAP_VCPU_EVENTS:
2071         case KVM_CAP_HYPERV:
2072         case KVM_CAP_HYPERV_VAPIC:
2073         case KVM_CAP_HYPERV_SPIN:
2074         case KVM_CAP_PCI_SEGMENT:
2075         case KVM_CAP_DEBUGREGS:
2076         case KVM_CAP_X86_ROBUST_SINGLESTEP:
2077         case KVM_CAP_XSAVE:
2078         case KVM_CAP_ASYNC_PF:
2079         case KVM_CAP_GET_TSC_KHZ:
2080                 r = 1;
2081                 break;
2082         case KVM_CAP_COALESCED_MMIO:
2083                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2084                 break;
2085         case KVM_CAP_VAPIC:
2086                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2087                 break;
2088         case KVM_CAP_NR_VCPUS:
2089                 r = KVM_SOFT_MAX_VCPUS;
2090                 break;
2091         case KVM_CAP_MAX_VCPUS:
2092                 r = KVM_MAX_VCPUS;
2093                 break;
2094         case KVM_CAP_NR_MEMSLOTS:
2095                 r = KVM_MEMORY_SLOTS;
2096                 break;
2097         case KVM_CAP_PV_MMU:    /* obsolete */
2098                 r = 0;
2099                 break;
2100         case KVM_CAP_IOMMU:
2101                 r = iommu_found();
2102                 break;
2103         case KVM_CAP_MCE:
2104                 r = KVM_MAX_MCE_BANKS;
2105                 break;
2106         case KVM_CAP_XCRS:
2107                 r = cpu_has_xsave;
2108                 break;
2109         case KVM_CAP_TSC_CONTROL:
2110                 r = kvm_has_tsc_control;
2111                 break;
2112         default:
2113                 r = 0;
2114                 break;
2115         }
2116         return r;
2117
2118 }
2119
2120 long kvm_arch_dev_ioctl(struct file *filp,
2121                         unsigned int ioctl, unsigned long arg)
2122 {
2123         void __user *argp = (void __user *)arg;
2124         long r;
2125
2126         switch (ioctl) {
2127         case KVM_GET_MSR_INDEX_LIST: {
2128                 struct kvm_msr_list __user *user_msr_list = argp;
2129                 struct kvm_msr_list msr_list;
2130                 unsigned n;
2131
2132                 r = -EFAULT;
2133                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2134                         goto out;
2135                 n = msr_list.nmsrs;
2136                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2137                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2138                         goto out;
2139                 r = -E2BIG;
2140                 if (n < msr_list.nmsrs)
2141                         goto out;
2142                 r = -EFAULT;
2143                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2144                                  num_msrs_to_save * sizeof(u32)))
2145                         goto out;
2146                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2147                                  &emulated_msrs,
2148                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2149                         goto out;
2150                 r = 0;
2151                 break;
2152         }
2153         case KVM_GET_SUPPORTED_CPUID: {
2154                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2155                 struct kvm_cpuid2 cpuid;
2156
2157                 r = -EFAULT;
2158                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2159                         goto out;
2160                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2161                                                       cpuid_arg->entries);
2162                 if (r)
2163                         goto out;
2164
2165                 r = -EFAULT;
2166                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2167                         goto out;
2168                 r = 0;
2169                 break;
2170         }
2171         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2172                 u64 mce_cap;
2173
2174                 mce_cap = KVM_MCE_CAP_SUPPORTED;
2175                 r = -EFAULT;
2176                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2177                         goto out;
2178                 r = 0;
2179                 break;
2180         }
2181         default:
2182                 r = -EINVAL;
2183         }
2184 out:
2185         return r;
2186 }
2187
2188 static void wbinvd_ipi(void *garbage)
2189 {
2190         wbinvd();
2191 }
2192
2193 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2194 {
2195         return vcpu->kvm->arch.iommu_domain &&
2196                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2197 }
2198
2199 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2200 {
2201         /* Address WBINVD may be executed by guest */
2202         if (need_emulate_wbinvd(vcpu)) {
2203                 if (kvm_x86_ops->has_wbinvd_exit())
2204                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2205                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2206                         smp_call_function_single(vcpu->cpu,
2207                                         wbinvd_ipi, NULL, 1);
2208         }
2209
2210         kvm_x86_ops->vcpu_load(vcpu, cpu);
2211         if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2212                 /* Make sure TSC doesn't go backwards */
2213                 s64 tsc_delta;
2214                 u64 tsc;
2215
2216                 kvm_get_msr(vcpu, MSR_IA32_TSC, &tsc);
2217                 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2218                              tsc - vcpu->arch.last_guest_tsc;
2219
2220                 if (tsc_delta < 0)
2221                         mark_tsc_unstable("KVM discovered backwards TSC");
2222                 if (check_tsc_unstable()) {
2223                         kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2224                         vcpu->arch.tsc_catchup = 1;
2225                 }
2226                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2227                 if (vcpu->cpu != cpu)
2228                         kvm_migrate_timers(vcpu);
2229                 vcpu->cpu = cpu;
2230         }
2231
2232         accumulate_steal_time(vcpu);
2233         kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2234 }
2235
2236 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2237 {
2238         kvm_x86_ops->vcpu_put(vcpu);
2239         kvm_put_guest_fpu(vcpu);
2240         kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
2241 }
2242
2243 static int is_efer_nx(void)
2244 {
2245         unsigned long long efer = 0;
2246
2247         rdmsrl_safe(MSR_EFER, &efer);
2248         return efer & EFER_NX;
2249 }
2250
2251 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2252 {
2253         int i;
2254         struct kvm_cpuid_entry2 *e, *entry;
2255
2256         entry = NULL;
2257         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2258                 e = &vcpu->arch.cpuid_entries[i];
2259                 if (e->function == 0x80000001) {
2260                         entry = e;
2261                         break;
2262                 }
2263         }
2264         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2265                 entry->edx &= ~(1 << 20);
2266                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2267         }
2268 }
2269
2270 /* when an old userspace process fills a new kernel module */
2271 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2272                                     struct kvm_cpuid *cpuid,
2273                                     struct kvm_cpuid_entry __user *entries)
2274 {
2275         int r, i;
2276         struct kvm_cpuid_entry *cpuid_entries;
2277
2278         r = -E2BIG;
2279         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2280                 goto out;
2281         r = -ENOMEM;
2282         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2283         if (!cpuid_entries)
2284                 goto out;
2285         r = -EFAULT;
2286         if (copy_from_user(cpuid_entries, entries,
2287                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2288                 goto out_free;
2289         for (i = 0; i < cpuid->nent; i++) {
2290                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2291                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2292                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2293                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2294                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2295                 vcpu->arch.cpuid_entries[i].index = 0;
2296                 vcpu->arch.cpuid_entries[i].flags = 0;
2297                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2298                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2299                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2300         }
2301         vcpu->arch.cpuid_nent = cpuid->nent;
2302         cpuid_fix_nx_cap(vcpu);
2303         r = 0;
2304         kvm_apic_set_version(vcpu);
2305         kvm_x86_ops->cpuid_update(vcpu);
2306         update_cpuid(vcpu);
2307
2308 out_free:
2309         vfree(cpuid_entries);
2310 out:
2311         return r;
2312 }
2313
2314 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2315                                      struct kvm_cpuid2 *cpuid,
2316                                      struct kvm_cpuid_entry2 __user *entries)
2317 {
2318         int r;
2319
2320         r = -E2BIG;
2321         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2322                 goto out;
2323         r = -EFAULT;
2324         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2325                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2326                 goto out;
2327         vcpu->arch.cpuid_nent = cpuid->nent;
2328         kvm_apic_set_version(vcpu);
2329         kvm_x86_ops->cpuid_update(vcpu);
2330         update_cpuid(vcpu);
2331         return 0;
2332
2333 out:
2334         return r;
2335 }
2336
2337 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2338                                      struct kvm_cpuid2 *cpuid,
2339                                      struct kvm_cpuid_entry2 __user *entries)
2340 {
2341         int r;
2342
2343         r = -E2BIG;
2344         if (cpuid->nent < vcpu->arch.cpuid_nent)
2345                 goto out;
2346         r = -EFAULT;
2347         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2348                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2349                 goto out;
2350         return 0;
2351
2352 out:
2353         cpuid->nent = vcpu->arch.cpuid_nent;
2354         return r;
2355 }
2356
2357 static void cpuid_mask(u32 *word, int wordnum)
2358 {
2359         *word &= boot_cpu_data.x86_capability[wordnum];
2360 }
2361
2362 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2363                            u32 index)
2364 {
2365         entry->function = function;
2366         entry->index = index;
2367         cpuid_count(entry->function, entry->index,
2368                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2369         entry->flags = 0;
2370 }
2371
2372 static bool supported_xcr0_bit(unsigned bit)
2373 {
2374         u64 mask = ((u64)1 << bit);
2375
2376         return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0;
2377 }
2378
2379 #define F(x) bit(X86_FEATURE_##x)
2380
2381 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2382                          u32 index, int *nent, int maxnent)
2383 {
2384         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2385 #ifdef CONFIG_X86_64
2386         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2387                                 ? F(GBPAGES) : 0;
2388         unsigned f_lm = F(LM);
2389 #else
2390         unsigned f_gbpages = 0;
2391         unsigned f_lm = 0;
2392 #endif
2393         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2394
2395         /* cpuid 1.edx */
2396         const u32 kvm_supported_word0_x86_features =
2397                 F(FPU) | F(VME) | F(DE) | F(PSE) |
2398                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2399                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2400                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2401                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2402                 0 /* Reserved, DS, ACPI */ | F(MMX) |
2403                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2404                 0 /* HTT, TM, Reserved, PBE */;
2405         /* cpuid 0x80000001.edx */
2406         const u32 kvm_supported_word1_x86_features =
2407                 F(FPU) | F(VME) | F(DE) | F(PSE) |
2408                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2409                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2410                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2411                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2412                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2413                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2414                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2415         /* cpuid 1.ecx */
2416         const u32 kvm_supported_word4_x86_features =
2417                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2418                 0 /* DS-CPL, VMX, SMX, EST */ |
2419                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2420                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2421                 0 /* Reserved, DCA */ | F(XMM4_1) |
2422                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2423                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2424                 F(F16C) | F(RDRAND);
2425         /* cpuid 0x80000001.ecx */
2426         const u32 kvm_supported_word6_x86_features =
2427                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2428                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2429                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2430                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2431
2432         /* cpuid 0xC0000001.edx */
2433         const u32 kvm_supported_word5_x86_features =
2434                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
2435                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
2436                 F(PMM) | F(PMM_EN);
2437
2438         /* cpuid 7.0.ebx */
2439         const u32 kvm_supported_word9_x86_features =
2440                 F(SMEP) | F(FSGSBASE) | F(ERMS);
2441
2442         /* all calls to cpuid_count() should be made on the same cpu */
2443         get_cpu();
2444         do_cpuid_1_ent(entry, function, index);
2445         ++*nent;
2446
2447         switch (function) {
2448         case 0:
2449                 entry->eax = min(entry->eax, (u32)0xd);
2450                 break;
2451         case 1:
2452                 entry->edx &= kvm_supported_word0_x86_features;
2453                 cpuid_mask(&entry->edx, 0);
2454                 entry->ecx &= kvm_supported_word4_x86_features;
2455                 cpuid_mask(&entry->ecx, 4);
2456                 /* we support x2apic emulation even if host does not support
2457                  * it since we emulate x2apic in software */
2458                 entry->ecx |= F(X2APIC);
2459                 break;
2460         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2461          * may return different values. This forces us to get_cpu() before
2462          * issuing the first command, and also to emulate this annoying behavior
2463          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2464         case 2: {
2465                 int t, times = entry->eax & 0xff;
2466
2467                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2468                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2469                 for (t = 1; t < times && *nent < maxnent; ++t) {
2470                         do_cpuid_1_ent(&entry[t], function, 0);
2471                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2472                         ++*nent;
2473                 }
2474                 break;
2475         }
2476         /* function 4 has additional index. */
2477         case 4: {
2478                 int i, cache_type;
2479
2480                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2481                 /* read more entries until cache_type is zero */
2482                 for (i = 1; *nent < maxnent; ++i) {
2483                         cache_type = entry[i - 1].eax & 0x1f;
2484                         if (!cache_type)
2485                                 break;
2486                         do_cpuid_1_ent(&entry[i], function, i);
2487                         entry[i].flags |=
2488                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2489                         ++*nent;
2490                 }
2491                 break;
2492         }
2493         case 7: {
2494                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2495                 /* Mask ebx against host capbability word 9 */
2496                 if (index == 0) {
2497                         entry->ebx &= kvm_supported_word9_x86_features;
2498                         cpuid_mask(&entry->ebx, 9);
2499                 } else
2500                         entry->ebx = 0;
2501                 entry->eax = 0;
2502                 entry->ecx = 0;
2503                 entry->edx = 0;
2504                 break;
2505         }
2506         case 9:
2507                 break;
2508         /* function 0xb has additional index. */
2509         case 0xb: {
2510                 int i, level_type;
2511
2512                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2513                 /* read more entries until level_type is zero */
2514                 for (i = 1; *nent < maxnent; ++i) {
2515                         level_type = entry[i - 1].ecx & 0xff00;
2516                         if (!level_type)
2517                                 break;
2518                         do_cpuid_1_ent(&entry[i], function, i);
2519                         entry[i].flags |=
2520                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2521                         ++*nent;
2522                 }
2523                 break;
2524         }
2525         case 0xd: {
2526                 int idx, i;
2527
2528                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2529                 for (idx = 1, i = 1; *nent < maxnent && idx < 64; ++idx) {
2530                         do_cpuid_1_ent(&entry[i], function, idx);
2531                         if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
2532                                 continue;
2533                         entry[i].flags |=
2534                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2535                         ++*nent;
2536                         ++i;
2537                 }
2538                 break;
2539         }
2540         case KVM_CPUID_SIGNATURE: {
2541                 char signature[12] = "KVMKVMKVM\0\0";
2542                 u32 *sigptr = (u32 *)signature;
2543                 entry->eax = 0;
2544                 entry->ebx = sigptr[0];
2545                 entry->ecx = sigptr[1];
2546                 entry->edx = sigptr[2];
2547                 break;
2548         }
2549         case KVM_CPUID_FEATURES:
2550                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2551                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
2552                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
2553                              (1 << KVM_FEATURE_ASYNC_PF) |
2554                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2555
2556                 if (sched_info_on())
2557                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
2558
2559                 entry->ebx = 0;
2560                 entry->ecx = 0;
2561                 entry->edx = 0;
2562                 break;
2563         case 0x80000000:
2564                 entry->eax = min(entry->eax, 0x8000001a);
2565                 break;
2566         case 0x80000001:
2567                 entry->edx &= kvm_supported_word1_x86_features;
2568                 cpuid_mask(&entry->edx, 1);
2569                 entry->ecx &= kvm_supported_word6_x86_features;
2570                 cpuid_mask(&entry->ecx, 6);
2571                 break;
2572         case 0x80000008: {
2573                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
2574                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
2575                 unsigned phys_as = entry->eax & 0xff;
2576
2577                 if (!g_phys_as)
2578                         g_phys_as = phys_as;
2579                 entry->eax = g_phys_as | (virt_as << 8);
2580                 entry->ebx = entry->edx = 0;
2581                 break;
2582         }
2583         case 0x80000019:
2584                 entry->ecx = entry->edx = 0;
2585                 break;
2586         case 0x8000001a:
2587                 break;
2588         case 0x8000001d:
2589                 break;
2590         /*Add support for Centaur's CPUID instruction*/
2591         case 0xC0000000:
2592                 /*Just support up to 0xC0000004 now*/
2593                 entry->eax = min(entry->eax, 0xC0000004);
2594                 break;
2595         case 0xC0000001:
2596                 entry->edx &= kvm_supported_word5_x86_features;
2597                 cpuid_mask(&entry->edx, 5);
2598                 break;
2599         case 3: /* Processor serial number */
2600         case 5: /* MONITOR/MWAIT */
2601         case 6: /* Thermal management */
2602         case 0xA: /* Architectural Performance Monitoring */
2603         case 0x80000007: /* Advanced power management */
2604         case 0xC0000002:
2605         case 0xC0000003:
2606         case 0xC0000004:
2607         default:
2608                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
2609                 break;
2610         }
2611
2612         kvm_x86_ops->set_supported_cpuid(function, entry);
2613
2614         put_cpu();
2615 }
2616
2617 #undef F
2618
2619 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2620                                      struct kvm_cpuid_entry2 __user *entries)
2621 {
2622         struct kvm_cpuid_entry2 *cpuid_entries;
2623         int limit, nent = 0, r = -E2BIG;
2624         u32 func;
2625
2626         if (cpuid->nent < 1)
2627                 goto out;
2628         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2629                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2630         r = -ENOMEM;
2631         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2632         if (!cpuid_entries)
2633                 goto out;
2634
2635         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2636         limit = cpuid_entries[0].eax;
2637         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2638                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2639                              &nent, cpuid->nent);
2640         r = -E2BIG;
2641         if (nent >= cpuid->nent)
2642                 goto out_free;
2643
2644         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2645         limit = cpuid_entries[nent - 1].eax;
2646         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2647                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2648                              &nent, cpuid->nent);
2649
2650
2651
2652         r = -E2BIG;
2653         if (nent >= cpuid->nent)
2654                 goto out_free;
2655
2656         /* Add support for Centaur's CPUID instruction. */
2657         if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) {
2658                 do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0,
2659                                 &nent, cpuid->nent);
2660
2661                 r = -E2BIG;
2662                 if (nent >= cpuid->nent)
2663                         goto out_free;
2664
2665                 limit = cpuid_entries[nent - 1].eax;
2666                 for (func = 0xC0000001;
2667                         func <= limit && nent < cpuid->nent; ++func)
2668                         do_cpuid_ent(&cpuid_entries[nent], func, 0,
2669                                         &nent, cpuid->nent);
2670
2671                 r = -E2BIG;
2672                 if (nent >= cpuid->nent)
2673                         goto out_free;
2674         }
2675
2676         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2677                      cpuid->nent);
2678
2679         r = -E2BIG;
2680         if (nent >= cpuid->nent)
2681                 goto out_free;
2682
2683         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2684                      cpuid->nent);
2685
2686         r = -E2BIG;
2687         if (nent >= cpuid->nent)
2688                 goto out_free;
2689
2690         r = -EFAULT;
2691         if (copy_to_user(entries, cpuid_entries,
2692                          nent * sizeof(struct kvm_cpuid_entry2)))
2693                 goto out_free;
2694         cpuid->nent = nent;
2695         r = 0;
2696
2697 out_free:
2698         vfree(cpuid_entries);
2699 out:
2700         return r;
2701 }
2702
2703 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2704                                     struct kvm_lapic_state *s)
2705 {
2706         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2707
2708         return 0;
2709 }
2710
2711 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2712                                     struct kvm_lapic_state *s)
2713 {
2714         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2715         kvm_apic_post_state_restore(vcpu);
2716         update_cr8_intercept(vcpu);
2717
2718         return 0;
2719 }
2720
2721 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2722                                     struct kvm_interrupt *irq)
2723 {
2724         if (irq->irq < 0 || irq->irq >= 256)
2725                 return -EINVAL;
2726         if (irqchip_in_kernel(vcpu->kvm))
2727                 return -ENXIO;
2728
2729         kvm_queue_interrupt(vcpu, irq->irq, false);
2730         kvm_make_request(KVM_REQ_EVENT, vcpu);
2731
2732         return 0;
2733 }
2734
2735 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2736 {
2737         kvm_inject_nmi(vcpu);
2738
2739         return 0;
2740 }
2741
2742 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2743                                            struct kvm_tpr_access_ctl *tac)
2744 {
2745         if (tac->flags)
2746                 return -EINVAL;
2747         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2748         return 0;
2749 }
2750
2751 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2752                                         u64 mcg_cap)
2753 {
2754         int r;
2755         unsigned bank_num = mcg_cap & 0xff, bank;
2756
2757         r = -EINVAL;
2758         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2759                 goto out;
2760         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2761                 goto out;
2762         r = 0;
2763         vcpu->arch.mcg_cap = mcg_cap;
2764         /* Init IA32_MCG_CTL to all 1s */
2765         if (mcg_cap & MCG_CTL_P)
2766                 vcpu->arch.mcg_ctl = ~(u64)0;
2767         /* Init IA32_MCi_CTL to all 1s */
2768         for (bank = 0; bank < bank_num; bank++)
2769                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2770 out:
2771         return r;
2772 }
2773
2774 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2775                                       struct kvm_x86_mce *mce)
2776 {
2777         u64 mcg_cap = vcpu->arch.mcg_cap;
2778         unsigned bank_num = mcg_cap & 0xff;
2779         u64 *banks = vcpu->arch.mce_banks;
2780
2781         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2782                 return -EINVAL;
2783         /*
2784          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2785          * reporting is disabled
2786          */
2787         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2788             vcpu->arch.mcg_ctl != ~(u64)0)
2789                 return 0;
2790         banks += 4 * mce->bank;
2791         /*
2792          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2793          * reporting is disabled for the bank
2794          */
2795         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2796                 return 0;
2797         if (mce->status & MCI_STATUS_UC) {
2798                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2799                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2800                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2801                         return 0;
2802                 }
2803                 if (banks[1] & MCI_STATUS_VAL)
2804                         mce->status |= MCI_STATUS_OVER;
2805                 banks[2] = mce->addr;
2806                 banks[3] = mce->misc;
2807                 vcpu->arch.mcg_status = mce->mcg_status;
2808                 banks[1] = mce->status;
2809                 kvm_queue_exception(vcpu, MC_VECTOR);
2810         } else if (!(banks[1] & MCI_STATUS_VAL)
2811                    || !(banks[1] & MCI_STATUS_UC)) {
2812                 if (banks[1] & MCI_STATUS_VAL)
2813                         mce->status |= MCI_STATUS_OVER;
2814                 banks[2] = mce->addr;
2815                 banks[3] = mce->misc;
2816                 banks[1] = mce->status;
2817         } else
2818                 banks[1] |= MCI_STATUS_OVER;
2819         return 0;
2820 }
2821
2822 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2823                                                struct kvm_vcpu_events *events)
2824 {
2825         events->exception.injected =
2826                 vcpu->arch.exception.pending &&
2827                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2828         events->exception.nr = vcpu->arch.exception.nr;
2829         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2830         events->exception.pad = 0;
2831         events->exception.error_code = vcpu->arch.exception.error_code;
2832
2833         events->interrupt.injected =
2834                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2835         events->interrupt.nr = vcpu->arch.interrupt.nr;
2836         events->interrupt.soft = 0;
2837         events->interrupt.shadow =
2838                 kvm_x86_ops->get_interrupt_shadow(vcpu,
2839                         KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2840
2841         events->nmi.injected = vcpu->arch.nmi_injected;
2842         events->nmi.pending = vcpu->arch.nmi_pending;
2843         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2844         events->nmi.pad = 0;
2845
2846         events->sipi_vector = vcpu->arch.sipi_vector;
2847
2848         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2849                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2850                          | KVM_VCPUEVENT_VALID_SHADOW);
2851         memset(&events->reserved, 0, sizeof(events->reserved));
2852 }
2853
2854 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2855                                               struct kvm_vcpu_events *events)
2856 {
2857         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2858                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2859                               | KVM_VCPUEVENT_VALID_SHADOW))
2860                 return -EINVAL;
2861
2862         vcpu->arch.exception.pending = events->exception.injected;
2863         vcpu->arch.exception.nr = events->exception.nr;
2864         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2865         vcpu->arch.exception.error_code = events->exception.error_code;
2866
2867         vcpu->arch.interrupt.pending = events->interrupt.injected;
2868         vcpu->arch.interrupt.nr = events->interrupt.nr;
2869         vcpu->arch.interrupt.soft = events->interrupt.soft;
2870         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2871                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2872                                                   events->interrupt.shadow);
2873
2874         vcpu->arch.nmi_injected = events->nmi.injected;
2875         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2876                 vcpu->arch.nmi_pending = events->nmi.pending;
2877         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2878
2879         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2880                 vcpu->arch.sipi_vector = events->sipi_vector;
2881
2882         kvm_make_request(KVM_REQ_EVENT, vcpu);
2883
2884         return 0;
2885 }
2886
2887 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2888                                              struct kvm_debugregs *dbgregs)
2889 {
2890         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2891         dbgregs->dr6 = vcpu->arch.dr6;
2892         dbgregs->dr7 = vcpu->arch.dr7;
2893         dbgregs->flags = 0;
2894         memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2895 }
2896
2897 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2898                                             struct kvm_debugregs *dbgregs)
2899 {
2900         if (dbgregs->flags)
2901                 return -EINVAL;
2902
2903         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2904         vcpu->arch.dr6 = dbgregs->dr6;
2905         vcpu->arch.dr7 = dbgregs->dr7;
2906
2907         return 0;
2908 }
2909
2910 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2911                                          struct kvm_xsave *guest_xsave)
2912 {
2913         if (cpu_has_xsave)
2914                 memcpy(guest_xsave->region,
2915                         &vcpu->arch.guest_fpu.state->xsave,
2916                         xstate_size);
2917         else {
2918                 memcpy(guest_xsave->region,
2919                         &vcpu->arch.guest_fpu.state->fxsave,
2920                         sizeof(struct i387_fxsave_struct));
2921                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2922                         XSTATE_FPSSE;
2923         }
2924 }
2925
2926 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2927                                         struct kvm_xsave *guest_xsave)
2928 {
2929         u64 xstate_bv =
2930                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2931
2932         if (cpu_has_xsave)
2933                 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2934                         guest_xsave->region, xstate_size);
2935         else {
2936                 if (xstate_bv & ~XSTATE_FPSSE)
2937                         return -EINVAL;
2938                 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2939                         guest_xsave->region, sizeof(struct i387_fxsave_struct));
2940         }
2941         return 0;
2942 }
2943
2944 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2945                                         struct kvm_xcrs *guest_xcrs)
2946 {
2947         if (!cpu_has_xsave) {
2948                 guest_xcrs->nr_xcrs = 0;
2949                 return;
2950         }
2951
2952         guest_xcrs->nr_xcrs = 1;
2953         guest_xcrs->flags = 0;
2954         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2955         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2956 }
2957
2958 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2959                                        struct kvm_xcrs *guest_xcrs)
2960 {
2961         int i, r = 0;
2962
2963         if (!cpu_has_xsave)
2964                 return -EINVAL;
2965
2966         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2967                 return -EINVAL;
2968
2969         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2970                 /* Only support XCR0 currently */
2971                 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2972                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2973                                 guest_xcrs->xcrs[0].value);
2974                         break;
2975                 }
2976         if (r)
2977                 r = -EINVAL;
2978         return r;
2979 }
2980
2981 long kvm_arch_vcpu_ioctl(struct file *filp,
2982                          unsigned int ioctl, unsigned long arg)
2983 {
2984         struct kvm_vcpu *vcpu = filp->private_data;
2985         void __user *argp = (void __user *)arg;
2986         int r;
2987         union {
2988                 struct kvm_lapic_state *lapic;
2989                 struct kvm_xsave *xsave;
2990                 struct kvm_xcrs *xcrs;
2991                 void *buffer;
2992         } u;
2993
2994         u.buffer = NULL;
2995         switch (ioctl) {
2996         case KVM_GET_LAPIC: {
2997                 r = -EINVAL;
2998                 if (!vcpu->arch.apic)
2999                         goto out;
3000                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3001
3002                 r = -ENOMEM;
3003                 if (!u.lapic)
3004                         goto out;
3005                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3006                 if (r)
3007                         goto out;
3008                 r = -EFAULT;
3009                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3010                         goto out;
3011                 r = 0;
3012                 break;
3013         }
3014         case KVM_SET_LAPIC: {
3015                 r = -EINVAL;
3016                 if (!vcpu->arch.apic)
3017                         goto out;
3018                 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3019                 r = -ENOMEM;
3020                 if (!u.lapic)
3021                         goto out;
3022                 r = -EFAULT;
3023                 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
3024                         goto out;
3025                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3026                 if (r)
3027                         goto out;
3028                 r = 0;
3029                 break;
3030         }
3031         case KVM_INTERRUPT: {
3032                 struct kvm_interrupt irq;
3033
3034                 r = -EFAULT;
3035                 if (copy_from_user(&irq, argp, sizeof irq))
3036                         goto out;
3037                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3038                 if (r)
3039                         goto out;
3040                 r = 0;
3041                 break;
3042         }
3043         case KVM_NMI: {
3044                 r = kvm_vcpu_ioctl_nmi(vcpu);
3045                 if (r)
3046                         goto out;
3047                 r = 0;
3048                 break;
3049         }
3050         case KVM_SET_CPUID: {
3051                 struct kvm_cpuid __user *cpuid_arg = argp;
3052                 struct kvm_cpuid cpuid;
3053
3054                 r = -EFAULT;
3055                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3056                         goto out;
3057                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3058                 if (r)
3059                         goto out;
3060                 break;
3061         }
3062         case KVM_SET_CPUID2: {
3063                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3064                 struct kvm_cpuid2 cpuid;
3065
3066                 r = -EFAULT;
3067                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3068                         goto out;
3069                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3070                                               cpuid_arg->entries);
3071                 if (r)
3072                         goto out;
3073                 break;
3074         }
3075         case KVM_GET_CPUID2: {
3076                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3077                 struct kvm_cpuid2 cpuid;
3078
3079                 r = -EFAULT;
3080                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3081                         goto out;
3082                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3083                                               cpuid_arg->entries);
3084                 if (r)
3085                         goto out;
3086                 r = -EFAULT;
3087                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3088                         goto out;
3089                 r = 0;
3090                 break;
3091         }
3092         case KVM_GET_MSRS:
3093                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3094                 break;
3095         case KVM_SET_MSRS:
3096                 r = msr_io(vcpu, argp, do_set_msr, 0);
3097                 break;
3098         case KVM_TPR_ACCESS_REPORTING: {
3099                 struct kvm_tpr_access_ctl tac;
3100
3101                 r = -EFAULT;
3102                 if (copy_from_user(&tac, argp, sizeof tac))
3103                         goto out;
3104                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3105                 if (r)
3106                         goto out;
3107                 r = -EFAULT;
3108                 if (copy_to_user(argp, &tac, sizeof tac))
3109                         goto out;
3110                 r = 0;
3111                 break;
3112         };
3113         case KVM_SET_VAPIC_ADDR: {
3114                 struct kvm_vapic_addr va;
3115
3116                 r = -EINVAL;
3117                 if (!irqchip_in_kernel(vcpu->kvm))
3118                         goto out;
3119                 r = -EFAULT;
3120                 if (copy_from_user(&va, argp, sizeof va))
3121                         goto out;
3122                 r = 0;
3123                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3124                 break;
3125         }
3126         case KVM_X86_SETUP_MCE: {
3127                 u64 mcg_cap;
3128
3129                 r = -EFAULT;
3130                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3131                         goto out;
3132                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3133                 break;
3134         }
3135         case KVM_X86_SET_MCE: {
3136                 struct kvm_x86_mce mce;
3137
3138                 r = -EFAULT;
3139                 if (copy_from_user(&mce, argp, sizeof mce))
3140                         goto out;
3141                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3142                 break;
3143         }
3144         case KVM_GET_VCPU_EVENTS: {
3145                 struct kvm_vcpu_events events;
3146
3147                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3148
3149                 r = -EFAULT;
3150                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3151                         break;
3152                 r = 0;
3153                 break;
3154         }
3155         case KVM_SET_VCPU_EVENTS: {
3156                 struct kvm_vcpu_events events;
3157
3158                 r = -EFAULT;
3159                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3160                         break;
3161
3162                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3163                 break;
3164         }
3165         case KVM_GET_DEBUGREGS: {
3166                 struct kvm_debugregs dbgregs;
3167
3168                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3169
3170                 r = -EFAULT;
3171                 if (copy_to_user(argp, &dbgregs,
3172                                  sizeof(struct kvm_debugregs)))
3173                         break;
3174                 r = 0;
3175                 break;
3176         }
3177         case KVM_SET_DEBUGREGS: {
3178                 struct kvm_debugregs dbgregs;
3179
3180                 r = -EFAULT;
3181                 if (copy_from_user(&dbgregs, argp,
3182                                    sizeof(struct kvm_debugregs)))
3183                         break;
3184
3185                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3186                 break;
3187         }
3188         case KVM_GET_XSAVE: {
3189                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3190                 r = -ENOMEM;
3191                 if (!u.xsave)
3192                         break;
3193
3194                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3195
3196                 r = -EFAULT;
3197                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3198                         break;
3199                 r = 0;
3200                 break;
3201         }
3202         case KVM_SET_XSAVE: {
3203                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3204                 r = -ENOMEM;
3205                 if (!u.xsave)
3206                         break;
3207
3208                 r = -EFAULT;
3209                 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3210                         break;
3211
3212                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3213                 break;
3214         }
3215         case KVM_GET_XCRS: {
3216                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3217                 r = -ENOMEM;
3218                 if (!u.xcrs)
3219                         break;
3220
3221                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3222
3223                 r = -EFAULT;
3224                 if (copy_to_user(argp, u.xcrs,
3225                                  sizeof(struct kvm_xcrs)))
3226                         break;
3227                 r = 0;
3228                 break;
3229         }
3230         case KVM_SET_XCRS: {
3231                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3232                 r = -ENOMEM;
3233                 if (!u.xcrs)
3234                         break;
3235
3236                 r = -EFAULT;
3237                 if (copy_from_user(u.xcrs, argp,
3238                                    sizeof(struct kvm_xcrs)))
3239                         break;
3240
3241                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3242                 break;
3243         }
3244         case KVM_SET_TSC_KHZ: {
3245                 u32 user_tsc_khz;
3246
3247                 r = -EINVAL;
3248                 if (!kvm_has_tsc_control)
3249                         break;
3250
3251                 user_tsc_khz = (u32)arg;
3252
3253                 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3254                         goto out;
3255
3256                 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3257
3258                 r = 0;
3259                 goto out;
3260         }
3261         case KVM_GET_TSC_KHZ: {
3262                 r = -EIO;
3263                 if (check_tsc_unstable())
3264                         goto out;
3265
3266                 r = vcpu_tsc_khz(vcpu);
3267
3268                 goto out;
3269         }
3270         default:
3271                 r = -EINVAL;
3272         }
3273 out:
3274         kfree(u.buffer);
3275         return r;
3276 }
3277
3278 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3279 {
3280         int ret;
3281
3282         if (addr > (unsigned int)(-3 * PAGE_SIZE))
3283                 return -1;
3284         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3285         return ret;
3286 }
3287
3288 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3289                                               u64 ident_addr)
3290 {
3291         kvm->arch.ept_identity_map_addr = ident_addr;
3292         return 0;
3293 }
3294
3295 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3296                                           u32 kvm_nr_mmu_pages)
3297 {
3298         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3299                 return -EINVAL;
3300
3301         mutex_lock(&kvm->slots_lock);
3302         spin_lock(&kvm->mmu_lock);
3303
3304         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3305         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3306
3307         spin_unlock(&kvm->mmu_lock);
3308         mutex_unlock(&kvm->slots_lock);
3309         return 0;
3310 }
3311
3312 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3313 {
3314         return kvm->arch.n_max_mmu_pages;
3315 }
3316
3317 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3318 {
3319         int r;
3320
3321         r = 0;
3322         switch (chip->chip_id) {
3323         case KVM_IRQCHIP_PIC_MASTER:
3324                 memcpy(&chip->chip.pic,
3325                         &pic_irqchip(kvm)->pics[0],
3326                         sizeof(struct kvm_pic_state));
3327                 break;
3328         case KVM_IRQCHIP_PIC_SLAVE:
3329                 memcpy(&chip->chip.pic,
3330                         &pic_irqchip(kvm)->pics[1],
3331                         sizeof(struct kvm_pic_state));
3332                 break;
3333         case KVM_IRQCHIP_IOAPIC:
3334                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3335                 break;
3336         default:
3337                 r = -EINVAL;
3338                 break;
3339         }
3340         return r;
3341 }
3342
3343 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3344 {
3345         int r;
3346
3347         r = 0;
3348         switch (chip->chip_id) {
3349         case KVM_IRQCHIP_PIC_MASTER:
3350                 spin_lock(&pic_irqchip(kvm)->lock);
3351                 memcpy(&pic_irqchip(kvm)->pics[0],
3352                         &chip->chip.pic,
3353                         sizeof(struct kvm_pic_state));
3354                 spin_unlock(&pic_irqchip(kvm)->lock);
3355                 break;
3356         case KVM_IRQCHIP_PIC_SLAVE:
3357                 spin_lock(&pic_irqchip(kvm)->lock);
3358                 memcpy(&pic_irqchip(kvm)->pics[1],
3359                         &chip->chip.pic,
3360                         sizeof(struct kvm_pic_state));
3361                 spin_unlock(&pic_irqchip(kvm)->lock);
3362                 break;
3363         case KVM_IRQCHIP_IOAPIC:
3364                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3365                 break;
3366         default:
3367                 r = -EINVAL;
3368                 break;
3369         }
3370         kvm_pic_update_irq(pic_irqchip(kvm));
3371         return r;
3372 }
3373
3374 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3375 {
3376         int r = 0;
3377
3378         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3379         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3380         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3381         return r;
3382 }
3383
3384 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3385 {
3386         int r = 0;
3387
3388         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3389         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3390         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3391         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3392         return r;
3393 }
3394
3395 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3396 {
3397         int r = 0;
3398
3399         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3400         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3401                 sizeof(ps->channels));
3402         ps->flags = kvm->arch.vpit->pit_state.flags;
3403         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3404         memset(&ps->reserved, 0, sizeof(ps->reserved));
3405         return r;
3406 }
3407
3408 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3409 {
3410         int r = 0, start = 0;
3411         u32 prev_legacy, cur_legacy;
3412         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3413         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3414         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3415         if (!prev_legacy && cur_legacy)
3416                 start = 1;
3417         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3418                sizeof(kvm->arch.vpit->pit_state.channels));
3419         kvm->arch.vpit->pit_state.flags = ps->flags;
3420         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3421         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3422         return r;
3423 }
3424
3425 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3426                                  struct kvm_reinject_control *control)
3427 {
3428         if (!kvm->arch.vpit)
3429                 return -ENXIO;
3430         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3431         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3432         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3433         return 0;
3434 }
3435
3436 /*
3437  * Get (and clear) the dirty memory log for a memory slot.
3438  */
3439 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3440                                       struct kvm_dirty_log *log)
3441 {
3442         int r, i;
3443         struct kvm_memory_slot *memslot;
3444         unsigned long n;
3445         unsigned long is_dirty = 0;
3446
3447         mutex_lock(&kvm->slots_lock);
3448
3449         r = -EINVAL;
3450         if (log->slot >= KVM_MEMORY_SLOTS)
3451                 goto out;
3452
3453         memslot = &kvm->memslots->memslots[log->slot];
3454         r = -ENOENT;
3455         if (!memslot->dirty_bitmap)
3456                 goto out;
3457
3458         n = kvm_dirty_bitmap_bytes(memslot);
3459
3460         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3461                 is_dirty = memslot->dirty_bitmap[i];
3462
3463         /* If nothing is dirty, don't bother messing with page tables. */
3464         if (is_dirty) {
3465                 struct kvm_memslots *slots, *old_slots;
3466                 unsigned long *dirty_bitmap;
3467
3468                 dirty_bitmap = memslot->dirty_bitmap_head;
3469                 if (memslot->dirty_bitmap == dirty_bitmap)
3470                         dirty_bitmap += n / sizeof(long);
3471                 memset(dirty_bitmap, 0, n);
3472
3473                 r = -ENOMEM;
3474                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3475                 if (!slots)
3476                         goto out;
3477                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3478                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3479                 slots->generation++;
3480
3481                 old_slots = kvm->memslots;
3482                 rcu_assign_pointer(kvm->memslots, slots);
3483                 synchronize_srcu_expedited(&kvm->srcu);
3484                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3485                 kfree(old_slots);
3486
3487                 spin_lock(&kvm->mmu_lock);
3488                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3489                 spin_unlock(&kvm->mmu_lock);
3490
3491                 r = -EFAULT;
3492                 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3493                         goto out;
3494         } else {
3495                 r = -EFAULT;
3496                 if (clear_user(log->dirty_bitmap, n))
3497                         goto out;
3498         }
3499
3500         r = 0;
3501 out:
3502         mutex_unlock(&kvm->slots_lock);
3503         return r;
3504 }
3505
3506 long kvm_arch_vm_ioctl(struct file *filp,
3507                        unsigned int ioctl, unsigned long arg)
3508 {
3509         struct kvm *kvm = filp->private_data;
3510         void __user *argp = (void __user *)arg;
3511         int r = -ENOTTY;
3512         /*
3513          * This union makes it completely explicit to gcc-3.x
3514          * that these two variables' stack usage should be
3515          * combined, not added together.
3516          */
3517         union {
3518                 struct kvm_pit_state ps;
3519                 struct kvm_pit_state2 ps2;
3520                 struct kvm_pit_config pit_config;
3521         } u;
3522
3523         switch (ioctl) {
3524         case KVM_SET_TSS_ADDR:
3525                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3526                 if (r < 0)
3527                         goto out;
3528                 break;
3529         case KVM_SET_IDENTITY_MAP_ADDR: {
3530                 u64 ident_addr;
3531
3532                 r = -EFAULT;
3533                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3534                         goto out;
3535                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3536                 if (r < 0)
3537                         goto out;
3538                 break;
3539         }
3540         case KVM_SET_NR_MMU_PAGES:
3541                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3542                 if (r)
3543                         goto out;
3544                 break;
3545         case KVM_GET_NR_MMU_PAGES:
3546                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3547                 break;
3548         case KVM_CREATE_IRQCHIP: {
3549                 struct kvm_pic *vpic;
3550
3551                 mutex_lock(&kvm->lock);
3552                 r = -EEXIST;
3553                 if (kvm->arch.vpic)
3554                         goto create_irqchip_unlock;
3555                 r = -ENOMEM;
3556                 vpic = kvm_create_pic(kvm);
3557                 if (vpic) {
3558                         r = kvm_ioapic_init(kvm);
3559                         if (r) {
3560                                 mutex_lock(&kvm->slots_lock);
3561                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3562                                                           &vpic->dev);
3563                                 mutex_unlock(&kvm->slots_lock);
3564                                 kfree(vpic);
3565                                 goto create_irqchip_unlock;
3566                         }
3567                 } else
3568                         goto create_irqchip_unlock;
3569                 smp_wmb();
3570                 kvm->arch.vpic = vpic;
3571                 smp_wmb();
3572                 r = kvm_setup_default_irq_routing(kvm);
3573                 if (r) {
3574                         mutex_lock(&kvm->slots_lock);
3575                         mutex_lock(&kvm->irq_lock);
3576                         kvm_ioapic_destroy(kvm);
3577                         kvm_destroy_pic(kvm);
3578                         mutex_unlock(&kvm->irq_lock);
3579                         mutex_unlock(&kvm->slots_lock);
3580                 }
3581         create_irqchip_unlock:
3582                 mutex_unlock(&kvm->lock);
3583                 break;
3584         }
3585         case KVM_CREATE_PIT:
3586                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3587                 goto create_pit;
3588         case KVM_CREATE_PIT2:
3589                 r = -EFAULT;
3590                 if (copy_from_user(&u.pit_config, argp,
3591                                    sizeof(struct kvm_pit_config)))
3592                         goto out;
3593         create_pit:
3594                 mutex_lock(&kvm->slots_lock);
3595                 r = -EEXIST;
3596                 if (kvm->arch.vpit)
3597                         goto create_pit_unlock;
3598                 r = -ENOMEM;
3599                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3600                 if (kvm->arch.vpit)
3601                         r = 0;
3602         create_pit_unlock:
3603                 mutex_unlock(&kvm->slots_lock);
3604                 break;
3605         case KVM_IRQ_LINE_STATUS:
3606         case KVM_IRQ_LINE: {
3607                 struct kvm_irq_level irq_event;
3608
3609                 r = -EFAULT;
3610                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3611                         goto out;
3612                 r = -ENXIO;
3613                 if (irqchip_in_kernel(kvm)) {
3614                         __s32 status;
3615                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3616                                         irq_event.irq, irq_event.level);
3617                         if (ioctl == KVM_IRQ_LINE_STATUS) {
3618                                 r = -EFAULT;
3619                                 irq_event.status = status;
3620                                 if (copy_to_user(argp, &irq_event,
3621                                                         sizeof irq_event))
3622                                         goto out;
3623                         }
3624                         r = 0;
3625                 }
3626                 break;
3627         }
3628         case KVM_GET_IRQCHIP: {
3629                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3630                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3631
3632                 r = -ENOMEM;
3633                 if (!chip)
3634                         goto out;
3635                 r = -EFAULT;
3636                 if (copy_from_user(chip, argp, sizeof *chip))
3637                         goto get_irqchip_out;
3638                 r = -ENXIO;
3639                 if (!irqchip_in_kernel(kvm))
3640                         goto get_irqchip_out;
3641                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3642                 if (r)
3643                         goto get_irqchip_out;
3644                 r = -EFAULT;
3645                 if (copy_to_user(argp, chip, sizeof *chip))
3646                         goto get_irqchip_out;
3647                 r = 0;
3648         get_irqchip_out:
3649                 kfree(chip);
3650                 if (r)
3651                         goto out;
3652                 break;
3653         }
3654         case KVM_SET_IRQCHIP: {
3655                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3656                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3657
3658                 r = -ENOMEM;
3659                 if (!chip)
3660                         goto out;
3661                 r = -EFAULT;
3662                 if (copy_from_user(chip, argp, sizeof *chip))
3663                         goto set_irqchip_out;
3664                 r = -ENXIO;
3665                 if (!irqchip_in_kernel(kvm))
3666                         goto set_irqchip_out;
3667                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3668                 if (r)
3669                         goto set_irqchip_out;
3670                 r = 0;
3671         set_irqchip_out:
3672                 kfree(chip);
3673                 if (r)
3674                         goto out;
3675                 break;
3676         }
3677         case KVM_GET_PIT: {
3678                 r = -EFAULT;
3679                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3680                         goto out;
3681                 r = -ENXIO;
3682                 if (!kvm->arch.vpit)
3683                         goto out;
3684                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3685                 if (r)
3686                         goto out;
3687                 r = -EFAULT;
3688                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3689                         goto out;
3690                 r = 0;
3691                 break;
3692         }
3693         case KVM_SET_PIT: {
3694                 r = -EFAULT;
3695                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3696                         goto out;
3697                 r = -ENXIO;
3698                 if (!kvm->arch.vpit)
3699                         goto out;
3700                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3701                 if (r)
3702                         goto out;
3703                 r = 0;
3704                 break;
3705         }
3706         case KVM_GET_PIT2: {
3707                 r = -ENXIO;
3708                 if (!kvm->arch.vpit)
3709                         goto out;
3710                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3711                 if (r)
3712                         goto out;
3713                 r = -EFAULT;
3714                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3715                         goto out;
3716                 r = 0;
3717                 break;
3718         }
3719         case KVM_SET_PIT2: {
3720                 r = -EFAULT;
3721                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3722                         goto out;
3723                 r = -ENXIO;
3724                 if (!kvm->arch.vpit)
3725                         goto out;
3726                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3727                 if (r)
3728                         goto out;
3729                 r = 0;
3730                 break;
3731         }
3732         case KVM_REINJECT_CONTROL: {
3733                 struct kvm_reinject_control control;
3734                 r =  -EFAULT;
3735                 if (copy_from_user(&control, argp, sizeof(control)))
3736                         goto out;
3737                 r = kvm_vm_ioctl_reinject(kvm, &control);
3738                 if (r)
3739                         goto out;
3740                 r = 0;
3741                 break;
3742         }
3743         case KVM_XEN_HVM_CONFIG: {
3744                 r = -EFAULT;
3745                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3746                                    sizeof(struct kvm_xen_hvm_config)))
3747                         goto out;
3748                 r = -EINVAL;
3749                 if (kvm->arch.xen_hvm_config.flags)
3750                         goto out;
3751                 r = 0;
3752                 break;
3753         }
3754         case KVM_SET_CLOCK: {
3755                 struct kvm_clock_data user_ns;
3756                 u64 now_ns;
3757                 s64 delta;
3758
3759                 r = -EFAULT;
3760                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3761                         goto out;
3762
3763                 r = -EINVAL;
3764                 if (user_ns.flags)
3765                         goto out;
3766
3767                 r = 0;
3768                 local_irq_disable();
3769                 now_ns = get_kernel_ns();
3770                 delta = user_ns.clock - now_ns;
3771                 local_irq_enable();
3772                 kvm->arch.kvmclock_offset = delta;
3773                 break;
3774         }
3775         case KVM_GET_CLOCK: {
3776                 struct kvm_clock_data user_ns;
3777                 u64 now_ns;
3778
3779                 local_irq_disable();
3780                 now_ns = get_kernel_ns();
3781                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3782                 local_irq_enable();
3783                 user_ns.flags = 0;
3784                 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3785
3786                 r = -EFAULT;
3787                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3788                         goto out;
3789                 r = 0;
3790                 break;
3791         }
3792
3793         default:
3794                 ;
3795         }
3796 out:
3797         return r;
3798 }
3799
3800 static void kvm_init_msr_list(void)
3801 {
3802         u32 dummy[2];
3803         unsigned i, j;
3804
3805         /* skip the first msrs in the list. KVM-specific */
3806         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3807                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3808                         continue;
3809                 if (j < i)
3810                         msrs_to_save[j] = msrs_to_save[i];
3811                 j++;
3812         }
3813         num_msrs_to_save = j;
3814 }
3815
3816 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3817                            const void *v)
3818 {
3819         int handled = 0;
3820         int n;
3821
3822         do {
3823                 n = min(len, 8);
3824                 if (!(vcpu->arch.apic &&
3825                       !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3826                     && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3827                         break;
3828                 handled += n;
3829                 addr += n;
3830                 len -= n;
3831                 v += n;
3832         } while (len);
3833
3834         return handled;
3835 }
3836
3837 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3838 {
3839         int handled = 0;
3840         int n;
3841
3842         do {
3843                 n = min(len, 8);
3844                 if (!(vcpu->arch.apic &&
3845                       !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3846                     && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3847                         break;
3848                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3849                 handled += n;
3850                 addr += n;
3851                 len -= n;
3852                 v += n;
3853         } while (len);
3854
3855         return handled;
3856 }
3857
3858 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3859                         struct kvm_segment *var, int seg)
3860 {
3861         kvm_x86_ops->set_segment(vcpu, var, seg);
3862 }
3863
3864 void kvm_get_segment(struct kvm_vcpu *vcpu,
3865                      struct kvm_segment *var, int seg)
3866 {
3867         kvm_x86_ops->get_segment(vcpu, var, seg);
3868 }
3869
3870 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3871 {
3872         return gpa;
3873 }
3874
3875 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3876 {
3877         gpa_t t_gpa;
3878         struct x86_exception exception;
3879
3880         BUG_ON(!mmu_is_nested(vcpu));
3881
3882         /* NPT walks are always user-walks */
3883         access |= PFERR_USER_MASK;
3884         t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3885
3886         return t_gpa;
3887 }
3888
3889 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3890                               struct x86_exception *exception)
3891 {
3892         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3893         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3894 }
3895
3896  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3897                                 struct x86_exception *exception)
3898 {
3899         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3900         access |= PFERR_FETCH_MASK;
3901         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3902 }
3903
3904 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3905                                struct x86_exception *exception)
3906 {
3907         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3908         access |= PFERR_WRITE_MASK;
3909         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3910 }
3911
3912 /* uses this to access any guest's mapped memory without checking CPL */
3913 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3914                                 struct x86_exception *exception)
3915 {
3916         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3917 }
3918
3919 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3920                                       struct kvm_vcpu *vcpu, u32 access,
3921                                       struct x86_exception *exception)
3922 {
3923         void *data = val;
3924         int r = X86EMUL_CONTINUE;
3925
3926         while (bytes) {
3927                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3928                                                             exception);
3929                 unsigned offset = addr & (PAGE_SIZE-1);
3930                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3931                 int ret;
3932
3933                 if (gpa == UNMAPPED_GVA)
3934                         return X86EMUL_PROPAGATE_FAULT;
3935                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3936                 if (ret < 0) {
3937                         r = X86EMUL_IO_NEEDED;
3938                         goto out;
3939                 }
3940
3941                 bytes -= toread;
3942                 data += toread;
3943                 addr += toread;
3944         }
3945 out:
3946         return r;
3947 }
3948
3949 /* used for instruction fetching */
3950 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3951                                 gva_t addr, void *val, unsigned int bytes,
3952                                 struct x86_exception *exception)
3953 {
3954         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3955         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3956
3957         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3958                                           access | PFERR_FETCH_MASK,
3959                                           exception);
3960 }
3961
3962 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3963                                gva_t addr, void *val, unsigned int bytes,
3964                                struct x86_exception *exception)
3965 {
3966         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3967         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3968
3969         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3970                                           exception);
3971 }
3972 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3973
3974 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3975                                       gva_t addr, void *val, unsigned int bytes,
3976                                       struct x86_exception *exception)
3977 {
3978         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3979         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3980 }
3981
3982 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3983                                        gva_t addr, void *val,
3984                                        unsigned int bytes,
3985                                        struct x86_exception *exception)
3986 {
3987         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3988         void *data = val;
3989         int r = X86EMUL_CONTINUE;
3990
3991         while (bytes) {
3992                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3993                                                              PFERR_WRITE_MASK,
3994                                                              exception);
3995                 unsigned offset = addr & (PAGE_SIZE-1);
3996                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3997                 int ret;
3998
3999                 if (gpa == UNMAPPED_GVA)
4000                         return X86EMUL_PROPAGATE_FAULT;
4001                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4002                 if (ret < 0) {
4003                         r = X86EMUL_IO_NEEDED;
4004                         goto out;
4005                 }
4006
4007                 bytes -= towrite;
4008                 data += towrite;
4009                 addr += towrite;
4010         }
4011 out:
4012         return r;
4013 }
4014 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4015
4016 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4017                                 gpa_t *gpa, struct x86_exception *exception,
4018                                 bool write)
4019 {
4020         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4021
4022         if (vcpu_match_mmio_gva(vcpu, gva) &&
4023                   check_write_user_access(vcpu, write, access,
4024                   vcpu->arch.access)) {
4025                 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4026                                         (gva & (PAGE_SIZE - 1));
4027                 trace_vcpu_match_mmio(gva, *gpa, write, false);
4028                 return 1;
4029         }
4030
4031         if (write)
4032                 access |= PFERR_WRITE_MASK;
4033
4034         *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4035
4036         if (*gpa == UNMAPPED_GVA)
4037                 return -1;
4038
4039         /* For APIC access vmexit */
4040         if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4041                 return 1;
4042
4043         if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4044                 trace_vcpu_match_mmio(gva, *gpa, write, true);
4045                 return 1;
4046         }
4047
4048         return 0;
4049 }
4050
4051 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4052                         const void *val, int bytes)
4053 {
4054         int ret;
4055
4056         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4057         if (ret < 0)
4058                 return 0;
4059         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
4060         return 1;
4061 }
4062
4063 struct read_write_emulator_ops {
4064         int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4065                                   int bytes);
4066         int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4067                                   void *val, int bytes);
4068         int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4069                                int bytes, void *val);
4070         int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4071                                     void *val, int bytes);
4072         bool write;
4073 };
4074
4075 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4076 {
4077         if (vcpu->mmio_read_completed) {
4078                 memcpy(val, vcpu->mmio_data, bytes);
4079                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4080                                vcpu->mmio_phys_addr, *(u64 *)val);
4081                 vcpu->mmio_read_completed = 0;
4082                 return 1;
4083         }
4084
4085         return 0;
4086 }
4087
4088 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4089                         void *val, int bytes)
4090 {
4091         return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4092 }
4093
4094 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4095                          void *val, int bytes)
4096 {
4097         return emulator_write_phys(vcpu, gpa, val, bytes);
4098 }
4099
4100 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4101 {
4102         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4103         return vcpu_mmio_write(vcpu, gpa, bytes, val);
4104 }
4105
4106 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4107                           void *val, int bytes)
4108 {
4109         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4110         return X86EMUL_IO_NEEDED;
4111 }
4112
4113 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4114                            void *val, int bytes)
4115 {
4116         memcpy(vcpu->mmio_data, val, bytes);
4117         memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
4118         return X86EMUL_CONTINUE;
4119 }
4120
4121 static struct read_write_emulator_ops read_emultor = {
4122         .read_write_prepare = read_prepare,
4123         .read_write_emulate = read_emulate,
4124         .read_write_mmio = vcpu_mmio_read,
4125         .read_write_exit_mmio = read_exit_mmio,
4126 };
4127
4128 static struct read_write_emulator_ops write_emultor = {
4129         .read_write_emulate = write_emulate,
4130         .read_write_mmio = write_mmio,
4131         .read_write_exit_mmio = write_exit_mmio,
4132         .write = true,
4133 };
4134
4135 static int emulator_read_write_onepage(unsigned long addr, void *val,
4136                                        unsigned int bytes,
4137                                        struct x86_exception *exception,
4138                                        struct kvm_vcpu *vcpu,
4139                                        struct read_write_emulator_ops *ops)
4140 {
4141         gpa_t gpa;
4142         int handled, ret;
4143         bool write = ops->write;
4144
4145         if (ops->read_write_prepare &&
4146                   ops->read_write_prepare(vcpu, val, bytes))
4147                 return X86EMUL_CONTINUE;
4148
4149         ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4150
4151         if (ret < 0)
4152                 return X86EMUL_PROPAGATE_FAULT;
4153
4154         /* For APIC access vmexit */
4155         if (ret)
4156                 goto mmio;
4157
4158         if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4159                 return X86EMUL_CONTINUE;
4160
4161 mmio:
4162         /*
4163          * Is this MMIO handled locally?
4164          */
4165         handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4166         if (handled == bytes)
4167                 return X86EMUL_CONTINUE;
4168
4169         gpa += handled;
4170         bytes -= handled;
4171         val += handled;
4172
4173         vcpu->mmio_needed = 1;
4174         vcpu->run->exit_reason = KVM_EXIT_MMIO;
4175         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
4176         vcpu->mmio_size = bytes;
4177         vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
4178         vcpu->run->mmio.is_write = vcpu->mmio_is_write = write;
4179         vcpu->mmio_index = 0;
4180
4181         return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4182 }
4183
4184 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4185                         void *val, unsigned int bytes,
4186                         struct x86_exception *exception,
4187                         struct read_write_emulator_ops *ops)
4188 {
4189         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4190
4191         /* Crossing a page boundary? */
4192         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4193                 int rc, now;
4194
4195                 now = -addr & ~PAGE_MASK;
4196                 rc = emulator_read_write_onepage(addr, val, now, exception,
4197                                                  vcpu, ops);
4198
4199                 if (rc != X86EMUL_CONTINUE)
4200                         return rc;
4201                 addr += now;
4202                 val += now;
4203                 bytes -= now;
4204         }
4205
4206         return emulator_read_write_onepage(addr, val, bytes, exception,
4207                                            vcpu, ops);
4208 }
4209
4210 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4211                                   unsigned long addr,
4212                                   void *val,
4213                                   unsigned int bytes,
4214                                   struct x86_exception *exception)
4215 {
4216         return emulator_read_write(ctxt, addr, val, bytes,
4217                                    exception, &read_emultor);
4218 }
4219
4220 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4221                             unsigned long addr,
4222                             const void *val,
4223                             unsigned int bytes,
4224                             struct x86_exception *exception)
4225 {
4226         return emulator_read_write(ctxt, addr, (void *)val, bytes,
4227                                    exception, &write_emultor);
4228 }
4229
4230 #define CMPXCHG_TYPE(t, ptr, old, new) \
4231         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4232
4233 #ifdef CONFIG_X86_64
4234 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4235 #else
4236 #  define CMPXCHG64(ptr, old, new) \
4237         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4238 #endif
4239
4240 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4241                                      unsigned long addr,
4242                                      const void *old,
4243                                      const void *new,
4244                                      unsigned int bytes,
4245                                      struct x86_exception *exception)
4246 {
4247         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4248         gpa_t gpa;
4249         struct page *page;
4250         char *kaddr;
4251         bool exchanged;
4252
4253         /* guests cmpxchg8b have to be emulated atomically */
4254         if (bytes > 8 || (bytes & (bytes - 1)))
4255                 goto emul_write;
4256
4257         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4258
4259         if (gpa == UNMAPPED_GVA ||
4260             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4261                 goto emul_write;
4262
4263         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4264                 goto emul_write;
4265
4266         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4267         if (is_error_page(page)) {
4268                 kvm_release_page_clean(page);
4269                 goto emul_write;
4270         }
4271
4272         kaddr = kmap_atomic(page, KM_USER0);
4273         kaddr += offset_in_page(gpa);
4274         switch (bytes) {
4275         case 1:
4276                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4277                 break;
4278         case 2:
4279                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4280                 break;
4281         case 4:
4282                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4283                 break;
4284         case 8:
4285                 exchanged = CMPXCHG64(kaddr, old, new);
4286                 break;
4287         default:
4288                 BUG();
4289         }
4290         kunmap_atomic(kaddr, KM_USER0);
4291         kvm_release_page_dirty(page);
4292
4293         if (!exchanged)
4294                 return X86EMUL_CMPXCHG_FAILED;
4295
4296         kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4297
4298         return X86EMUL_CONTINUE;
4299
4300 emul_write:
4301         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4302
4303         return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4304 }
4305
4306 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4307 {
4308         /* TODO: String I/O for in kernel device */
4309         int r;
4310
4311         if (vcpu->arch.pio.in)
4312                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4313                                     vcpu->arch.pio.size, pd);
4314         else
4315                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4316                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
4317                                      pd);
4318         return r;
4319 }
4320
4321
4322 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4323                                     int size, unsigned short port, void *val,
4324                                     unsigned int count)
4325 {
4326         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4327
4328         if (vcpu->arch.pio.count)
4329                 goto data_avail;
4330
4331         trace_kvm_pio(0, port, size, count);
4332
4333         vcpu->arch.pio.port = port;
4334         vcpu->arch.pio.in = 1;
4335         vcpu->arch.pio.count  = count;
4336         vcpu->arch.pio.size = size;
4337
4338         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4339         data_avail:
4340                 memcpy(val, vcpu->arch.pio_data, size * count);
4341                 vcpu->arch.pio.count = 0;
4342                 return 1;
4343         }
4344
4345         vcpu->run->exit_reason = KVM_EXIT_IO;
4346         vcpu->run->io.direction = KVM_EXIT_IO_IN;
4347         vcpu->run->io.size = size;
4348         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4349         vcpu->run->io.count = count;
4350         vcpu->run->io.port = port;
4351
4352         return 0;
4353 }
4354
4355 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4356                                      int size, unsigned short port,
4357                                      const void *val, unsigned int count)
4358 {
4359         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4360
4361         trace_kvm_pio(1, port, size, count);
4362
4363         vcpu->arch.pio.port = port;
4364         vcpu->arch.pio.in = 0;
4365         vcpu->arch.pio.count = count;
4366         vcpu->arch.pio.size = size;
4367
4368         memcpy(vcpu->arch.pio_data, val, size * count);
4369
4370         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4371                 vcpu->arch.pio.count = 0;
4372                 return 1;
4373         }
4374
4375         vcpu->run->exit_reason = KVM_EXIT_IO;
4376         vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4377         vcpu->run->io.size = size;
4378         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4379         vcpu->run->io.count = count;
4380         vcpu->run->io.port = port;
4381
4382         return 0;
4383 }
4384
4385 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4386 {
4387         return kvm_x86_ops->get_segment_base(vcpu, seg);
4388 }
4389
4390 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4391 {
4392         kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4393 }
4394
4395 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4396 {
4397         if (!need_emulate_wbinvd(vcpu))
4398                 return X86EMUL_CONTINUE;
4399
4400         if (kvm_x86_ops->has_wbinvd_exit()) {
4401                 int cpu = get_cpu();
4402
4403                 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4404                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4405                                 wbinvd_ipi, NULL, 1);
4406                 put_cpu();
4407                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4408         } else
4409                 wbinvd();
4410         return X86EMUL_CONTINUE;
4411 }
4412 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4413
4414 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4415 {
4416         kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4417 }
4418
4419 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4420 {
4421         return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4422 }
4423
4424 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4425 {
4426
4427         return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4428 }
4429
4430 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4431 {
4432         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4433 }
4434
4435 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4436 {
4437         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4438         unsigned long value;
4439
4440         switch (cr) {
4441         case 0:
4442                 value = kvm_read_cr0(vcpu);
4443                 break;
4444         case 2:
4445                 value = vcpu->arch.cr2;
4446                 break;
4447         case 3:
4448                 value = kvm_read_cr3(vcpu);
4449                 break;
4450         case 4:
4451                 value = kvm_read_cr4(vcpu);
4452                 break;
4453         case 8:
4454                 value = kvm_get_cr8(vcpu);
4455                 break;
4456         default:
4457                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4458                 return 0;
4459         }
4460
4461         return value;
4462 }
4463
4464 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4465 {
4466         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4467         int res = 0;
4468
4469         switch (cr) {
4470         case 0:
4471                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4472                 break;
4473         case 2:
4474                 vcpu->arch.cr2 = val;
4475                 break;
4476         case 3:
4477                 res = kvm_set_cr3(vcpu, val);
4478                 break;
4479         case 4:
4480                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4481                 break;
4482         case 8:
4483                 res = kvm_set_cr8(vcpu, val);
4484                 break;
4485         default:
4486                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4487                 res = -1;
4488         }
4489
4490         return res;
4491 }
4492
4493 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4494 {
4495         return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4496 }
4497
4498 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4499 {
4500         kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4501 }
4502
4503 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4504 {
4505         kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4506 }
4507
4508 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4509 {
4510         kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4511 }
4512
4513 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4514 {
4515         kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4516 }
4517
4518 static unsigned long emulator_get_cached_segment_base(
4519         struct x86_emulate_ctxt *ctxt, int seg)
4520 {
4521         return get_segment_base(emul_to_vcpu(ctxt), seg);
4522 }
4523
4524 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4525                                  struct desc_struct *desc, u32 *base3,
4526                                  int seg)
4527 {
4528         struct kvm_segment var;
4529
4530         kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4531         *selector = var.selector;
4532
4533         if (var.unusable)
4534                 return false;
4535
4536         if (var.g)
4537                 var.limit >>= 12;
4538         set_desc_limit(desc, var.limit);
4539         set_desc_base(desc, (unsigned long)var.base);
4540 #ifdef CONFIG_X86_64
4541         if (base3)
4542                 *base3 = var.base >> 32;
4543 #endif
4544         desc->type = var.type;
4545         desc->s = var.s;
4546         desc->dpl = var.dpl;
4547         desc->p = var.present;
4548         desc->avl = var.avl;
4549         desc->l = var.l;
4550         desc->d = var.db;
4551         desc->g = var.g;
4552
4553         return true;
4554 }
4555
4556 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4557                                  struct desc_struct *desc, u32 base3,
4558                                  int seg)
4559 {
4560         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4561         struct kvm_segment var;
4562
4563         var.selector = selector;
4564         var.base = get_desc_base(desc);
4565 #ifdef CONFIG_X86_64
4566         var.base |= ((u64)base3) << 32;
4567 #endif
4568         var.limit = get_desc_limit(desc);
4569         if (desc->g)
4570                 var.limit = (var.limit << 12) | 0xfff;
4571         var.type = desc->type;
4572         var.present = desc->p;
4573         var.dpl = desc->dpl;
4574         var.db = desc->d;
4575         var.s = desc->s;
4576         var.l = desc->l;
4577         var.g = desc->g;
4578         var.avl = desc->avl;
4579         var.present = desc->p;
4580         var.unusable = !var.present;
4581         var.padding = 0;
4582
4583         kvm_set_segment(vcpu, &var, seg);
4584         return;
4585 }
4586
4587 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4588                             u32 msr_index, u64 *pdata)
4589 {
4590         return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4591 }
4592
4593 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4594                             u32 msr_index, u64 data)
4595 {
4596         return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4597 }
4598
4599 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4600 {
4601         emul_to_vcpu(ctxt)->arch.halt_request = 1;
4602 }
4603
4604 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4605 {
4606         preempt_disable();
4607         kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4608         /*
4609          * CR0.TS may reference the host fpu state, not the guest fpu state,
4610          * so it may be clear at this point.
4611          */
4612         clts();
4613 }
4614
4615 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4616 {
4617         preempt_enable();
4618 }
4619
4620 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4621                               struct x86_instruction_info *info,
4622                               enum x86_intercept_stage stage)
4623 {
4624         return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4625 }
4626
4627 static struct x86_emulate_ops emulate_ops = {
4628         .read_std            = kvm_read_guest_virt_system,
4629         .write_std           = kvm_write_guest_virt_system,
4630         .fetch               = kvm_fetch_guest_virt,
4631         .read_emulated       = emulator_read_emulated,
4632         .write_emulated      = emulator_write_emulated,
4633         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
4634         .invlpg              = emulator_invlpg,
4635         .pio_in_emulated     = emulator_pio_in_emulated,
4636         .pio_out_emulated    = emulator_pio_out_emulated,
4637         .get_segment         = emulator_get_segment,
4638         .set_segment         = emulator_set_segment,
4639         .get_cached_segment_base = emulator_get_cached_segment_base,
4640         .get_gdt             = emulator_get_gdt,
4641         .get_idt             = emulator_get_idt,
4642         .set_gdt             = emulator_set_gdt,
4643         .set_idt             = emulator_set_idt,
4644         .get_cr              = emulator_get_cr,
4645         .set_cr              = emulator_set_cr,
4646         .cpl                 = emulator_get_cpl,
4647         .get_dr              = emulator_get_dr,
4648         .set_dr              = emulator_set_dr,
4649         .set_msr             = emulator_set_msr,
4650         .get_msr             = emulator_get_msr,
4651         .halt                = emulator_halt,
4652         .wbinvd              = emulator_wbinvd,
4653         .fix_hypercall       = emulator_fix_hypercall,
4654         .get_fpu             = emulator_get_fpu,
4655         .put_fpu             = emulator_put_fpu,
4656         .intercept           = emulator_intercept,
4657 };
4658
4659 static void cache_all_regs(struct kvm_vcpu *vcpu)
4660 {
4661         kvm_register_read(vcpu, VCPU_REGS_RAX);
4662         kvm_register_read(vcpu, VCPU_REGS_RSP);
4663         kvm_register_read(vcpu, VCPU_REGS_RIP);
4664         vcpu->arch.regs_dirty = ~0;
4665 }
4666
4667 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4668 {
4669         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4670         /*
4671          * an sti; sti; sequence only disable interrupts for the first
4672          * instruction. So, if the last instruction, be it emulated or
4673          * not, left the system with the INT_STI flag enabled, it
4674          * means that the last instruction is an sti. We should not
4675          * leave the flag on in this case. The same goes for mov ss
4676          */
4677         if (!(int_shadow & mask))
4678                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4679 }
4680
4681 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4682 {
4683         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4684         if (ctxt->exception.vector == PF_VECTOR)
4685                 kvm_propagate_fault(vcpu, &ctxt->exception);
4686         else if (ctxt->exception.error_code_valid)
4687                 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4688                                       ctxt->exception.error_code);
4689         else
4690                 kvm_queue_exception(vcpu, ctxt->exception.vector);
4691 }
4692
4693 static void init_decode_cache(struct x86_emulate_ctxt *ctxt,
4694                               const unsigned long *regs)
4695 {
4696         memset(&ctxt->twobyte, 0,
4697                (void *)&ctxt->regs - (void *)&ctxt->twobyte);
4698         memcpy(ctxt->regs, regs, sizeof(ctxt->regs));
4699
4700         ctxt->fetch.start = 0;
4701         ctxt->fetch.end = 0;
4702         ctxt->io_read.pos = 0;
4703         ctxt->io_read.end = 0;
4704         ctxt->mem_read.pos = 0;
4705         ctxt->mem_read.end = 0;
4706 }
4707
4708 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4709 {
4710         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4711         int cs_db, cs_l;
4712
4713         /*
4714          * TODO: fix emulate.c to use guest_read/write_register
4715          * instead of direct ->regs accesses, can save hundred cycles
4716          * on Intel for instructions that don't read/change RSP, for
4717          * for example.
4718          */
4719         cache_all_regs(vcpu);
4720
4721         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4722
4723         ctxt->eflags = kvm_get_rflags(vcpu);
4724         ctxt->eip = kvm_rip_read(vcpu);
4725         ctxt->mode = (!is_protmode(vcpu))               ? X86EMUL_MODE_REAL :
4726                      (ctxt->eflags & X86_EFLAGS_VM)     ? X86EMUL_MODE_VM86 :
4727                      cs_l                               ? X86EMUL_MODE_PROT64 :
4728                      cs_db                              ? X86EMUL_MODE_PROT32 :
4729                                                           X86EMUL_MODE_PROT16;
4730         ctxt->guest_mode = is_guest_mode(vcpu);
4731
4732         init_decode_cache(ctxt, vcpu->arch.regs);
4733         vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4734 }
4735
4736 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4737 {
4738         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4739         int ret;
4740
4741         init_emulate_ctxt(vcpu);
4742
4743         ctxt->op_bytes = 2;
4744         ctxt->ad_bytes = 2;
4745         ctxt->_eip = ctxt->eip + inc_eip;
4746         ret = emulate_int_real(ctxt, irq);
4747
4748         if (ret != X86EMUL_CONTINUE)
4749                 return EMULATE_FAIL;
4750
4751         ctxt->eip = ctxt->_eip;
4752         memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4753         kvm_rip_write(vcpu, ctxt->eip);
4754         kvm_set_rflags(vcpu, ctxt->eflags);
4755
4756         if (irq == NMI_VECTOR)
4757                 vcpu->arch.nmi_pending = false;
4758         else
4759                 vcpu->arch.interrupt.pending = false;
4760
4761         return EMULATE_DONE;
4762 }
4763 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4764
4765 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4766 {
4767         int r = EMULATE_DONE;
4768
4769         ++vcpu->stat.insn_emulation_fail;
4770         trace_kvm_emulate_insn_failed(vcpu);
4771         if (!is_guest_mode(vcpu)) {
4772                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4773                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4774                 vcpu->run->internal.ndata = 0;
4775                 r = EMULATE_FAIL;
4776         }
4777         kvm_queue_exception(vcpu, UD_VECTOR);
4778
4779         return r;
4780 }
4781
4782 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4783 {
4784         gpa_t gpa;
4785
4786         if (tdp_enabled)
4787                 return false;
4788
4789         /*
4790          * if emulation was due to access to shadowed page table
4791          * and it failed try to unshadow page and re-entetr the
4792          * guest to let CPU execute the instruction.
4793          */
4794         if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4795                 return true;
4796
4797         gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4798
4799         if (gpa == UNMAPPED_GVA)
4800                 return true; /* let cpu generate fault */
4801
4802         if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4803                 return true;
4804
4805         return false;
4806 }
4807
4808 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4809                             unsigned long cr2,
4810                             int emulation_type,
4811                             void *insn,
4812                             int insn_len)
4813 {
4814         int r;
4815         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4816         bool writeback = true;
4817
4818         kvm_clear_exception_queue(vcpu);
4819
4820         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4821                 init_emulate_ctxt(vcpu);
4822                 ctxt->interruptibility = 0;
4823                 ctxt->have_exception = false;
4824                 ctxt->perm_ok = false;
4825
4826                 ctxt->only_vendor_specific_insn
4827                         = emulation_type & EMULTYPE_TRAP_UD;
4828
4829                 r = x86_decode_insn(ctxt, insn, insn_len);
4830
4831                 trace_kvm_emulate_insn_start(vcpu);
4832                 ++vcpu->stat.insn_emulation;
4833                 if (r)  {
4834                         if (emulation_type & EMULTYPE_TRAP_UD)
4835                                 return EMULATE_FAIL;
4836                         if (reexecute_instruction(vcpu, cr2))
4837                                 return EMULATE_DONE;
4838                         if (emulation_type & EMULTYPE_SKIP)
4839                                 return EMULATE_FAIL;
4840                         return handle_emulation_failure(vcpu);
4841                 }
4842         }
4843
4844         if (emulation_type & EMULTYPE_SKIP) {
4845                 kvm_rip_write(vcpu, ctxt->_eip);
4846                 return EMULATE_DONE;
4847         }
4848
4849         /* this is needed for vmware backdoor interface to work since it
4850            changes registers values  during IO operation */
4851         if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4852                 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4853                 memcpy(ctxt->regs, vcpu->arch.regs, sizeof ctxt->regs);
4854         }
4855
4856 restart:
4857         r = x86_emulate_insn(ctxt);
4858
4859         if (r == EMULATION_INTERCEPTED)
4860                 return EMULATE_DONE;
4861
4862         if (r == EMULATION_FAILED) {
4863                 if (reexecute_instruction(vcpu, cr2))
4864                         return EMULATE_DONE;
4865
4866                 return handle_emulation_failure(vcpu);
4867         }
4868
4869         if (ctxt->have_exception) {
4870                 inject_emulated_exception(vcpu);
4871                 r = EMULATE_DONE;
4872         } else if (vcpu->arch.pio.count) {
4873                 if (!vcpu->arch.pio.in)
4874                         vcpu->arch.pio.count = 0;
4875                 else
4876                         writeback = false;
4877                 r = EMULATE_DO_MMIO;
4878         } else if (vcpu->mmio_needed) {
4879                 if (!vcpu->mmio_is_write)
4880                         writeback = false;
4881                 r = EMULATE_DO_MMIO;
4882         } else if (r == EMULATION_RESTART)
4883                 goto restart;
4884         else
4885                 r = EMULATE_DONE;
4886
4887         if (writeback) {
4888                 toggle_interruptibility(vcpu, ctxt->interruptibility);
4889                 kvm_set_rflags(vcpu, ctxt->eflags);
4890                 kvm_make_request(KVM_REQ_EVENT, vcpu);
4891                 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4892                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4893                 kvm_rip_write(vcpu, ctxt->eip);
4894         } else
4895                 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4896
4897         return r;
4898 }
4899 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4900
4901 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4902 {
4903         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4904         int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4905                                             size, port, &val, 1);
4906         /* do not return to emulator after return from userspace */
4907         vcpu->arch.pio.count = 0;
4908         return ret;
4909 }
4910 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4911
4912 static void tsc_bad(void *info)
4913 {
4914         __this_cpu_write(cpu_tsc_khz, 0);
4915 }
4916
4917 static void tsc_khz_changed(void *data)
4918 {
4919         struct cpufreq_freqs *freq = data;
4920         unsigned long khz = 0;
4921
4922         if (data)
4923                 khz = freq->new;
4924         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4925                 khz = cpufreq_quick_get(raw_smp_processor_id());
4926         if (!khz)
4927                 khz = tsc_khz;
4928         __this_cpu_write(cpu_tsc_khz, khz);
4929 }
4930
4931 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4932                                      void *data)
4933 {
4934         struct cpufreq_freqs *freq = data;
4935         struct kvm *kvm;
4936         struct kvm_vcpu *vcpu;
4937         int i, send_ipi = 0;
4938
4939         /*
4940          * We allow guests to temporarily run on slowing clocks,
4941          * provided we notify them after, or to run on accelerating
4942          * clocks, provided we notify them before.  Thus time never
4943          * goes backwards.
4944          *
4945          * However, we have a problem.  We can't atomically update
4946          * the frequency of a given CPU from this function; it is
4947          * merely a notifier, which can be called from any CPU.
4948          * Changing the TSC frequency at arbitrary points in time
4949          * requires a recomputation of local variables related to
4950          * the TSC for each VCPU.  We must flag these local variables
4951          * to be updated and be sure the update takes place with the
4952          * new frequency before any guests proceed.
4953          *
4954          * Unfortunately, the combination of hotplug CPU and frequency
4955          * change creates an intractable locking scenario; the order
4956          * of when these callouts happen is undefined with respect to
4957          * CPU hotplug, and they can race with each other.  As such,
4958          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4959          * undefined; you can actually have a CPU frequency change take
4960          * place in between the computation of X and the setting of the
4961          * variable.  To protect against this problem, all updates of
4962          * the per_cpu tsc_khz variable are done in an interrupt
4963          * protected IPI, and all callers wishing to update the value
4964          * must wait for a synchronous IPI to complete (which is trivial
4965          * if the caller is on the CPU already).  This establishes the
4966          * necessary total order on variable updates.
4967          *
4968          * Note that because a guest time update may take place
4969          * anytime after the setting of the VCPU's request bit, the
4970          * correct TSC value must be set before the request.  However,
4971          * to ensure the update actually makes it to any guest which
4972          * starts running in hardware virtualization between the set
4973          * and the acquisition of the spinlock, we must also ping the
4974          * CPU after setting the request bit.
4975          *
4976          */
4977
4978         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4979                 return 0;
4980         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4981                 return 0;
4982
4983         smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4984
4985         raw_spin_lock(&kvm_lock);
4986         list_for_each_entry(kvm, &vm_list, vm_list) {
4987                 kvm_for_each_vcpu(i, vcpu, kvm) {
4988                         if (vcpu->cpu != freq->cpu)
4989                                 continue;
4990                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4991                         if (vcpu->cpu != smp_processor_id())
4992                                 send_ipi = 1;
4993                 }
4994         }
4995         raw_spin_unlock(&kvm_lock);
4996
4997         if (freq->old < freq->new && send_ipi) {
4998                 /*
4999                  * We upscale the frequency.  Must make the guest
5000                  * doesn't see old kvmclock values while running with
5001                  * the new frequency, otherwise we risk the guest sees
5002                  * time go backwards.
5003                  *
5004                  * In case we update the frequency for another cpu
5005                  * (which might be in guest context) send an interrupt
5006                  * to kick the cpu out of guest context.  Next time
5007                  * guest context is entered kvmclock will be updated,
5008                  * so the guest will not see stale values.
5009                  */
5010                 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5011         }
5012         return 0;
5013 }
5014
5015 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5016         .notifier_call  = kvmclock_cpufreq_notifier
5017 };
5018
5019 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5020                                         unsigned long action, void *hcpu)
5021 {
5022         unsigned int cpu = (unsigned long)hcpu;
5023
5024         switch (action) {
5025                 case CPU_ONLINE:
5026                 case CPU_DOWN_FAILED:
5027                         smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5028                         break;
5029                 case CPU_DOWN_PREPARE:
5030                         smp_call_function_single(cpu, tsc_bad, NULL, 1);
5031                         break;
5032         }
5033         return NOTIFY_OK;
5034 }
5035
5036 static struct notifier_block kvmclock_cpu_notifier_block = {
5037         .notifier_call  = kvmclock_cpu_notifier,
5038         .priority = -INT_MAX
5039 };
5040
5041 static void kvm_timer_init(void)
5042 {
5043         int cpu;
5044
5045         max_tsc_khz = tsc_khz;
5046         register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5047         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5048 #ifdef CONFIG_CPU_FREQ
5049                 struct cpufreq_policy policy;
5050                 memset(&policy, 0, sizeof(policy));
5051                 cpu = get_cpu();
5052                 cpufreq_get_policy(&policy, cpu);
5053                 if (policy.cpuinfo.max_freq)
5054                         max_tsc_khz = policy.cpuinfo.max_freq;
5055                 put_cpu();
5056 #endif
5057                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5058                                           CPUFREQ_TRANSITION_NOTIFIER);
5059         }
5060         pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5061         for_each_online_cpu(cpu)
5062                 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5063 }
5064
5065 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5066
5067 static int kvm_is_in_guest(void)
5068 {
5069         return percpu_read(current_vcpu) != NULL;
5070 }
5071
5072 static int kvm_is_user_mode(void)
5073 {
5074         int user_mode = 3;
5075
5076         if (percpu_read(current_vcpu))
5077                 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
5078
5079         return user_mode != 0;
5080 }
5081
5082 static unsigned long kvm_get_guest_ip(void)
5083 {
5084         unsigned long ip = 0;
5085
5086         if (percpu_read(current_vcpu))
5087                 ip = kvm_rip_read(percpu_read(current_vcpu));
5088
5089         return ip;
5090 }
5091
5092 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5093         .is_in_guest            = kvm_is_in_guest,
5094         .is_user_mode           = kvm_is_user_mode,
5095         .get_guest_ip           = kvm_get_guest_ip,
5096 };
5097
5098 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5099 {
5100         percpu_write(current_vcpu, vcpu);
5101 }
5102 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5103
5104 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5105 {
5106         percpu_write(current_vcpu, NULL);
5107 }
5108 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5109
5110 static void kvm_set_mmio_spte_mask(void)
5111 {
5112         u64 mask;
5113         int maxphyaddr = boot_cpu_data.x86_phys_bits;
5114
5115         /*
5116          * Set the reserved bits and the present bit of an paging-structure
5117          * entry to generate page fault with PFER.RSV = 1.
5118          */
5119         mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5120         mask |= 1ull;
5121
5122 #ifdef CONFIG_X86_64
5123         /*
5124          * If reserved bit is not supported, clear the present bit to disable
5125          * mmio page fault.
5126          */
5127         if (maxphyaddr == 52)
5128                 mask &= ~1ull;
5129 #endif
5130
5131         kvm_mmu_set_mmio_spte_mask(mask);
5132 }
5133
5134 int kvm_arch_init(void *opaque)
5135 {
5136         int r;
5137         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5138
5139         if (kvm_x86_ops) {
5140                 printk(KERN_ERR "kvm: already loaded the other module\n");
5141                 r = -EEXIST;
5142                 goto out;
5143         }
5144
5145         if (!ops->cpu_has_kvm_support()) {
5146                 printk(KERN_ERR "kvm: no hardware support\n");
5147                 r = -EOPNOTSUPP;
5148                 goto out;
5149         }
5150         if (ops->disabled_by_bios()) {
5151                 printk(KERN_ERR "kvm: disabled by bios\n");
5152                 r = -EOPNOTSUPP;
5153                 goto out;
5154         }
5155
5156         r = kvm_mmu_module_init();
5157         if (r)
5158                 goto out;
5159
5160         kvm_set_mmio_spte_mask();
5161         kvm_init_msr_list();
5162
5163         kvm_x86_ops = ops;
5164         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5165                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
5166
5167         kvm_timer_init();
5168
5169         perf_register_guest_info_callbacks(&kvm_guest_cbs);
5170
5171         if (cpu_has_xsave)
5172                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5173
5174         return 0;
5175
5176 out:
5177         return r;
5178 }
5179
5180 void kvm_arch_exit(void)
5181 {
5182         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5183
5184         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5185                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5186                                             CPUFREQ_TRANSITION_NOTIFIER);
5187         unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5188         kvm_x86_ops = NULL;
5189         kvm_mmu_module_exit();
5190 }
5191
5192 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5193 {
5194         ++vcpu->stat.halt_exits;
5195         if (irqchip_in_kernel(vcpu->kvm)) {
5196                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5197                 return 1;
5198         } else {
5199                 vcpu->run->exit_reason = KVM_EXIT_HLT;
5200                 return 0;
5201         }
5202 }
5203 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5204
5205 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
5206                            unsigned long a1)
5207 {
5208         if (is_long_mode(vcpu))
5209                 return a0;
5210         else
5211                 return a0 | ((gpa_t)a1 << 32);
5212 }
5213
5214 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5215 {
5216         u64 param, ingpa, outgpa, ret;
5217         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5218         bool fast, longmode;
5219         int cs_db, cs_l;
5220
5221         /*
5222          * hypercall generates UD from non zero cpl and real mode
5223          * per HYPER-V spec
5224          */
5225         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5226                 kvm_queue_exception(vcpu, UD_VECTOR);
5227                 return 0;
5228         }
5229
5230         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5231         longmode = is_long_mode(vcpu) && cs_l == 1;
5232
5233         if (!longmode) {
5234                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5235                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5236                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5237                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5238                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5239                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5240         }
5241 #ifdef CONFIG_X86_64
5242         else {
5243                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5244                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5245                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5246         }
5247 #endif
5248
5249         code = param & 0xffff;
5250         fast = (param >> 16) & 0x1;
5251         rep_cnt = (param >> 32) & 0xfff;
5252         rep_idx = (param >> 48) & 0xfff;
5253
5254         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5255
5256         switch (code) {
5257         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5258                 kvm_vcpu_on_spin(vcpu);
5259                 break;
5260         default:
5261                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5262                 break;
5263         }
5264
5265         ret = res | (((u64)rep_done & 0xfff) << 32);
5266         if (longmode) {
5267                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5268         } else {
5269                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5270                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5271         }
5272
5273         return 1;
5274 }
5275
5276 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5277 {
5278         unsigned long nr, a0, a1, a2, a3, ret;
5279         int r = 1;
5280
5281         if (kvm_hv_hypercall_enabled(vcpu->kvm))
5282                 return kvm_hv_hypercall(vcpu);
5283
5284         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5285         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5286         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5287         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5288         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5289
5290         trace_kvm_hypercall(nr, a0, a1, a2, a3);
5291
5292         if (!is_long_mode(vcpu)) {
5293                 nr &= 0xFFFFFFFF;
5294                 a0 &= 0xFFFFFFFF;
5295                 a1 &= 0xFFFFFFFF;
5296                 a2 &= 0xFFFFFFFF;
5297                 a3 &= 0xFFFFFFFF;
5298         }
5299
5300         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5301                 ret = -KVM_EPERM;
5302                 goto out;
5303         }
5304
5305         switch (nr) {
5306         case KVM_HC_VAPIC_POLL_IRQ:
5307                 ret = 0;
5308                 break;
5309         case KVM_HC_MMU_OP:
5310                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5311                 break;
5312         default:
5313                 ret = -KVM_ENOSYS;
5314                 break;
5315         }
5316 out:
5317         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5318         ++vcpu->stat.hypercalls;
5319         return r;
5320 }
5321 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5322
5323 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5324 {
5325         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5326         char instruction[3];
5327         unsigned long rip = kvm_rip_read(vcpu);
5328
5329         /*
5330          * Blow out the MMU to ensure that no other VCPU has an active mapping
5331          * to ensure that the updated hypercall appears atomically across all
5332          * VCPUs.
5333          */
5334         kvm_mmu_zap_all(vcpu->kvm);
5335
5336         kvm_x86_ops->patch_hypercall(vcpu, instruction);
5337
5338         return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5339 }
5340
5341 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5342 {
5343         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5344         int j, nent = vcpu->arch.cpuid_nent;
5345
5346         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5347         /* when no next entry is found, the current entry[i] is reselected */
5348         for (j = i + 1; ; j = (j + 1) % nent) {
5349                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5350                 if (ej->function == e->function) {
5351                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5352                         return j;
5353                 }
5354         }
5355         return 0; /* silence gcc, even though control never reaches here */
5356 }
5357
5358 /* find an entry with matching function, matching index (if needed), and that
5359  * should be read next (if it's stateful) */
5360 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5361         u32 function, u32 index)
5362 {
5363         if (e->function != function)
5364                 return 0;
5365         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5366                 return 0;
5367         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5368             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5369                 return 0;
5370         return 1;
5371 }
5372
5373 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5374                                               u32 function, u32 index)
5375 {
5376         int i;
5377         struct kvm_cpuid_entry2 *best = NULL;
5378
5379         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5380                 struct kvm_cpuid_entry2 *e;
5381
5382                 e = &vcpu->arch.cpuid_entries[i];
5383                 if (is_matching_cpuid_entry(e, function, index)) {
5384                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5385                                 move_to_next_stateful_cpuid_entry(vcpu, i);
5386                         best = e;
5387                         break;
5388                 }
5389         }
5390         return best;
5391 }
5392 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5393
5394 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5395 {
5396         struct kvm_cpuid_entry2 *best;
5397
5398         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5399         if (!best || best->eax < 0x80000008)
5400                 goto not_found;
5401         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5402         if (best)
5403                 return best->eax & 0xff;
5404 not_found:
5405         return 36;
5406 }
5407
5408 /*
5409  * If no match is found, check whether we exceed the vCPU's limit
5410  * and return the content of the highest valid _standard_ leaf instead.
5411  * This is to satisfy the CPUID specification.
5412  */
5413 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5414                                                   u32 function, u32 index)
5415 {
5416         struct kvm_cpuid_entry2 *maxlevel;
5417
5418         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5419         if (!maxlevel || maxlevel->eax >= function)
5420                 return NULL;
5421         if (function & 0x80000000) {
5422                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5423                 if (!maxlevel)
5424                         return NULL;
5425         }
5426         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5427 }
5428
5429 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5430 {
5431         u32 function, index;
5432         struct kvm_cpuid_entry2 *best;
5433
5434         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5435         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5436         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5437         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5438         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5439         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5440         best = kvm_find_cpuid_entry(vcpu, function, index);
5441
5442         if (!best)
5443                 best = check_cpuid_limit(vcpu, function, index);
5444
5445         if (best) {
5446                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5447                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5448                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5449                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5450         }
5451         kvm_x86_ops->skip_emulated_instruction(vcpu);
5452         trace_kvm_cpuid(function,
5453                         kvm_register_read(vcpu, VCPU_REGS_RAX),
5454                         kvm_register_read(vcpu, VCPU_REGS_RBX),
5455                         kvm_register_read(vcpu, VCPU_REGS_RCX),
5456                         kvm_register_read(vcpu, VCPU_REGS_RDX));
5457 }
5458 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5459
5460 /*
5461  * Check if userspace requested an interrupt window, and that the
5462  * interrupt window is open.
5463  *
5464  * No need to exit to userspace if we already have an interrupt queued.
5465  */
5466 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5467 {
5468         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5469                 vcpu->run->request_interrupt_window &&
5470                 kvm_arch_interrupt_allowed(vcpu));
5471 }
5472
5473 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5474 {
5475         struct kvm_run *kvm_run = vcpu->run;
5476
5477         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5478         kvm_run->cr8 = kvm_get_cr8(vcpu);
5479         kvm_run->apic_base = kvm_get_apic_base(vcpu);
5480         if (irqchip_in_kernel(vcpu->kvm))
5481                 kvm_run->ready_for_interrupt_injection = 1;
5482         else
5483                 kvm_run->ready_for_interrupt_injection =
5484                         kvm_arch_interrupt_allowed(vcpu) &&
5485                         !kvm_cpu_has_interrupt(vcpu) &&
5486                         !kvm_event_needs_reinjection(vcpu);
5487 }
5488
5489 static void vapic_enter(struct kvm_vcpu *vcpu)
5490 {
5491         struct kvm_lapic *apic = vcpu->arch.apic;
5492         struct page *page;
5493
5494         if (!apic || !apic->vapic_addr)
5495                 return;
5496
5497         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5498
5499         vcpu->arch.apic->vapic_page = page;
5500 }
5501
5502 static void vapic_exit(struct kvm_vcpu *vcpu)
5503 {
5504         struct kvm_lapic *apic = vcpu->arch.apic;
5505         int idx;
5506
5507         if (!apic || !apic->vapic_addr)
5508                 return;
5509
5510         idx = srcu_read_lock(&vcpu->kvm->srcu);
5511         kvm_release_page_dirty(apic->vapic_page);
5512         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5513         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5514 }
5515
5516 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5517 {
5518         int max_irr, tpr;
5519
5520         if (!kvm_x86_ops->update_cr8_intercept)
5521                 return;
5522
5523         if (!vcpu->arch.apic)
5524                 return;
5525
5526         if (!vcpu->arch.apic->vapic_addr)
5527                 max_irr = kvm_lapic_find_highest_irr(vcpu);
5528         else
5529                 max_irr = -1;
5530
5531         if (max_irr != -1)
5532                 max_irr >>= 4;
5533
5534         tpr = kvm_lapic_get_cr8(vcpu);
5535
5536         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5537 }
5538
5539 static void inject_pending_event(struct kvm_vcpu *vcpu)
5540 {
5541         /* try to reinject previous events if any */
5542         if (vcpu->arch.exception.pending) {
5543                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5544                                         vcpu->arch.exception.has_error_code,
5545                                         vcpu->arch.exception.error_code);
5546                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5547                                           vcpu->arch.exception.has_error_code,
5548                                           vcpu->arch.exception.error_code,
5549                                           vcpu->arch.exception.reinject);
5550                 return;
5551         }
5552
5553         if (vcpu->arch.nmi_injected) {
5554                 kvm_x86_ops->set_nmi(vcpu);
5555                 return;
5556         }
5557
5558         if (vcpu->arch.interrupt.pending) {
5559                 kvm_x86_ops->set_irq(vcpu);
5560                 return;
5561         }
5562
5563         /* try to inject new event if pending */
5564         if (vcpu->arch.nmi_pending) {
5565                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5566                         vcpu->arch.nmi_pending = false;
5567                         vcpu->arch.nmi_injected = true;
5568                         kvm_x86_ops->set_nmi(vcpu);
5569                 }
5570         } else if (kvm_cpu_has_interrupt(vcpu)) {
5571                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5572                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5573                                             false);
5574                         kvm_x86_ops->set_irq(vcpu);
5575                 }
5576         }
5577 }
5578
5579 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5580 {
5581         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5582                         !vcpu->guest_xcr0_loaded) {
5583                 /* kvm_set_xcr() also depends on this */
5584                 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5585                 vcpu->guest_xcr0_loaded = 1;
5586         }
5587 }
5588
5589 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5590 {
5591         if (vcpu->guest_xcr0_loaded) {
5592                 if (vcpu->arch.xcr0 != host_xcr0)
5593                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5594                 vcpu->guest_xcr0_loaded = 0;
5595         }
5596 }
5597
5598 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5599 {
5600         int r;
5601         bool nmi_pending;
5602         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5603                 vcpu->run->request_interrupt_window;
5604
5605         if (vcpu->requests) {
5606                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5607                         kvm_mmu_unload(vcpu);
5608                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5609                         __kvm_migrate_timers(vcpu);
5610                 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5611                         r = kvm_guest_time_update(vcpu);
5612                         if (unlikely(r))
5613                                 goto out;
5614                 }
5615                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5616                         kvm_mmu_sync_roots(vcpu);
5617                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5618                         kvm_x86_ops->tlb_flush(vcpu);
5619                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5620                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5621                         r = 0;
5622                         goto out;
5623                 }
5624                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5625                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5626                         r = 0;
5627                         goto out;
5628                 }
5629                 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5630                         vcpu->fpu_active = 0;
5631                         kvm_x86_ops->fpu_deactivate(vcpu);
5632                 }
5633                 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5634                         /* Page is swapped out. Do synthetic halt */
5635                         vcpu->arch.apf.halted = true;
5636                         r = 1;
5637                         goto out;
5638                 }
5639                 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5640                         record_steal_time(vcpu);
5641
5642         }
5643
5644         r = kvm_mmu_reload(vcpu);
5645         if (unlikely(r))
5646                 goto out;
5647
5648         /*
5649          * An NMI can be injected between local nmi_pending read and
5650          * vcpu->arch.nmi_pending read inside inject_pending_event().
5651          * But in that case, KVM_REQ_EVENT will be set, which makes
5652          * the race described above benign.
5653          */
5654         nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5655
5656         if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5657                 inject_pending_event(vcpu);
5658
5659                 /* enable NMI/IRQ window open exits if needed */
5660                 if (nmi_pending)
5661                         kvm_x86_ops->enable_nmi_window(vcpu);
5662                 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5663                         kvm_x86_ops->enable_irq_window(vcpu);
5664
5665                 if (kvm_lapic_enabled(vcpu)) {
5666                         update_cr8_intercept(vcpu);
5667                         kvm_lapic_sync_to_vapic(vcpu);
5668                 }
5669         }
5670
5671         preempt_disable();
5672
5673         kvm_x86_ops->prepare_guest_switch(vcpu);
5674         if (vcpu->fpu_active)
5675                 kvm_load_guest_fpu(vcpu);
5676         kvm_load_guest_xcr0(vcpu);
5677
5678         vcpu->mode = IN_GUEST_MODE;
5679
5680         /* We should set ->mode before check ->requests,
5681          * see the comment in make_all_cpus_request.
5682          */
5683         smp_mb();
5684
5685         local_irq_disable();
5686
5687         if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5688             || need_resched() || signal_pending(current)) {
5689                 vcpu->mode = OUTSIDE_GUEST_MODE;
5690                 smp_wmb();
5691                 local_irq_enable();
5692                 preempt_enable();
5693                 kvm_x86_ops->cancel_injection(vcpu);
5694                 r = 1;
5695                 goto out;
5696         }
5697
5698         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5699
5700         kvm_guest_enter();
5701
5702         if (unlikely(vcpu->arch.switch_db_regs)) {
5703                 set_debugreg(0, 7);
5704                 set_debugreg(vcpu->arch.eff_db[0], 0);
5705                 set_debugreg(vcpu->arch.eff_db[1], 1);
5706                 set_debugreg(vcpu->arch.eff_db[2], 2);
5707                 set_debugreg(vcpu->arch.eff_db[3], 3);
5708         }
5709
5710         trace_kvm_entry(vcpu->vcpu_id);
5711         kvm_x86_ops->run(vcpu);
5712
5713         /*
5714          * If the guest has used debug registers, at least dr7
5715          * will be disabled while returning to the host.
5716          * If we don't have active breakpoints in the host, we don't
5717          * care about the messed up debug address registers. But if
5718          * we have some of them active, restore the old state.
5719          */
5720         if (hw_breakpoint_active())
5721                 hw_breakpoint_restore();
5722
5723         kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5724
5725         vcpu->mode = OUTSIDE_GUEST_MODE;
5726         smp_wmb();
5727         local_irq_enable();
5728
5729         ++vcpu->stat.exits;
5730
5731         /*
5732          * We must have an instruction between local_irq_enable() and
5733          * kvm_guest_exit(), so the timer interrupt isn't delayed by
5734          * the interrupt shadow.  The stat.exits increment will do nicely.
5735          * But we need to prevent reordering, hence this barrier():
5736          */
5737         barrier();
5738
5739         kvm_guest_exit();
5740
5741         preempt_enable();
5742
5743         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5744
5745         /*
5746          * Profile KVM exit RIPs:
5747          */
5748         if (unlikely(prof_on == KVM_PROFILING)) {
5749                 unsigned long rip = kvm_rip_read(vcpu);
5750                 profile_hit(KVM_PROFILING, (void *)rip);
5751         }
5752
5753
5754         kvm_lapic_sync_from_vapic(vcpu);
5755
5756         r = kvm_x86_ops->handle_exit(vcpu);
5757 out:
5758         return r;
5759 }
5760
5761
5762 static int __vcpu_run(struct kvm_vcpu *vcpu)
5763 {
5764         int r;
5765         struct kvm *kvm = vcpu->kvm;
5766
5767         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5768                 pr_debug("vcpu %d received sipi with vector # %x\n",
5769                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
5770                 kvm_lapic_reset(vcpu);
5771                 r = kvm_arch_vcpu_reset(vcpu);
5772                 if (r)
5773                         return r;
5774                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5775         }
5776
5777         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5778         vapic_enter(vcpu);
5779
5780         r = 1;
5781         while (r > 0) {
5782                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5783                     !vcpu->arch.apf.halted)
5784                         r = vcpu_enter_guest(vcpu);
5785                 else {
5786                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5787                         kvm_vcpu_block(vcpu);
5788                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5789                         if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5790                         {
5791                                 switch(vcpu->arch.mp_state) {
5792                                 case KVM_MP_STATE_HALTED:
5793                                         vcpu->arch.mp_state =
5794                                                 KVM_MP_STATE_RUNNABLE;
5795                                 case KVM_MP_STATE_RUNNABLE:
5796                                         vcpu->arch.apf.halted = false;
5797                                         break;
5798                                 case KVM_MP_STATE_SIPI_RECEIVED:
5799                                 default:
5800                                         r = -EINTR;
5801                                         break;
5802                                 }
5803                         }
5804                 }
5805
5806                 if (r <= 0)
5807                         break;
5808
5809                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5810                 if (kvm_cpu_has_pending_timer(vcpu))
5811                         kvm_inject_pending_timer_irqs(vcpu);
5812
5813                 if (dm_request_for_irq_injection(vcpu)) {
5814                         r = -EINTR;
5815                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5816                         ++vcpu->stat.request_irq_exits;
5817                 }
5818
5819                 kvm_check_async_pf_completion(vcpu);
5820
5821                 if (signal_pending(current)) {
5822                         r = -EINTR;
5823                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5824                         ++vcpu->stat.signal_exits;
5825                 }
5826                 if (need_resched()) {
5827                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5828                         kvm_resched(vcpu);
5829                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5830                 }
5831         }
5832
5833         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5834
5835         vapic_exit(vcpu);
5836
5837         return r;
5838 }
5839
5840 static int complete_mmio(struct kvm_vcpu *vcpu)
5841 {
5842         struct kvm_run *run = vcpu->run;
5843         int r;
5844
5845         if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5846                 return 1;
5847
5848         if (vcpu->mmio_needed) {
5849                 vcpu->mmio_needed = 0;
5850                 if (!vcpu->mmio_is_write)
5851                         memcpy(vcpu->mmio_data + vcpu->mmio_index,
5852                                run->mmio.data, 8);
5853                 vcpu->mmio_index += 8;
5854                 if (vcpu->mmio_index < vcpu->mmio_size) {
5855                         run->exit_reason = KVM_EXIT_MMIO;
5856                         run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5857                         memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5858                         run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5859                         run->mmio.is_write = vcpu->mmio_is_write;
5860                         vcpu->mmio_needed = 1;
5861                         return 0;
5862                 }
5863                 if (vcpu->mmio_is_write)
5864                         return 1;
5865                 vcpu->mmio_read_completed = 1;
5866         }
5867         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5868         r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5869         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5870         if (r != EMULATE_DONE)
5871                 return 0;
5872         return 1;
5873 }
5874
5875 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5876 {
5877         int r;
5878         sigset_t sigsaved;
5879
5880         if (!tsk_used_math(current) && init_fpu(current))
5881                 return -ENOMEM;
5882
5883         if (vcpu->sigset_active)
5884                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5885
5886         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5887                 kvm_vcpu_block(vcpu);
5888                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5889                 r = -EAGAIN;
5890                 goto out;
5891         }
5892
5893         /* re-sync apic's tpr */
5894         if (!irqchip_in_kernel(vcpu->kvm)) {
5895                 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5896                         r = -EINVAL;
5897                         goto out;
5898                 }
5899         }
5900
5901         r = complete_mmio(vcpu);
5902         if (r <= 0)
5903                 goto out;
5904
5905         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5906                 kvm_register_write(vcpu, VCPU_REGS_RAX,
5907                                      kvm_run->hypercall.ret);
5908
5909         r = __vcpu_run(vcpu);
5910
5911 out:
5912         post_kvm_run_save(vcpu);
5913         if (vcpu->sigset_active)
5914                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5915
5916         return r;
5917 }
5918
5919 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5920 {
5921         if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5922                 /*
5923                  * We are here if userspace calls get_regs() in the middle of
5924                  * instruction emulation. Registers state needs to be copied
5925                  * back from emulation context to vcpu. Usrapace shouldn't do
5926                  * that usually, but some bad designed PV devices (vmware
5927                  * backdoor interface) need this to work
5928                  */
5929                 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5930                 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5931                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5932         }
5933         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5934         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5935         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5936         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5937         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5938         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5939         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5940         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5941 #ifdef CONFIG_X86_64
5942         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5943         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5944         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5945         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5946         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5947         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5948         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5949         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5950 #endif
5951
5952         regs->rip = kvm_rip_read(vcpu);
5953         regs->rflags = kvm_get_rflags(vcpu);
5954
5955         return 0;
5956 }
5957
5958 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5959 {
5960         vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5961         vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5962
5963         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5964         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5965         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5966         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5967         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5968         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5969         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5970         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5971 #ifdef CONFIG_X86_64
5972         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5973         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5974         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5975         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5976         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5977         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5978         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5979         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5980 #endif
5981
5982         kvm_rip_write(vcpu, regs->rip);
5983         kvm_set_rflags(vcpu, regs->rflags);
5984
5985         vcpu->arch.exception.pending = false;
5986
5987         kvm_make_request(KVM_REQ_EVENT, vcpu);
5988
5989         return 0;
5990 }
5991
5992 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5993 {
5994         struct kvm_segment cs;
5995
5996         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5997         *db = cs.db;
5998         *l = cs.l;
5999 }
6000 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6001
6002 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6003                                   struct kvm_sregs *sregs)
6004 {
6005         struct desc_ptr dt;
6006
6007         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6008         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6009         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6010         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6011         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6012         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6013
6014         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6015         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6016
6017         kvm_x86_ops->get_idt(vcpu, &dt);
6018         sregs->idt.limit = dt.size;
6019         sregs->idt.base = dt.address;
6020         kvm_x86_ops->get_gdt(vcpu, &dt);
6021         sregs->gdt.limit = dt.size;
6022         sregs->gdt.base = dt.address;
6023
6024         sregs->cr0 = kvm_read_cr0(vcpu);
6025         sregs->cr2 = vcpu->arch.cr2;
6026         sregs->cr3 = kvm_read_cr3(vcpu);
6027         sregs->cr4 = kvm_read_cr4(vcpu);
6028         sregs->cr8 = kvm_get_cr8(vcpu);
6029         sregs->efer = vcpu->arch.efer;
6030         sregs->apic_base = kvm_get_apic_base(vcpu);
6031
6032         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6033
6034         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6035                 set_bit(vcpu->arch.interrupt.nr,
6036                         (unsigned long *)sregs->interrupt_bitmap);
6037
6038         return 0;
6039 }
6040
6041 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6042                                     struct kvm_mp_state *mp_state)
6043 {
6044         mp_state->mp_state = vcpu->arch.mp_state;
6045         return 0;
6046 }
6047
6048 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6049                                     struct kvm_mp_state *mp_state)
6050 {
6051         vcpu->arch.mp_state = mp_state->mp_state;
6052         kvm_make_request(KVM_REQ_EVENT, vcpu);
6053         return 0;
6054 }
6055
6056 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
6057                     bool has_error_code, u32 error_code)
6058 {
6059         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6060         int ret;
6061
6062         init_emulate_ctxt(vcpu);
6063
6064         ret = emulator_task_switch(ctxt, tss_selector, reason,
6065                                    has_error_code, error_code);
6066
6067         if (ret)
6068                 return EMULATE_FAIL;
6069
6070         memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
6071         kvm_rip_write(vcpu, ctxt->eip);
6072         kvm_set_rflags(vcpu, ctxt->eflags);
6073         kvm_make_request(KVM_REQ_EVENT, vcpu);
6074         return EMULATE_DONE;
6075 }
6076 EXPORT_SYMBOL_GPL(kvm_task_switch);
6077
6078 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6079                                   struct kvm_sregs *sregs)
6080 {
6081         int mmu_reset_needed = 0;
6082         int pending_vec, max_bits, idx;
6083         struct desc_ptr dt;
6084
6085         dt.size = sregs->idt.limit;
6086         dt.address = sregs->idt.base;
6087         kvm_x86_ops->set_idt(vcpu, &dt);
6088         dt.size = sregs->gdt.limit;
6089         dt.address = sregs->gdt.base;
6090         kvm_x86_ops->set_gdt(vcpu, &dt);
6091
6092         vcpu->arch.cr2 = sregs->cr2;
6093         mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6094         vcpu->arch.cr3 = sregs->cr3;
6095         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6096
6097         kvm_set_cr8(vcpu, sregs->cr8);
6098
6099         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6100         kvm_x86_ops->set_efer(vcpu, sregs->efer);
6101         kvm_set_apic_base(vcpu, sregs->apic_base);
6102
6103         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6104         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6105         vcpu->arch.cr0 = sregs->cr0;
6106
6107         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6108         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6109         if (sregs->cr4 & X86_CR4_OSXSAVE)
6110                 update_cpuid(vcpu);
6111
6112         idx = srcu_read_lock(&vcpu->kvm->srcu);
6113         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6114                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6115                 mmu_reset_needed = 1;
6116         }
6117         srcu_read_unlock(&vcpu->kvm->srcu, idx);
6118
6119         if (mmu_reset_needed)
6120                 kvm_mmu_reset_context(vcpu);
6121
6122         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
6123         pending_vec = find_first_bit(
6124                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6125         if (pending_vec < max_bits) {
6126                 kvm_queue_interrupt(vcpu, pending_vec, false);
6127                 pr_debug("Set back pending irq %d\n", pending_vec);
6128         }
6129
6130         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6131         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6132         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6133         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6134         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6135         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6136
6137         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6138         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6139
6140         update_cr8_intercept(vcpu);
6141
6142         /* Older userspace won't unhalt the vcpu on reset. */
6143         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6144             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6145             !is_protmode(vcpu))
6146                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6147
6148         kvm_make_request(KVM_REQ_EVENT, vcpu);
6149
6150         return 0;
6151 }
6152
6153 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6154                                         struct kvm_guest_debug *dbg)
6155 {
6156         unsigned long rflags;
6157         int i, r;
6158
6159         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6160                 r = -EBUSY;
6161                 if (vcpu->arch.exception.pending)
6162                         goto out;
6163                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6164                         kvm_queue_exception(vcpu, DB_VECTOR);
6165                 else
6166                         kvm_queue_exception(vcpu, BP_VECTOR);
6167         }
6168
6169         /*
6170          * Read rflags as long as potentially injected trace flags are still
6171          * filtered out.
6172          */
6173         rflags = kvm_get_rflags(vcpu);
6174
6175         vcpu->guest_debug = dbg->control;
6176         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6177                 vcpu->guest_debug = 0;
6178
6179         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6180                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6181                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6182                 vcpu->arch.switch_db_regs =
6183                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
6184         } else {
6185                 for (i = 0; i < KVM_NR_DB_REGS; i++)
6186                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6187                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
6188         }
6189
6190         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6191                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6192                         get_segment_base(vcpu, VCPU_SREG_CS);
6193
6194         /*
6195          * Trigger an rflags update that will inject or remove the trace
6196          * flags.
6197          */
6198         kvm_set_rflags(vcpu, rflags);
6199
6200         kvm_x86_ops->set_guest_debug(vcpu, dbg);
6201
6202         r = 0;
6203
6204 out:
6205
6206         return r;
6207 }
6208
6209 /*
6210  * Translate a guest virtual address to a guest physical address.
6211  */
6212 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6213                                     struct kvm_translation *tr)
6214 {
6215         unsigned long vaddr = tr->linear_address;
6216         gpa_t gpa;
6217         int idx;
6218
6219         idx = srcu_read_lock(&vcpu->kvm->srcu);
6220         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6221         srcu_read_unlock(&vcpu->kvm->srcu, idx);
6222         tr->physical_address = gpa;
6223         tr->valid = gpa != UNMAPPED_GVA;
6224         tr->writeable = 1;
6225         tr->usermode = 0;
6226
6227         return 0;
6228 }
6229
6230 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6231 {
6232         struct i387_fxsave_struct *fxsave =
6233                         &vcpu->arch.guest_fpu.state->fxsave;
6234
6235         memcpy(fpu->fpr, fxsave->st_space, 128);
6236         fpu->fcw = fxsave->cwd;
6237         fpu->fsw = fxsave->swd;
6238         fpu->ftwx = fxsave->twd;
6239         fpu->last_opcode = fxsave->fop;
6240         fpu->last_ip = fxsave->rip;
6241         fpu->last_dp = fxsave->rdp;
6242         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6243
6244         return 0;
6245 }
6246
6247 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6248 {
6249         struct i387_fxsave_struct *fxsave =
6250                         &vcpu->arch.guest_fpu.state->fxsave;
6251
6252         memcpy(fxsave->st_space, fpu->fpr, 128);
6253         fxsave->cwd = fpu->fcw;
6254         fxsave->swd = fpu->fsw;
6255         fxsave->twd = fpu->ftwx;
6256         fxsave->fop = fpu->last_opcode;
6257         fxsave->rip = fpu->last_ip;
6258         fxsave->rdp = fpu->last_dp;
6259         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6260
6261         return 0;
6262 }
6263
6264 int fx_init(struct kvm_vcpu *vcpu)
6265 {
6266         int err;
6267
6268         err = fpu_alloc(&vcpu->arch.guest_fpu);
6269         if (err)
6270                 return err;
6271
6272         fpu_finit(&vcpu->arch.guest_fpu);
6273
6274         /*
6275          * Ensure guest xcr0 is valid for loading
6276          */
6277         vcpu->arch.xcr0 = XSTATE_FP;
6278
6279         vcpu->arch.cr0 |= X86_CR0_ET;
6280
6281         return 0;
6282 }
6283 EXPORT_SYMBOL_GPL(fx_init);
6284
6285 static void fx_free(struct kvm_vcpu *vcpu)
6286 {
6287         fpu_free(&vcpu->arch.guest_fpu);
6288 }
6289
6290 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6291 {
6292         if (vcpu->guest_fpu_loaded)
6293                 return;
6294
6295         /*
6296          * Restore all possible states in the guest,
6297          * and assume host would use all available bits.
6298          * Guest xcr0 would be loaded later.
6299          */
6300         kvm_put_guest_xcr0(vcpu);
6301         vcpu->guest_fpu_loaded = 1;
6302         unlazy_fpu(current);
6303         fpu_restore_checking(&vcpu->arch.guest_fpu);
6304         trace_kvm_fpu(1);
6305 }
6306
6307 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6308 {
6309         kvm_put_guest_xcr0(vcpu);
6310
6311         if (!vcpu->guest_fpu_loaded)
6312                 return;
6313
6314         vcpu->guest_fpu_loaded = 0;
6315         fpu_save_init(&vcpu->arch.guest_fpu);
6316         ++vcpu->stat.fpu_reload;
6317         kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6318         trace_kvm_fpu(0);
6319 }
6320
6321 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6322 {
6323         kvmclock_reset(vcpu);
6324
6325         free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6326         fx_free(vcpu);
6327         kvm_x86_ops->vcpu_free(vcpu);
6328 }
6329
6330 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6331                                                 unsigned int id)
6332 {
6333         if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6334                 printk_once(KERN_WARNING
6335                 "kvm: SMP vm created on host with unstable TSC; "
6336                 "guest TSC will not be reliable\n");
6337         return kvm_x86_ops->vcpu_create(kvm, id);
6338 }
6339
6340 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6341 {
6342         int r;
6343
6344         vcpu->arch.mtrr_state.have_fixed = 1;
6345         vcpu_load(vcpu);
6346         r = kvm_arch_vcpu_reset(vcpu);
6347         if (r == 0)
6348                 r = kvm_mmu_setup(vcpu);
6349         vcpu_put(vcpu);
6350
6351         return r;
6352 }
6353
6354 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6355 {
6356         vcpu->arch.apf.msr_val = 0;
6357
6358         vcpu_load(vcpu);
6359         kvm_mmu_unload(vcpu);
6360         vcpu_put(vcpu);
6361
6362         fx_free(vcpu);
6363         kvm_x86_ops->vcpu_free(vcpu);
6364 }
6365
6366 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6367 {
6368         vcpu->arch.nmi_pending = false;
6369         vcpu->arch.nmi_injected = false;
6370
6371         vcpu->arch.switch_db_regs = 0;
6372         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6373         vcpu->arch.dr6 = DR6_FIXED_1;
6374         vcpu->arch.dr7 = DR7_FIXED_1;
6375
6376         kvm_make_request(KVM_REQ_EVENT, vcpu);
6377         vcpu->arch.apf.msr_val = 0;
6378         vcpu->arch.st.msr_val = 0;
6379
6380         kvmclock_reset(vcpu);
6381
6382         kvm_clear_async_pf_completion_queue(vcpu);
6383         kvm_async_pf_hash_reset(vcpu);
6384         vcpu->arch.apf.halted = false;
6385
6386         return kvm_x86_ops->vcpu_reset(vcpu);
6387 }
6388
6389 int kvm_arch_hardware_enable(void *garbage)
6390 {
6391         struct kvm *kvm;
6392         struct kvm_vcpu *vcpu;
6393         int i;
6394
6395         kvm_shared_msr_cpu_online();
6396         list_for_each_entry(kvm, &vm_list, vm_list)
6397                 kvm_for_each_vcpu(i, vcpu, kvm)
6398                         if (vcpu->cpu == smp_processor_id())
6399                                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6400         return kvm_x86_ops->hardware_enable(garbage);
6401 }
6402
6403 void kvm_arch_hardware_disable(void *garbage)
6404 {
6405         kvm_x86_ops->hardware_disable(garbage);
6406         drop_user_return_notifiers(garbage);
6407 }
6408
6409 int kvm_arch_hardware_setup(void)
6410 {
6411         return kvm_x86_ops->hardware_setup();
6412 }
6413
6414 void kvm_arch_hardware_unsetup(void)
6415 {
6416         kvm_x86_ops->hardware_unsetup();
6417 }
6418
6419 void kvm_arch_check_processor_compat(void *rtn)
6420 {
6421         kvm_x86_ops->check_processor_compatibility(rtn);
6422 }
6423
6424 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6425 {
6426         struct page *page;
6427         struct kvm *kvm;
6428         int r;
6429
6430         BUG_ON(vcpu->kvm == NULL);
6431         kvm = vcpu->kvm;
6432
6433         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6434         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6435         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6436         vcpu->arch.mmu.translate_gpa = translate_gpa;
6437         vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6438         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6439                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6440         else
6441                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6442
6443         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6444         if (!page) {
6445                 r = -ENOMEM;
6446                 goto fail;
6447         }
6448         vcpu->arch.pio_data = page_address(page);
6449
6450         kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6451
6452         r = kvm_mmu_create(vcpu);
6453         if (r < 0)
6454                 goto fail_free_pio_data;
6455
6456         if (irqchip_in_kernel(kvm)) {
6457                 r = kvm_create_lapic(vcpu);
6458                 if (r < 0)
6459                         goto fail_mmu_destroy;
6460         }
6461
6462         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6463                                        GFP_KERNEL);
6464         if (!vcpu->arch.mce_banks) {
6465                 r = -ENOMEM;
6466                 goto fail_free_lapic;
6467         }
6468         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6469
6470         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6471                 goto fail_free_mce_banks;
6472
6473         kvm_async_pf_hash_reset(vcpu);
6474
6475         return 0;
6476 fail_free_mce_banks:
6477         kfree(vcpu->arch.mce_banks);
6478 fail_free_lapic:
6479         kvm_free_lapic(vcpu);
6480 fail_mmu_destroy:
6481         kvm_mmu_destroy(vcpu);
6482 fail_free_pio_data:
6483         free_page((unsigned long)vcpu->arch.pio_data);
6484 fail:
6485         return r;
6486 }
6487
6488 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6489 {
6490         int idx;
6491
6492         kfree(vcpu->arch.mce_banks);
6493         kvm_free_lapic(vcpu);
6494         idx = srcu_read_lock(&vcpu->kvm->srcu);
6495         kvm_mmu_destroy(vcpu);
6496         srcu_read_unlock(&vcpu->kvm->srcu, idx);
6497         free_page((unsigned long)vcpu->arch.pio_data);
6498 }
6499
6500 int kvm_arch_init_vm(struct kvm *kvm)
6501 {
6502         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6503         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6504
6505         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6506         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6507
6508         raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6509
6510         return 0;
6511 }
6512
6513 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6514 {
6515         vcpu_load(vcpu);
6516         kvm_mmu_unload(vcpu);
6517         vcpu_put(vcpu);
6518 }
6519
6520 static void kvm_free_vcpus(struct kvm *kvm)
6521 {
6522         unsigned int i;
6523         struct kvm_vcpu *vcpu;
6524
6525         /*
6526          * Unpin any mmu pages first.
6527          */
6528         kvm_for_each_vcpu(i, vcpu, kvm) {
6529                 kvm_clear_async_pf_completion_queue(vcpu);
6530                 kvm_unload_vcpu_mmu(vcpu);
6531         }
6532         kvm_for_each_vcpu(i, vcpu, kvm)
6533                 kvm_arch_vcpu_free(vcpu);
6534
6535         mutex_lock(&kvm->lock);
6536         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6537                 kvm->vcpus[i] = NULL;
6538
6539         atomic_set(&kvm->online_vcpus, 0);
6540         mutex_unlock(&kvm->lock);
6541 }
6542
6543 void kvm_arch_sync_events(struct kvm *kvm)
6544 {
6545         kvm_free_all_assigned_devices(kvm);
6546         kvm_free_pit(kvm);
6547 }
6548
6549 void kvm_arch_destroy_vm(struct kvm *kvm)
6550 {
6551         kvm_iommu_unmap_guest(kvm);
6552         kfree(kvm->arch.vpic);
6553         kfree(kvm->arch.vioapic);
6554         kvm_free_vcpus(kvm);
6555         if (kvm->arch.apic_access_page)
6556                 put_page(kvm->arch.apic_access_page);
6557         if (kvm->arch.ept_identity_pagetable)
6558                 put_page(kvm->arch.ept_identity_pagetable);
6559 }
6560
6561 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6562                                 struct kvm_memory_slot *memslot,
6563                                 struct kvm_memory_slot old,
6564                                 struct kvm_userspace_memory_region *mem,
6565                                 int user_alloc)
6566 {
6567         int npages = memslot->npages;
6568         int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6569
6570         /* Prevent internal slot pages from being moved by fork()/COW. */
6571         if (memslot->id >= KVM_MEMORY_SLOTS)
6572                 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6573
6574         /*To keep backward compatibility with older userspace,
6575          *x86 needs to hanlde !user_alloc case.
6576          */
6577         if (!user_alloc) {
6578                 if (npages && !old.rmap) {
6579                         unsigned long userspace_addr;
6580
6581                         down_write(&current->mm->mmap_sem);
6582                         userspace_addr = do_mmap(NULL, 0,
6583                                                  npages * PAGE_SIZE,
6584                                                  PROT_READ | PROT_WRITE,
6585                                                  map_flags,
6586                                                  0);
6587                         up_write(&current->mm->mmap_sem);
6588
6589                         if (IS_ERR((void *)userspace_addr))
6590                                 return PTR_ERR((void *)userspace_addr);
6591
6592                         memslot->userspace_addr = userspace_addr;
6593                 }
6594         }
6595
6596
6597         return 0;
6598 }
6599
6600 void kvm_arch_commit_memory_region(struct kvm *kvm,
6601                                 struct kvm_userspace_memory_region *mem,
6602                                 struct kvm_memory_slot old,
6603                                 int user_alloc)
6604 {
6605
6606         int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6607
6608         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6609                 int ret;
6610
6611                 down_write(&current->mm->mmap_sem);
6612                 ret = do_munmap(current->mm, old.userspace_addr,
6613                                 old.npages * PAGE_SIZE);
6614                 up_write(&current->mm->mmap_sem);
6615                 if (ret < 0)
6616                         printk(KERN_WARNING
6617                                "kvm_vm_ioctl_set_memory_region: "
6618                                "failed to munmap memory\n");
6619         }
6620
6621         if (!kvm->arch.n_requested_mmu_pages)
6622                 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6623
6624         spin_lock(&kvm->mmu_lock);
6625         if (nr_mmu_pages)
6626                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6627         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6628         spin_unlock(&kvm->mmu_lock);
6629 }
6630
6631 void kvm_arch_flush_shadow(struct kvm *kvm)
6632 {
6633         kvm_mmu_zap_all(kvm);
6634         kvm_reload_remote_mmus(kvm);
6635 }
6636
6637 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6638 {
6639         return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6640                 !vcpu->arch.apf.halted)
6641                 || !list_empty_careful(&vcpu->async_pf.done)
6642                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6643                 || vcpu->arch.nmi_pending ||
6644                 (kvm_arch_interrupt_allowed(vcpu) &&
6645                  kvm_cpu_has_interrupt(vcpu));
6646 }
6647
6648 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6649 {
6650         int me;
6651         int cpu = vcpu->cpu;
6652
6653         if (waitqueue_active(&vcpu->wq)) {
6654                 wake_up_interruptible(&vcpu->wq);
6655                 ++vcpu->stat.halt_wakeup;
6656         }
6657
6658         me = get_cpu();
6659         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6660                 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6661                         smp_send_reschedule(cpu);
6662         put_cpu();
6663 }
6664
6665 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6666 {
6667         return kvm_x86_ops->interrupt_allowed(vcpu);
6668 }
6669
6670 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6671 {
6672         unsigned long current_rip = kvm_rip_read(vcpu) +
6673                 get_segment_base(vcpu, VCPU_SREG_CS);
6674
6675         return current_rip == linear_rip;
6676 }
6677 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6678
6679 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6680 {
6681         unsigned long rflags;
6682
6683         rflags = kvm_x86_ops->get_rflags(vcpu);
6684         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6685                 rflags &= ~X86_EFLAGS_TF;
6686         return rflags;
6687 }
6688 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6689
6690 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6691 {
6692         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6693             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6694                 rflags |= X86_EFLAGS_TF;
6695         kvm_x86_ops->set_rflags(vcpu, rflags);
6696         kvm_make_request(KVM_REQ_EVENT, vcpu);
6697 }
6698 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6699
6700 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6701 {
6702         int r;
6703
6704         if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6705               is_error_page(work->page))
6706                 return;
6707
6708         r = kvm_mmu_reload(vcpu);
6709         if (unlikely(r))
6710                 return;
6711
6712         if (!vcpu->arch.mmu.direct_map &&
6713               work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6714                 return;
6715
6716         vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6717 }
6718
6719 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6720 {
6721         return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6722 }
6723
6724 static inline u32 kvm_async_pf_next_probe(u32 key)
6725 {
6726         return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6727 }
6728
6729 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6730 {
6731         u32 key = kvm_async_pf_hash_fn(gfn);
6732
6733         while (vcpu->arch.apf.gfns[key] != ~0)
6734                 key = kvm_async_pf_next_probe(key);
6735
6736         vcpu->arch.apf.gfns[key] = gfn;
6737 }
6738
6739 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6740 {
6741         int i;
6742         u32 key = kvm_async_pf_hash_fn(gfn);
6743
6744         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6745                      (vcpu->arch.apf.gfns[key] != gfn &&
6746                       vcpu->arch.apf.gfns[key] != ~0); i++)
6747                 key = kvm_async_pf_next_probe(key);
6748
6749         return key;
6750 }
6751
6752 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6753 {
6754         return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6755 }
6756
6757 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6758 {
6759         u32 i, j, k;
6760
6761         i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6762         while (true) {
6763                 vcpu->arch.apf.gfns[i] = ~0;
6764                 do {
6765                         j = kvm_async_pf_next_probe(j);
6766                         if (vcpu->arch.apf.gfns[j] == ~0)
6767                                 return;
6768                         k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6769                         /*
6770                          * k lies cyclically in ]i,j]
6771                          * |    i.k.j |
6772                          * |....j i.k.| or  |.k..j i...|
6773                          */
6774                 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6775                 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6776                 i = j;
6777         }
6778 }
6779
6780 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6781 {
6782
6783         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6784                                       sizeof(val));
6785 }
6786
6787 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6788                                      struct kvm_async_pf *work)
6789 {
6790         struct x86_exception fault;
6791
6792         trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6793         kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6794
6795         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6796             (vcpu->arch.apf.send_user_only &&
6797              kvm_x86_ops->get_cpl(vcpu) == 0))
6798                 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6799         else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6800                 fault.vector = PF_VECTOR;
6801                 fault.error_code_valid = true;
6802                 fault.error_code = 0;
6803                 fault.nested_page_fault = false;
6804                 fault.address = work->arch.token;
6805                 kvm_inject_page_fault(vcpu, &fault);
6806         }
6807 }
6808
6809 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6810                                  struct kvm_async_pf *work)
6811 {
6812         struct x86_exception fault;
6813
6814         trace_kvm_async_pf_ready(work->arch.token, work->gva);
6815         if (is_error_page(work->page))
6816                 work->arch.token = ~0; /* broadcast wakeup */
6817         else
6818                 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6819
6820         if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6821             !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6822                 fault.vector = PF_VECTOR;
6823                 fault.error_code_valid = true;
6824                 fault.error_code = 0;
6825                 fault.nested_page_fault = false;
6826                 fault.address = work->arch.token;
6827                 kvm_inject_page_fault(vcpu, &fault);
6828         }
6829         vcpu->arch.apf.halted = false;
6830 }
6831
6832 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6833 {
6834         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6835                 return true;
6836         else
6837                 return !kvm_event_needs_reinjection(vcpu) &&
6838                         kvm_x86_ops->interrupt_allowed(vcpu);
6839 }
6840
6841 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6842 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6843 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6844 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6845 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6846 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6847 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6848 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6849 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6850 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6851 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6852 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);